mirror of
https://github.com/facebookresearch/faiss.git
synced 2026-10-12 02:13:36 +00:00
Gergely Szilvasy
56ae01622f
Summary: Pull Request resolved: https://github.com/facebookresearch/faiss/pull/4780 Introduces Dynamic Dispatch (DD) infrastructure for FAISS, enabling runtime SIMD level selection via buckconfig parameterization. Build Configuration: - Enable with: buck build -c faiss.dynamic_dispatch=true //faiss:faiss - is_dd_enabled() function in fbcode.bzl reads buckconfig - DD mode is purely additive - all static build options remain available Core Components: simd_levels.h/cpp - Runtime SIMD detection: - SIMDLevel enum: NONE, AVX2, AVX512, AVX512_SPR, ARM_NEON - SIMDConfig class: auto-detects CPU capabilities via CPUID - FAISS_SIMD_LEVEL env var override for testing/debugging - Guarded by FAISS_ENABLE_DD preprocessor flag DISPATCH_SIMDLevel macro - Dual-mode dispatch: - DD mode: runtime switch on SIMDConfig::level - Static mode: compiles to direct call (zero overhead) simd_dispatch.bzl - Build configuration: - Per-architecture SIMD level configuration - Per-file compiler flags (SIMD files get -mavx2/-mavx512f, common files get baseline) - Generates source file lists and flags for each SIMD level Test Infrastructure: - DD-only tests use target_compatible_with to exclude from static builds - Static tests continue to work unchanged - Both modes share the same test logic where applicable Reviewed By: mdouze, mnorris11 Differential Revision: D72937710 fbshipit-source-id: c9f06595f1ac6966ab1d2f3b5bc25afcf25638a4
76 lines
2.3 KiB
C++
76 lines
2.3 KiB
C++
/*
|
|
* Copyright (c) Meta Platforms, Inc. and affiliates.
|
|
*
|
|
* This source code is licensed under the MIT license found in the
|
|
* LICENSE file in the root directory of this source tree.
|
|
*/
|
|
|
|
// AVX2 hardware execution tests - this file is compiled with AVX2 flags.
|
|
// Verifies that AVX2 instructions execute correctly when SIMDConfig reports
|
|
// AVX2 support.
|
|
|
|
#if defined(__x86_64__) || defined(_M_X64)
|
|
|
|
#include <gtest/gtest.h>
|
|
#include <immintrin.h>
|
|
#include <setjmp.h>
|
|
#include <signal.h>
|
|
#include <vector>
|
|
|
|
#include <faiss/utils/simd_levels.h>
|
|
|
|
namespace {
|
|
|
|
// NOLINTNEXTLINE(facebook-avoid-non-const-global-variables)
|
|
static jmp_buf jmpbuf;
|
|
|
|
[[noreturn]] static void sigill_handler(int /* sig */) {
|
|
longjmp(jmpbuf, 1);
|
|
}
|
|
|
|
std::pair<bool, std::vector<int>> try_execute_avx2(std::vector<int> (*func)()) {
|
|
signal(SIGILL, sigill_handler);
|
|
if (setjmp(jmpbuf) == 0) {
|
|
auto result = func();
|
|
signal(SIGILL, SIG_DFL);
|
|
return std::make_pair(true, result);
|
|
} else {
|
|
signal(SIGILL, SIG_DFL);
|
|
return std::make_pair(false, std::vector<int>());
|
|
}
|
|
}
|
|
|
|
std::vector<int> run_avx2_computation() {
|
|
alignas(32) int result[8];
|
|
alignas(32) int input1[8] = {1, 2, 3, 4, 5, 6, 7, 8};
|
|
alignas(32) int input2[8] = {8, 7, 6, 5, 4, 3, 2, 1};
|
|
|
|
__m256i vec1 = _mm256_load_si256(reinterpret_cast<__m256i*>(input1));
|
|
__m256i vec2 = _mm256_load_si256(reinterpret_cast<__m256i*>(input2));
|
|
__m256i vec_result = _mm256_add_epi32(vec1, vec2);
|
|
_mm256_store_si256(reinterpret_cast<__m256i*>(result), vec_result);
|
|
|
|
return {result, result + 8};
|
|
}
|
|
|
|
} // namespace
|
|
|
|
TEST(SIMDConfig, successful_avx2_execution_on_x86arch) {
|
|
if (faiss::SIMDConfig::is_simd_level_available(faiss::SIMDLevel::AVX2)) {
|
|
auto actual_result = try_execute_avx2(run_avx2_computation);
|
|
EXPECT_TRUE(actual_result.first);
|
|
auto expected_result_vector = std::vector<int>(8, 9);
|
|
EXPECT_EQ(actual_result.second, expected_result_vector);
|
|
}
|
|
}
|
|
|
|
TEST(SIMDConfig, on_avx512f_supported_we_should_have_avx2_support_as_well) {
|
|
if (faiss::SIMDConfig::is_simd_level_available(faiss::SIMDLevel::AVX512)) {
|
|
EXPECT_TRUE(
|
|
faiss::SIMDConfig::is_simd_level_available(
|
|
faiss::SIMDLevel::AVX2));
|
|
}
|
|
}
|
|
|
|
#endif // defined(__x86_64__) || defined(_M_X64)
|