gitea/SDL3_fork
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

cmake: Detect AVX + allow build system to disable Intel intrinsics

Browse Source
This commit is contained in:
Anonymous Maarten
2023-02-25 00:21:15 +01:00
committed by Anonymous Maarten
parent 683411e96f
commit 4681240241
16 changed files with 175 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
src/SDL_internal.h
View File

@@ -185,6 +185,46 @@
#include <SDL3/SDL.h>
#include <SDL3/SDL_intrin.h>
#ifdef __ARM_NEON
#define HAVE_NEON_INTRINSICS 1
#endif
#if defined(__MMX__) && !defined(SDL_DISABLE_MMX)
#define HAVE_MMX_INTRINSICS 1
#endif
#if defined(__SSE__) && !defined(SDL_DISABLE_SSE)
#define HAVE_SSE_INTRINSICS 1
#endif
#if defined(__SSE2__) && !defined(SDL_DISABLE_SSE2)
#define HAVE_SSE2_INTRINSICS 1
#endif
#if defined(__SSE3__) && !defined(SDL_DISABLE_SSE3)
#define HAVE_SSE3_INTRINSICS 1
#endif
#if defined(__AVX__) && !defined(SDL_DISABLE_AVX)
#define HAVE_AVX_INTRINSICS 1
#endif
#if defined __clang__
#if (!__has_attribute(target))
#undef HAVE_AVX_INTRINSICS
#endif
#if (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX__)
#undef HAVE_AVX_INTRINSICS
#endif
#elif defined __GNUC__
#if (__GNUC__ < 4) || (__GNUC__ == 4 && __GNUC_MINOR__ < 9)
#undef HAVE_AVX_INTRINSICS
#endif
#endif
#define SDL_MAIN_NOIMPL /* don't drag in header-only implementation of SDL_main */
#include <SDL3/SDL_main.h>

29
src/audio/SDL_audiocvt.c
View File

@@ -29,35 +29,6 @@
#define DEBUG_AUDIOSTREAM 0
#ifdef __ARM_NEON
#define HAVE_NEON_INTRINSICS 1
#endif
#ifdef __SSE__
#define HAVE_SSE_INTRINSICS 1
#endif
#ifdef __SSE3__
#define HAVE_SSE3_INTRINSICS 1
#endif
#if defined(HAVE_IMMINTRIN_H) && !defined(SDL_DISABLE_IMMINTRIN_H)
#define HAVE_AVX_INTRINSICS 1
#endif
#if defined __clang__
#if (!__has_attribute(target))
#undef HAVE_AVX_INTRINSICS
#endif
#if (defined(_MSC_VER) || defined(__SCE__)) && !defined(__AVX__)
#undef HAVE_AVX_INTRINSICS
#endif
#elif defined __GNUC__
#if (__GNUC__ < 4) || (__GNUC__ == 4 && __GNUC_MINOR__ < 9)
#undef HAVE_AVX_INTRINSICS
#endif
#endif
/**
* Initialize an SDL_AudioCVT structure for conversion.
*

4
src/audio/SDL_audiotypecvt.c
View File

@@ -27,10 +27,6 @@
#define HAVE_NEON_INTRINSICS 1
#endif
#ifdef __SSE2__
#define HAVE_SSE2_INTRINSICS 1
#endif
#if defined(__x86_64__) && HAVE_SSE2_INTRINSICS
#define NEED_SCALAR_CONVERTER_FALLBACKS 0 /* x86_64 guarantees SSE2. */
#elif __MACOS__ && HAVE_SSE2_INTRINSICS

