ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

[CK_TILE] FMHA BWD Add D96 Instances (#2916)

Browse Source
This commit is contained in:
Yi DING
2025-09-24 17:04:23 +08:00
committed by GitHub
parent 15fff74503
commit fe0a47a011
5 changed files with 35 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
example/ck_tile/01_fmha/script/run_full_test.sh
View File

@@ -34,15 +34,15 @@ function print_log_header(){
}
#run verification tests
example/ck_tile/01_fmha/script/smoke_test_fwd.sh
example/ck_tile/01_fmha/script/smoke_test_bwd.sh
time example/ck_tile/01_fmha/script/smoke_test_fwd.sh
time example/ck_tile/01_fmha/script/smoke_test_bwd.sh
#run performance benchmarks
export fmha_fwd_log="perf_fmha_fwd_$GPU_arch.log"
print_log_header $fmha_fwd_log $env_type $branch $host_name
example/ck_tile/01_fmha/script/benchmark_fwd.sh 2>&1 | tee -a $fmha_fwd_log
time example/ck_tile/01_fmha/script/benchmark_fwd.sh 2>&1 | tee -a $fmha_fwd_log
export fmha_bwd_log="perf_fmha_bwd_$GPU_arch.log"
print_log_header $fmha_bwd_log $env_type $branch $host_name
example/ck_tile/01_fmha/script/benchmark_bwd.sh 2>&1 | tee -a $fmha_bwd_log
time example/ck_tile/01_fmha/script/benchmark_bwd.sh 2>&1 | tee -a $fmha_bwd_log

40
example/ck_tile/01_fmha/script/smoke_test_bwd.sh
View File

@@ -6,7 +6,7 @@ SCRIPT_DIR=$(cd $(dirname "${BASH_SOURCE[0]}") && pwd)
EXE_NAME=tile_example_fmha_bwd
EXE="$(find . -name $EXE_NAME -type f | head -n 1)"
KNAME=1
GPU_arch=$GPU_arch
GPU_arch=${GPU_arch:-""}
if [ -z "$GPU_arch" ] ; then
GPU_arch=$(rocminfo | grep -E 'Name:\s+gfx' | head -n1 | awk '{print $2}')
fi
@@ -31,7 +31,17 @@ run_exe() {
set -ex
}
test_h_s_mask() {
run_exe -b=1 -h=4 -h_k=2 -s=259 $@
run_exe -b=2 -h=2 -s=516 -s_k=253 $@
run_exe -b=1 -h=4 -h_k=1 -s=500 -s_k=251 -mask=1 $@
run_exe -b=1 -h=2 -s=900 -s_k=258 -mask=2 $@
run_exe -b=2 -h=1 -s=987 -s_k=219 -mask=t:128,30 $@
run_exe -b=2 -h=3 -h_k=1 -s=244 -s_k=499 -mask=b:4,35 $@
}
set -x
# main tests
for prec in "fp16" "bf16" ; do
for perm in 0 1 ; do
for hdim in 32 64 128 256 ; do
@@ -40,21 +50,21 @@ for bias in "n" "a" ; do
for dbias in 0 ; do
for p_drop in 0.0 0.2 ; do
for deterministic in 0 ; do
test_h_s_mask -prec=$prec -d=$hdim -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
done
done
done
done
done
done
done
done
run_exe -prec=$prec -b=1 -h=4 -h_k=2 -d=$hdim -s=259 -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
run_exe -prec=$prec -b=2 -h=2 -d=$hdim -s=516 -s_k=253 -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
run_exe -prec=$prec -b=1 -h=4 -h_k=1 -d=$hdim -s=500 -s_k=251 -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -mask=1 -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
run_exe -prec=$prec -b=1 -h=2 -d=$hdim -s=900 -s_k=258 -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -mask=2 -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
run_exe -prec=$prec -b=2 -h=1 -d=$hdim -s=987 -s_k=219 -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -mask=t:128,30 -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
run_exe -prec=$prec -b=2 -h=3 -h_k=1 -d=$hdim -s=244 -s_k=499 -bias=$bias -dbias=$dbias -p_drop=$p_drop -iperm=$perm -operm=$perm -mask=b:4,35 -deterministic=$deterministic -v=1 -mode=$mode -kname=$KNAME $COMMON_ARGS
done
done
done
done
done
done
done
# additional cases
for hdim in 72 96 ; do
test_h_s_mask -prec=fp16 -d=$hdim -bias=a -dbias=0 -p_drop=0.2 -iperm=0 -operm=0 -deterministic=0 -v=1 -mode=1 -kname=$KNAME $COMMON_ARGS
test_h_s_mask -prec=bf16 -d=$hdim -bias=n -dbias=0 -p_drop=0 -iperm=1 -operm=1 -deterministic=0 -v=1 -mode=1 -kname=$KNAME $COMMON_ARGS
test_h_s_mask -prec=bf16 -d=$hdim -bias=a -dbias=0 -p_drop=0.2 -iperm=1 -operm=1 -deterministic=0 -v=1 -mode=1 -kname=$KNAME $COMMON_ARGS
done
set +x