ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-15 18:42:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit 'b0ee317d83b77741022997265d4125697e7f7804' into develop

Browse Source
This commit is contained in:
assistant-librarian[bot]
2025-09-12 20:11:58 +00:00
parent facbc883fa
commit 302aa809ea
65 changed files with 2301 additions and 232 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
profiler/src/CMakeLists.txt
View File

@@ -68,7 +68,6 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9")
list(APPEND PROFILER_OPS profile_gemm_splitk.cpp)
list(APPEND PROFILER_OPS profile_batched_gemm_b_scale.cpp)
list(APPEND PROFILER_OPS profile_gemm_universal_batched.cpp)
list(APPEND PROFILER_OPS profile_gemm_universal_reduce.cpp)
list(APPEND PROFILER_OPS profile_gemm_universal_streamk.cpp)
list(APPEND PROFILER_OPS profile_conv_fwd_bias_relu.cpp)
list(APPEND PROFILER_OPS profile_conv_fwd_bias_relu_add.cpp)
@@ -90,6 +89,7 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[1
list(APPEND PROFILER_OPS profile_gemm_universal.cpp)
list(APPEND PROFILER_OPS profile_batched_gemm.cpp)
list(APPEND PROFILER_OPS profile_gemm_b_scale.cpp)
list(APPEND PROFILER_OPS profile_gemm_universal_reduce.cpp)
list(APPEND PROFILER_OPS profile_grouped_conv_fwd.cpp)
list(APPEND PROFILER_OPS profile_grouped_conv_fwd_bias_clamp.cpp)
list(APPEND PROFILER_OPS profile_grouped_conv_fwd_clamp.cpp)
@@ -185,7 +185,6 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9")
list(APPEND DEVICE_INSTANCES device_gemm_splitk_instance)
list(APPEND DEVICE_INSTANCES device_batched_gemm_b_scale_instance)
list(APPEND DEVICE_INSTANCES device_gemm_universal_batched_instance)
list(APPEND DEVICE_INSTANCES device_gemm_universal_reduce_instance)
list(APPEND DEVICE_INSTANCES device_gemm_universal_streamk_instance)
list(APPEND DEVICE_INSTANCES device_gemm_add_multiply_instance)
list(APPEND DEVICE_INSTANCES device_gemm_add_instance)
@@ -221,6 +220,7 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[1
list(APPEND DEVICE_INSTANCES device_gemm_universal_instance)
list(APPEND DEVICE_INSTANCES device_batched_gemm_instance)
list(APPEND DEVICE_INSTANCES device_gemm_b_scale_instance)
list(APPEND DEVICE_INSTANCES device_gemm_universal_reduce_instance)
list(APPEND DEVICE_INSTANCES device_grouped_conv3d_fwd_instance)
list(APPEND DEVICE_INSTANCES device_grouped_conv2d_bwd_data_instance)
list(APPEND DEVICE_INSTANCES device_grouped_conv3d_bwd_data_instance)