ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-05 22:22:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

Export ROCm/rocm-libraries@2d4a3223cb

Browse Source
This commit is contained in:
Ding, Yi
2026-03-11 23:03:20 -04:00
commit e6cd3f1e3f
6330 changed files with 1132789 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
example/ck_tile/35_batched_transpose/script/perf_test.sh Executable file
View File

@@ -0,0 +1,18 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE=./build/bin/tile_example_batched_transpose
for C in "64" "256" "1024" "4096" "16384"; do
for W in "64" "256" "1024" "4096" "16384"; do
for pr in "fp8" "fp16" "bf16"; do
for pipeline in "0" "1"; do
$EXE -pipeline=$pipeline -pr=$pr -N=1 -C=$C -H=1 -W=$W -layout_in='NCHW' -layout_out='NHWC'
done
done
done
done

42
example/ck_tile/35_batched_transpose/script/run_full_test.sh Executable file
View File

@@ -0,0 +1,42 @@
#!/bin/bash
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
#
# in order to run this script you'd first need to build the tile_example_batched_transpose executables in ../build/bin/
#
# run the script as "./run_full_test.sh <tag for your test environment> <branch name> <host name> <gpu_arch>
# input arguments:
# environment tag : a string describing the specifics of your test environment
# branch name : name of the branch in git repo (git status | grep -e 'On branch')
# host name : $hostname
# gpu architecture: e.g., gfx90a, or gfx942, etc.
#get the command line arguments:
export env_type=$1
echo 'Environment type: ' $env_type
export branch=$2
echo 'Branch name: ' $branch
export host_name=$3
echo 'Host name: ' $host_name
export GPU_arch=$4
echo 'GPU_arch: ' $GPU_arch
function print_log_header(){
rm -f $1;
echo 'On branch ' $3 &> $1;
echo 'Node name: ' $4 >> $1;
#get GPU_arch and number of compute units from rocminfo
echo -n "GPU_arch: " >> $1; rocminfo | grep "Name:" | grep "gfx" >> $1;
rocminfo | grep "Compute Unit:" >> $1;
hipcc --version | grep -e 'HIP version' >> $1;
echo 'Environment type: ' $2 >> $1;
/opt/rocm/bin/amdclang++ --version | grep -e 'InstalledDir' >> $1;
}
#run verification tests
example/ck_tile/35_batched_transpose/script/smoke_test.sh
#run performance benchmarks

32
example/ck_tile/35_batched_transpose/script/smoke_test.sh Executable file
View File

@@ -0,0 +1,32 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE=./build/bin/tile_example_batched_transpose
for pr in "fp8" "fp16" "bf16"; do
for pipeline in "0" "1"; do
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=32 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=64 -H=1 -W=64 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=2 -C=12 -H=1 -W=32 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=3 -C=1334 -H=1 -W=37 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=4 -C=27 -H=1 -W=32 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=5 -C=1234 -H=1 -W=12 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=128 -C=1024 -H=64 -W=64 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=128 -C=1024 -H=64 -W=64 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=16 -C=64 -H=32 -W=128 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=16 -C=64 -H=128 -W=32 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=2048 -H=1 -W=1 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=2048 -H=1 -W=1 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1024 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=1 -H=1024 -W=1024 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=8 -C=16 -H=8 -W=16 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=8 -C=16 -H=8 -W=16 -layout_in='NHWC' -layout_out='NCHW'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=64 -H=1 -W=1024 -layout_in='NCHW' -layout_out='NHWC'
$EXE -pr=$pr -pipeline=$pipeline -N=1 -C=64 -H=1024 -W=1 -layout_in='NHWC' -layout_out='NCHW'
done
done