20
src/video/SDL_blit_A.c
View File

@@ -166,7 +166,7 @@ static void BlitNto1SurfaceAlphaKey(SDL_BlitInfo *info)
}
}
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
/* fast RGB888->(A)RGB888 blending with surface alpha=128 special case */
static void BlitRGBtoRGBSurfaceAlpha128MMX(SDL_BlitInfo *info)
@@ -409,7 +409,7 @@ static void BlitRGBtoRGBPixelAlphaMMX(SDL_BlitInfo *info)
_mm_empty();
}
#endif /* __MMX__ */
#endif /* HAVE_MMX_INTRINSICS */
#if SDL_ARM_SIMD_BLITTERS
void BlitARGBto565PixelAlphaARMSIMDAsm(int32_t w, int32_t h, uint16_t *dst, int32_t dst_stride, uint32_t *src, int32_t src_stride);
@@ -750,7 +750,7 @@ static void Blit16to16SurfaceAlpha128(SDL_BlitInfo *info, Uint16 mask)
}
}
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
/* fast RGB565->RGB565 blending with surface alpha */
static void Blit565to565SurfaceAlphaMMX(SDL_BlitInfo *info)
@@ -1025,7 +1025,7 @@ static void Blit555to555SurfaceAlphaMMX(SDL_BlitInfo *info)
}
}
#endif /* __MMX__ */
#endif /* HAVE_MMX_INTRINSICS */
/* fast RGB565->RGB565 blending with surface alpha */
static void Blit565to565SurfaceAlpha(SDL_BlitInfo *info)
@@ -1357,15 +1357,13 @@ SDL_CalculateBlitA(SDL_Surface *surface)
case 4:
if (sf->Rmask == df->Rmask && sf->Gmask == df->Gmask && sf->Bmask == df->Bmask && sf->BytesPerPixel == 4) {
#if defined(__MMX__)
#if HAVE_MMX_INTRINSICS
if (sf->Rshift % 8 == 0 && sf->Gshift % 8 == 0 && sf->Bshift % 8 == 0 && sf->Ashift % 8 == 0 && sf->Aloss == 0) {
#ifdef __MMX__
if (SDL_HasMMX()) {
return BlitRGBtoRGBPixelAlphaMMX;
}
#endif
}
#endif /* __MMX__ */
#endif /* HAVE_MMX_INTRINSICS */
if (sf->Amask == 0xff000000) {
#if SDL_ARM_NEON_BLITTERS
if (SDL_HasNEON()) {
@@ -1407,7 +1405,7 @@ SDL_CalculateBlitA(SDL_Surface *surface)
case 2:
if (surface->map->identity) {
if (df->Gmask == 0x7e0) {
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
if (SDL_HasMMX()) {
return Blit565to565SurfaceAlphaMMX;
} else
@@ -1416,7 +1414,7 @@ SDL_CalculateBlitA(SDL_Surface *surface)
return Blit565to565SurfaceAlpha;
}
} else if (df->Gmask == 0x3e0) {
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
if (SDL_HasMMX()) {
return Blit555to555SurfaceAlphaMMX;
} else
@@ -1430,7 +1428,7 @@ SDL_CalculateBlitA(SDL_Surface *surface)
case 4:
if (sf->Rmask == df->Rmask && sf->Gmask == df->Gmask && sf->Bmask == df->Bmask && sf->BytesPerPixel == 4) {
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
if (sf->Rshift % 8 == 0 && sf->Gshift % 8 == 0 && sf->Bshift % 8 == 0 && SDL_HasMMX()) {
return BlitRGBtoRGBSurfaceAlphaMMX;
}

12
src/video/SDL_blit_copy.c
View File

@@ -23,7 +23,7 @@
#include "SDL_blit.h"
#include "SDL_blit_copy.h"
#ifdef __SSE__
#if HAVE_SSE_INTRINSICS
/* This assumes 16-byte aligned src and dst */
static SDL_INLINE void SDL_memcpySSE(Uint8 *dst, const Uint8 *src, int len)
{
@@ -48,9 +48,9 @@ static SDL_INLINE void SDL_memcpySSE(Uint8 *dst, const Uint8 *src, int len)
SDL_memcpy(dst, src, len & 63);
}
}
#endif /* __SSE__ */
#endif /* HAVE_SSE_INTRINSICS */
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
#ifdef _MSC_VER
#pragma warning(disable : 4799)
#endif
@@ -81,7 +81,7 @@ static SDL_INLINE void SDL_memcpyMMX(Uint8 *dst, const Uint8 *src, int len)
SDL_memcpy(dst + skip, src + skip, remain);
}
}
#endif /* __MMX__ */
#endif /* HAVE_MMX_INTRINSICS */
void SDL_BlitCopy(SDL_BlitInfo *info)
{
@@ -122,7 +122,7 @@ void SDL_BlitCopy(SDL_BlitInfo *info)
return;
}
#ifdef __SSE__
#if HAVE_SSE_INTRINSICS
if (SDL_HasSSE() &&
!((uintptr_t)src & 15) && !(srcskip & 15) &&
!((uintptr_t)dst & 15) && !(dstskip & 15)) {
@@ -135,7 +135,7 @@ void SDL_BlitCopy(SDL_BlitInfo *info)
}
#endif
#ifdef __MMX__
#if HAVE_MMX_INTRINSICS
if (SDL_HasMMX() && !(srcskip & 7) && !(dstskip & 7)) {
while (h--) {
SDL_memcpyMMX(dst, src, w);

