ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-20 04:49:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

implement device batched gemm b scale for wmma (#2825)

Browse Source 
* rebased on top of develop

* fixed missing shuffeling and wrong indexing

* added tests for batched_b_scale

* added missing files

* fixed wrong stride computation and removed k batching (for now) due to precision issues

* reinstated k-batching with PRNG constrained to -1..1

* added specialization of GeneratorTensor_3 for int4 and fixed internal overflow

* added k-batching to reference and increased tolerances for test

* changed gemm_b_scale and gemm_universal tests to use correct parameters

* adressed review commentsd

* ported fixes back to non-batched version of b_scale

* adressed review comments

* run clang-format on older commits

* add type-conversion to AccDataType and then to CDataType to exactly mimic GPU's behavior

* added newline at end of file

* reflected changes from muitl-abd branch in batched b_scale

* fixed gfx11 issue

* changed range for pki4 to -1...1 (-0.5...0.5 never really made sense for i4 anyway and always should have caused compiler errors, but since there was no int4 specialization of GeneratorTensor3 until now, this passed

* run clang format

* set range of i4 generation to 0...1 for upstream tests to pass. This replicated previous behavior, which however means that it is NOT properly tested.

* reduced range for pk_i4 even further to 0..0

* removed failing xld instances. Failure now uncovered now that tests were fixed

* removed generation of int4 values entierly

* divide B buffer by BPackedSize

---------

Co-authored-by: Kevin Abraham <kevin.abraham@streamhpc.com>

[ROCm/composable_kernel commit: c4b2da9cbd]
This commit is contained in:
kabrahamAMD
2025-10-16 20:00:42 +02:00
committed by GitHub
parent 62afd9eb14
commit 06d76b160e
22 changed files with 1352 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
profiler/src/CMakeLists.txt
View File

@@ -67,7 +67,6 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9")
list(APPEND PROFILER_OPS profile_gemm_add.cpp)
list(APPEND PROFILER_OPS profile_gemm_bias_add_reduce.cpp)
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_streamk.cpp)
list(APPEND PROFILER_OPS profile_conv_fwd_bias_relu.cpp)
@@ -89,6 +88,7 @@ endif()
if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]")
list(APPEND PROFILER_OPS profile_gemm_universal.cpp)
list(APPEND PROFILER_OPS profile_batched_gemm.cpp)
list(APPEND PROFILER_OPS profile_batched_gemm_b_scale.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)
@@ -191,7 +191,6 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[1
list(APPEND DEVICE_INSTANCES device_gemm_mx_instance)
endif()
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_streamk_instance)
list(APPEND DEVICE_INSTANCES device_gemm_add_multiply_instance)
@@ -229,6 +228,7 @@ if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[1
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_batched_gemm_b_scale_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)

2
profiler/src/profile_batched_gemm_b_scale.cpp
View File

@@ -57,7 +57,7 @@ int profile_batched_gemm_b_scale(int argc, char* argv[])
printf("arg6: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg7: print tensor value (0: no; 1: yes)\n");
printf("arg8: time kernel (0=no, 1=yes)\n");
printf("arg9 to 15: M, N, K, StrideA, StrideB, StrideC, BatachCount\n");
printf("arg9 to 15: M, N, K, StrideA, StrideB, StrideC, BatchCount\n");
printf("arg16: split k into mulitiple batch\n");
printf("optional:\n");
printf("arg17: number of warm-up cycles (default 1)\n");