[
  {
    "id": 0,
    "metadata": "ProgramInfo(global_size=(1024, 512, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=2419064832, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      512,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 1,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 2,
    "metadata": "ProgramInfo(global_size=(512, 1024, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=2420113408, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      512,
      1024,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 3,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 4,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 5,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(3, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 6,
    "metadata": "ProgramInfo(global_size=(1024, 256, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=2015887360, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      256,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 7,
    "metadata": "ProgramInfo(global_size=(256, 1024, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=2017460224, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      256,
      1024,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 8,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 9,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(4, AxisType.UPCAST)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 10,
    "metadata": "ProgramInfo(global_size=(1024, 1, 1), local_size=(1024, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(0, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1,
      1
    ],
    "local_size": [
      1024,
      1,
      1
    ]
  },
  {
    "id": 11,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 12,
    "metadata": "ProgramInfo(global_size=(1024, 1, 1), local_size=(1024, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1,
      1
    ],
    "local_size": [
      1024,
      1,
      1
    ]
  },
  {
    "id": 13,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(768, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(0, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=620891209728, lds=2481955799040, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      768,
      1,
      1
    ]
  },
  {
    "id": 14,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 15,
    "metadata": "ProgramInfo(global_size=(1024, 512, 1), local_size=(2, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      512,
      1
    ],
    "local_size": [
      2,
      1,
      1
    ]
  },
  {
    "id": 16,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 17,
    "metadata": "ProgramInfo(global_size=(512, 1024, 1), local_size=(2, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      512,
      1024,
      1
    ],
    "local_size": [
      2,
      1,
      1
    ]
  },
  {
    "id": 18,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 19,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(2, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1619001344, lds=3254779904, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      2,
      1,
      1
    ]
  },
  {
    "id": 20,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(3, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(3, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1626341376, lds=3284140032, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      3,
      1,
      1
    ]
  },
  {
    "id": 21,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 22,
    "metadata": "ProgramInfo(global_size=(1024, 256, 1), local_size=(4, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      256,
      1
    ],
    "local_size": [
      4,
      1,
      1
    ]
  },
  {
    "id": 23,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 24,
    "metadata": "ProgramInfo(global_size=(256, 1024, 1), local_size=(4, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      256,
      1024,
      1
    ],
    "local_size": [
      4,
      1,
      1
    ]
  },
  {
    "id": 25,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 26,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(4, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(4, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1635778560, lds=3321888768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      4,
      1,
      1
    ]
  },
  {
    "id": 27,
    "metadata": "ProgramInfo(global_size=(1024, 128, 1), local_size=(8, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      128,
      1
    ],
    "local_size": [
      8,
      1,
      1
    ]
  },
  {
    "id": 28,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 29,
    "metadata": "ProgramInfo(global_size=(128, 1024, 1), local_size=(8, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      1024,
      1
    ],
    "local_size": [
      8,
      1,
      1
    ]
  },
  {
    "id": 30,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 31,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(8, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(8, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1694498816, lds=3556769792, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      8,
      1,
      1
    ]
  },
  {
    "id": 32,
    "metadata": "ProgramInfo(global_size=(1024, 64, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      64,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 33,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 34,
    "metadata": "ProgramInfo(global_size=(64, 1024, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      1024,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 35,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 36,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(16, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1912602624, lds=4429185024, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 37,
    "metadata": "ProgramInfo(global_size=(1024, 64, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      64,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 38,
    "metadata": "ProgramInfo(global_size=(64, 1024, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      1024,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 39,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(16, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1912602624, lds=4429185024, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 40,
    "metadata": "ProgramInfo(global_size=(1024, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 41,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 42,
    "metadata": "ProgramInfo(global_size=(32, 1024, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(32, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1024,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 43,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 44,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(32, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=2751463424, lds=7784628224, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 45,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 46,
    "metadata": "ProgramInfo(global_size=(1024, 16, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(64, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      16,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 47,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 48,
    "metadata": "ProgramInfo(global_size=(16, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(64, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 49,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 50,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 51,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 52,
    "metadata": "ProgramInfo(global_size=(1024, 4, 1), local_size=(256, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(256, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      4,
      1
    ],
    "local_size": [
      256,
      1,
      1
    ]
  },
  {
    "id": 53,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 54,
    "metadata": "ProgramInfo(global_size=(4, 1024, 1), local_size=(256, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=1, arg=(256, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      1024,
      1
    ],
    "local_size": [
      256,
      1,
      1
    ]
  },
  {