8
src/video/SDL_fillrect.c
View File

@@ -22,7 +22,7 @@
#include "SDL_blit.h"
#ifdef __SSE__
#if HAVE_SSE_INTRINSICS
/* *INDENT-OFF* */ /* clang-format off */
#if defined(_MSC_VER) && !defined(__clang__)
@@ -376,7 +376,7 @@ int SDL_FillSurfaceRects(SDL_Surface *dst, const SDL_Rect *rects, int count,
{
color |= (color << 8);
color |= (color << 16);
#ifdef __SSE__
#if HAVE_SSE_INTRINSICS
if (SDL_HasSSE()) {
fill_function = SDL_FillSurfaceRect1SSE;
break;
@@ -389,7 +389,7 @@ int SDL_FillSurfaceRects(SDL_Surface *dst, const SDL_Rect *rects, int count,
case 2:
{
color |= (color << 16);
#ifdef __SSE__
#if HAVE_SSE_INTRINSICS
if (SDL_HasSSE()) {
fill_function = SDL_FillSurfaceRect2SSE;
break;
@@ -408,7 +408,7 @@ int SDL_FillSurfaceRects(SDL_Surface *dst, const SDL_Rect *rects, int count,
case 4:
{
#ifdef __SSE__
#if HAVE_SSE_INTRINSICS
if (SDL_HasSSE()) {
fill_function = SDL_FillSurfaceRect4SSE;
break;

4
src/video/SDL_stretch.c
View File

@@ -332,10 +332,6 @@ static int scale_mat(const Uint32 *src, int src_w, int src_h, int src_pitch,
return 0;
}
#if defined(__SSE2__)
#define HAVE_SSE2_INTRINSICS 1
#endif
#if defined(__ARM_NEON)
#define HAVE_NEON_INTRINSICS 1
#define CAST_uint8x8_t (uint8x8_t)

