Skip to content

GitLab

Commit 67bd84ff authored by Chao Liu's avatar Chao Liu
Browse files

update test script

parent a1841d55
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 298 additions and 584 deletions
script/conv2d_fwd.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
make -j $1
DRIVER=example/$1
VERIFY=$2
INIT=$3
REPEAT=$4
# test
######## verify init repeat N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads Desired_grid_size__
$DRIVER $VERIFY $INIT $REPEAT 128 256 192 3 3 71 71 2 2 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT 128 256 64 1 1 1 1 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT 256 64 3 7 7 230 230 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT 128 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT 256 64 3 7 7 224 224 2 2 1 1 3 3 3 3
N=$5
# Resnet50
######## verify init repeat N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads Desired_grid_size__
#$DRIVER $VERIFY $INIT $REPEAT $N 2048 1024 1 1 14 14 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 128 128 3 3 58 58 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 256 256 3 3 30 30 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 512 256 1 1 56 56 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 512 512 3 3 16 16 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 1024 512 1 1 28 28 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $VERIFY $INIT $REPEAT $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
script/conv_driver.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
make -j conv_fwd_driver_offline
#make -j conv_bwd_driver_offline
#make -j conv_wrw_driver_offline
DRIVER="./host/driver_offline/conv_fwd_driver_offline"
#DRIVER="./host/driver_offline/conv_bwd_driver_offline"
#DRIVER="./host/driver_offline/conv_wrw_driver_offline"
LAYOUT=$1
ALGO=$2
VERIFY=$3
INIT=$4
LOG=$5
REPEAT=$6
DESIRED_GRID_SIZE=$7
######### layout algo verify init log repeat N__ K___ C___ Y X Hi_ Wi__ Strides Dilations LeftPads RightPads Desired_grid_size__
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 128 192 3 3 71 71 2 2 1 1 1 1 1 1 $DESIRED_GRID_SIZE
$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 256 192 3 3 71 71 2 2 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 256 1024 1 7 17 17 1 1 1 1 0 3 0 3 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 128 128 3 3 14 14 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
$DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 512 192 3 3 35 35 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 256 3 3 30 30 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 512 3 3 16 16 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
$DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 2048 1024 1 1 14 14 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 2048 1 1 7 7 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
$DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 256 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
$DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 32 256 3 3 1 1 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 32 256 1 1 1 1 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
$DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 256 64 1 1 2 2 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 128 256 128 1 1 2 2 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
# Resnet50
######### layout algo verify init log repeat N__ K___ C___ Y X Hi_ Wi__ Strides Dilations LeftPads RightPads Desired_grid_size__
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 2048 1024 1 1 14 14 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 1024 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 128 128 3 3 28 28 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 128 1 1 28 28 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 128 128 3 3 58 58 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 2048 1 1 7 7 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 1024 256 1 1 14 14 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 256 3 3 14 14 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 256 3 3 30 30 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 128 256 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 256 1 1 56 56 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 64 256 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 512 3 3 16 16 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 1024 512 1 1 28 28 2 2 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 128 512 1 1 28 28 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 512 1 1 28 28 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 2048 512 1 1 7 7 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 512 512 3 3 7 7 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 256 64 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 64 64 1 1 56 56 1 1 1 1 0 0 0 0 $DESIRED_GRID_SIZE
##DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 256 64 64 3 3 56 56 1 1 1 1 1 1 1 1 $DESIRED_GRID_SIZE
script/example_gemm_xdl.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=1