    "id": 55,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 56,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(256, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(256, AxisType.LOCAL, True)),), opts_to_apply=None, estimates=Estimates(ops=70866960384, lds=280246616064, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      256,
      1,
      1
    ]
  },
  {
    "id": 57,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 58,
    "metadata": "ProgramInfo(global_size=(1024, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL)),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 59,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)),), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 60,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 61,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 2, 1)),), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 62,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SWAP, axis=0, arg=1),), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=3225419776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 63,
    "metadata": "ProgramInfo(global_size=(128, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=254279680, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 64,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 65,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 66,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 67,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 68,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 69,
    "metadata": "ProgramInfo(global_size=(128, 16, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=228851712, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      16,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 70,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 71,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 72,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 73,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 74,
    "metadata": "ProgramInfo(global_size=(128, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 75,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 76,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 77,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 78,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1667760128, lds=333447168, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 79,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 80,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 3, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(3, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1674313728, lds=361758720, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      3,
      1
    ]
  },
  {
    "id": 81,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 82,
    "metadata": "ProgramInfo(global_size=(128, 16, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      16,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 83,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 84,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 85,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 86,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1682964480, lds=398458880, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 87,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 88,
    "metadata": "ProgramInfo(global_size=(128, 8, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      8,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 89,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 90,
    "metadata": "ProgramInfo(global_size=(16, 64, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      64,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 91,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 92,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1738539008, lds=629145600, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 93,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 94,
    "metadata": "ProgramInfo(global_size=(128, 4, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      4,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 95,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 96,
    "metadata": "ProgramInfo(global_size=(8, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 97,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 98,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1950351360, lds=1493172224, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 99,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 100,
    "metadata": "ProgramInfo(global_size=(128, 4, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      4,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 101,
    "metadata": "ProgramInfo(global_size=(8, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 102,
    "metadata": "ProgramInfo(global_size=(128, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=6, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1950351360, lds=1493172224, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 103,
    "metadata": "ProgramInfo(global_size=(128, 2, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      2,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 104,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 105,
    "metadata": "ProgramInfo(global_size=(4, 64, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      64,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 106,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 107,
    "metadata": "ProgramInfo(global_size=(128, 2, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      2,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 108,
    "metadata": "ProgramInfo(global_size=(64, 128, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=305135616, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      128,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 109,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 110,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=5972688896, lds=19328925696, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 111,
    "metadata": "ProgramInfo(global_size=(1024, 512, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=6006243328, lds=20065550336, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      512,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 112,
    "metadata": "ProgramInfo(global_size=(512, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=6006243328, lds=20132659200, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      512,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 113,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 114,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=2751463424, lds=7784628224, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 115,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 116,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 117,
    "metadata": "ProgramInfo(global_size=(1024, 256, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=5989466112, lds=19629342720, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      256,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 118,
    "metadata": "ProgramInfo(global_size=(256, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=5989466112, lds=19730006016, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      256,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 119,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 120,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(16, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=2, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1912602624, lds=4429185024, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      16,
      1,
      1
    ]
  },
  {
    "id": 121,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 122,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 12, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=622501822464, lds=2481955799040, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      12,
      1
    ]
  },
  {
    "id": 123,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 124,
    "metadata": "ProgramInfo(global_size=(1024, 512, 1), local_size=(64, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      512,
      1
    ],
    "local_size": [
      64,
      2,
      1
    ]
  },
  {