40
src/video/SDL_yuv.c
View File

@@ -310,7 +310,7 @@ static SDL_bool yuv_rgb_sse(
Uint8 *rgb, Uint32 rgb_stride,
YCbCrType yuv_type)
{
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (!SDL_HasSSE2()) {
return SDL_FALSE;
}
@@ -1114,7 +1114,7 @@ static int SDL_ConvertPixels_PackUVPlanes_to_NV(int width, int height, const voi
const Uint8 *src1, *src2;
Uint8 *dstUV;
Uint8 *tmp = NULL;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
@@ -1144,7 +1144,7 @@ static int SDL_ConvertPixels_PackUVPlanes_to_NV(int width, int height, const voi
y = UVheight;
while (y--) {
x = UVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
while (x >= 16) {
__m128i u = _mm_loadu_si128((__m128i *)src1);
@@ -1187,7 +1187,7 @@ static int SDL_ConvertPixels_SplitNV_to_UVPlanes(int width, int height, const vo
const Uint8 *srcUV;
Uint8 *dst1, *dst2;
Uint8 *tmp = NULL;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
@@ -1217,7 +1217,7 @@ static int SDL_ConvertPixels_SplitNV_to_UVPlanes(int width, int height, const vo
y = UVheight;
while (y--) {
x = UVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
__m128i mask = _mm_set1_epi16(0x00FF);
while (x >= 16) {
@@ -1264,7 +1264,7 @@ static int SDL_ConvertPixels_SwapNV(int width, int height, const void *src, int
const int dstUVPitchLeft = (dstUVPitch - UVwidth * 2) / sizeof(Uint16);
const Uint16 *srcUV;
Uint16 *dstUV;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
@@ -1277,7 +1277,7 @@ static int SDL_ConvertPixels_SwapNV(int width, int height, const void *src, int
y = UVheight;
while (y--) {
x = UVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
while (x >= 8) {
__m128i uv = _mm_loadu_si128((__m128i *)srcUV);
@@ -1372,7 +1372,7 @@ static int SDL_ConvertPixels_Planar2x2_to_Planar2x2(int width, int height,
SDL_GetPixelFormatName(dst_format));
}
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
#define PACKED4_TO_PACKED4_ROW_SSE2(shuffle) \
while (x >= 4) { \
__m128i yuv = _mm_loadu_si128((__m128i *)srcYUV); \
@@ -1399,14 +1399,14 @@ static int SDL_ConvertPixels_YUY2_to_UYVY(int width, int height, const void *src
const int dstYUVPitchLeft = (dst_pitch - YUVwidth * 4);
const Uint8 *srcYUV = (const Uint8 *)src;
Uint8 *dstYUV = (Uint8 *)dst;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
y = height;
while (y--) {
x = YUVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
PACKED4_TO_PACKED4_ROW_SSE2(_MM_SHUFFLE(2, 3, 0, 1));
}
@@ -1440,14 +1440,14 @@ static int SDL_ConvertPixels_YUY2_to_YVYU(int width, int height, const void *src
const int dstYUVPitchLeft = (dst_pitch - YUVwidth * 4);
const Uint8 *srcYUV = (const Uint8 *)src;
Uint8 *dstYUV = (Uint8 *)dst;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
y = height;
while (y--) {
x = YUVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
PACKED4_TO_PACKED4_ROW_SSE2(_MM_SHUFFLE(1, 2, 3, 0));
}
@@ -1481,14 +1481,14 @@ static int SDL_ConvertPixels_UYVY_to_YUY2(int width, int height, const void *src
const int dstYUVPitchLeft = (dst_pitch - YUVwidth * 4);
const Uint8 *srcYUV = (const Uint8 *)src;
Uint8 *dstYUV = (Uint8 *)dst;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
y = height;
while (y--) {
x = YUVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
PACKED4_TO_PACKED4_ROW_SSE2(_MM_SHUFFLE(2, 3, 0, 1));
}
@@ -1522,14 +1522,14 @@ static int SDL_ConvertPixels_UYVY_to_YVYU(int width, int height, const void *src
const int dstYUVPitchLeft = (dst_pitch - YUVwidth * 4);
const Uint8 *srcYUV = (const Uint8 *)src;
Uint8 *dstYUV = (Uint8 *)dst;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
y = height;
while (y--) {
x = YUVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
PACKED4_TO_PACKED4_ROW_SSE2(_MM_SHUFFLE(0, 3, 2, 1));
}
@@ -1563,14 +1563,14 @@ static int SDL_ConvertPixels_YVYU_to_YUY2(int width, int height, const void *src
const int dstYUVPitchLeft = (dst_pitch - YUVwidth * 4);
const Uint8 *srcYUV = (const Uint8 *)src;
Uint8 *dstYUV = (Uint8 *)dst;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
y = height;
while (y--) {
x = YUVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
PACKED4_TO_PACKED4_ROW_SSE2(_MM_SHUFFLE(1, 2, 3, 0));
}
@@ -1604,14 +1604,14 @@ static int SDL_ConvertPixels_YVYU_to_UYVY(int width, int height, const void *src
const int dstYUVPitchLeft = (dst_pitch - YUVwidth * 4);
const Uint8 *srcYUV = (const Uint8 *)src;
Uint8 *dstYUV = (Uint8 *)dst;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
const SDL_bool use_SSE2 = SDL_HasSSE2();
#endif
y = height;
while (y--) {
x = YUVwidth;
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
if (use_SSE2) {
PACKED4_TO_PACKED4_ROW_SSE2(_MM_SHUFFLE(2, 1, 0, 3));
}

5
src/video/yuv2rgb/yuv_rgb.c
View File

@@ -6,7 +6,6 @@
#include "yuv_rgb.h"
#define PRECISION 6
#define PRECISION_FACTOR (1<<PRECISION)
@@ -240,7 +239,7 @@ void rgb24_yuv420_std(
}
}
#ifdef __SSE2__
#if HAVE_SSE2_INTRINSICS
#define SSE_FUNCTION_NAME yuv420_rgb565_sse
#define STD_FUNCTION_NAME yuv420_rgb565_std
@@ -683,7 +682,7 @@ void rgb24_yuv420_sseu(uint32_t width, uint32_t height,
}
#endif //__SSE2__
#endif //HAVE_SSE2_INTRINSICS
#ifdef __loongarch_sx