[clang][tools] Add LevelZero support to offload-arch #160570
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Member
|
@llvm/pr-subscribers-clang Author: Alex Duran (adurang) ChangesFull diff: https://github.com/llvm/llvm-project/pull/160570.diff 3 Files Affected:
diff --git a/clang/tools/offload-arch/CMakeLists.txt b/clang/tools/offload-arch/CMakeLists.txt
index cb50b9c1d6dde..187c4d9517551 100644
--- a/clang/tools/offload-arch/CMakeLists.txt
+++ b/clang/tools/offload-arch/CMakeLists.txt
@@ -1,8 +1,16 @@
set(LLVM_LINK_COMPONENTS Support)
-add_clang_tool(offload-arch OffloadArch.cpp NVPTXArch.cpp AMDGPUArchByKFD.cpp AMDGPUArchByHIP.cpp)
+add_clang_tool(offload-arch OffloadArch.cpp NVPTXArch.cpp AMDGPUArchByKFD.cpp
+ AMDGPUArchByHIP.cpp LevelZeroArch.cpp)
+
+find_path(OFFLOAD_ARCH_LEVEL_ZERO_INCLUDE_DIR NAMES level_zero/ze_api.h)
+if (OFFLOAD_ARCH_LEVEL_ZERO_INCLUDE_DIR)
+ target_include_directories(offload-arch PRIVATE ${OFFLOAD_ARCH_LEVEL_ZERO_INCLUDE_DIR})
+ target_compile_definitions(offload-arch PRIVATE HAVE_LEVEL_ZERO_HEADERS)
+endif()
add_clang_symlink(amdgpu-arch offload-arch)
add_clang_symlink(nvptx-arch offload-arch)
+add_clang_symlink(intelgpu-arch offload-arch)
target_link_libraries(offload-arch PRIVATE clangBasic)
diff --git a/clang/tools/offload-arch/LevelZeroArch.cpp b/clang/tools/offload-arch/LevelZeroArch.cpp
new file mode 100644
index 0000000000000..8b6b393f6f9f6
--- /dev/null
+++ b/clang/tools/offload-arch/LevelZeroArch.cpp
@@ -0,0 +1,133 @@
+//===- LevelZeroArch.cpp - list installed Level Zero devices ---*- C++ -*--===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements a tool for detecting Level Zero devices installed in the
+// system
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef HAVE_LEVEL_ZERO_HEADERS
+
+int printGPUsByLevelZero() {
+ return 0;
+}
+
+#else
+
+#include "clang/Basic/Version.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/DynamicLibrary.h"
+#include "llvm/Support/Error.h"
+#include <cstdint>
+#include <cstdio>
+#include <memory>
+#include <level_zero/ze_api.h>
+
+using namespace llvm;
+extern cl::opt<bool> Verbose;
+
+#define DEFINE_WRAPPER(NAME) \
+ using NAME##_ty = decltype(NAME); \
+ void *NAME##Ptr = nullptr; \
+ template <class... Ts> ze_result_t NAME##_wrap(Ts... args) { \
+ if (!NAME##Ptr) { \
+ return ZE_RESULT_ERROR_UNKNOWN; \
+ } \
+ return reinterpret_cast<NAME##_ty *>(NAME##Ptr)(args...); \
+ };
+
+DEFINE_WRAPPER(zeInitDrivers)
+DEFINE_WRAPPER(zeDeviceGet)
+DEFINE_WRAPPER(zeDeviceGetProperties)
+
+static bool loadLevelZero() {
+ const char *L0Library = "libze_loader.so";
+ std::string ErrMsg;
+
+ auto DynlibHandle = std::make_unique<llvm::sys::DynamicLibrary>(
+ llvm::sys::DynamicLibrary::getPermanentLibrary(L0Library, &ErrMsg));
+ if (!DynlibHandle->isValid()) {
+ if (ErrMsg.empty())
+ ErrMsg = "unknown error";
+ if (Verbose)
+ llvm::errs() << "Unable to load library '" << L0Library << "': " << ErrMsg
+ << "!\n";
+ return false;
+ }
+
+ constexpr struct {
+ const char *name;
+ void **fptr;
+ } dlwrap[] = {
+ {"zeInitDrivers", &zeInitDriversPtr},
+ {"zeDeviceGet", &zeDeviceGetPtr},
+ {"zeDeviceGetProperties", &zeDeviceGetPropertiesPtr},
