Files
composable_kernel/script/profile_reduce_with_index.sh
Qianfeng 0875df0f9a Pr82 followup (#115)
* Use thread cluster descriptor and explicit M_K 2d descriptor to simply Blockwise Reduction

* Change by replacing ReduceDims by NumReduceDims as Device Reduce interface template parameter

* Rename the folder name for the pool2d and reduce examples

* Update to reduction test scripts

* Add Readme for pool2d_fwd and reduce_blockwise examples

* Tiny fix in reduce profiler and tiny update in reduce testing scripts

* Tiny fix in testing script profile_reduce_no_index.sh

* Tiny change in script/profile_reduce_with_index.sh

* Renaming and refining in Reduction profiler/device layer/examples

* Renaming and refining in Reduction profiler/device layer/examples

* Renaming all NumReduceDims to NumReduceDim

[ROCm/composable_kernel commit: 827301d95a]
2022-03-10 10:14:43 -06:00

71 lines
4.4 KiB
Bash
Executable File

#!/bin/bash
PRECISION=
##PRECISION=--half
##PRECISION=--double
driver="./bin/ckProfiler"
VERIFY="-v $1"
INIT=$2
NREPEAT=$3
#### 2 - MIN, 3 - MAX, 4 - AMAX
Operations="2 4"
## for generic validation
for op in $Operations; do
for use_idx in 0 1; do
set -x
####### datatype layout reduce dims op use index verify init repeats
$driver reduce $PRECISION -D 64,4,280,82 -R 0 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 1 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 3 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 1,2,3 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 0,2,3 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 64,4,280,82 -R 0,1,3 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,22960 -R 0 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,22960 -R 1 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 4,1469440 -R 0 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 4,1469440 -R 1 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
set +x
done
done
Operations=2
## for performance evaluation (resnet50 NHWC => C)
for op in $Operations; do
for use_idx in 0 1; do
set -x
####### datatype layout reduce dims op use index verify init repeats
$driver reduce $PRECISION -D 256,14,14,1024 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,28,28,128 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,58,58,128 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,7,7,2048 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,14,14,256 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,30,30,256 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,56,56,256 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,16,16,512 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,28,28,512 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,7,7,512 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,56,56,64 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 256,230,230,3 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,14,14,1024 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,28,28,128 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,58,58,128 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,7,7,2048 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,14,14,256 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,30,30,256 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,56,56,256 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,16,16,512 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,28,28,512 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,7,7,512 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
$driver reduce $PRECISION -D 128,56,56,64 -R 0,1,2 -O $op -I $use_idx $VERIFY $INIT $NREPEAT
set +x
done
done