[Offload][Conformance] Add tests for single-precision math functions #152013
+1,603
−17
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
|
@llvm/pr-subscribers-offload Author: Leandro Lacerda (leandrolcampos) ChangesThis patch adds a new set of conformance tests for single-precision math functions provided by the LLVM libm for GPUs. The functions included in this set were selected based on the following criteria:
Patch is 62.08 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/152013.diff 30 Files Affected:
diff --git a/offload/unittests/Conformance/device_code/CMakeLists.txt b/offload/unittests/Conformance/device_code/CMakeLists.txt
index 9cbd11096292c..a7f504dbadc9b 100644
--- a/offload/unittests/Conformance/device_code/CMakeLists.txt
+++ b/offload/unittests/Conformance/device_code/CMakeLists.txt
@@ -1,4 +1,4 @@
-add_offload_test_device_code(LLVMLibm.cpp llvm-libm -stdlib -fno-builtin)
+add_offload_test_device_code(LLVMLibm.cpp llvm-libm -O3 -stdlib -fno-builtin -I.)
add_custom_target(conformance_device_binaries DEPENDS llvm-libm.bin)
set(OFFLOAD_CONFORMANCE_DEVICE_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE)
diff --git a/offload/unittests/Conformance/device_code/Common.hpp b/offload/unittests/Conformance/device_code/Common.hpp
new file mode 100644
index 0000000000000..5f9c3bf673559
--- /dev/null
+++ b/offload/unittests/Conformance/device_code/Common.hpp
@@ -0,0 +1,38 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains common utilities for defining device kernel wrappers to
+/// math functions.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef COMMON_HPP
+#define COMMON_HPP
+
+#include <gpuintrin.h>
+#include <stddef.h>
+#include <stdint.h>
+
+namespace common {
+
+typedef _Float16 float16;
+
+template <auto Func, typename OutType, typename... InTypes>
+__attribute__((always_inline)) void
+runKernelBody(size_t NumElements, OutType *Out, const InTypes *...Ins) {
+ uint32_t Index =
+ __gpu_num_threads_x() * __gpu_block_id_x() + __gpu_thread_id_x();
+
+ if (Index < NumElements) {
+ Out[Index] = Func(Ins[Index]...);
+ }
+}
+} // namespace common
+
+#endif // COMMON_HPP
diff --git a/offload/unittests/Conformance/device_code/LLVMLibm.cpp b/offload/unittests/Conformance/device_code/LLVMLibm.cpp
index 2c3d9bc5bf5cf..8d9412ee73c44 100644
--- a/offload/unittests/Conformance/device_code/LLVMLibm.cpp
+++ b/offload/unittests/Conformance/device_code/LLVMLibm.cpp
@@ -12,29 +12,173 @@
///
//===----------------------------------------------------------------------===//
+#include "Common.hpp"
+
#include <gpuintrin.h>
#include <math.h>
#include <stddef.h>
-#include <stdint.h>
-typedef _Float16 float16;
+using namespace common;
+
+//===----------------------------------------------------------------------===//
+// Helpers
+//===----------------------------------------------------------------------===//
+
+static inline float sincosfSin(float X) {
+ float SinX, CosX;
+ sincosf(X, &SinX, &CosX);
+ return SinX;
+}
+
+static inline float sincosfCos(float X) {
+ float SinX, CosX;
+ sincosf(X, &SinX, &CosX);
+ return CosX;
+}
+
+//===----------------------------------------------------------------------===//
+// Kernels
+//===----------------------------------------------------------------------===//
extern "C" {
+__gpu_kernel void acosfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<acosf>(NumElements, Out, X);
+}
+
+__gpu_kernel void acoshfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<acoshf>(NumElements, Out, X);
+}
+
+__gpu_kernel void asinfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<asinf>(NumElements, Out, X);
+}
+
+__gpu_kernel void asinhfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<asinhf>(NumElements, Out, X);
+}
+
+__gpu_kernel void atanfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<atanf>(NumElements, Out, X);
+}
+
+__gpu_kernel void atanhfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<atanhf>(NumElements, Out, X);
+}
+
+__gpu_kernel void cbrtfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<cbrtf>(NumElements, Out, X);
+}
+
+__gpu_kernel void cosfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<cosf>(NumElements, Out, X);
+}
+
+__gpu_kernel void coshfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<coshf>(NumElements, Out, X);
+}
+
+__gpu_kernel void cospifKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<cospif>(NumElements, Out, X);
+}
+
+__gpu_kernel void erffKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<erff>(NumElements, Out, X);