    "id": 125,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 126,
    "metadata": "ProgramInfo(global_size=(512, 1024, 1), local_size=(64, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      512,
      1024,
      1
    ],
    "local_size": [
      64,
      2,
      1
    ]
  },
  {
    "id": 127,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 128,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=19327352832, lds=73014444032, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      2,
      1
    ]
  },
  {
    "id": 129,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 3, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(3, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=41070624768, lds=159450660864, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      3,
      1
    ]
  },
  {
    "id": 130,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 131,
    "metadata": "ProgramInfo(global_size=(1024, 256, 1), local_size=(64, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      256,
      1
    ],
    "local_size": [
      64,
      4,
      1
    ]
  },
  {
    "id": 132,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 133,
    "metadata": "ProgramInfo(global_size=(256, 1024, 1), local_size=(64, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      256,
      1024,
      1
    ],
    "local_size": [
      64,
      4,
      1
    ]
  },
  {
    "id": 134,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 135,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=3, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=71403831296, lds=280246616064, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      4,
      1
    ]
  },
  {
    "id": 136,
    "metadata": "ProgramInfo(global_size=(1024, 128, 1), local_size=(64, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      128,
      1
    ],
    "local_size": [
      64,
      8,
      1
    ]
  },
  {
    "id": 137,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 138,
    "metadata": "ProgramInfo(global_size=(128, 1024, 1), local_size=(64, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      128,
      1024,
      1
    ],
    "local_size": [
      64,
      8,
      1
    ]
  },
  {
    "id": 139,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 140,
    "metadata": "ProgramInfo(global_size=(1024, 64, 1), local_size=(64, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      64,
      1
    ],
    "local_size": [
      64,
      16,
      1
    ]
  },
  {
    "id": 141,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 142,
    "metadata": "ProgramInfo(global_size=(64, 1024, 1), local_size=(64, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      1024,
      1
    ],
    "local_size": [
      64,
      16,
      1
    ]
  },
  {
    "id": 143,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 144,
    "metadata": "ProgramInfo(global_size=(1024, 64, 1), local_size=(64, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      64,
      1
    ],
    "local_size": [
      64,
      16,
      1
    ]
  },
  {
    "id": 145,
    "metadata": "ProgramInfo(global_size=(64, 1024, 1), local_size=(64, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      1024,
      1
    ],
    "local_size": [
      64,
      16,
      1
    ]
  },
  {
    "id": 146,
    "metadata": "ProgramInfo(global_size=(1024, 1024, 1), local_size=(64, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.SPLIT, axis=2, arg=(64, AxisType.LOCAL, True)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=6039797760, lds=20937965568, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1024,
      1024,
      1
    ],
    "local_size": [
      64,
      1,
      1
    ]
  },
  {
    "id": 147,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 148,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 149,
    "metadata": "ProgramInfo(global_size=(16, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=128974848, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 150,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 151,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 152,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 153,
    "metadata": "ProgramInfo(global_size=(32, 16, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      16,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 154,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 155,
    "metadata": "ProgramInfo(global_size=(8, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=116391936, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 156,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 157,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 158,
    "metadata": "ProgramInfo(global_size=(64, 1, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      1,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 159,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 160,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 161,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 162,
    "metadata": "ProgramInfo(global_size=(16, 64, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      64,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 163,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 164,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1617035264, lds=182452224, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 165,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 166,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 3, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(3, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1623392256, lds=210763776, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      3,
      1
    ]
  },
  {
    "id": 167,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 168,
    "metadata": "ProgramInfo(global_size=(32, 16, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      16,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 169,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 170,
    "metadata": "ProgramInfo(global_size=(8, 64, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      64,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 171,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 172,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1631846400, lds=247463936, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 173,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 174,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 175,
    "metadata": "ProgramInfo(global_size=(4, 64, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      64,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 176,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 177,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1686634496, lds=478150656, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 178,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 179,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 180,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 181,
    "metadata": "ProgramInfo(global_size=(2, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 182,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 183,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 184,
    "metadata": "ProgramInfo(global_size=(2, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 185,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 186,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 187,
    "metadata": "ProgramInfo(global_size=(64, 1, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      1,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 188,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 189,
    "metadata": "ProgramInfo(global_size=(64, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 190,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 191,
    "metadata": "ProgramInfo(global_size=(64, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=153616384, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 192,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 193,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 194,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 195,
    "metadata": "ProgramInfo(global_size=(64, 16, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=128188416, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      16,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 196,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 197,
