Add attn sink (#2892)

* enable attn sink

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* update attn_sink script

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* fix some error

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* clang-format

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* update fmha_bwd mask

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* update fmha_bwd_kernel'mask

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* update block_fmha_pipeline_qr_ks_vs.hpp

Signed-off-by: JL-underdog <Jun.Lin@amd.com>

* fix ci error

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* fix format error

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* Update block_fmha_bwd_pipeline_default_policy.hpp

* Update fmha_fwd_runner.hpp

* Update block_fmha_batch_prefill_pipeline_qr_ks_vs_async.hpp

* Update fmha_fwd_runner.hpp

* Update fmha_fwd_runner.hpp

* Update fmha_fwd_runner.hpp

* update splitkv_pipline

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* update splitkv&pagedkv pipeline

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* add sink test

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* update attn_sink result log

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* update smoke_test_fwd_sink.sh

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* update test file

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* update test script

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* Update block_fmha_fwd_splitkv_pipeline_qr_ks_vs.hpp

* use constexpr kHasSink for sink in fmha pipeline

Signed-off-by: Linjun-AMD <Jun.Lin@amd.com>

* update by pre-commit

Signed-off-by: Linjun-AMD <Jun.Lin@amd.com>

* Update include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs.hpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Update include/ck_tile/ops/fmha/pipeline/block_fmha_pipeline_qr_ks_vs.hpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Update include/ck_tile/ops/fmha/kernel/fmha_fwd_pagedkv_kernel.hpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Update fmha_fwd.py

* Update example/ck_tile/01_fmha/codegen/ops/fmha_fwd_splitkv.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Update include/ck_tile/ops/fmha/pipeline/block_fmha_fwd_splitkv_pipeline_nwarp_sshuffle_qr_ks_vs.hpp

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

* Remove causal mask setting logic from mask.hpp

Removed the mask setting logic for causal masks.

* fix ci error that some usage of lamada not support in c++17

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* Update remod.py

* add smoke sink test

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* Update fmha_pagedkv_prefill.py

* Update FmhaFwdPipeline parameters in fmha_fwd.py

* update block_fmha_pipeline_qr_ks_vs_async_trload.hpp

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* fix c++17 unsupprot error

Signed-off-by: LJ-underdog <Jun.Lin@amd.com>

* Update block_fmha_fwd_pagedkv_pipeline_qr_ks_vs.hpp

* Fix formatting of sink_seq_end assignment

* Fix indentation for sink_seq_end assignment

* Update block_fmha_fwd_pagedkv_pipeline_qr_ks_vs.hpp

---------

Signed-off-by: JL-underdog <Jun.Lin@amd.com>
Signed-off-by: LJ-underdog <Jun.Lin@amd.com>
Signed-off-by: Linjun-AMD <Jun.Lin@amd.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This commit is contained in:
Linjun-AMD
2025-11-20 19:24:05 +08:00
committed by GitHub
parent 84540edff3
commit 9fa4e8d5ab
25 changed files with 940 additions and 195 deletions

