LoadStoreOp can have scalar inputs, so type check is required

[naoyam]

Type error was detected with #3263 while I was testing it with a Debug build.

pytest -v tests/python/test_python_frontend.py -k test_pad_dynamic

It has a fusion of:

Inputs:
  T0_g_float[ bS0{1}, iS1{i1}, iS2{i2} ]
Outputs:
  T1_g_float[ bS11{1}, iS12{( ( i1 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )}, iS13{( ( i2 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )} ]

%kernel_math {
f7 = (float)(7);
f9 = float(2.5) * f7;
i11 = (int64_t)(f9);
i14 = (nvfuser_index_t)(i11);
i16 = (nvfuser_index_t)(i11);
i18 = (nvfuser_index_t)(i11);
i20 = (nvfuser_index_t)(i11);
T2_l_float[ bS3{1}, iS5{( ( i1 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )}rf, iS7{( ( i2 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )}rf ]
   = pad( T0_g_float[ bS0{1}, iS1{i1}, iS2{i2} ], {0, 0, i14, i16, i18, i20} )
T1_g_float[ bS11{1}, iS12{( ( i1 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )}, iS13{( ( i2 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )} ]
   = Set( T2_l_float[ bS3{1}, iS5{( ( i1 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )}rf, iS7{( ( i2 + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) ) + ( (nvfuser_index_t)(( (int64_t)(( float(2.5) * ( (float)(7) ) )) )) ) )}rf ], cache_op=Streaming )
} // %kernel_math

Stack trace:

#0  __cxxabiv1::__cxa_throw (obj=0xabc6ea0, tinfo=0x7ffeba81c248 <typeinfo for nvfuser::nvfError>, dest=0x7ffeba059370 <nvfuser::nvfError::~nvfError()>) at ../../../../libstdc++-v3/libsupc++/eh_throw.cc:80
#1  0x00007ffeba058665 in nvfuser::nvfCheckFail (func=0x7ffeb9927c33 "as", file=0x7ffeb99d9a1b "/raid/nmaruyama/debug1/csrc/utils.h", line=119,
    msg=0x7ffeb99c36e6 " INTERNAL ASSERT FAILED at \"/raid/nmaruyama/debug1/csrc/utils.h\":119, please report a bug with repro script to NVFuser at https://github.com/NVIDIA/Fuser/issues. ") at /raid/nmaruyama/debug1/csrc/exceptions.cpp:283
#2  0x00007ffeb9c1be4b in nvfuser::nvfErrorFail (func=0x7ffeb9927c33 "as", file=0x7ffeb99d9a1b "/raid/nmaruyama/debug1/csrc/utils.h", line=119,
    condMsg=0x7ffeb99c36e6 " INTERNAL ASSERT FAILED at \"/raid/nmaruyama/debug1/csrc/utils.h\":119, please report a bug with repro script to NVFuser at https://github.com/NVIDIA/Fuser/issues. ") at /raid/nmaruyama/debug1/csrc/exceptions.h:229
#3  0x00007ffeb9c1bbe4 in nvfuser::PolymorphicBase::as<nvfuser::TensorView> (this=0xac07490) at /raid/nmaruyama/debug1/csrc/utils.h:119
#4  0x00007ffeba54c67f in nvfuser::(anonymous namespace)::isLoadGlobalToLocal (expr=0xabd7c50) at /raid/nmaruyama/debug1/csrc/scheduler/cache_policy_refiner.cpp:61
#5  0x00007ffeba54c599 in nvfuser::refineCachePolicy (fusion=0xabef940) at /raid/nmaruyama/debug1/csrc/scheduler/cache_policy_refiner.cpp:153

[naoyam]

!test

[naoyam]

I'd try to add a unit test to catch potential regression. Fine with me as is if that's complicated.

I'm not sure if that would be possible with the Release build mode since we use Release for CI runs.