make -j gemm_xdl
DRIVER="./example/gemm_xdl"
VERIFY=$1
INIT=$2
LOG=$3
REPEAT=$4
######### verify init log repeat M___ N___ K___ StrideA StrideB StrideC
#$DRIVER $VERIFY $INIT $LOG $REPEAT 960 1024 1024 1024 1024 1024
#$DRIVER $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1024 1024 1024
#$DRIVER $VERIFY $INIT $LOG $REPEAT 1920 2048 2048 2048 2048 2048
$DRIVER $VERIFY $INIT $LOG $REPEAT 3840 4096 4096 4096 4096 4096
#$DRIVER $VERIFY $INIT $LOG $REPEAT 7680 8192 8192 8192 8192 8192
script/gemm.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
make -j $1
DRIVER=example/$1
VERIFY=$2
INIT=$3
REPEAT=$4
######## verify init repeat M___ N___ K___ StrideA StrideB StrideC StrideC1
#$DRIVER $VERIFY $INIT $REPEAT 256 256 256 256 256 256 256
#$DRIVER $VERIFY $INIT $REPEAT 960 1024 1024 1024 1024 1024 1024
#$DRIVER $VERIFY $INIT $REPEAT 1920 2048 2048 2048 2048 2048 2048
$DRIVER $VERIFY $INIT $REPEAT 3840 4096 4096 4096 4096 4096 4096
#$DRIVER $VERIFY $INIT $REPEAT 7680 8192 8192 8192 8192 8192 8192
#$DRIVER $VERIFY $INIT $REPEAT 1024 1024 1024 1024 1024 1024 1024
#$DRIVER $VERIFY $INIT $REPEAT 2048 2048 2048 2048 2048 2048 2048
script/gemm_driver.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
make -j gemm_driver_offline
DRIVER="./host/driver_offline/gemm_driver_offline"
LAYOUT=$1
ALGO=$2
VERIFY=$3
INIT=$4
LOG=$5
REPEAT=$6
M01=$7
N01=$8
######### layout algo verify init log repeat M___ N___ K___ M01_ N01_
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 960 1024 1024 $M01 $N01
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 $M01 $N01
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 1920 2048 2048 $M01 $N01
$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 3840 4096 4096 $M01 $N01
#$DRIVER $LAYOUT $ALGO $VERIFY $INIT $LOG $REPEAT 7680 8192 8192 $M01 $N01
script/pool2d_fwd.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
make -j $1
DRIVER=example/$1
VERIFY=$2
INIT=$3
REPEAT=$4
# test
######## verify init repeat N__ C___ Y X Hi__ Wi__ Strides LeftPads RightPads
#$DRIVER $VERIFY $INIT $REPEAT 128 192 3 3 71 71 2 2 1 1 1 1
#$DRIVER $VERIFY $INIT $REPEAT 128 64 1 1 1 1 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT 256 3 7 7 230 230 2 2 0 0 0 0
$DRIVER $VERIFY $INIT $REPEAT 256 1024 14 14 14 14 1 1 0 0 0 0
N=$5
# Resnet50
######## verify init repeat N__ C___ Y X Hi__ Wi__ Strides LeftPads RightPads
#$DRIVER $VERIFY $INIT $REPEAT $N 1024 1 1 14 14 2 2 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 1024 1 1 14 14 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 1024 1 1 14 14 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 128 3 3 28 28 1 1 1 1 1 1
#$DRIVER $VERIFY $INIT $REPEAT $N 128 1 1 28 28 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 128 3 3 58 58 2 2 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 2048 1 1 7 7 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 256 1 1 14 14 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 256 3 3 14 14 1 1 1 1 1 1
#$DRIVER $VERIFY $INIT $REPEAT $N 256 3 3 30 30 2 2 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 256 1 1 56 56 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 256 1 1 56 56 2 2 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 256 1 1 56 56 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 512 3 3 16 16 2 2 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 512 1 1 28 28 2 2 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 512 1 1 28 28 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 512 1 1 28 28 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 512 1 1 7 7 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 512 3 3 7 7 1 1 1 1 1 1
#$DRIVER $VERIFY $INIT $REPEAT $N 64 1 1 56 56 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 64 1 1 56 56 1 1 0 0 0 0
#$DRIVER $VERIFY $INIT $REPEAT $N 64 3 3 56 56 1 1 1 1 1 1
#$DRIVER $VERIFY $INIT $REPEAT $N 3 7 7 230 230 2 2 0 0 0 0
script/process_perf_data.py
View file @ 67bd84ff
......@@ -120,14 +120,14 @@ def parse_logfile(logfile):
res = [x for _,x in sorted(zip(tests,tflops))]
#sorted_kernels = [x for _,x in sorted(zip(tests,kernels))]
test_list=list(range(1,len(tests)+1))
#parse fwd_conv performance tests:
elif 'fwd_conv' in logfile:
#parse conv_fwd performance tests:
elif 'conv_fwd' in logfile:
for line in open(logfile):
if 'tflops:' in line:
lst=line.split()
res.append(lst[1])
#parse all other performance tests:
elif 'resnet50' or 'batched_gemm' or 'grouped_gemm' or 'bwd_conv' or 'fusion' or 'reduction' in logfile:
elif 'resnet50' or 'batched_gemm' or 'grouped_gemm' or 'conv_bwd_data' or 'gemm_bilinear' or 'reduction' in logfile:
for line in open(logfile):
if 'Best Perf' in line:
lst=line.split()
......@@ -257,18 +257,18 @@ def main():
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_grouped_gemm_tflops"
if 'fwd_conv' in filename:
if 'conv_fwd' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_fwd_conv_tflops"
if 'bwd_conv' in filename:
table_name="ck_conv_fwd_tflops"
if 'conv_bwd_data' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_bwd_conv_tflops"
if 'fusion' in filename:
table_name="ck_conv_bwd_data_tflops"
