From a1e5dae0f0d78d9c2801fd08e2a87f863833a573 Mon Sep 17 00:00:00 2001 From: James Zern Date: Sat, 3 Jul 2021 17:43:11 -0700 Subject: [PATCH] alpha_processing*: use WEBP_RESTRICT qualifier this helps both auto-vectorization in the C code and the optimized code generation Change-Id: Ide570d6be45125ffef7248bdc40e9eb08f00e832 --- src/dsp/alpha_processing.c | 39 ++++++++++++++++++++------------ src/dsp/alpha_processing_neon.c | 22 ++++++++++-------- src/dsp/alpha_processing_sse2.c | 22 ++++++++++-------- src/dsp/alpha_processing_sse41.c | 8 ++++--- 4 files changed, 54 insertions(+), 37 deletions(-) diff --git a/src/dsp/alpha_processing.c b/src/dsp/alpha_processing.c index 3a27990d..de38bcb9 100644 --- a/src/dsp/alpha_processing.c +++ b/src/dsp/alpha_processing.c @@ -13,6 +13,7 @@ #include #include "src/dsp/dsp.h" +#include "src/utils/utils.h" // Tables can be faster on some platform but incur some extra binary size (~2k). #if !defined(USE_TABLES_FOR_ALPHA_MULT) @@ -157,7 +158,8 @@ void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse) { } } -void WebPMultRow_C(uint8_t* const ptr, const uint8_t* const alpha, +void WebPMultRow_C(uint8_t* WEBP_RESTRICT const ptr, + const uint8_t* WEBP_RESTRICT const alpha, int width, int inverse) { int x; for (x = 0; x < width; ++x) { @@ -193,8 +195,8 @@ void WebPMultARGBRows(uint8_t* ptr, int stride, int width, int num_rows, } } -void WebPMultRows(uint8_t* ptr, int stride, - const uint8_t* alpha, int alpha_stride, +void WebPMultRows(uint8_t* WEBP_RESTRICT ptr, int stride, + const uint8_t* WEBP_RESTRICT alpha, int alpha_stride, int width, int num_rows, int inverse) { int n; for (n = 0; n < num_rows; ++n) { @@ -290,9 +292,9 @@ static void ApplyAlphaMultiply_16b_C(uint8_t* rgba4444, } #if !WEBP_NEON_OMIT_C_CODE -static int DispatchAlpha_C(const uint8_t* alpha, int alpha_stride, +static int DispatchAlpha_C(const uint8_t* WEBP_RESTRICT alpha, int alpha_stride, int width, int height, - uint8_t* dst, int dst_stride) { + uint8_t* WEBP_RESTRICT dst, int dst_stride) { uint32_t alpha_mask = 0xff; int i, j; @@ -309,9 +311,10 @@ static int DispatchAlpha_C(const uint8_t* alpha, int alpha_stride, return (alpha_mask != 0xff); } -static void DispatchAlphaToGreen_C(const uint8_t* alpha, int alpha_stride, - int width, int height, - uint32_t* dst, int dst_stride) { +static void DispatchAlphaToGreen_C(const uint8_t* WEBP_RESTRICT alpha, + int alpha_stride, int width, int height, + uint32_t* WEBP_RESTRICT dst, + int dst_stride) { int i, j; for (j = 0; j < height; ++j) { for (i = 0; i < width; ++i) { @@ -322,9 +325,9 @@ static void DispatchAlphaToGreen_C(const uint8_t* alpha, int alpha_stride, } } -static int ExtractAlpha_C(const uint8_t* argb, int argb_stride, +static int ExtractAlpha_C(const uint8_t* WEBP_RESTRICT argb, int argb_stride, int width, int height, - uint8_t* alpha, int alpha_stride) { + uint8_t* WEBP_RESTRICT alpha, int alpha_stride) { uint8_t alpha_mask = 0xff; int i, j; @@ -340,7 +343,8 @@ static int ExtractAlpha_C(const uint8_t* argb, int argb_stride, return (alpha_mask == 0xff); } -static void ExtractGreen_C(const uint32_t* argb, uint8_t* alpha, int size) { +static void ExtractGreen_C(const uint32_t* WEBP_RESTRICT argb, + uint8_t* WEBP_RESTRICT