[Issue]: Calling min(size_t, int) in ContractionSolution.cpp causes ambiguity

[yxsamliu]

Problem Description

when we try to fix a compiler bug about function min by llvm/llvm-project#82956 we encountered build failure in Tensile/rocBLAS. Then we found the line that causes the compilation failure is

https://github.com/ROCm/Tensile/blame/6aa9c335a5cecf573dc129adda79e284ac898904/Tensile/Source/lib/source/ContractionSolution.cpp#L1874

Basically this line is calling function min defined in clang header with (unsigned long, int). Previously the compiler only defined min(int, int). As we introduce min(unsigned, unsigned), min(long, long), min(unsigned long, unsigned long). The compiler could not choose among these candidates since there is no exact match.

We do not want to define min(unsigned long, int) because that would do implicit conversion from int to unsigned long then compare. This may cause unexpected results, e.g. min(1UL, -1) will return 1UL. We want users to do explicit cast to indicate their intention. We think it is better than doing silent implicit conversion.

Therefore we would like Tensile to modify the line https://github.com/ROCm/Tensile/blame/6aa9c335a5cecf573dc129adda79e284ac898904/Tensile/Source/lib/source/ContractionSolution.cpp#L1874

to be

return min(cuCount, static_cast<size_t>(pAMDGPU->skMaxCUs));

Operating System

Ubuntu 22.04

CPU

any

GPU

AMD Instinct MI300X

ROCm Version

ROCm 6.1.0

ROCm Component

Tensile

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

[bstefanuk]

Updates have been applied to address this issue in #2003.