View File

@@ -0,0 +1,74 @@
#!/bin/bash
# TODO: run this script from CK root or build directory
EXE="$(find . -name tile_example_fmha_fwd -type f | head -n 1)"
KNAME=1
export CK_WARMUP=0
export CK_REPEAT=1
COMMON_ARGS='-v=1 -warmup=0 -repeat=1'
# mode=0
# export HIP_VISIBLE_DEVICES=4
TEST_SPLITKV=0
TEST_APPENDKV=0
# options:
# -s: run splitkv tests
# -a: run appendkv tests
while getopts ":sa" opt; do
case "${opt}" in
s)
TEST_SPLITKV=1
;;
a)
TEST_APPENDKV=1
;;
*)
;;
esac
done
run_fp16_bf16_tests() {
local NUM_SPLITS="1"
local PAGE_BLOCK_SIZE="0"
local CACHE_BATCH_IDX="0"
if [ $TEST_SPLITKV -eq 1 ] ; then
NUM_SPLITS="$NUM_SPLITS 2 3"
PAGE_BLOCK_SIZE="$PAGE_BLOCK_SIZE 128"
CACHE_BATCH_IDX="$CACHE_BATCH_IDX 1"
fi
for prec in "fp16"; do
for mode in 1 0 ; do
for perm in 0 1 ; do
for vlayout in "r" "c" ; do
for batch in 1 4; do
for head in 1; do
for h_k in 1; do
for q_seq in 128 512 ; do
for kv_seq in 128 1024; do
for hdim in 32 64 128 256; do #256
for lse in 0 1 ; do
for bias in "e" ; do
for p_drop in 0.0 0.2; do # 0.0
for mask in "t:2,0,4" "b:1,0,2"; do
for num_splits in $NUM_SPLITS ; do
for page_block_size in $PAGE_BLOCK_SIZE ; do
for cache_batch_idx in $CACHE_BATCH_IDX ; do
# $EXE -prec=$prec -mode=$mode -b=1 -h=1 -d=$hdim -s=1024 -bias=$bias -p_drop=$p_drop -lse=$lse -iperm=$perm -operm=$perm -vlayout=$vlayout -num_splits=$num_splits -page_block_size=$page_block_size -kname=$KNAME $COMMON_ARGS
$EXE -prec=$prec -mode=$mode -b=$batch -h=$head -h_k=$h_k -d=16, -d_v=$hdim -s=$q_seq -s_k=$kv_seq -bias=$bias -p_drop=$p_drop -lse=$lse -iperm=$perm -operm=$perm -vlayout=$vlayout -num_splits=$num_splits -page_block_size=$page_block_size -cache_batch_idx=$cache_batch_idx -kname=$KNAME $COMMON_ARGS -mask=$mask
done ; done ; done ; done ; done
done ; done ; done ; done ; done
done ; done ; done ; done ; done
done ; done
}
set -x
run_fp16_bf16_tests
set +x

View File

@@ -36,6 +36,7 @@ function print_log_header(){
#run verification tests
time example/ck_tile/01_fmha/script/smoke_test_fwd.sh
time example/ck_tile/01_fmha/script/smoke_test_bwd.sh
time example/ck_tile/01_fmha/script/smoke_test_fwd_sink.sh
#run performance benchmarks
export fmha_fwd_log="perf_fmha_fwd_$GPU_arch.log"