alpha, int size) { int i; for (i = 0; i < size; ++i) alpha[i] = argb[i] >> 8; } @@ -372,8 +376,11 @@ static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) { } #ifdef WORDS_BIGENDIAN -static void PackARGB_C(const uint8_t* a, const uint8_t* r, const uint8_t* g, - const uint8_t* b, int len, uint32_t* out) { +static void PackARGB_C(const uint8_t* WEBP_RESTRICT a, + const uint8_t* WEBP_RESTRICT r, + const uint8_t* WEBP_RESTRICT g, + const uint8_t* WEBP_RESTRICT b, + int len, uint32_t* WEBP_RESTRICT out) { int i; for (i = 0; i < len; ++i) { out[i] = MakeARGB32(a[4 * i], r[4 * i], g[4 * i], b[4 * i]); @@ -381,8 +388,10 @@ static void PackARGB_C(const uint8_t* a, const uint8_t* r, const uint8_t* g, } #endif -static void PackRGB_C(const uint8_t* r, const uint8_t* g, const uint8_t* b, - int len, int step, uint32_t* out) { +static void PackRGB_C(const uint8_t* WEBP_RESTRICT r, + const uint8_t* WEBP_RESTRICT g, + const uint8_t* WEBP_RESTRICT b, + int len, int step, uint32_t* WEBP_RESTRICT out) { int i, offset = 0; for (i = 0; i < len; ++i) { out[i] = MakeARGB32(0xff, r[offset], g[offset], b[offset]); diff --git a/src/dsp/alpha_processing_neon.c b/src/dsp/alpha_processing_neon.c index 9d554217..073d7965 100644 --- a/src/dsp/alpha_processing_neon.c +++ b/src/dsp/alpha_processing_neon.c @@ -16,6 +16,7 @@ #if defined(WEBP_USE_NEON) #include "src/dsp/neon.h" +#include "src/utils/utils.h" //------------------------------------------------------------------------------ @@ -80,9 +81,9 @@ static void ApplyAlphaMultiply_NEON(uint8_t* rgba, int alpha_first, //------------------------------------------------------------------------------ -static int DispatchAlpha_NEON(const uint8_t* alpha, int alpha_stride, - int width, int height, - uint8_t* dst, int dst_stride) { +static int DispatchAlpha_NEON(const uint8_t* WEBP_RESTRICT alpha, + int alpha_stride, int width, int height, + uint8_t* WEBP_RESTRICT dst, int dst_stride) { uint32_t alpha_mask = 0xffffffffu; uint8x8_t mask8 = vdup_n_u8(0xff); uint32_t tmp[2]; @@ -112,9 +113,10 @@ static int DispatchAlpha_NEON(const uint8_t* alpha, int alpha_stride, return (alpha_mask != 0xffffffffu); } -static void DispatchAlphaToGreen_NEON(const uint8_t* alpha, int alpha_stride, - int width, int height, - uint32_t* dst, int dst_stride) { +static void DispatchAlphaToGreen_NEON(const uint8_t* WEBP_RESTRICT alpha, + int alpha_stride, int width, int height, + uint32_t* WEBP_RESTRICT dst, + int dst_stride) { int i, j; uint8x8x4_t greens; // leave A/R/B channels zero'd. greens.val[0] = vdup_n_u8(0); @@ -131,9 +133,9 @@ static void DispatchAlphaToGreen_NEON(const uint8_t* alpha, int alpha_stride, } } -static int ExtractAlpha_NEON(const uint8_t* argb, int argb_stride, +static int ExtractAlpha_NEON(const uint8_t* WEBP_RESTRICT argb, int argb_stride, int width, int height, - uint8_t* alpha, int alpha_stride) { + uint8_t* WEBP_RESTRICT alpha, int alpha_stride) { uint32_t alpha_mask = 0xffffffffu; uint8x8_t mask8 = vdup_n_u8(0xff); uint32_t tmp[2]; @@ -161,8 +163,8 @@ static int ExtractAlpha_NEON(const uint8_t* argb, int argb_stride, return (alpha_mask == 0xffffffffu); } -static void ExtractGreen_NEON(const uint32_t* argb, - uint8_t* alpha, int size) { +static void