facebookresearch/faiss
1
0
Fork 0
mirror of https://github.com/facebookresearch/faiss.git synced 2026-10-12 02:13:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

Win+AVX2 compat (#1681)

Browse Source 
Summary:
Upstreaming patches from conda-forge/faiss-split-feedstock#27, follow-up (sorta) to https://github.com/facebookresearch/faiss/issues/1600.

All these should be fairly uncontroversial, I think (mostly just oversights or stuff that never got
triggered on windows due to not having https://github.com/facebookresearch/faiss/issues/1680 so far).

Things work without `#include <faiss/impl/platform_macros.h>`, but I preferred to be explicit here,
because the `__SSEx__` macros (that are used in the affected files) are only defined there for windows.

Pull Request resolved: https://github.com/facebookresearch/faiss/pull/1681

Reviewed By: beauby

Differential Revision: D26454427

Pulled By: mdouze

fbshipit-source-id: 345e0ef45888f338e71bba004454a701572f9afb
This commit is contained in:
H. Vetinari
2021-02-17 08:05:55 -08:00
committed by Facebook GitHub Bot
parent 96d0d331a7
commit 6d8afffadd
3 changed files with 23 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
faiss/impl/ScalarQuantizer.cpp
+1
View File
@@ -13,6 +13,7 @@
#include <algorithm>
#include <omp.h>
#include <faiss/impl/platform_macros.h>
#ifdef __SSE__
#include <immintrin.h>
faiss/impl/platform_macros.h
+19
View File
@@ -49,8 +49,27 @@ inline int __builtin_clzll(uint64_t x) {
return (int)__lzcnt64(x);
}
#define __builtin_popcount __popcnt
#define __builtin_popcountl __popcnt64
// MSVC does not define __SSEx__, and _M_IX86_FP is only defined on 32-bit processors
// cf. https://docs.microsoft.com/en-us/cpp/preprocessor/predefined-macros
#ifdef __AVX__
#define __SSE__ 1
#define __SSE2__ 1
#define __SSE3__ 1
#define __SSE4_1__ 1
#define __SSE4_2__ 1
#endif
// MSVC sets FMA and F16C automatically when using AVX2
// Ref. FMA (under /arch:AVX2): https://docs.microsoft.com/en-us/cpp/build/reference/arch-x64
// Ref. F16C (2nd paragraph): https://walbourn.github.io/directxmath-avx2/
#ifdef __AVX2__
#define __FMA__ 1
#define __F16C__ 1
#endif
#else
/*******************************************************
* Linux and OSX
faiss/utils/distances_simd.cpp
+3 -2
View File
@@ -16,6 +16,7 @@
#include <faiss/utils/simdlib.h>
#include <faiss/impl/FaissAssert.h>
#include <faiss/impl/platform_macros.h>
#ifdef __SSE3__
#include <immintrin.h>
@@ -165,7 +166,7 @@ void fvec_inner_products_ny_ref (float * ip,
static inline __m128 masked_read (int d, const float *x)
{
assert (0 <= d && d < 4);
__attribute__((__aligned__(16))) float buf[4] = {0, 0, 0, 0};
ALIGNED(16) float buf[4] = {0, 0, 0, 0};
switch (d) {
case 3:
buf[2] = x[2];
@@ -987,7 +988,7 @@ void compute_PQ_dis_tables_dsub2(
simd8float32 centroids[8];
for (int k = 0; k < 8; k++) {
float centroid[8] __attribute__((aligned(32)));
ALIGNED(32) float centroid[8];
size_t wp = 0;
size_t rp = (m0 * ksub + k + k0) * 2;
for (int m = m0; m < m1; m++) {