    "metadata": "ProgramInfo(global_size=(16, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=128974848, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 198,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 199,
    "metadata": "ProgramInfo(global_size=(64, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 200,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 201,
    "metadata": "ProgramInfo(global_size=(32, 64, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      64,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 202,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1617166336, lds=232783872, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 203,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 3, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(3, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1623588864, lds=261095424, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      3,
      1
    ]
  },
  {
    "id": 204,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 205,
    "metadata": "ProgramInfo(global_size=(64, 16, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      16,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 206,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 207,
    "metadata": "ProgramInfo(global_size=(16, 64, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      64,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 208,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 209,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1632108544, lds=297795584, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 210,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 211,
    "metadata": "ProgramInfo(global_size=(64, 8, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      8,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 212,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 213,
    "metadata": "ProgramInfo(global_size=(8, 64, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      64,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 214,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 215,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1687158784, lds=528482304, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 216,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 217,
    "metadata": "ProgramInfo(global_size=(64, 4, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      4,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 218,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 219,
    "metadata": "ProgramInfo(global_size=(4, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 220,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 221,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1897922560, lds=1392508928, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 222,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 223,
    "metadata": "ProgramInfo(global_size=(64, 4, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      4,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 224,
    "metadata": "ProgramInfo(global_size=(4, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 225,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=7, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1897922560, lds=1392508928, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 226,
    "metadata": "ProgramInfo(global_size=(64, 2, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      2,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 227,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 228,
    "metadata": "ProgramInfo(global_size=(2, 64, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      64,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 229,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 230,
    "metadata": "ProgramInfo(global_size=(64, 2, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      2,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 231,
    "metadata": "ProgramInfo(global_size=(64, 64, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=204472320, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      64,
      64,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 232,
    "metadata": "ProgramInfo(global_size=(32, 16, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      16,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 233,
    "metadata": "ProgramInfo(global_size=(16, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 234,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 235,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=8, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 236,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=8, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 237,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65142784, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 238,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 239,
    "metadata": "ProgramInfo(global_size=(8, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65536000, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 240,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 241,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=8, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 242,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 243,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 244,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 245,
    "metadata": "ProgramInfo(global_size=(32, 16, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      16,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 246,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 247,
    "metadata": "ProgramInfo(global_size=(16, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 248,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 249,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 250,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 251,
    "metadata": "ProgramInfo(global_size=(8, 32, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      32,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 252,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 253,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 254,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 255,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 256,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 257,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 258,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 259,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 260,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 261,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 262,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 16, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      16,
      1
    ]
  },
  {
    "id": 263,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 264,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 32, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(32, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      32,
      1
    ]
  },
  {
    "id": 265,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 266,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 267,
    "metadata": "ProgramInfo(global_size=(32, 16, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      16,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 268,
    "metadata": "ProgramInfo(global_size=(16, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=128974848, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 269,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=8, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 270,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=8, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 271,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 272,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 273,
    "metadata": "ProgramInfo(global_size=(8, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=116391936, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 274,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 275,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=8, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 276,