View File

@@ -0,0 +1,83 @@
#!/bin/bash
# TODO: run this script from CK root or build directory
#EXE="/code/composable_kernel/build/bin/tile_example_fmha_fwd"
set -euo pipefail
SCRIPT_DIR=$(cd $(dirname "${BASH_SOURCE[0]}") && pwd)
EXE_NAME=tile_example_fmha_fwd
EXE="$(find . -name $EXE_NAME -type f | head -n 1)"
KNAME=1
GPU_arch=$GPU_arch
if [ -z "$GPU_arch" ] ; then
GPU_arch=$(rocminfo | grep -E 'Name:\s+gfx' | head -n1 | awk '{print $2}')
fi
set -x
COMMON_ARGS='-v=1 -warmup=0 -repeat=1'
$EXE -prec=fp16 -mode=0 -b=1 -h=1 -d=128 -d_v=128 -s=512 -s_k=512 -bias=n -lse=0 -iperm=0 -operm=0 -vlayout=r -num_splits=1 -page_block_size=128 -cache_batch_idx=0 -kname=1 -v=1 -warmup=0 -repeat=1 -mask=t:2,0,2
# window_size[2,0], sink_size = 2
# x=1/y=3
# 1 * * * * * * * 1 * * * * * * *
# 1 1 * * * * * * 1 1 * * * * * *
# 1 1 1 * * * * * ----> 1 1 1 * * * * *
# * 1 1 1 * * * * 1 1 1 1 * * * *
# * * 1 1 1 * * * 1 1 1 1 1 * * *
# * * * 1 1 1 * * 1 1 * 1 1 1 * *
# * * * * 1 1 1 * 1 1 * * 1 1 1 *
# * * * * * 1 1 1 1 1 * * * 1 1 1
# l=2/r=0(tl) l=2/r=0/s=2(tl)
$EXE -prec=fp16 -mode=0 -b=1 -h=1 -d=128 -d_v=128 -s=1024 -s_k=1024 -bias=n -lse=0 -iperm=0 -operm=0 -vlayout=r -num_splits=1 -page_block_size=128 -cache_batch_idx=0 -kname=1 -v=1 -warmup=0 -repeat=1 -mask=t:0,3,2 #-mask=b:3,0,2
# x=4/y=1
# 1 1 1 1 * * * * 1 1 1 1 * * * *
# * 1 1 1 1 * * * 1 1 1 1 1 * * *
# * * 1 1 1 1 * * ----> 1 1 1 1 1 1 * *
# * * * 1 1 1 1 * 1 1 * 1 1 1 1 *
# * * * * 1 1 1 1 1 1 * * 1 1 1 1
# l=0/r=3(tl) l=0/r=3/s=2(tl)
# l=3/r=0(br) l=3/r=0/s=2(br)
$EXE -prec=fp16 -mode=0 -b=1 -h=1 -d=128 -d_v=128 -s=4096 -s_k=4096 -bias=n -lse=0 -iperm=0 -operm=0 -vlayout=r -num_splits=1 -page_block_size=128 -cache_batch_idx=0 -kname=1 -v=1 -warmup=0 -repeat=1 -mask=b:1,0,2
# x=4/y=-1
# * * 1 1 * * * * 1 1 1 1 * * * *
# * * * 1 1 * * * 1 1 * 1 1 * * *
# * * * * 1 1 * * ----> 1 1 * * 1 1 * *
# * * * * * 1 1 * 1 1 * * * 1 1 *
# * * * * * * 1 1 1 1 * * * * 1 1
# l=1/r=0(br) l=1/r=0/s=2(br)
$EXE -prec=fp16 -mode=1 -b=1 -h=1 -d=128 -d_v=128 -s=8192 -s_k=8192 -bias=n -lse=0 -iperm=0 -operm=0 -vlayout=r -num_splits=1 -page_block_size=128 -cache_batch_idx=0 -kname=1 -v=1 -warmup=0 -repeat=1 -mask=b:2,0,2
# x=-1/y=5
# * * * * * * * * * * * *
# * * * * * * * * * * * *
# 1 * * * * * 1 * * * * *
# 1 1 * * * * 1 1 * * * *
# 1 1 1 * * * ----> 1 1 1 * * *
# * 1 1 1 * * 1 1 1 1 * *
# * * 1 1 1 * 1 1 1 1 1 *
# * * * 1 1 1 1 1 * 1 1 1
# l=2/r=0(br) l=2/r=0/s=2(br)
$EXE -prec=fp16 -mode=1 -b=1 -h=1 -d=128 -d_v=128 -s=16384 -s_k=16384 -bias=n -lse=0 -iperm=0 -operm=0 -vlayout=r -num_splits=1 -page_block_size=128 -cache_batch_idx=0 -kname=1 -v=1 -warmup=0 -repeat=1 -mask=b:-1,1,2
# x=-1/y=8
# * * * * * * * * * *
# * * * * * * * * * *
# 1 * * * * ----> 1 * * * *
# 1 1 * * * 1 1 * * *
# 1 1 1 * * 1 1 1 * *
# 1 1 1 1 * 1 1 1 1 *
# 1 1 1 1 1 1 1 1 1 1
# 1 1 1 1 1 1 1 1 1 1
# l=2/r=0(br) l=2/r=0/s=2(br)