[Joint Matrix] Enable different accumulator and output types in spirv. Add tests to cover bfloat16 and half floating-point sizes.
Updated matrix_compare and matrix_multiply_ref functions to better match bfloat16 calculations on device.
@ggojska , please, resolve conflicts.
@intel/llvm-gatekeepers Are those test fails in pre-commit know issue or is it on my side?
@YuriPlyakhin, can you approve the requested changes if you are happy with the changes made?
@intel/llvm-reviewers-runtime, please help review the non matrix parts
@intel/llvm-gatekeepers, this PR is ready to be merged.
@intel/llvm-gatekeepers, this PR is ready to be merged.
Need @YuriPlyakhin to unblock.
This commit will be authored by [email protected].
@ggojska you need to make your e-mail public on Github, I think.
@aelovikov-intel it should be public now