if 'gemm_bilinear' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
table_name="ck_fusion_tflops"
table_name="ck_gemm_bilinear_tflops"
if 'reduction' in filename:
for i in range(1,len(results)+1):
testlist.append("Test%i"%i)
......
script/process_qa_data.sh
View file @ 67bd84ff
......@@ -16,7 +16,7 @@ python3 process_perf_data.py perf_resnet50_N265_"$gpu_arch".log
python3 process_perf_data.py perf_resnet50_N4_"$gpu_arch".log
python3 process_perf_data.py perf_batched_gemm_"$gpu_arch".log
python3 process_perf_data.py perf_grouped_gemm_"$gpu_arch".log
python3 process_perf_data.py perf_fwd_conv_"$gpu_arch".log
python3 process_perf_data.py perf_bwd_conv_"$gpu_arch".log
python3 process_perf_data.py perf_fusion_"$gpu_arch".log
python3 process_perf_data.py perf_conv_fwd_"$gpu_arch".log
python3 process_perf_data.py perf_conv_bwd_data_"$gpu_arch".log
python3 process_perf_data.py perf_gemm_bilinear_"$gpu_arch".log
python3 process_perf_data.py perf_reduction_"$gpu_arch".log
\ No newline at end of file
script/profile_batched_gemm.sh
View file @ 67bd84ff
......@@ -9,7 +9,7 @@ LAYOUT=$3
VERIFY=$4
INIT=$5
LOG=$6
REPEAT=$7
TIME=$7
OP=$1
DATATYPE=$2
......@@ -17,28 +17,28 @@ LAYOUT=$3
VERIFY=$4
INIT=$5
LOG=$6
REPEAT=$7
TIME=$7
######## op datatype layout verify init log repeat M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 960 1024 1024 -1 -1 -1 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1920 2048 2048 -1 -1 -1 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 3840 4096 4096 -1 -1 -1 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 7680 8192 8192 -1 -1 -1 -1 -1 -1 2
######## op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 960 1024 1024 -1 -1 -1 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1920 2048 2048 -1 -1 -1 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 3840 4096 4096 -1 -1 -1 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 7680 8192 8192 -1 -1 -1 -1 -1 -1 2
####### op datatype layout verify init log repeat M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1024 1024 1024 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2048 2048 2048 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 4096 4096 4096 4096 4096 4096 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 8192 8192 8192 8192 8192 8192 -1 -1 -1 2
####### op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1024 1024 1024 1024 1024 1024 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2048 2048 2048 2048 2048 2048 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 4096 4096 4096 4096 4096 4096 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 8192 8192 8192 8192 8192 8192 -1 -1 -1 2
####### op datatype layout verify init log repeat M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1056 1056 1056 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2080 2080 2080 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 4096 4096 4096 4128 4128 4128 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 8192 8192 8192 8224 8224 8224 -1 -1 -1 2
####### op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1024 1024 1024 1056 1056 1056 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2048 2048 2048 2080 2080 2080 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 4096 4096 4096 4128 4128 4128 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 8192 8192 8192 8224 8224 8224 -1 -1 -1 2
####### op datatype layout verify init log repeat M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1088 1088 1088 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2112 2112 2112 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 4096 4096 4096 4160 4160 4160 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 8192 8192 8192 8256 8256 8256 -1 -1 -1 2
\ No newline at end of file
####### op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1024 1024 1024 1088 1088 1088 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2048 2048 2048 2112 2112 2112 -1 -1 -1 8
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 4096 4096 4096 4160 4160 4160 -1 -1 -1 4
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 8192 8192 8192 8256 8256 8256 -1 -1 -1 2
script/profile_conv.sh deleted 100755 → 0
View file @ a1841d55
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
DRIVER="../build/bin/ckProfiler"
OP=$1
DATATYPE=$2
IN_LAYOUT=$3
WEI_LAYOUT=$4
OUT_LAYOUT=$5
VERIFY=$6
INIT=$7
LOG=$8
REPEAT=$9
N=${10}
######## op datatype in_layout wei_layout out_layout verify init log repeat N__ K___ C___ Y X Hi__ Wi__ Strides Dilations LeftPads RightPads
