ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-25 15:24:39 +00:00
Code Issues Packages Projects Releases Wiki Activity

sync with upstream

Browse Source
This commit is contained in:
carlushuang
2024-03-26 16:05:54 +00:00
parent 04ee01191a
commit 1c92c5d83d
268 changed files with 16113 additions and 2241 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
script/process_perf_data.py
View File

@@ -133,11 +133,16 @@ def parse_logfile(logfile):
if 'Best Perf' in line:
lst=line.split()
res.append(lst[4])
elif 'onnx_gemm' in logfile or 'splitK_gemm' in logfile or 'mixed_gemm' in logfile:
elif 'onnx_gemm' in logfile or 'mixed_gemm' in logfile:
for line in open(logfile):
if 'Best Perf' in line:
lst=line.split()
res.append(lst[33])
elif 'splitK_gemm' in logfile:
for line in open(logfile):
if 'Best Perf' in line:
lst=line.split()
res.append(lst[36])
return res
@@ -231,7 +236,7 @@ def main():
sql_hostname = '127.0.0.1'
sql_username = os.environ["dbuser"]
sql_password = os.environ["dbpassword"]
sql_main_database = 'miopen_perf'
sql_main_database = os.environ["ck_perf_db"]
sql_port = 3306
ssh_host = os.environ["dbsship"]
ssh_user = os.environ["dbsshuser"]

43
script/profile_permute_scale.sh Executable file
View File

@@ -0,0 +1,43 @@
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
DRIVER="../build/bin/ckProfiler"
echo $DRIVER
OP=$1
DATATYPE=$2
VERIFY=$3
INIT=$4
LOG=$5
TIME=$6
# 1D
######## op datatype verify init log time dims in_strides_order out_strides_order
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 67108864 0 0
# # 2D
# ######## op datatype verify init log time dims in_strides_order out_strides_order
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8192 8192 0 1 1 0
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8192 8192 1 0 0 1
# 3D
######## op datatype verify init log time dims in_strides_order out_strides_order
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 1024 8192 0 1 2 2 1 0
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 1024 8192 2 1 0 0 1 2
# 4D
######## op datatype verify init log time dims in_strides_order out_strides_order
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 2 512 8192 0 1 2 3 3 2 1 0
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 2 512 8192 3 2 1 0 0 1 2 3
# 5D
######## op datatype verify init log time dims in_strides_order out_strides_order
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 2 2 256 8192 0 1 2 3 4 4 3 2 1 0
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 2 2 256 8192 4 3 2 1 0 0 1 2 3 4
# 6D
######## op datatype verify init log time dims in_strides_order out_strides_order
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 2 2 2 128 8192 0 1 2 3 4 5 5 4 3 2 1 0
$DRIVER $OP $DATATYPE $VERIFY $INIT $LOG $TIME 8 2 2 2 128 8192 5 4 3 2 1 0 0 1 2 3 4 5

26
script/run_full_performance_tests.sh
View File

@@ -121,26 +121,16 @@ print_log_header $reduction_log $env_type $branch $host_name
./profile_reduce_no_index.sh $verify 2 10 --half 2>&1 | tee -a $reduction_log
#run splitK_gemm tests, first correctness verification, then performance
export splitK_gemm_ver_log="perf_splitK_gemm_verify.log"
print_log_header $splitK_gemm_ver_log $env_type $branch $host_name
./profile_splitK_gemm.sh gemm_splitk 0 0 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 0 1 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 0 2 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 0 3 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 1 0 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 1 1 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 1 2 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
./profile_splitK_gemm.sh gemm_splitk 1 3 $verify 1 0 0 4 2>&1 | tee -a $splitK_gemm_ver_log
export splitK_gemm_log="perf_splitK_gemm.log"
print_log_header $splitK_gemm_log $env_type $branch $host_name
./profile_splitK_gemm.sh gemm_splitk 0 0 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 1 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 2 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 3 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 0 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 1 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 2 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 3 0 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 0 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 1 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 2 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 0 3 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 0 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 1 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 2 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
./profile_splitK_gemm.sh gemm_splitk 1 3 $verify 1 0 1 4 2>&1 | tee -a $splitK_gemm_log
#run ONNX gemm tests
export onnx_log="perf_onnx_gemm.log"