From 09ac14604c74827c6fefca067d616a6b4c88428c Mon Sep 17 00:00:00 2001 From: Qianfeng Zhang Date: Wed, 18 Jun 2025 15:31:47 +0000 Subject: [PATCH] Align the -seqlens=xxx in the mattn0_full0 and mattn256_full256 scripts with the required benchmarks --- .../bench_jagged_causal_mattn0_full0.sh | 36 +++++++++---------- .../bench_jagged_causal_mattn256_full256.sh | 36 +++++++++---------- 2 files changed, 36 insertions(+), 36 deletions(-) diff --git a/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn0_full0.sh b/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn0_full0.sh index 90a1c8bcff..51972213cd 100644 --- a/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn0_full0.sh +++ b/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn0_full0.sh @@ -11,79 +11,79 @@ set -x target8="10,10,14,17,16,12,14,9" ## seqlen 1024 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1024 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1004 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 2048 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2048 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2028 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 3072 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3072 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3052 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 4096 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4096 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4076 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 8192 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8192 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8172 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 16384 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16384 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16364 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" target16="13,17,16,13,7,14,3,18,15,15,1,9,18,18,7,10" ## seqlen 1024 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1024 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1004 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 2048 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2048 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2028 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 3072 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3072 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3052 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 4096 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4096 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4076 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 8192 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8192 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8172 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 16384 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16384 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16364 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" target32="13,17,16,13,7,14,3,18,15,15,1,9,18,18,7,10,11,0,4,8,2,10,20,14,11,7,4,6,9,7,14,17" ## seqlen 1024 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1024 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1004 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 2048 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2048 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2028 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 3072 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3072 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3052 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 4096 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4096 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4076 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 8192 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8192 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8172 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 16384 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16384 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16364 -causal=1 -local_len=0 -context_len=0 -minfull_len=0 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" set +x diff --git a/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn256_full256.sh b/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn256_full256.sh index 4099faf7dd..92ec23deab 100644 --- a/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn256_full256.sh +++ b/example/ck_tile/18_hstu_attention/scripts/bench_jagged_causal_mattn256_full256.sh @@ -11,79 +11,79 @@ set -x target8="10,10,14,17,16,12,14,9" ## seqlen 1024 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1024 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1004 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 2048 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2048 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2028 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 3072 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3072 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3052 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 4096 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4096 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4076 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 8192 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8192 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8172 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 16384 -$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16384 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=8 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16364 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target8 -max_target=20 -perf=1 -alpha=2.0 echo -e "" target16="13,17,16,13,7,14,3,18,15,15,1,9,18,18,7,10" ## seqlen 1024 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1024 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1004 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 2048 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2048 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2028 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 3072 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3072 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3052 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 4096 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4096 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4076 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 8192 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8192 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8172 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 16384 -$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16384 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=16 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16364 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target16 -max_target=20 -perf=1 -alpha=2.0 echo -e "" target32="13,17,16,13,7,14,3,18,15,15,1,9,18,18,7,10,11,0,4,8,2,10,20,14,11,7,4,6,9,7,14,17" ## seqlen 1024 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1024 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=1004 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 2048 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2048 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=2028 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 3072 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3072 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=3052 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 4096 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4096 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=4076 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 8192 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8192 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=8172 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" ## seqlen 16384 -$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16384 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 +$EXE -v=0 -prec=$dtype -b=32 -jagged=1 -nhead=4 -hdim_qk=128 -hdim_v=128 -seqlens=16364 -causal=1 -local_len=256 -context_len=0 -minfull_len=256 -targets=$target32 -max_target=20 -perf=1 -alpha=2.0 echo -e "" set +x