+ };
+
+ for (auto entry : dlwrap) {
+ void *P = DynlibHandle->getAddressOfSymbol(entry.name);
+ if (P == nullptr) {
+ if (Verbose)
+ llvm::errs() << "Unable to find '" << entry.name << "' in '" << L0Library
+ << "'!\n";
+ return false;
+ }
+ *(entry.fptr) = P;
+ }
+
+ return true;
+}
+
+#define CALL_ZE_AND_CHECK(Fn, ...) \
+ do { \
+ ze_result_t Rc = Fn##_wrap(__VA_ARGS__); \
+ if (Rc != ZE_RESULT_SUCCESS) { \
+ if (Verbose) \
+ llvm::errs() << "Error: " << __func__ << ":" << #Fn \
+ << " failed with error code " << Rc << "\n"; \
+ return 1; \
+ } \
+ } while (0)
+
+int printGPUsByLevelZero() {
+ if (!loadLevelZero())
+ return 1;
+
+ ze_init_driver_type_desc_t driver_type = {};
+ driver_type.stype = ZE_STRUCTURE_TYPE_INIT_DRIVER_TYPE_DESC;
+ driver_type.flags = ZE_INIT_DRIVER_TYPE_FLAG_GPU;
+ driver_type.pNext = nullptr;
+ uint32_t driverCount{0};
+
+ // Initialize and find all drivers
+ CALL_ZE_AND_CHECK(zeInitDrivers, &driverCount, nullptr, &driver_type);
+
+ llvm::SmallVector<ze_driver_handle_t> drivers(driverCount);
+ CALL_ZE_AND_CHECK(zeInitDrivers, &driverCount, drivers.data(), &driver_type);
+
+ for (auto driver : drivers) {
+ // Discover all the devices for a given driver
+ uint32_t deviceCount = 0;
+ CALL_ZE_AND_CHECK(zeDeviceGet, driver, &deviceCount, nullptr);
+
+ llvm::SmallVector<ze_device_handle_t> devices(deviceCount);
+ CALL_ZE_AND_CHECK(zeDeviceGet, driver, &deviceCount, devices.data());
+
+ for (auto device : devices) {
+ // Get device properties
+ ze_device_properties_t deviceProperties;
+ CALL_ZE_AND_CHECK(zeDeviceGetProperties, device, &deviceProperties);
+ // Print device name
+ llvm::outs() << deviceProperties.name << '\n';
+ }
+ }
+
+ return 0;
+}
+
+#endif // HAVE_LEVEL_ZERO_HEADERS
diff --git a/clang/tools/offload-arch/OffloadArch.cpp b/clang/tools/offload-arch/OffloadArch.cpp
index 74be40214a0ec..5043c00986a9b 100644
--- a/clang/tools/offload-arch/OffloadArch.cpp
+++ b/clang/tools/offload-arch/OffloadArch.cpp
@@ -21,6 +21,7 @@ enum VendorName {
all,
amdgpu,
nvptx,
+ intel,
};
static cl::opt<VendorName>
@@ -28,7 +29,8 @@ static cl::opt<VendorName>
cl::init(all),
cl::values(clEnumVal(all, "Print all GPUs (default)"),
clEnumVal(amdgpu, "Only print AMD GPUs"),
- clEnumVal(nvptx, "Only print NVIDIA GPUs")));
+ clEnumVal(nvptx, "Only print NVIDIA GPUs"),
+ clEnumVal(intel, "Only print Intel GPUs")));
cl::opt<bool> Verbose("verbose", cl::desc("Enable verbose output"),
cl::init(false), cl::cat(OffloadArchCategory));
@@ -40,6 +42,7 @@ static void PrintVersion(raw_ostream &OS) {
int printGPUsByKFD();
int printGPUsByHIP();
int printGPUsByCUDA();
+int printGPUsByLevelZero();
static int printAMD() {
#ifndef _WIN32
@@ -51,6 +54,7 @@ static int printAMD() {
}
static int printNVIDIA() { return printGPUsByCUDA(); }
+static int printIntel() { return printGPUsByLevelZero(); }
int main(int argc, char *argv[]) {
cl::HideUnrelatedOptions(OffloadArchCategory);
@@ -73,15 +77,22 @@ int main(int argc, char *argv[]) {
sys::path::stem(argv[0]).starts_with("amdgpu-arch");
bool NVIDIAOnly = Only == VendorName::nvptx ||
sys::path::stem(argv[0]).starts_with("nvptx-arch");
+ bool IntelOnly = Only == VendorName::intel ||