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 256 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 512 1024 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 128 128 3 3 28 28 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 512 128 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 128 128 3 3 56 56 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 512 2048 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 1024 256 1 1 14 14 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 256 256 3 3 14 14 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 256 256 3 3 28 28 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 128 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 64 256 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 512 512 3 3 14 14 2 2 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 128 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 256 512 1 1 28 28 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 2048 512 1 1 7 7 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 512 512 3 3 7 7 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 256 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 64 64 1 1 56 56 1 1 1 1 0 0 0 0
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 64 64 3 3 56 56 1 1 1 1 1 1 1 1
$DRIVER $OP $DATATYPE $IN_LAYOUT $WEI_LAYOUT $OUT_LAYOUT $VERIFY $INIT $LOG $REPEAT $N 64 3 7 7 224 224 2 2 1 1 3 3 3 3
script/profile_conv_bwd_data.sh 0 → 100755
View file @ 67bd84ff
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
DRIVER="../build/bin/ckProfiler"
OP=$1
DATATYPE=$2
LAYOUT=$3
VERIFY=$4
INIT=$5
LOG=$6
TIME=$7
N=$8
# Resnet50
######## 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 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 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 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 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 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 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 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 28 28 2 2 1 1 1 1 1 1
$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 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 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 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 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 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 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 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 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 3 3 56 56 1 1 1 1 1 1 1 1
$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
script/profile_conv_fwd.sh 0 → 100755
View file @ 67bd84ff
#!/bin/bash
## GPU visibility
export HIP_VISIBLE_DEVICES=0
DRIVER="../build/bin/ckProfiler"
OP=$1
DATATYPE=$2
LAYOUT=$3
VERIFY=$4
INIT=$5
LOG=$6
TIME=$7
N=$8
# Resnet50
######## 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 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 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 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 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 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 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 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 28 28 2 2 1 1 1 1 1 1
$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 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 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 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 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 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 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 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 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 3 3 56 56 1 1 1 1 1 1 1 1
$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
script/profile_gemm.sh
View file @ 67bd84ff
......@@ -2,7 +2,6 @@
## GPU visibility
export HIP_VISIBLE_DEVICES=0
#make -j ckProfiler
DRIVER="../build/bin/ckProfiler"
echo $DRIVER
OP=$1
......@@ -11,43 +10,49 @@ LAYOUT=$3
VERIFY=$4
INIT=$5
LOG=$6
REPEAT=$7
######## op datatype layout verify init log repeat M___ N___ K___ StrideA StrideB StrideC
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 256 256 256 256 256 256
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 960 1024 1024 1024 1024 1024
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1920 2048 2048 2048 2048 2048
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 3840 4096 4096 4096 4096 4096
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 7680 8192 8192 8192 8192 8192
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1024 1024 1024
#$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2048 2048 2048
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 960 1024 1024 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1920 2048 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 3840 4096 4096 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 7680 8192 8192 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1024 1024 1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2048 2048 2048
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 4096 4096 4096 4096 4096 4096
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 8192 8192 8192 8192 8192 8192
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1056 1056 1056
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2080 2080 2080
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 4096 4096 4096 4128 4128 4128
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 8192 8192 8192 8224 8224 8224
