Skip to content

[CI] Automatically detect AMD architecture #16071

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 10 commits into from
Nov 14, 2024
Merged

[CI] Automatically detect AMD architecture #16071

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
2d28503
[CI] Automatically detect AMD architecture
sarnex Nov 12, 2024
9f48bd6
do it in the testing infra instead
sarnex Nov 13, 2024
22641d9
Merge remote-tracking branch 'upstream/sycl' into sycl
sarnex Nov 13, 2024
a01f872
fix UNSUPPORTED
sarnex Nov 13, 2024
8494ca5
format
sarnex Nov 13, 2024
48b6502
add -fsycl to test
sarnex Nov 13, 2024
7db9bff
move to existing loop
sarnex Nov 13, 2024
0061488
Merge remote-tracking branch 'upstream/sycl' into sycl
sarnex Nov 13, 2024
b666f7b
format
sarnex Nov 13, 2024
75f90f9
clean up set extraction
sarnex Nov 14, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 0 additions & 1 deletion .github/workflows/sycl-linux-precommit.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -162,7 +162,6 @@ jobs:
runner: '["Linux", "amdgpu"]'
image: ghcr.io/intel/llvm/ubuntu2204_build:latest
image_extra_opts: --device=/dev/dri --device=/dev/kfd
extra_cmake_args: -DHIP_PLATFORM="AMD" -DAMD_ARCH="gfx1031"
- name: CUDA system
runner: '["Linux", "cuda"]'
image: ghcr.io/intel/llvm/ubuntu2204_build:latest
Expand Down
13 changes: 0 additions & 13 deletions .github/workflows/sycl-linux-run-tests.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,9 +21,6 @@ on:
type: string
required: True
extra_cmake_args:
description: |
If empty, then HIP_PLATFORM and AMD_ARCH would be automatically added
if inputs.target_devices contains 'ext_oneapi_hip'
type: string
required: False
tests_selector:
Expand Down Expand Up @@ -282,16 +279,6 @@ jobs:
run: |
if [ -n "$CMAKE_EXTRA_ARGS" ]; then
echo "opts=$CMAKE_EXTRA_ARGS" >> $GITHUB_OUTPUT
else
if [ "${{ contains(inputs.target_devices, 'ext_oneapi_hip') }}" == "true" ]; then
if [ "${{ runner.name }}" == "cp-amd-runner" ]; then
echo 'opts=-DHIP_PLATFORM="AMD" -DAMD_ARCH="gfx1030"' >> $GITHUB_OUTPUT
else
echo 'opts=-DHIP_PLATFORM="AMD" -DAMD_ARCH="gfx1031"' >> $GITHUB_OUTPUT
fi
else
echo 'opts=' >> $GITHUB_OUTPUT
fi
fi
- name: Configure E2E tests
if: inputs.tests_selector == 'e2e'
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/Matrix/joint_matrix_hip_gfx90a.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,10 +6,10 @@
//
//===----------------------------------------------------------------------===//

// RUN: %{build} -fsycl -fsycl-targets=amd_gpu_gfx90a %s -o %t.out
// RUN: %clangxx -fsycl -fsycl-targets=amd_gpu_gfx90a %s -o %t.out
Copy link
Contributor Author

@sarnex sarnex Nov 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This change is a bug fix, we shouldn't use %{build} if we're setting -fsycl-targets/choosing file and output

// RUN: %{run} %t.out

// REQUIRES: gpu-amd-gfx90a
// REQUIRES: arch-amd_gpu_gfx90a
Copy link
Contributor Author

@sarnex sarnex Nov 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

this change is just to remove a hack in the testing infra, should be exactly the same functionally

Copy link
Contributor

@ayylol ayylol Nov 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

hey, just wondering if you were able to build and run these gfx90a tests. AFAIK we don't have these in our CI, so I don't think they ever ran.

Copy link
Contributor Author

@sarnex sarnex Nov 13, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yeah i actually tested on the one AMD machine i know not in our CI and it happens to be gfx90a (MI250)


#include "joint_matrix_hip_apply.hpp"
#include "joint_matrix_hip_copy.hpp"
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/Matrix/joint_matrix_hip_half_gfx90a.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,10 +6,10 @@
//
//===----------------------------------------------------------------------===//

// RUN: %{build} -fsycl -fsycl-targets=amd_gpu_gfx90a %s -o %t.out
// RUN: %clangxx -fsycl -fsycl-targets=amd_gpu_gfx90a %s -o %t.out
// RUN: %{run} %t.out

// REQUIRES: gpu-amd-gfx90a
// REQUIRES: arch-amd_gpu_gfx90a
// REQUIRES: aspect-fp16

#include "joint_matrix_hip_apply.hpp"
Expand Down
4 changes: 2 additions & 2 deletions sycl/test-e2e/Matrix/runtime_query_hip_gfx90a.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,8 +6,8 @@
//
//===----------------------------------------------------------------------===//

// REQUIRES: gpu-amd-gfx90a
// RUN: %{build} -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=gfx90a -o %t.out
// REQUIRES: arch-amd_gpu_gfx90a
// RUN: %clangxx -Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=gfx90a -o %t.out
Copy link
Contributor

@YuriPlyakhin YuriPlyakhin Nov 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

should we add -fsycl?