ExtractGreen_NEON(const uint32_t* WEBP_RESTRICT argb, + uint8_t* WEBP_RESTRICT alpha, int size) { int i; for (i = 0; i + 16 <= size; i += 16) { const uint8x16x4_t rgbX = vld4q_u8((const uint8_t*)(argb + i)); diff --git a/src/dsp/alpha_processing_sse2.c b/src/dsp/alpha_processing_sse2.c index f6c6e0fb..269e784d 100644 --- a/src/dsp/alpha_processing_sse2.c +++ b/src/dsp/alpha_processing_sse2.c @@ -16,11 +16,13 @@ #if defined(WEBP_USE_SSE2) #include +#include "src/utils/utils.h" + //------------------------------------------------------------------------------ -static int DispatchAlpha_SSE2(const uint8_t* alpha, int alpha_stride, - int width, int height, - uint8_t* dst, int dst_stride) { +static int DispatchAlpha_SSE2(const uint8_t* WEBP_RESTRICT alpha, + int alpha_stride, int width, int height, + uint8_t* WEBP_RESTRICT dst, int dst_stride) { // alpha_and stores an 'and' operation of all the alpha[] values. The final // value is not 0xff if any of the alpha[] is not equal to 0xff. uint32_t alpha_and = 0xff; @@ -72,9 +74,10 @@ static int DispatchAlpha_SSE2(const uint8_t* alpha, int alpha_stride, return (alpha_and != 0xff); } -static void DispatchAlphaToGreen_SSE2(const uint8_t* alpha, int alpha_stride, - int width, int height, - uint32_t* dst, int dst_stride) { +static void DispatchAlphaToGreen_SSE2(const uint8_t* WEBP_RESTRICT alpha, + int alpha_stride, int width, int height, + uint32_t* WEBP_RESTRICT dst, + int dst_stride) { int i, j; const __m128i zero = _mm_setzero_si128(); const int limit = width & ~15; @@ -98,9 +101,9 @@ static void DispatchAlphaToGreen_SSE2(const uint8_t* alpha, int alpha_stride, } } -static int ExtractAlpha_SSE2(const uint8_t* argb, int argb_stride, +static int ExtractAlpha_SSE2(const uint8_t* WEBP_RESTRICT argb, int argb_stride, int width, int height, - uint8_t* alpha, int alpha_stride) { + uint8_t* WEBP_RESTRICT alpha, int alpha_stride) { // alpha_and stores an 'and' operation of all the alpha[] values. The final // value is not 0xff if any of the alpha[] is not equal to 0xff. uint32_t alpha_and = 0xff; @@ -317,7 +320,8 @@ static void MultARGBRow_SSE2(uint32_t* const ptr, int width, int inverse) { if (width > 0) WebPMultARGBRow_C(ptr + x, width, inverse); } -static void MultRow_SSE2(uint8_t* const ptr, const uint8_t* const alpha, +static void MultRow_SSE2(uint8_t* WEBP_RESTRICT const ptr, + const uint8_t* WEBP_RESTRICT const alpha, int width, int inverse) { int x = 0; if (!inverse) { diff --git a/src/dsp/alpha_processing_sse41.c b/src/dsp/alpha_processing_sse41.c index 56040f9c..8205e5f6 100644 --- a/src/dsp/alpha_processing_sse41.c +++ b/src/dsp/alpha_processing_sse41.c @@ -17,11 +17,13 @@ #include +#include "src/utils/utils.h" + //------------------------------------------------------------------------------ -static int ExtractAlpha_SSE41(const uint8_t* argb, int argb_stride, - int width, int height, - uint8_t* alpha, int alpha_stride) { +static int ExtractAlpha_SSE41(const uint8_t* WEBP_RESTRICT argb, + int argb_stride, int width, int height, + uint8_t* WEBP_RESTRICT alpha, int alpha_stride) { // alpha_and stores an 'and' operation of all the alpha[] values. The final // value is not 0xff if any of the alpha[] is not equal to 0xff. uint32_t alpha_and = 0xff;