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1024 1024 1024 1088 1088 1088
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2048 2048 2048 2112 2112 2112
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 4096 4096 4096 4160 4160 4160
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 8192 8192 8192 8256 8256 8256
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 6656 8192 8192 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 3328 4096 4096 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1664 2048 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 832 1024 1024 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 7040 8192 8192 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 5120 5632 4096 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 2560 2816 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 1280 1408 1024 -1 -1 -1
TIME=$7
# 120 CU
######## op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 960 1024 1024 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 960 2048 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1920 1024 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1920 2048 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 3840 4096 4096 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 7680 8192 8192 -1 -1 -1
# 104 CU
######## op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 832 1024 1024 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 832 2048 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1664 1024 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1664 2048 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 3328 4096 4096 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 6656 8192 8192 -1 -1 -1
# 110 CU
######## op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1280 1408 1024 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1280 2816 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2560 1408 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2560 2816 2048 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 5120 5632 4096 -1 -1 -1
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 7040 8192 8192 -1 -1 -1
# testing different strides
######## op datatype layout verify init log time M___ N___ K___ StrideA StrideB StrideC
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1024 1024 1024 1024 1024 1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2048 2048 2048 2048 2048 2048
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 4096 4096 4096 4096 4096 4096
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 8192 8192 8192 8192 8192 8192
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1024 1024 1024 1056 1056 1056
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2048 2048 2048 2080 2080 2080
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 4096 4096 4096 4128 4128 4128
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 8192 8192 8192 8224 8224 8224
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 1024 1024 1024 1088 1088 1088
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 2048 2048 2048 2112 2112 2112
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 4096 4096 4096 4160 4160 4160
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 8192 8192 8192 8256 8256 8256
script/profile_grouped_gemm.sh
View file @ 67bd84ff
......@@ -9,10 +9,10 @@ LAYOUT=$3
VERIFY=$4
INIT=$5
LOG=$6
REPEAT=$7
TIME=$7
######## op datatype layout verify init log repeat Ms______________ Ns______________ Ks_____________ StrideAs___________ StrideBs__________ StrideCs___________
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 256,512,1024,768 128,256,384,1024 128,192,256,512 1024,1025,1044,1026 1024,1024,1024,1024 1025,1024,1028,1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 512,768,2048,128 128,256,384,1024 128,192,256,512 1024,1025,2053,1026 1024,1024,1024,1024 1025,1024,2054,1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 256,512,1024,768 512,256,768,1024 128,192,256,512 1024,1045,1034,1026 1024,1024,1024,1024 1025,1063,1028,1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $REPEAT 512,768,4096,768 128,768,512,2048 128,192,256,512 1024,1027,4096,2050 1024,1024,1024,2048 1025,1024,4099,2049
\ No newline at end of file
######## op datatype layout verify init log time Ms______________ Ns______________ Ks_____________ StrideAs___________ StrideBs__________ StrideCs___________
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 256,512,1024,768 128,256,384,1024 128,192,256,512 1024,1025,1044,1026 1024,1024,1024,1024 1025,1024,1028,1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 512,768,2048,128 128,256,384,1024 128,192,256,512 1024,1025,2053,1026 1024,1024,1024,1024 1025,1024,2054,1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 256,512,1024,768 512,256,768,1024 128,192,256,512 1024,1045,1034,1026 1024,1024,1024,1024 1025,1063,1028,1024
$DRIVER $OP $DATATYPE $LAYOUT $VERIFY $INIT $LOG $TIME 512,768,4096,768 128,768,512,2048 128,192,256,512 1024,1027,4096,2050 1024,1024,1024,2048 1025,1024,4099,2049
script/profile_resnet50.sh
View file @ 67bd84ff
This diff is collapsed.