    "metadata": "ProgramInfo(global_size=(32, 16, 1), local_size=(32, 2, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      16,
      1
    ],
    "local_size": [
      32,
      2,
      2
    ]
  },
  {
    "id": 277,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 278,
    "metadata": "ProgramInfo(global_size=(16, 32, 1), local_size=(32, 2, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      2
    ]
  },
  {
    "id": 279,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 280,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      4
    ]
  },
  {
    "id": 281,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 282,
    "metadata": "ProgramInfo(global_size=(8, 32, 1), local_size=(32, 2, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      4
    ]
  },
  {
    "id": 283,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 284,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 2, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      2,
      8
    ]
  },
  {
    "id": 285,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 286,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 2, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      8
    ]
  },
  {
    "id": 287,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 288,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 289,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 290,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 291,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 292,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 293,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 294,
    "metadata": "ProgramInfo(global_size=(32, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=154140672, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 295,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65536000, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 296,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 297,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=59244544, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 298,
    "metadata": "ProgramInfo(global_size=(4, 16, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      16,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 299,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 300,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 301,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 302,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 303,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 304,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 305,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=9, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 306,
    "metadata": "ProgramInfo(global_size=(4, 8, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65142784, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      8,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 307,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 308,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65536000, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 309,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 310,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=9, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 311,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 8, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      8,
      4
    ]
  },
  {
    "id": 312,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 313,
    "metadata": "ProgramInfo(global_size=(4, 16, 1), local_size=(32, 8, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      16,
      1
    ],
    "local_size": [
      32,
      8,
      2
    ]
  },
  {
    "id": 314,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 315,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 8, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      8,
      2
    ]
  },
  {
    "id": 316,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 317,
    "metadata": "ProgramInfo(global_size=(4, 8, 1), local_size=(32, 8, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      8,
      1
    ],
    "local_size": [
      32,
      8,
      4
    ]
  },
  {
    "id": 318,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 319,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 8, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=103284736, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      8,
      1
    ]
  },
  {
    "id": 320,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=55607296, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 321,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=2, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65142784, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 322,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65142784, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 323,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 324,
    "metadata": "ProgramInfo(global_size=(16, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=52690944, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 325,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 326,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 327,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 328,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=58785792, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 329,
    "metadata": "ProgramInfo(global_size=(8, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=40108032, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 330,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 331,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 2, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      2,
      8
    ]
  },
  {
    "id": 332,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 333,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 2, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      2,
      2
    ]
  },
  {
    "id": 334,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 335,
    "metadata": "ProgramInfo(global_size=(16, 8, 1), local_size=(32, 2, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      2
    ]
  },
  {
    "id": 336,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 337,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 2, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      2,
      4
    ]
  },
  {
    "id": 338,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 339,
    "metadata": "ProgramInfo(global_size=(8, 8, 1), local_size=(32, 2, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      4
    ]
  },
  {
    "id": 340,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 341,
    "metadata": "ProgramInfo(global_size=(4, 8, 1), local_size=(32, 2, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      8
    ]
  },
  {
    "id": 342,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 343,
    "metadata": "ProgramInfo(global_size=(2, 8, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 344,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 345,
    "metadata": "ProgramInfo(global_size=(2, 8, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 346,
    "metadata": "ProgramInfo(global_size=(8, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 347,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=55607296, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 348,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=2, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=65142784, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 349,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=65142784, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 350,
    "metadata": "ProgramInfo(global_size=(16, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=52690944, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 351,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 352,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(3, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 353,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=58785792, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 354,
