Skip to content

GitLab

Unverified Commit 2581727d authored by Bartłomiej Kocot's avatar Bartłomiej Kocot Committed by GitHub
Browse files

Add performance and large tensor tests for grouped conv (#1456) 



* Add performance and large tensor tests for grouped conv

* Resize tests

* Resize tests

* update the python script to parse the grouped_conv results

* Remove int8 tests

* change bwd wei layout

---------
Co-authored-by: default avatarillsilin <Illia.Silin@amd.com>
parent 76bd0af6
Hide whitespace changes
Inline Side-by-side
Showing with 307 additions and 92 deletions
Jenkinsfile
View file @ 2581727d
...@@ -426,8 +426,9 @@ def runCKProfiler(Map conf=[:]){ ...@@ -426,8 +426,9 @@ def runCKProfiler(Map conf=[:]){
archiveArtifacts "perf_resnet50_N4.log" archiveArtifacts "perf_resnet50_N4.log"
archiveArtifacts "perf_batched_gemm.log" archiveArtifacts "perf_batched_gemm.log"
archiveArtifacts "perf_grouped_gemm.log" archiveArtifacts "perf_grouped_gemm.log"
archiveArtifacts "perf_conv_fwd.log" archiveArtifacts "perf_grouped_conv_fwd.log"
archiveArtifacts "perf_conv_bwd_data.log" archiveArtifacts "perf_grouped_conv_bwd_data.log"
archiveArtifacts "perf_grouped_conv_bwd_weight.log"
archiveArtifacts "perf_gemm_bilinear.log" archiveArtifacts "perf_gemm_bilinear.log"
archiveArtifacts "perf_reduction.log" archiveArtifacts "perf_reduction.log"
archiveArtifacts "perf_splitK_gemm.log" archiveArtifacts "perf_splitK_gemm.log"
...@@ -439,8 +440,9 @@ def runCKProfiler(Map conf=[:]){ ...@@ -439,8 +440,9 @@ def runCKProfiler(Map conf=[:]){
stash name: "perf_resnet50_N4.log" stash name: "perf_resnet50_N4.log"
stash name: "perf_batched_gemm.log" stash name: "perf_batched_gemm.log"
stash name: "perf_grouped_gemm.log" stash name: "perf_grouped_gemm.log"
stash name: "perf_conv_fwd.log" stash name: "perf_grouped_conv_fwd.log"
stash name: "perf_conv_bwd_data.log" stash name: "perf_grouped_conv_bwd_data.log"
stash name: "perf_grouped_conv_bwd_weight.log"
stash name: "perf_gemm_bilinear.log" stash name: "perf_gemm_bilinear.log"
stash name: "perf_reduction.log" stash name: "perf_reduction.log"
stash name: "perf_splitK_gemm.log" stash name: "perf_splitK_gemm.log"
...@@ -648,8 +650,9 @@ def process_results(Map conf=[:]){ ...@@ -648,8 +650,9 @@ def process_results(Map conf=[:]){
unstash "perf_resnet50_N4.log" unstash "perf_resnet50_N4.log"
unstash "perf_batched_gemm.log" unstash "perf_batched_gemm.log"
unstash "perf_grouped_gemm.log" unstash "perf_grouped_gemm.log"
unstash "perf_conv_fwd.log" unstash "perf_grouped_conv_fwd.log"
unstash "perf_conv_bwd_data.log" unstash "perf_grouped_conv_bwd_data.log"
unstash "perf_grouped_conv_bwd_weight.log"
unstash "perf_gemm_bilinear.log" unstash "perf_gemm_bilinear.log"
unstash "perf_reduction.log" unstash "perf_reduction.log"
unstash "perf_splitK_gemm.log" unstash "perf_splitK_gemm.log"
...@@ -746,6 +749,10 @@ pipeline { ...@@ -746,6 +749,10 @@ pipeline {
name: "RUN_PERFORMANCE_TESTS", name: "RUN_PERFORMANCE_TESTS",
defaultValue: true, defaultValue: true,
description: "Run the performance tests (default: ON)") description: "Run the performance tests (default: ON)")
booleanParam(
name: "RUN_GROUPED_CONV_LARGE_CASES_TESTS",
defaultValue: false,
description: "Run the grouped conv large cases tests (default: OFF)")
booleanParam( booleanParam(
name: "RUN_CK_TILE_TESTS", name: "RUN_CK_TILE_TESTS",
defaultValue: false, defaultValue: false,
...@@ -837,6 +844,30 @@ pipeline { ...@@ -837,6 +844,30 @@ pipeline {
} }
} }
} }
stage("Run Grouped Conv Large Case Tests")
{
parallel
{
stage("Run Grouped Conv Large Case Tests on gfx90a")
{
when {
beforeAgent true
expression { params.RUN_GROUPED_CONV_LARGE_CASES_TESTS.toBoolean() }
}
agent{ label rocmnode("gfx90a")}
environment{
setup_args = "NO_CK_BUILD"
execute_args = """ ../script/cmake-ck-dev.sh ../ gfx90a && \
make -j64 test_grouped_convnd_fwd_large_cases_xdl && \
./bin/test_grouped_convnd_fwd_large_cases_xdl"""
}
steps{
buildHipClangJobAndReboot(setup_args:setup_args, no_reboot:true, build_type: 'Release', execute_cmd: execute_args)
cleanWs()
}
}
}
}
stage("Run CK_TILE Tests") stage("Run CK_TILE Tests")
{ {
parallel parallel
......
