ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-21 13:29:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

[TILE ENGINE] Restructure to Base class of GEMM (#3434)

Browse Source 
[ROCm/composable_kernel commit: e22622f0ec]
This commit is contained in:
Thrupti Raj Lakshmana Gowda
2025-12-19 09:53:56 -06:00
committed by GitHub
parent a6821f428f
commit a571bf9e3a
41 changed files with 2246 additions and 3458 deletions
Download Patch File Download Diff File Expand all files Collapse all files

32
Jenkinsfile vendored
View File

@@ -1637,18 +1637,18 @@ pipeline {
-D CMAKE_CXX_COMPILER="${params.BUILD_COMPILER}" \
-D CMAKE_BUILD_TYPE=Release \
-D GPU_TARGETS="gfx90a" \
-D GEMM_DATATYPE="fp8;fp16" \
-D GEMM_LAYOUT="rcr;rrr;crr;ccr" \
-D GEMM_UNIVERSAL_DATATYPE="fp8;fp16" \
-D GEMM_UNIVERSAL_LAYOUT="rcr;rrr;crr;ccr" \
-D GEMM_STREAMK_DATATYPE="fp8;fp16" \
-D GEMM_STREAMK_LAYOUT="rcr" \
-D GEMM_MULTI_D_DATATYPE="fp16" \
-D GEMM_MULTI_D_LAYOUT="rcrr;rrrr;crrr;ccrr" \
-D GEMM_PRESHUFFLE_DATATYPE="fp16;fp8;bf16;bf8" \
-D GEMM_PRESHUFFLE_LAYOUT="rcr" .. && \
ninja -j64 benchmark_gemm_all benchmark_gemm_preshuffle_all benchmark_gemm_multi_d_all benchmark_gemm_streamk_all && \
python3 ../tile_engine/ops/gemm/gemm_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm_preshuffle/gemm_preshuffle_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm_multi_d/gemm_multi_d_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json """
ninja -j${nthreads()} benchmark_gemm_universal_all benchmark_gemm_preshuffle_all benchmark_gemm_multi_d_all benchmark_gemm_streamk_all && \
python3 ../tile_engine/ops/gemm/gemm_universal/gemm_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm/gemm_preshuffle/gemm_preshuffle_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm/gemm_multi_d/gemm_multi_d_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json """
}
steps{
buildHipClangJobAndReboot(setup_args:setup_args, build_type: 'Release', execute_cmd: execute_args)
@@ -1668,18 +1668,18 @@ pipeline {
-D CMAKE_CXX_COMPILER="${params.BUILD_COMPILER}" \
-D CMAKE_BUILD_TYPE=Release \
-D GPU_TARGETS="gfx942" \
-D GEMM_DATATYPE="fp8;fp16" \
-D GEMM_LAYOUT="rcr;rrr;crr;ccr" \
-D GEMM_UNIVERSAL_DATATYPE="fp8;fp16" \
-D GEMM_UNIVERSAL_LAYOUT="rcr;rrr;crr;ccr" \
-D GEMM_STREAMK_DATATYPE="fp8;fp16" \
-D GEMM_STREAMK_LAYOUT="rcr" \
-D GEMM_MULTI_D_DATATYPE="fp16" \
-D GEMM_MULTI_D_LAYOUT="rcrr;rrrr;crrr;ccrr" \
-D GEMM_PRESHUFFLE_DATATYPE="fp16;fp8;bf16;bf8" \
-D GEMM_PRESHUFFLE_LAYOUT="rcr" .. && \
ninja -j64 benchmark_gemm_all benchmark_gemm_preshuffle_all benchmark_gemm_multi_d_all benchmark_gemm_streamk_all && \
python3 ../tile_engine/ops/gemm/gemm_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm_preshuffle/gemm_preshuffle_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm_multi_d/gemm_multi_d_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json """
ninja -j${nthreads()} benchmark_gemm_universal_all benchmark_gemm_preshuffle_all benchmark_gemm_multi_d_all benchmark_gemm_streamk_all && \
python3 ../tile_engine/ops/gemm/gemm_universal/gemm_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm/gemm_preshuffle/gemm_preshuffle_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json && \
python3 ../tile_engine/ops/gemm/gemm_multi_d/gemm_multi_d_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json """
}
steps{
buildHipClangJobAndReboot(setup_args:setup_args, build_type: 'Release', execute_cmd: execute_args)
@@ -1699,10 +1699,10 @@ pipeline {
-D CMAKE_CXX_COMPILER="${params.BUILD_COMPILER}" \
-D CMAKE_BUILD_TYPE=Release \
-D GPU_TARGETS="gfx1201" \
-D GEMM_DATATYPE="fp16" \
-D GEMM_LAYOUT="rcr;rrr;crr;ccr" .. && \
ninja -j64 benchmark_gemm_all && \
python3 ../tile_engine/ops/gemm/gemm_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json """
-D GEMM_UNIVERSAL_DATATYPE="fp16" \
-D GEMM_UNIVERSAL_LAYOUT="rcr;rrr;crr;ccr" .. && \
ninja -j${nthreads()} benchmark_gemm_universal_all && \
python3 ../tile_engine/ops/gemm/gemm_universal/gemm_benchmark.py . --problem-sizes "1024,1024,1024" --warmup 5 --repeat 5 --verbose --json results.json """
}
steps{
buildHipClangJobAndReboot(setup_args:setup_args, build_type: 'Release', execute_cmd: execute_args)