+ sys::path::stem(argv[0]).starts_with("intelgpu-arch");
+ bool All = !AMDGPUOnly && !NVIDIAOnly && !IntelOnly;
int NVIDIAResult = 0;
- if (!AMDGPUOnly)
+ if (NVIDIAOnly || All)
NVIDIAResult = printNVIDIA();
int AMDResult = 0;
- if (!NVIDIAOnly)
+ if (AMDGPUOnly || All)
AMDResult = printAMD();
+ int IntelResult = 0;
+ if (IntelOnly || All)
+ IntelResult = printIntel();
+
// We only failed if all cases returned an error.
- return AMDResult && NVIDIAResult;
+ return AMDResult && NVIDIAResult && IntelResult;
}
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
Contributor
Author
|
@sarnex @hansangbae @jhuber6 would you mind taking a look at this PR? |
jhuber6
reviewed
Sep 24, 2025
sarnex
reviewed
Sep 24, 2025
jhuber6
reviewed
Sep 24, 2025
jhuber6
reviewed
Sep 24, 2025
hansangbae
reviewed
Sep 24, 2025
jhuber6
reviewed
Sep 24, 2025
Co-authored-by: Joseph Huber <[email protected]>
jhuber6
reviewed
Sep 24, 2025
| if (!NVIDIAOnly) | ||
| AMDResult = printAMD(); | ||
| llvm::SmallVector<int> results(VendorTable.size()); | ||
| llvm::transform(VendorTable, results.begin(), [&](const auto &entry) { |
Contributor
There was a problem hiding this comment.
Can't we just do return llvm::all_of here now?
Contributor
Author
There was a problem hiding this comment.
As I mentioned llvm::all_of doesn't guarantee all elements will be visited. If one of them returns false the remaining ones won't be executed (as the result is already known to be false).
Contributor
There was a problem hiding this comment.
Oh yeah, then just a manual loop with result &= func() or use std::transform_reduce with logical and. I would assume the former is easier.
Contributor
There was a problem hiding this comment.
Also if you do a look remember you can do for (auto [Name, Func] : Table. Also I forgot to nag you about the LLVM naming convention, remember it's CamelCase.
Contributor
Author
There was a problem hiding this comment.
Did the loop, and I think I changed all the names too :)
Contributor
Author
|
@jhuber6 @sarnex @hansangbae any further comments? |
Contributor
Author
|
all done I think :) |
sarnex
reviewed
Sep 25, 2025
hansangbae
approved these changes
Sep 25, 2025
sarnex
approved these changes
Sep 25, 2025
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/66/builds/19756 Here is the relevant piece of the build log for the reference |
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/51/builds/24168 Here is the relevant piece of the build log for the reference |
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/27/builds/16608 Here is the relevant piece of the build log for the reference |
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/186/builds/12691 Here is the relevant piece of the build log for the reference |
Contributor
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/72/builds/15222 Here is the relevant piece of the build log for the reference |
Contributor
Author
|
many thanks @kazutakahirata ! |
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Oct 3, 2025
Co-authored-by: Joseph Huber <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
7 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.