+}
+
+__gpu_kernel void expfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<expf>(NumElements, Out, X);
+}
+
+__gpu_kernel void exp10fKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<exp10f>(NumElements, Out, X);
+}
+
+__gpu_kernel void exp2fKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<exp2f>(NumElements, Out, X);
+}
+
+__gpu_kernel void expm1fKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<expm1f>(NumElements, Out, X);
+}
+
__gpu_kernel void hypotf16Kernel(const float16 *X, float16 *Y, float16 *Out,
- size_t NumElements) {
- uint32_t Index =
- __gpu_num_threads_x() * __gpu_block_id_x() + __gpu_thread_id_x();
+ size_t NumElements) noexcept {
+ runKernelBody<hypotf16>(NumElements, Out, X, Y);
+}
+
+__gpu_kernel void logfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<logf>(NumElements, Out, X);
+}
+
+__gpu_kernel void log10fKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<log10f>(NumElements, Out, X);
+}
- if (Index < NumElements)
- Out[Index] = hypotf16(X[Index], Y[Index]);
+__gpu_kernel void log1pfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<log1pf>(NumElements, Out, X);
}
-__gpu_kernel void logfKernel(const float *X, float *Out, size_t NumElements) {
- uint32_t Index =
- __gpu_num_threads_x() * __gpu_block_id_x() + __gpu_thread_id_x();
+__gpu_kernel void log2fKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<log2f>(NumElements, Out, X);
+}
+
+__gpu_kernel void sinfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<sinf>(NumElements, Out, X);
+}
+
+__gpu_kernel void sincosfSinKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<sincosfSin>(NumElements, Out, X);
+}
+
+__gpu_kernel void sincosfCosKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<sincosfCos>(NumElements, Out, X);
+}
+
+__gpu_kernel void sinhfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<sinhf>(NumElements, Out, X);
+}
+
+__gpu_kernel void sinpifKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<sinpif>(NumElements, Out, X);
+}
+
+__gpu_kernel void tanfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<tanf>(NumElements, Out, X);
+}
+
+__gpu_kernel void tanhfKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<tanhf>(NumElements, Out, X);
+}
- if (Index < NumElements)
- Out[Index] = logf(X[Index]);
+__gpu_kernel void tanpifKernel(const float *X, float *Out,
+ size_t NumElements) noexcept {
+ runKernelBody<tanpif>(NumElements, Out, X);
}
} // extern "C"
diff --git a/offload/unittests/Conformance/tests/AcosfTest.cpp b/offload/unittests/Conformance/tests/AcosfTest.cpp
new file mode 100644
index 0000000000000..e69ee3b7d1fd7
--- /dev/null
+++ b/offload/unittests/Conformance/tests/AcosfTest.cpp
@@ -0,0 +1,53 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains the conformance test of the acosf function.
+///
+//===----------------------------------------------------------------------===//
+
+#include "mathtest/CommandLineExtras.hpp"
+#include "mathtest/ExhaustiveGenerator.hpp"
+#include "mathtest/IndexedRange.hpp"
+#include "mathtest/TestConfig.hpp"
+#include "mathtest/TestRunner.hpp"
+
+#include "llvm/ADT/StringRef.h"
+
+#include <cstdlib>
+#include <math.h>
+
+namespace mathtest {
+
+template <> struct FunctionConfig<acosf> {
+ static constexpr llvm::StringRef Name = "acosf";
+ static constexpr llvm::StringRef KernelName = "acosfKernel";
+
+ // Source: The Khronos Group, The OpenCL C Specification v3.0.19, Sec. 7.4,
+ // Table 65, Khronos Registry [July 10, 2025].
+ static constexpr uint64_t UlpTolerance = 4;
+};
+} // namespace mathtest
+
+int main(int argc, const char **argv) {
+ llvm::cl::ParseCommandLineOptions(argc, argv,
+ "Conformance test of the acosf function");
+
+ using namespace mathtest;
+
+ IndexedRange<float> Range;
+ ExhaustiveGenerator<float> Generator(Range);
+
+ const auto Configs = cl::getTestConfigs();
+ const llvm::StringRef DeviceBinaryDir = DEVICE_BINARY_DIR;
+ const bool IsVerbose = cl::IsVerbose;
+
+ bool Passed = runTests<acosf>(Generator, Configs, DeviceBinaryDir, IsVerbose);
+
+ return Passed ? EXIT_SUCCESS : EXIT_FAILURE;
+}
diff --git a/offload/unittests/Conformance/tests/AcoshfTest.cpp b/offload/unittests/Conformance/tests/AcoshfTest.cpp
new file mode 100644
index 0000000000000..4d43a661c3535
--- /dev/null
+++ b/offload/unittests/Conformance/tests/AcoshfTest.cpp
@@ -0,0 +1,57 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains the conformance test of the acoshf function.