script/process_perf_data.py
View file @ 2581727d
...@@ -122,7 +122,7 @@ def parse_logfile(logfile): ...@@ -122,7 +122,7 @@ def parse_logfile(logfile):
#sorted_kernels = [x for _,x in sorted(zip(tests,kernels))] #sorted_kernels = [x for _,x in sorted(zip(tests,kernels))]
test_list=list(range(1,len(tests)+1)) test_list=list(range(1,len(tests)+1))
#parse conv_fwd and conv_bwd performance tests: #parse conv_fwd and conv_bwd performance tests:
elif 'conv_fwd' in logfile or 'conv_bwd_data' in logfile: elif 'conv_fwd' in logfile or 'conv_bwd' in logfile:
for line in open(logfile): for line in open(logfile):
if 'tflops:' in line: if 'tflops:' in line:
lst=line.split() lst=line.split()
...@@ -274,14 +274,26 @@ def main(): ...@@ -274,14 +274,26 @@ def main():
for i in range(1,len(results)+1): for i in range(1,len(results)+1):
testlist.append("Test%i"%i) testlist.append("Test%i"%i)
table_name="ck_grouped_gemm_tflops" table_name="ck_grouped_gemm_tflops"
if 'conv_fwd' in filename: if 'perf_conv_fwd' in filename:
for i in range(1,len(results)+1): for i in range(1,len(results)+1):
testlist.append("Test%i"%i) testlist.append("Test%i"%i)
table_name="ck_conv_fwd_tflops" table_name="ck_conv_fwd_tflops"
if 'conv_bwd_data' in filename: if 'perf_conv_bwd_data' in filename:
for i in range(1,len(results)+1): for i in range(1,len(results)+1):
testlist.append("Test%i"%i) testlist.append("Test%i"%i)
table_name="ck_conv_bwd_data_tflops" table_name="ck_conv_bwd_data_tflops"
if 'grouped_conv_fwd' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_conv_fwd_tflops"
if 'grouped_conv_bwd_data' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_conv_bwd_data_tflops"
if 'grouped_conv_bwd_weight' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_conv_bwd_weight_tflops"
if 'gemm_bilinear' in filename: if 'gemm_bilinear' in filename:
for i in range(1,len(results)+1): for i in range(1,len(results)+1):
testlist.append("Test%i"%i) testlist.append("Test%i"%i)
......
script/process_qa_data.sh
View file @ 2581727d
...@@ -15,8 +15,9 @@ python3 process_perf_data.py perf_resnet50_N256.log ...@@ -15,8 +15,9 @@ python3 process_perf_data.py perf_resnet50_N256.log
python3 process_perf_data.py perf_resnet50_N4.log python3 process_perf_data.py perf_resnet50_N4.log
python3 process_perf_data.py perf_batched_gemm.log python3 process_perf_data.py perf_batched_gemm.log
python3 process_perf_data.py perf_grouped_gemm.log python3 process_perf_data.py perf_grouped_gemm.log
python3 process_perf_data.py perf_conv_fwd.log python3 process_perf_data.py perf_grouped_conv_fwd.log
python3 process_perf_data.py perf_conv_bwd_data.log python3 process_perf_data.py perf_grouped_conv_bwd_data.log
python3 process_perf_data.py perf_grouped_conv_bwd_weight.log
python3 process_perf_data.py perf_gemm_bilinear.log python3 process_perf_data.py perf_gemm_bilinear.log
python3 process_perf_data.py perf_reduction.log python3 process_perf_data.py perf_reduction.log
python3 process_perf_data.py perf_splitK_gemm.log python3 process_perf_data.py perf_splitK_gemm.log
......