script/run_full_performance_tests.sh
View file @ 67bd84ff
......@@ -40,82 +40,82 @@ function print_log_header(){
#run gemm tests
export gemm_log="perf_gemm_${gpu_arch}.log"
print_log_header $gemm_log $env_type $branch $host_name
./profile_gemm.sh gemm 0 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 3 $verify 1 0 5 | tee -a $gemm_log
#run resnet50 tests
export resnet256_log="perf_resnet50_N256_${gpu_arch}.log"
print_log_header $resnet256_log $env_type $branch $host_name
./profile_resnet50.sh conv_fwd_bias_relu 1 1 1 1 $verify 1 0 1 256 | tee -a $resnet256_log
export resnet4_log="perf_resnet50_N4_${gpu_arch}.log"
print_log_header $resnet4_log $env_type $branch $host_name
./profile_resnet50.sh conv_fwd_bias_relu 1 1 1 1 $verify 1 0 1 4 | tee -a $resnet4_log
./profile_gemm.sh gemm 0 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 2 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 0 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 2 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 0 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 2 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 0 3 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 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 $gemm_log
#run batched_gemm tests
export batched_gemm_log="perf_batched_gemm_${gpu_arch}.log"
print_log_header $batched_gemm_log $env_type $branch $host_name
./profile_batched_gemm.sh batched_gemm 0 0 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 1 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 2 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 3 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 0 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 1 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 2 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 3 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 0 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 1 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 2 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 3 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 0 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 1 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 2 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 3 $verify 1 0 5 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 0 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 1 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 2 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 0 3 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 0 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 1 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 2 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 1 3 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 0 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 1 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 2 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 2 3 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 0 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 1 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 2 $verify 1 0 1 | tee -a $batched_gemm_log
./profile_batched_gemm.sh batched_gemm 3 3 $verify 1 0 1 | tee -a $batched_gemm_log
#run grouped_gemm tests
export grouped_gemm_log="perf_grouped_gemm_${gpu_arch}.log"
print_log_header $grouped_gemm_log $env_type $branch $host_name
./profile_grouped_gemm.sh grouped_gemm 1 0 $verify 1 0 5 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 1 $verify 1 0 5 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 2 $verify 1 0 5 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 3 $verify 1 0 5 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 0 $verify 1 0 1 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 1 $verify 1 0 1 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 2 $verify 1 0 1 | tee -a $grouped_gemm_log
./profile_grouped_gemm.sh grouped_gemm 1 3 $verify 1 0 1 | tee -a $grouped_gemm_log
#run fwd_conv tests
export fwd_conv_log="perf_fwd_conv_${gpu_arch}.log"
print_log_header $fwd_conv_log $env_type $branch $host_name
./profile_conv.sh conv_fwd 0 1 $verify 1 0 5 2 256 | tee -a $fwd_conv_log
./profile_conv.sh conv_fwd 1 1 $verify 1 0 5 2 256 | tee -a $fwd_conv_log
./profile_conv.sh conv_fwd 2 1 $verify 1 0 5 2 256 | tee -a $fwd_conv_log
./profile_conv.sh conv_fwd 3 1 $verify 1 0 5 2 256 | tee -a $fwd_conv_log
#run GEMM+Bilinear tests
export gemm_bilinear_log="perf_gemm_bilinear_${gpu_arch}.log"
