[AMD] Implement RepOrder for AMD MMA layouts and change kMajor notation to kMinor

[oplavsic]

Implement RepOrder methods for MFMA and WMMA layouts. Both layouts have row major rep layout. Also,
isTranspose flag in MFMA layout does not affect RepOrder, meaning RepOrder is row major in both cases.
Also, this commit changes kMajor to kMinor notation. For a long time, kMajor is mis-used to indicate K is the fastest changing dimension. If we follow the same convention of row-major, it actually means elements on the same row are contiguous in memory.

[zhanglx13]

Thank you @oplavsic
Since you are here, can you also change kMajor to kMinor? For a long time, kMajor is mis-used to indicate K is the fastest changing dimension. If we follow the same convention of row-major, it actually means elements on the same row are contiguous in memory. So k-major means elements of the same k index are contiguous in memory, which is not what we mean here.

[zhanglx13]

Also AMD sharedToDotOperand:

triton/third_party/amd/lib/TritonAMDGPUToLLVM/ConvertLayoutOpToLLVM/SharedToDotOperandHelper.cpp

Line 79 in 9d6736a

bool isKMajor(llvm::ArrayRef<unsigned> order, int opIdx) {

[oplavsic]

Also AMD sharedToDotOperand:

triton/third_party/amd/lib/TritonAMDGPUToLLVM/ConvertLayoutOpToLLVM/SharedToDotOperandHelper.cpp

Line 79 in 9d6736a

bool isKMajor(llvm::ArrayRef<unsigned> order, int opIdx) {

Done!

[zhanglx13]

LGTM
@oplavsic Can you also include the fix of kmajor -> kminor in the PR title and description?