script/profile_conv_fwd.sh → script/profile_grouped_conv_bwd_weight.sh
View file @ 2581727d
...@@ -12,27 +12,28 @@ INIT=$5 ...@@ -12,27 +12,28 @@ INIT=$5
LOG=$6 LOG=$6
TIME=$7 TIME=$7
N=$8 N=$8
SplitK=$9
# Resnet50 # Resnet50
######## op datatype layout verify init log time conv_dim G__ N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads ######## op datatype layout verify init log time conv_dim G__ N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1 $SplitK
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3 $DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3 $SplitK
script/profile_grouped_conv_fwd.sh 0 → 100755
View file @ 2581727d
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
DRIVER="../build/bin/ckProfiler"
OP=$1
DATATYPE=$2
LAYOUT=$3
INDEXTYPE=$4
VERIFY=$5
INIT=$6
LOG=$7
TIME=$8
N=$9
# Resnet50
######## op datatype indextype layout verify init log time conv_dim G__ N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $INDEXTYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2 1 $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3
script/run_full_performance_tests.sh
View file @ 2581727d
...@@ -90,21 +90,27 @@ print_log_header $gemm_bilinear_log $env_type $branch $host_name ...@@ -90,21 +90,27 @@ print_log_header $gemm_bilinear_log $env_type $branch $host_name
./profile_gemm_bilinear.sh gemm_bilinear 1 2 $verify 1 0 1 2>&1 | tee -a $gemm_bilinear_log ./profile_gemm_bilinear.sh gemm_bilinear 1 2 $verify 1 0 1 2>&1 | tee -a $gemm_bilinear_log
./profile_gemm_bilinear.sh gemm_bilinear 1 3 $verify 1 0 1 2>&1 | tee -a $gemm_bilinear_log ./profile_gemm_bilinear.sh gemm_bilinear 1 3 $verify 1 0 1 2>&1 | tee -a $gemm_bilinear_log
#run conv_fwd tests #run grouped_fwd tests
export conv_fwd_log="perf_conv_fwd.log" export grouped_conv_fwd_log="perf_grouped_conv_fwd.log"
print_log_header $conv_fwd_log $env_type $branch $host_name print_log_header $grouped_conv_fwd_log $env_type $branch $host_name
./profile_conv_fwd.sh conv_fwd 0 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log ./profile_grouped_conv_fwd.sh grouped_conv_fwd 0 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
./profile_conv_fwd.sh conv_fwd 1 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log ./profile_grouped_conv_fwd.sh grouped_conv_fwd 1 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
./profile_conv_fwd.sh conv_fwd 2 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log ./profile_grouped_conv_fwd.sh grouped_conv_fwd 2 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
./profile_conv_fwd.sh conv_fwd 3 1 $verify 1 0 1 256 2>&1 | tee -a $conv_fwd_log
#run conv_bwd_data tests #run grouped_bwd_data tests
export conv_bwd_data_log="perf_conv_bwd_data.log" export grouped_conv_bwd_data_log="perf_grouped_conv_bwd_data.log"
print_log_header $conv_bwd_data_log $env_type $branch $host_name print_log_header $grouped_conv_bwd_data_log $env_type $branch $host_name
./profile_conv_bwd_data.sh conv_bwd_data 0 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log ./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 0 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 1 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log ./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 1 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 2 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log ./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 2 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 3 1 $verify 1 0 1 256 2>&1 | tee -a $conv_bwd_data_log
#run grouped_bwd_weight tests
export grouped_conv_bwd_weight_log="perf_grouped_conv_bwd_weight.log"
print_log_header $grouped_conv_bwd_weight_log $env_type $branch $host_name
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 0 2 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 1 2 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 2 2 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 1 2 $verify 1 0 1 256 4 2>&1 | tee -a $grouped_conv_bwd_weight_log
#run resnet50 tests #run resnet50 tests
export resnet256_log="perf_resnet50_N256.log" export resnet256_log="perf_resnet50_N256.log"
......