+///
+//===----------------------------------------------------------------------===//
+
+#include "mathtest/CommandLineExtras.hpp"
+#include "mathtest/ExhaustiveGenerator.hpp"
+#include "mathtest/IndexedRange.hpp"
+#include "mathtest/TestConfig.hpp"
+#include "mathtest/TestRunner.hpp"
+
+#include "llvm/ADT/StringRef.h"
+
+#include <cstdlib>
+#include <limits>
+#include <math.h>
+
+namespace mathtest {
+
+template <> struct FunctionConfig<acoshf> {
+ static constexpr llvm::StringRef Name = "acoshf";
+ static constexpr llvm::StringRef KernelName = "acoshfKernel";
+
+ // Source: The Khronos Group, The OpenCL C Specification v3.0.19, Sec. 7.4,
+ // Table 65, Khronos Registry [July 10, 2025].
+ static constexpr uint64_t UlpTolerance = 4;
+};
+} // namespace mathtest
+
+int main(int argc, const char **argv) {
+ llvm::cl::ParseCommandLineOptions(argc, argv,
+ "Conformance test of the acoshf function");
+
+ using namespace mathtest;
+
+ IndexedRange<float> Range(/*Begin=*/1.0f,
+ /*End=*/std::numeric_limits<float>::infinity(),
+ /*Inclusive=*/true);
+ ExhaustiveGenerator<float> Generator(Range);
+
+ const auto Configs = cl::getTestConfigs();
+ const llvm::StringRef DeviceBinaryDir = DEVICE_BINARY_DIR;
+ const bool IsVerbose = cl::IsVerbose;
+
+ bool Passed =
+ runTests<acoshf>(Generator, Configs, DeviceBinaryDir, IsVerbose);
+
+ return Passed ? EXIT_SUCCESS : EXIT_FAILURE;
+}
diff --git a/offload/unittests/Conformance/tests/AsinfTest.cpp b/offload/unittests/Conformance/tests/AsinfTest.cpp
new file mode 100644
index 0000000000000..991f79b111efe
--- /dev/null
+++ b/offload/unittests/Conformance/tests/AsinfTest.cpp
@@ -0,0 +1,53 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains the conformance test of the asinf function.
+///
+//===----------------------------------------------------------------------===//
+
+#include "mathtest/CommandLineExtras.hpp"
+#include "mathtest/ExhaustiveGenerator.hpp"
+#include "mathtest/IndexedRange.hpp"
+#include "mathtest/TestConfig.hpp"
+#include "mathtest/TestRunner.hpp"
+
+#include "llvm/ADT/StringRef.h"
+
+#include <cstdlib>
+#include <math.h>
+
+namespace mathtest {
+
+template <> struct FunctionConfig<asinf> {
+ static constexpr llvm::StringRef Name = "asinf";
+ static constexpr llvm::StringRef KernelName = "asinfKernel";
+
+ // Source: The Khronos Group, The OpenCL C Specification v3.0.19, Sec. 7.4,
+ // Table 65, Khronos Registry [July 10, 2025].
+ static constexpr uint64_t UlpTolerance = 4;
+};
+} // namespace mathtest
+
+int main(int argc, const char **argv) {
+ llvm::cl::ParseCommandLineOptions(argc, argv,
+ "Conformance test of the asinf function");
+
+ using namespace mathtest;
+
+ IndexedRange<float> Range;
+ ExhaustiveGenerator<float> Generator(Range);
+
+ const auto Configs = cl::getTestConfigs();
+ const llvm::StringRef DeviceBinaryDir = DEVICE_BINARY_DIR;
+ const bool IsVerbose = cl::IsVerbose;
+
+ bool Passed = runTests<asinf>(Generator, Configs, DeviceBinaryDir, IsVerbose);
+
+ return Passed ? EXIT_SUCCESS : EXIT_FAILURE;
+}
diff --git a/offload/unittests/Conformance/tests/AsinhfTest.cpp b/offload/unittests/Conformance/tests/AsinhfTest.cpp
new file mode 100644
index 0000000000000..9439383772314
--- /dev/null
+++ b/offload/unittests/Conformance/tests/AsinhfTest.cpp
@@ -0,0 +1,54 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains the conformance test of the asinhf function.
+///
+//===----------------------------------------------------------------------===//
+
+#include "mathtest/CommandLineExtras.hpp"
+#include "mathtest/ExhaustiveGenerator.hpp"
+#include "mathtest/IndexedRange.hpp"
+#include "mathtest/TestConfig.hpp"
+#include "mathtest/TestRunner.hpp"
+
+#include "llvm/ADT/StringRef.h"
+
+#include <cstdlib>
+#include <math.h>
+
+namespace mathtest {
+
+template <> struct FunctionConfig<asinhf> {
+ static constexpr llvm::StringRef Name = "asinhf";
+ static constexpr llvm::StringRef KernelName = "asinhfKernel";
+
+ // Source: The Khronos Group, The OpenCL C Specification v3.0.19, Sec. 7.4,
+ // Table 65, Khronos Registry [July 10, 2025].