    "metadata": "ProgramInfo(global_size=(8, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=40108032, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 355,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 356,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 2, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      2,
      8
    ]
  },
  {
    "id": 357,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 2, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      2,
      2
    ]
  },
  {
    "id": 358,
    "metadata": "ProgramInfo(global_size=(16, 8, 1), local_size=(32, 2, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      16,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      2
    ]
  },
  {
    "id": 359,
    "metadata": "ProgramInfo(global_size=(32, 2, 1), local_size=(32, 2, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      2,
      1
    ],
    "local_size": [
      32,
      2,
      4
    ]
  },
  {
    "id": 360,
    "metadata": "ProgramInfo(global_size=(8, 8, 1), local_size=(32, 2, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      4
    ]
  },
  {
    "id": 361,
    "metadata": "ProgramInfo(global_size=(4, 8, 1), local_size=(32, 2, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      8
    ]
  },
  {
    "id": 362,
    "metadata": "ProgramInfo(global_size=(2, 8, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 363,
    "metadata": "ProgramInfo(global_size=(2, 8, 1), local_size=(32, 2, 16), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(16, AxisType.LOCAL, True))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      16
    ]
  },
  {
    "id": 364,
    "metadata": "ProgramInfo(global_size=(8, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      8,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 365,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=59244544, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 366,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=2, arg=(0, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=59244544, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      1,
      1
    ]
  },
  {
    "id": 367,
    "metadata": "ProgramInfo(global_size=(4, 16, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=52690944, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      16,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 368,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 369,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65536000, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 370,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 371,
    "metadata": "ProgramInfo(global_size=(4, 32, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=65536000, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      32,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 372,
    "metadata": "ProgramInfo(global_size=(4, 8, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=39976960, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      8,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 373,
    "metadata": "ProgramInfo(global_size=(32, 1, 1), local_size=(32, 4, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(0, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      1,
      1
    ],
    "local_size": [
      32,
      4,
      4
    ]
  },
  {
    "id": 374,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 375,
    "metadata": "ProgramInfo(global_size=(4, 16, 1), local_size=(32, 4, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      16,
      1
    ],
    "local_size": [
      32,
      4,
      2
    ]
  },
  {
    "id": 376,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 377,
    "metadata": "ProgramInfo(global_size=(2, 32, 1), local_size=(32, 4, 2), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(2, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      2,
      32,
      1
    ],
    "local_size": [
      32,
      4,
      2
    ]
  },
  {
    "id": 378,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 379,
    "metadata": "ProgramInfo(global_size=(4, 8, 1), local_size=(32, 4, 4), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      8,
      1
    ],
    "local_size": [
      32,
      4,
      4
    ]
  },
  {
    "id": 380,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 381,
    "metadata": "ProgramInfo(global_size=(4, 4, 1), local_size=(32, 4, 8), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(8, AxisType.LOCAL))), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      4,
      4,
      1
    ],
    "local_size": [
      32,
      4,
      8
    ]
  },
  {
    "id": 382,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 383,
    "metadata": "ProgramInfo(global_size=(32, 4, 1), local_size=(32, 4, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='test', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=1, arg=(8, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=2, arg=(2, AxisType.UPCAST)), Opt(op=OptOps.SWAP, axis=0, arg=1)), opts_to_apply=None, estimates=Estimates(ops=1611661312, lds=78118912, mem=5244928), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      32,
      4,
      1
    ],
    "local_size": [
      32,
      4,
      1
    ]
  },
  {
    "id": 384,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='slots_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='cuda_cuda_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(6, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(7, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),)), UOp(Ops.PARAM, arg=ParamArg(8, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(0, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": false,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  },
  {
    "id": 385,
    "metadata": "ProgramInfo(global_size=(32, 8, 1), local_size=(32, 2, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1048576, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=786432, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.f16, device='CUDA'), src=(\n  UOp(Ops.CONST, arg=1024, src=()),))), globals=(0, 1, 2, 3), outs=(0,), ins=(1, 2, 3), target=CUDA::sm_90)",
    "kernel_info": "KernelInfo(name='r\\x1b[90m_\\x1b[0m\\x1b[34m8\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[34m32\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[36m2\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[96m32\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[33m2\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[33m2\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[33m2\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[33m4\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[33m4\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[31m24\\x1b[0m\\x1b[90m_\\x1b[0m\\x1b[33m2\\x1b[0m', applied_opts=(Opt(op=OptOps.TC, axis=0, arg=(-1, 0, 1)), Opt(op=OptOps.SPLIT, axis=1, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=0, arg=(2, AxisType.LOCAL)), Opt(op=OptOps.SPLIT, axis=0, arg=(4, AxisType.UPCAST)), Opt(op=OptOps.SPLIT, axis=9, arg=(2, AxisType.UPCAST))), opts_to_apply=None, estimates=Estimates(ops=1661992960, lds=77856768, mem=5244928), beam=0, optimized=True)",
    "selected": true,
    "global_size": [
      32,
      8,
      1
    ],
    "local_size": [
      32,
      2,
      1
    ]
  },
  {
    "id": 386,
    "metadata": "ProgramInfo(global_size=(1, 1, 1), local_size=(1, 1, 1), params=(UOp(Ops.PARAM, arg=ParamArg(0, dtypes.u64, name='cuda_cuda_0', device='CPU'), src=(\n  UOp(Ops.CONST, arg=3, src=()),)), UOp(Ops.PARAM, arg=ParamArg(1, dtypes.u64, name='slots_1', device='CPU'), src=(\n  UOp(Ops.CONST, arg=4, src=()),)), UOp(Ops.PARAM, arg=ParamArg(2, dtypes.u64, name='timeline_2', device='CPU'), src=(\n  UOp(Ops.CONST, arg=2, src=()),)), UOp(Ops.PARAM, arg=ParamArg(3, dtypes.u8, name='program_3', device='CPU'), src=(\n  UOp(Ops.CONST, arg=8, src=()),)), UOp(Ops.PARAM, arg=ParamArg(4, dtypes.u8, name='kernargs_4', device='CPU'), src=(\n  UOp(Ops.CONST, arg=80, src=()),)), UOp(Ops.PARAM, arg=ParamArg(5, dtypes.u64, addrspace=AddrSpace.ALU), src=(\n  UOp(Ops.STACK, arg=None, src=()),))), globals=(0, 1, 2, 3, 4), outs=(1, 2), ins=(2,), target=CPU:LLVM:x86_64,native)",
    "kernel_info": "KernelInfo(name='hcq_submit', applied_opts=(), opts_to_apply=None, estimates=Estimates(ops=0, lds=0, mem=0), beam=0, optimized=True)",
    "selected": true,
    "global_size": [
      1,
      1,
      1
    ],
    "local_size": [
      1,
      1,
      1
    ]
  }
]