script/run_performance_tests.sh
View file @ 2581727d
...@@ -51,6 +51,21 @@ print_log_header $gemm_log $env_type $branch $host_name ...@@ -51,6 +51,21 @@ print_log_header $gemm_log $env_type $branch $host_name
./profile_gemm.sh gemm 2 3 $verify 1 0 1 | tee -a $gemm_log ./profile_gemm.sh gemm 2 3 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 3 $verify 1 0 1 | tee -a $gemm_log ./profile_gemm.sh gemm 3 3 $verify 1 0 1 | tee -a $gemm_log
#run grouped_fwd fp16 tests
export grouped_conv_fwd_log="perf_grouped_conv_fwd_fp16.log"
print_log_header $conv_fwd_log $env_type $branch $host_name
./profile_grouped_conv_fwd.sh grouped_conv_fwd 1 1 0 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_fwd_log
#run grouped_bwd_data fp16 tests
export grouped_conv_bwd_data_log="perf_grouped_conv_bwd_data_fp16.log"
print_log_header $grouped_conv_bwd_data_log $env_type $branch $host_name
./profile_grouped_conv_bwd_data.sh grouped_conv_bwd_data 1 1 $verify 1 0 1 256 2>&1 | tee -a $grouped_conv_bwd_data_log
#run grouped_bwd_weight fp16 tests
export grouped_conv_bwd_weight_log="perf_grouped_conv_bwd_weight_fp16.log"
print_log_header $grouped_conv_bwd_weight_log $env_type $branch $host_name
./profile_grouped_conv_bwd_weight.sh grouped_conv_bwd_weight 1 1 $verify 1 0 1 256 1 2>&1 | tee -a $grouped_conv_bwd_weight_log
#run resnet50 tests #run resnet50 tests
export resnet256_log="perf_resnet50_N256.log" export resnet256_log="perf_resnet50_N256.log"
print_log_header $resnet256_log $env_type $branch $host_name print_log_header $resnet256_log $env_type $branch $host_name
......
test/grouped_convnd_fwd/CMakeLists.txt
View file @ 2581727d
...@@ -7,6 +7,12 @@ if(GPU_TARGETS MATCHES "gfx9" OR GPU_TARGETS MATCHES "gfx11") ...@@ -7,6 +7,12 @@ if(GPU_TARGETS MATCHES "gfx9" OR GPU_TARGETS MATCHES "gfx11")
endif() endif()
endif() endif()
if(GPU_TARGETS MATCHES "gfx9")
add_executable(test_grouped_convnd_fwd_large_cases_xdl test_grouped_convnd_fwd_large_cases_xdl.cpp)
target_compile_options(test_grouped_convnd_fwd_large_cases_xdl PRIVATE -Wno-global-constructors -Wno-undef)
target_link_libraries(test_grouped_convnd_fwd_large_cases_xdl PRIVATE gtest_main getopt::getopt utility device_grouped_conv1d_fwd_instance device_grouped_conv2d_fwd_instance device_grouped_conv3d_fwd_instance)
endif()
add_gtest_executable(test_grouped_convnd_fwd_multi_ab_interface test_grouped_convnd_fwd_multi_ab_interface.cpp) add_gtest_executable(test_grouped_convnd_fwd_multi_ab_interface test_grouped_convnd_fwd_multi_ab_interface.cpp)
if(result EQUAL 0) if(result EQUAL 0)
target_link_libraries(test_grouped_convnd_fwd_multi_ab_interface PRIVATE utility) target_link_libraries(test_grouped_convnd_fwd_multi_ab_interface PRIVATE utility)
......