print_log_header $gemm_bilinear_log $env_type $branch $host_name
./profile_gemm_bilinear.sh gemm_bilinear 1 0 $verify 1 0 1 | tee -a $gemm_bilinear_log
./profile_gemm_bilinear.sh gemm_bilinear 1 1 $verify 1 0 1 | tee -a $gemm_bilinear_log
./profile_gemm_bilinear.sh gemm_bilinear 1 2 $verify 1 0 1 | tee -a $gemm_bilinear_log
./profile_gemm_bilinear.sh gemm_bilinear 1 3 $verify 1 0 1 | tee -a $gemm_bilinear_log
#run bwd_conv tests
export bwd_conv_log="perf_bwd_conv_${gpu_arch}.log"
print_log_header $bwd_conv_log $env_type $branch $host_name
./profile_conv.sh conv2d_bwd_data 0 1 1 1 $verify 1 0 5 128 | tee -a $bwd_conv_log
./profile_conv.sh conv2d_bwd_data 1 1 1 1 $verify 1 0 5 128 | tee -a $bwd_conv_log
./profile_conv.sh conv2d_bwd_data 2 1 1 1 $verify 1 0 5 128 | tee -a $bwd_conv_log
./profile_conv.sh conv2d_bwd_data 3 1 1 1 $verify 1 0 5 128 | tee -a $bwd_conv_log
#run conv_fwd tests
export conv_fwd_log="perf_conv_fwd_${gpu_arch}.log"
print_log_header $conv_fwd_log $env_type $branch $host_name
./profile_conv_fwd.sh conv_fwd 0 1 $verify 1 0 1 256 | tee -a $conv_fwd_log
./profile_conv_fwd.sh conv_fwd 1 1 $verify 1 0 1 256 | tee -a $conv_fwd_log
./profile_conv_fwd.sh conv_fwd 2 1 $verify 1 0 1 256 | tee -a $conv_fwd_log
./profile_conv_fwd.sh conv_fwd 3 1 $verify 1 0 1 256 | tee -a $conv_fwd_log
#run fusion tests
export fusion_log="perf_fusion_${gpu_arch}.log"
print_log_header $fusion_log $env_type $branch $host_name
./profile_gemm_bilinear.sh gemm_bilinear 1 0 $verify 1 0 1 | tee -a $fusion_log
./profile_gemm_bilinear.sh gemm_bilinear 1 1 $verify 1 0 1 | tee -a $fusion_log
./profile_gemm_bilinear.sh gemm_bilinear 1 2 $verify 1 0 1 | tee -a $fusion_log
./profile_gemm_bilinear.sh gemm_bilinear 1 3 $verify 1 0 1 | tee -a $fusion_log
#run conv_bwd_data tests
export conv_bwd_data_log="perf_conv_bwd_data_${gpu_arch}.log"
print_log_header $conv_bwd_data_log $env_type $branch $host_name
./profile_conv_bwd_data.sh conv_bwd_data 0 1 $verify 1 0 1 256 | tee -a $conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 1 1 $verify 1 0 1 256 | tee -a $conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 2 1 $verify 1 0 1 256 | tee -a $conv_bwd_data_log
./profile_conv_bwd_data.sh conv_bwd_data 3 1 $verify 1 0 1 256 | tee -a $conv_bwd_data_log
#run resnet50 tests
export resnet256_log="perf_resnet50_N256_${gpu_arch}.log"
print_log_header $resnet256_log $env_type $branch $host_name
./profile_resnet50.sh conv_fwd_bias_relu 1 1 1 1 $verify 1 0 1 256 | tee -a $resnet256_log
export resnet4_log="perf_resnet50_N4_${gpu_arch}.log"
print_log_header $resnet4_log $env_type $branch $host_name
./profile_resnet50.sh conv_fwd_bias_relu 1 1 1 1 $verify 1 0 1 4 | tee -a $resnet4_log
#run reduction tests
export reduction_log="perf_reduction_${gpu_arch}.log"
......
script/run_performance_tests.sh
View file @ 67bd84ff
......@@ -33,27 +33,28 @@ function print_log_header(){
echo 'Environment type: ' $2 >> $1;
/opt/rocm/bin/amdclang++ --version | grep -e 'InstalledDir' >> $1;
}
#run gemm tests
export gemm_log="perf_gemm_${gpu_arch}.log"
print_log_header $gemm_log $env_type $branch $host_name
./profile_gemm.sh gemm 0 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 0 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 1 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 2 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 1 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 2 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 3 3 $verify 1 0 5 | tee -a $gemm_log
./profile_gemm.sh gemm 0 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 2 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 0 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 0 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 2 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 1 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 0 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 2 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 3 2 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 0 3 $verify 1 0 1 | tee -a $gemm_log
./profile_gemm.sh gemm 1 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 $gemm_log
#run resnet50 test
#run resnet50 tests
export resnet256_log="perf_resnet50_N256_${gpu_arch}.log"
print_log_header $resnet256_log $env_type $branch $host_name
./profile_resnet50.sh conv_fwd_bias_relu 1 1 1 1 $verify 1 0 1 256 | tee -a $resnet256_log
......
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