Copy link
Contributor Author

@sarnex sarnex Nov 13, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

good catch, fixed in latest commit

// RUN: %{run} %t.out

#include <sycl/detail/core.hpp>
Expand Down
81 changes: 43 additions & 38 deletions sycl/test-e2e/lit.cfg.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -442,8 +442,6 @@ def open_check_file(file_name):
)
sp = subprocess.check_output(cmd, text=True, shell=True)
for line in sp.splitlines():
if "gfx90a" in line:
config.available_features.add("gpu-amd-gfx90a")
if not line.startswith("["):
continue
(backend, device) = line[1:].split("]")[0].split(":")
Expand Down Expand Up @@ -540,42 +538,6 @@ def open_check_file(file_name):
config.cuda_libs_dir = os.path.join(os.environ["CUDA_PATH"], r"lib64")
config.cuda_include = os.path.join(os.environ["CUDA_PATH"], "include")

# FIXME: This needs to be made per-device as well, possibly with a helper.
if "hip:gpu" in config.sycl_devices and config.hip_platform == "AMD":
if not config.amd_arch:
lit_config.error(
"Cannot run tests for HIP without an offload-arch. Please "
+ "specify one via the 'amd_arch' parameter or 'AMD_ARCH' CMake "
+ "variable."
)
llvm_config.with_system_environment("ROCM_PATH")
config.available_features.add("hip_amd")
arch_flag = (
"-Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=" + config.amd_arch
)
config.substitutions.append(
("%rocm_path", os.environ.get("ROCM_PATH", "/opt/rocm"))
)
elif "hip:gpu" in config.sycl_devices and config.hip_platform == "NVIDIA":
config.available_features.add("hip_nvidia")
arch_flag = ""
else:
arch_flag = ""

if lit_config.params.get("compatibility_testing", False):
config.substitutions.append(("%clangxx", " true "))
config.substitutions.append(("%clang", " true "))
else:
config.substitutions.append(
(
"%clangxx",
" " + config.dpcpp_compiler + " " + config.cxx_flags + " " + arch_flag,
)
)
config.substitutions.append(
("%clang", " " + config.dpcpp_compiler + " " + config.c_flags)
)

config.substitutions.append(("%threads_lib", config.sycl_threads_lib))

if lit_config.params.get("ze_debug"):
Expand Down Expand Up @@ -805,6 +767,49 @@ def open_check_file(file_name):
else:
config.intel_driver_ver[sycl_device] = {}

if "hip:gpu" in config.sycl_devices and config.hip_platform == "AMD":
if not config.amd_arch:
amd_arch_prefix = "arch-amd_gpu_"
amd_device_arch = [i for i in config.sycl_dev_features["hip:gpu"] if amd_arch_prefix in i]
if len(amd_device_arch) == 0:
lit_config.error(
Copy link
Contributor Author

@sarnex sarnex Nov 13, 2024 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

i had to move this block down so the sycl-ls parsing we already do is available, but that meant i had to add a loop at the end to update the features

"Cannot detect architecture for AMD HIP device, specify it explicitly"
)
if len(amd_device_arch) > 1:
lit_config.warning("Multiple AMD HIP devices, using the architecture from the first one")
config.amd_arch = amd_device_arch[0].replace(amd_arch_prefix, "")
llvm_config.with_system_environment("ROCM_PATH")
config.available_features.add("hip_amd")
arch_flag = (
"-Xsycl-target-backend=amdgcn-amd-amdhsa --offload-arch=" + config.amd_arch
)
config.substitutions.append(
("%rocm_path", os.environ.get("ROCM_PATH", "/opt/rocm"))
)
elif "hip:gpu" in config.sycl_devices and config.hip_platform == "NVIDIA":
config.available_features.add("hip_nvidia")
arch_flag = ""
else:
arch_flag = ""

if lit_config.params.get("compatibility_testing", False):
config.substitutions.append(("%clangxx", " true "))
config.substitutions.append(("%clang", " true "))
else:
config.substitutions.append(
(
"%clangxx",
" " + config.dpcpp_compiler + " " + config.cxx_flags + " " + arch_flag,
)
)
config.substitutions.append(
("%clang", " " + config.dpcpp_compiler + " " + config.c_flags)
)

# Update each device's features with anything added during HIP device handling
for sycl_device in config.sycl_devices:
config.sycl_dev_features[sycl_device] = config.sycl_dev_features[sycl_device].union(config.available_features)

# Set timeout for a single test
try:
import psutil
Expand Down
Loading