test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp
View file @ 2581727d
...@@ -17,7 +17,7 @@ class TestGroupedConvndFwd : public ::testing::Test ...@@ -17,7 +17,7 @@ class TestGroupedConvndFwd : public ::testing::Test
using InLayout = std::tuple_element_t<1, Tuple>; using InLayout = std::tuple_element_t<1, Tuple>;
using WeiLayout = std::tuple_element_t<2, Tuple>; using WeiLayout = std::tuple_element_t<2, Tuple>;
using OutLayout = std::tuple_element_t<3, Tuple>; using OutLayout = std::tuple_element_t<3, Tuple>;
using IndexType = std::tuple_element_t<4, Tuple>; using IndexType = ck::index_t;
std::vector<ck::utils::conv::ConvParam> conv_params; std::vector<ck::utils::conv::ConvParam> conv_params;
...@@ -50,31 +50,28 @@ class TestGroupedConvndFwd : public ::testing::Test ...@@ -50,31 +50,28 @@ class TestGroupedConvndFwd : public ::testing::Test
using namespace ck::tensor_layout::convolution; using namespace ck::tensor_layout::convolution;
using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK, ck::index_t>, using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK>,
std::tuple<ck::half_t, GNWC, GKXC, GNWK, ck::index_t>, std::tuple<ck::half_t, GNWC, GKXC, GNWK>,
std::tuple<ck::bhalf_t, GNWC, GKXC, GNWK, ck::index_t>, std::tuple<ck::bhalf_t, GNWC, GKXC, GNWK>,
std::tuple<int8_t, GNWC, GKXC, GNWK, ck::index_t>>; std::tuple<int8_t, GNWC, GKXC, GNWK>>;
using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK, ck::index_t>, using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK, ck::index_t>, std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK>,
std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK, ck::index_t>, std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK>,
std::tuple<int8_t, GNHWC, GKYXC, GNHWK, ck::index_t>, std::tuple<int8_t, GNHWC, GKYXC, GNHWK>,
std::tuple<float, NHWGC, GKYXC, NHWGK, ck::index_t>, std::tuple<float, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK, ck::index_t>, std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK, ck::index_t>, std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>,
std::tuple<int8_t, NHWGC, GKYXC, NHWGK, ck::index_t>>; std::tuple<int8_t, NHWGC, GKYXC, NHWGK>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK, ck::index_t>, using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::index_t>, std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::index_t>, std::tuple<ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<int8_t, GNDHWC, GKZYXC, GNDHWK, ck::index_t>, std::tuple<int8_t, GNDHWC, GKZYXC, GNDHWK>,
std::tuple<float, NDHWGC, GKZYXC, NDHWGK, ck::index_t>, std::tuple<float, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK, ck::index_t>, std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK, ck::index_t>, std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<int8_t, NDHWGC, GKZYXC, NDHWGK, ck::index_t>>; std::tuple<int8_t, NDHWGC, GKZYXC, NDHWGK>>;
using KernelTypes2dLargeCases =
::testing::Types<std::tuple<float, NHWGC, GKYXC, NHWGK, ck::long_index_t>>;
template <typename Tuple> template <typename Tuple>
class TestGroupedConvndFwd1d : public TestGroupedConvndFwd<Tuple> class TestGroupedConvndFwd1d : public TestGroupedConvndFwd<Tuple>
...@@ -91,15 +88,9 @@ class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple> ...@@ -91,15 +88,9 @@ class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple>
{ {
}; };
template <typename Tuple>
class TestGroupedConvndFwd2dLargeCases : public TestGroupedConvndFwd<Tuple>
{
};
TYPED_TEST_SUITE(TestGroupedConvndFwd1d, KernelTypes1d); TYPED_TEST_SUITE(TestGroupedConvndFwd1d, KernelTypes1d);
TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d); TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d);
TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d); TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d);
TYPED_TEST_SUITE(TestGroupedConvndFwd2dLargeCases, KernelTypes2dLargeCases);
TYPED_TEST(TestGroupedConvndFwd1d, Test1D) TYPED_TEST(TestGroupedConvndFwd1d, Test1D)
{ {
...@@ -149,17 +140,3 @@ TYPED_TEST(TestGroupedConvndFwd3d, Test3D) ...@@ -149,17 +140,3 @@ TYPED_TEST(TestGroupedConvndFwd3d, Test3D)
{3, 96, 1, 1, 1, {3, 3, 3}, {4, 30, 160}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}); {3, 96, 1, 1, 1, {3, 3, 3}, {4, 30, 160}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
this->template Run<3>(); this->template Run<3>();
} }
TYPED_TEST(TestGroupedConvndFwd2dLargeCases, Test2DLargeCases)
{
// Case larger than 2GB
this->conv_params.push_back(
{2, 1, 64, 4, 192, {2, 2}, {224, 224}, {224, 224}, {1, 1}, {0, 0}, {0, 0}});
// With supported NumGroupsToMerge > 1
this->conv_params.push_back(
{2, 32, 64, 1, 1, {2, 2}, {672, 672}, {672, 672}, {1, 1}, {0, 0}, {0, 0}});
// When image is larger than 2GB
this->conv_params.push_back(
{2, 1, 1, 256, 256, {3, 3}, {4096, 2048}, {1024, 1024}, {3, 3}, {1, 1}, {1, 1}});
this->template Run<2>();
}
test/grouped_convnd_fwd/test_grouped_convnd_fwd_large_cases_xdl.cpp 0 → 100644
View file @ 2581727d
// SPDX-License-Identifier: MIT
// Copyright (c) 2024, Advanced Micro Devices, Inc. All rights reserved.