+ static constexpr uint64_t UlpTolerance = 4;
+};
+} // namespace mathtest
+
+int main(int argc, const char **argv) {
+ llvm::cl::ParseCommandLineOptions(argc, argv,
+ "Conformance test of the asinhf function");
+
+ using namespace mathtest;
+
+ IndexedRange<float> Range;
+ ExhaustiveGenerator<float> Generator(Range);
+
+ const auto Configs = cl::getTestConfigs();
+ const llvm::StringRef DeviceBinaryDir = DEVICE_BINARY_DIR;
+ const bool IsVerbose = cl::IsVerbose;
+
+ bool Passed =
+ runTests<asinhf>(Generator, Configs, DeviceBinaryDir, IsVerbose);
+
+ return Passed ? EXIT_SUCCESS : EXIT_FAILURE;
+}
diff --git a/offload/unittests/Conformance/tests/AtanfTest.cpp b/offload/unittests/Conformance/tests/AtanfTest.cpp
new file mode 100644
index 0000000000000..64068ef6bff0c
--- /dev/null
+++ b/offload/unittests/Conformance/tests/AtanfTest.cpp
@@ -0,0 +1,53 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains the conformance test of the atanf function.
+///
+//===----------------------------------------------------------------------===//
+
+#include "mathtest/CommandLineExtras.hpp"
+#include "mathtest/ExhaustiveGenerator.hpp"
+#include "mathtest/IndexedRange.hpp"
+#include "mathtest/TestConfig.hpp"
+#include "mathtest/TestRunner.hpp"
+
+#include "llvm/ADT/StringRef.h"
+
+#include <cstdlib>
+#include <math.h>
+
+namespace mathtest {
+
+template <> struct FunctionConfig<atanf> {
+ static constexpr llvm::StringRef Name = "atanf";
+ static constexpr llvm::StringRef KernelName = "atanfKernel";
+
+ // Source: The Khronos Group, The OpenCL C Specification v3.0.19, Sec. 7.4,
+ // Table 65, Khronos Registry [July 10, 2025].
+ static constexpr uint64_t UlpTolerance = 5;
+};
+} // namespace mathtest
+
+int main(int argc, const char **argv) {
+ llvm::cl::ParseCommandLineOptions(argc, argv,
+ "Conformance test of the atanf function");
+
+ using namespace mathtest;
+
+ IndexedRange<float> Range;
+ ExhaustiveGenerator<float> Generator(Range);
+
+ const auto Configs = cl::getTestConfigs();
+ const llvm::StringRef DeviceBinaryDir = DEVICE_BINARY_DIR;
+ const bool IsVerbose = cl::IsVerbose;
+
+ bool Passed = runTests<atanf>(Generator, Configs, DeviceBinaryDir, IsVerbose);
+
+ return Passed ? EXIT_SUCCESS : EXIT_FAILURE;
+}
diff --git a/offload/unittests/Conformance/tests/AtanhfTest.cpp b/offload/unittests/Conformance/tests/AtanhfTest.cpp
new file mode 100644
index 0000000000000..66b934c2f925e
--- /dev/null
+++ b/offload/unittests/Conformance/tests/AtanhfTest.cpp
@@ -0,0 +1,56 @@
+//===----------------------------------------------------------------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This file contains the conformance test of the atanhf function.
+///
+//===----------------------------------------------------------------------===//
+
+#include "mathtest/CommandLineExtras.hpp"
+#include "mathtest/ExhaustiveGenerator.hpp"
+#include "mathtest/IndexedRange.hpp"
+#include "mathtest/TestConfig.hpp"
+#include "mathtest/TestRunner.hpp"
+
+#include "llvm/ADT/StringRef.h"
+
+#include <cstdlib>
+#include <math.h>
+
+namespace mathtest {
+
+template <> struct FunctionConfig<atanhf> {
+ static constexpr llvm::StringRef Name = "atanhf";
+ static constexpr llvm::StringRef KernelName = "atanhfKernel";
+
+ // Source: The Khronos Group, The OpenCL C Specification v3.0.19, Sec. 7.4,
+ // Table 65, Khronos Registry [July 10, 2025].
+ static constexpr uint64_t UlpTolerance = 5;
+};
+} ...
[truncated]
|
|
Test Results for Single-Precision Math Functions in
Device: NVIDIA GeForce RTX 4070 Laptop GPU |
jhuber6
approved these changes
Aug 5, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
This patch adds a new set of conformance tests for single-precision math functions provided by the LLVM libm for GPUs.
The functions included in this set were selected based on the following criteria:
libc/src/math/generic(i.e., it is not just a wrapper around a compiler built-in).