Skip to content

Commit f7879c8

Browse files
EetusjoGoogle-ML-Automation
authored andcommitted
PR #35575: [ROCm] Add autotuner rocblas/hipblaslt backends
Imported from GitHub PR #35575 Adds ROCm-specific autotuner backends. Essentially copies the existing cublas/lt backends and renames, with minor changes. Comment from another PR as context, with the ask to separate CUDA/ROCm autotuner backends: #35280 (comment) Copybara import of the project: -- e5a5496 by Eetu Sjöblom <[email protected]>: copy cublas/cublaslt backends to create rocblas/hipblaslt ones -- a09edaf by Eetu Sjöblom <[email protected]>: Pass cc to GetBlasComputationType Merging this change closes #35575 FUTURE_COPYBARA_INTEGRATE_REVIEW=#35575 from ROCm:ci_rocm_autotuner_backends f9e8b77 PiperOrigin-RevId: 853116778
1 parent 84e869c commit f7879c8

File tree

1 file changed

+11
-0
lines changed
  • xla/backends/gpu/autotuner

1 file changed

+11
-0
lines changed

xla/backends/gpu/autotuner/BUILD

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -641,23 +641,34 @@ cc_library(
641641
deps = [
642642
":gpu_codegen_backend",
643643
"//xla:autotuning_proto_cc",
644+
"//xla:shape_util",
644645
"//xla:xla_proto_cc",
645646
"//xla/backends/autotuner:codegen_backend",
646647
"//xla/hlo/ir:hlo",
648+
"//xla/hlo/utils:hlo_query",
647649
"//xla/service:compiler",
650+
"//xla/service:hlo_cost_analysis",
648651
"//xla/service/gpu:backend_configs_cc",
649652
"//xla/service/gpu:cublas_cudnn",
650653
"//xla/service/gpu:matmul_utils",
654+
"//xla/service/gpu/autotuning:redzone_buffers",
655+
"//xla/service/gpu/transforms:dot_algorithm_rewriter",
656+
"//xla/service/gpu/transforms:gemm_rewriter",
651657
"//xla/stream_executor:blas",
652658
"//xla/stream_executor:device_address",
653659
"//xla/stream_executor:device_address_allocator",
654660
"//xla/stream_executor:device_description",
661+
"//xla/stream_executor:semantic_version",
655662
"//xla/stream_executor:stream_executor_h",
656663
"//xla/stream_executor:stream_executor_memory_allocator",
657664
"//xla/stream_executor/gpu:gpu_blas_lt",
665+
"//xla/stream_executor/gpu:redzone_allocator",
658666
"//xla/stream_executor/rocm:rocblas_plugin",
667+
"//xla/tools:hlo_decomposer_lib",
668+
"//xla/tsl/lib/gtl:iterator_range",
659669
"//xla/tsl/platform:errors",
660670
"//xla/tsl/platform:statusor",
671+
"@com_google_absl//absl/algorithm:container",
661672
"@com_google_absl//absl/log",
662673
"@com_google_absl//absl/status",
663674
"@com_google_absl//absl/status:statusor",

0 commit comments

Comments
 (0)