Add CudnnPadForConvolutions and CudnnVecotrizeConvolutions HLO pass#4

Open

hsharsha wants to merge 11 commits intomainfrom ci_convbf16_20240305

+228-37

Commits on Feb 27, 2024

use potrfDSCZ

i-chaochen
authored and
pemeliya
committed
fixing use_hgemm_alt_impl_ flag
pemeliya
committed

Commits on Mar 5, 2024

ncclCommSplit is from rocm6.0

i-chaochen
authored and
hsharsha
committed
fixed dd7604b on absl::Status

i-chaochen
authored and
hsharsha
committed
fixed c952055

i-chaochen
authored and
hsharsha
committed
add GraphNodeGetDependencies in rocm

i-chaochen
authored and
hsharsha
committed
no break

i-chaochen
authored and
hsharsha
committed
[ROCm] Use addresspace(5) for allocas on ROCM also

draganmladjenovic
authored and
hsharsha
committed
[ROCm] Don't use CUDA PTX for ROCM in ComputationIdCmd

draganmladjenovic
authored and
hsharsha
committed
added ConvBfloat16Support HLO pass

pemeliya
authored and
hsharsha
committed

Commits on Mar 7, 2024

Add CudnnPadForConvolutions and CudnnVecotrizeConvolutions HLO pass
hsharsha
committed