#include <cstdlib>
#include <iostream>
#include <initializer_list>
#include <vector>
#include <gtest/gtest.h>
#include "profiler/profile_grouped_conv_fwd_impl.hpp"
template <typename Tuple>
class TestGroupedConvndFwd : public ::testing::Test
{
protected:
using DataType = std::tuple_element_t<0, Tuple>;
using InLayout = std::tuple_element_t<1, Tuple>;
using WeiLayout = std::tuple_element_t<2, Tuple>;
using OutLayout = std::tuple_element_t<3, Tuple>;
using IndexType = ck::long_index_t;
std::vector<ck::utils::conv::ConvParam> conv_params;
template <ck::index_t NDimSpatial>
void Run()
{
EXPECT_FALSE(conv_params.empty());
bool pass = true;
for(auto& param : conv_params)
{
pass = pass && ck::profiler::profile_grouped_conv_fwd_impl<NDimSpatial,
InLayout,
WeiLayout,
OutLayout,
DataType,
DataType,
DataType,
DataType,
DataType,
IndexType>(
true, // do_verification
1, // init_method: integer value
false, // do_log
false, // time_kernel
param);
}
EXPECT_TRUE(pass);
}
};
using namespace ck::tensor_layout::convolution;
using KernelTypes2d = ::testing::Types<std::tuple<float, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>>;
using KernelTypes3d = ::testing::Types<std::tuple<float, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK>,
std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK>>;
template <typename Tuple>
class TestGroupedConvndFwd2d : public TestGroupedConvndFwd<Tuple>
{
};
template <typename Tuple>
class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple>
{
};
TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d);
TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d);
TYPED_TEST(TestGroupedConvndFwd2d, Test2D)
{
// Case larger than 2GB
this->conv_params.push_back(
{2, 1, 128, 4, 192, {2, 2}, {224, 224}, {224, 224}, {1, 1}, {0, 0}, {0, 0}});
// With supported NumGroupsToMerge > 1
this->conv_params.push_back(
{2, 32, 64, 1, 1, {2, 2}, {672, 672}, {672, 672}, {1, 1}, {0, 0}, {0, 0}});
// When image is larger than 2GB
this->conv_params.push_back(
{2, 2, 2, 128, 128, {3, 3}, {4096, 2048}, {300, 300}, {3, 3}, {1, 1}, {1, 1}});
this->template Run<2>();
}
TYPED_TEST(TestGroupedConvndFwd3d, Test3D)
{
// Case larger than 2GB
this->conv_params.push_back({3,
1,
128,
4,
192,
{2, 2, 2},
{2, 224, 224},
{1, 224, 224},
{1, 1, 1},
{0, 0, 0},
{0, 0, 0}});
// With supported NumGroupsToMerge > 1
this->conv_params.push_back({3,
32,
64,
1,
1,
{2, 2, 2},
{360, 2, 672},
{360, 2, 672},
{1, 1, 1},
{0, 0, 0},
{0, 0, 0}});
// When image is larger than 2GB
this->conv_params.push_back({3,
1,
2,
128,
128,
{3, 1, 3},
{900, 2, 2048},
{300, 1, 300},
{3, 2, 3},
{1, 1, 1},
{1, 1, 1}});
this->template Run<3>();
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment