ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-05 14:11:29 +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

17
example/ck_tile/03_gemm/script/benchmark_basic_bf16.sh Executable file
View File

@@ -0,0 +1,17 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_basic -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "64" "512" "1024" "2048"; do
for n in "512" "1024" "2048"; do
for k in "64" "512" "1024" "2048"; do
$EXE -prec=bf16 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

17
example/ck_tile/03_gemm/script/benchmark_basic_bf8.sh Executable file
View File

@@ -0,0 +1,17 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_basic -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "64" "512" "1024" "2048"; do
for n in "512" "1024" "2048"; do
for k in "64" "512" "1024" "2048"; do
$EXE -prec=bf8 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

16
example/ck_tile/03_gemm/script/benchmark_basic_fp16.sh Executable file
View File

@@ -0,0 +1,16 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_basic -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "64" "512" "1024" "2048"; do
for n in "512" "1024" "2048"; do
for k in "64" "512" "1024" "2048"; do
$EXE -prec=fp16 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

17
example/ck_tile/03_gemm/script/benchmark_basic_fp8.sh Executable file
View File

@@ -0,0 +1,17 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_basic -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "64" "512" "1024" "2048"; do
for n in "512" "1024" "2048"; do
for k in "64" "512" "1024" "2048"; do
$EXE -prec=fp8 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

16
example/ck_tile/03_gemm/script/benchmark_mem_pipeline_bf16.sh Executable file
View File

@@ -0,0 +1,16 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_universal -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "512" "1024" "2048" "4096"; do
for n in "512" "1024" "2048"; do
for k in "512" "1024" "2048"; do
$EXE -prec=bf16 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

16
example/ck_tile/03_gemm/script/benchmark_mem_pipeline_bf8.sh Executable file
View File

@@ -0,0 +1,16 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_universal -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "512" "1024" "2048" "4096"; do
for n in "512" "1024" "2048"; do
for k in "512" "1024" "2048"; do
$EXE -prec=bf8 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

16
example/ck_tile/03_gemm/script/benchmark_mem_pipeline_fp16.sh Executable file
View File

@@ -0,0 +1,16 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_universal -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "512" "1024" "2048" "4096"; do
for n in "512" "1024" "2048"; do
for k in "512" "1024" "2048"; do
$EXE -prec=fp16 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

16
example/ck_tile/03_gemm/script/benchmark_mem_pipeline_fp8.sh Executable file
View File

@@ -0,0 +1,16 @@
#!/bin/sh
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_universal -type f | head -n 1)"
VALID=1
for b_matrix_layout in "C"; do
for m in "512" "1024" "2048" "4096"; do
for n in "512" "1024" "2048"; do
for k in "512" "1024" "2048"; do
$EXE -prec=fp8 -m=$m -n=$n -k=$k -a_layout="R" -b_layout="$b_matrix_layout" -c_layout="R" -v=$VALID
done
done
done
done

48
example/ck_tile/03_gemm/script/run_full_test.sh Executable file
View File

@@ -0,0 +1,48 @@
#!/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_gemm 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 architecture and 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' >> $1;
}
# run verification tests
for dtype in fp16 bf16 fp8 bf8; do
example/ck_tile/03_gemm/script/benchmark_basic_$dtype.sh
done
example/ck_tile/03_gemm/script/smoke_test_mem_pipeline.sh
# run performance benchmarks
for dtype in fp16 bf16 fp8 bf8; do
export gemm_log="perf_tile_gemm_mem_pipeline_${dtype}_${GPU_arch}.log"
print_log_header $gemm_log $env_type $branch $host_name
example/ck_tile/03_gemm/script/benchmark_mem_pipeline_$dtype.sh 2>&1 | tee -a $gemm_log
done

39
example/ck_tile/03_gemm/script/smoke_test_basic.sh Executable file
View File

@@ -0,0 +1,39 @@
#!/bin/bash
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_basic -type f | head -n 1)"
KNAME=1
export CK_WARMUP=0
export CK_REPEAT=1
COMMON_ARGS='-v=2 -warmup=0 -repeat=1'
run_tests() {
for m in 128 1024; do
for n in 128 2048; do
for k in 64 128; do
$EXE -m=$m -n=$n -k=$k -stride_a=0 -stride_b=0 -stride_c=0 -prec=$1 $COMMON_ARGS
if [ $? -eq 0 ]; then
echo "Success: Test with m=$m, n=$n, k=$k executed successfully."
else
echo "Error: Test with m=$m, n=$n, k=$k failed to execute properly."
# Optionally, exit or break if you need to halt further execution
# exit 1
fi
done
done
done
}
set -x
run_tests "fp16"
run_tests "bf16"
run_tests "fp8"
run_tests "bf8"
set +x

42
example/ck_tile/03_gemm/script/smoke_test_mem_pipeline.sh Executable file
View File

@@ -0,0 +1,42 @@
#!/bin/bash
# Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
# SPDX-License-Identifier: MIT
EXE="$(find . -name tile_example_gemm_universal -type f | head -n 1)"
KNAME=1
export CK_WARMUP=0
export CK_REPEAT=1
COMMON_ARGS='-v=1 -warmup=0 -repeat=1'
run_tests() {
for m in 512 1024; do
for n in 512 2048; do
for k in 512 1024; do
$EXE -m=$m -n=$n -k=$k -stride_a=0 -stride_b=0 -stride_c=0 -prec=$1 $COMMON_ARGS
if [ $? -eq 0 ]; then
echo "Success: Test with batch=$batch, m=$m, n=$n, k=$k executed successfully."
else
echo "Error: Test with batch=$batch, m=$m, n=$n, k=$k failed to execute properly."
# Optionally, exit or break if you need to halt further execution
# exit 1
fi
done
done
done
}
set -x
run_tests "fp16"
run_tests "bf16"
run_tests "fp8"
run_tests "bf8"
run_tests "fp16i4"
run_tests "fp8i4"
run_tests "bf8i4"
set +x