From 06b4b2c5864c97624143183b18ff94fcb51bfd8f Mon Sep 17 00:00:00 2001
From: rinechran <rinechran@gmail.com>
Date: Sun, 7 Apr 2024 14:43:23 +0900
Subject: [PATCH] sw_engine: Implement AVX and NEON optimizations for
 RasterGrayscale

---
 src/renderer/sw_engine/tvgSwRaster.cpp   |  7 ++++++-
 src/renderer/sw_engine/tvgSwRasterAvx.h  | 19 ++++++++++++++++++-
 src/renderer/sw_engine/tvgSwRasterNeon.h | 17 +++++++++++++++++
 3 files changed, 41 insertions(+), 2 deletions(-)

diff --git a/src/renderer/sw_engine/tvgSwRaster.cpp b/src/renderer/sw_engine/tvgSwRaster.cpp
index 98374d3f..4e0020ee 100644
--- a/src/renderer/sw_engine/tvgSwRaster.cpp
+++ b/src/renderer/sw_engine/tvgSwRaster.cpp
@@ -1755,8 +1755,13 @@ static bool _rasterRadialGradientRle(SwSurface* surface, const SwRleData* rle, c
 
 void rasterGrayscale8(uint8_t *dst, uint8_t val, uint32_t offset, int32_t len)
 {
-    //OPTIMIZE_ME: Support SIMD
+#if defined(THORVG_AVX_VECTOR_SUPPORT)
+    avxRasterGrayscale8(dst, val, offset, len);
+#elif defined(THORVG_NEON_VECTOR_SUPPORT)
+    neonRasterGrayscale8(dst, val, offset, len);
+#else
     cRasterPixels(dst, val, offset, len);
+#endif
 }
 
 
diff --git a/src/renderer/sw_engine/tvgSwRasterAvx.h b/src/renderer/sw_engine/tvgSwRasterAvx.h
index 177c7b84..cbaec28f 100644
--- a/src/renderer/sw_engine/tvgSwRasterAvx.h
+++ b/src/renderer/sw_engine/tvgSwRasterAvx.h
@@ -1,4 +1,4 @@
-/*
+/*
  * Copyright (c) 2021 - 2024 the ThorVG project. All rights reserved.
 
  * Permission is hereby granted, free of charge, to any person obtaining a copy
@@ -62,6 +62,23 @@ static inline __m128i ALPHA_BLEND(__m128i c, __m128i a)
 }
 
 
+static void avxRasterGrayscale8(uint8_t* dst, uint8_t val, uint32_t offset, int32_t len) 
+{
+    dst += offset; 
+
+    __m256i vecVal = _mm256_set1_epi8(val);
+
+    int32_t i = 0;
+    for (; i <= len - 32; i += 32) {
+        _mm256_storeu_si256((__m256i*)(dst + i), vecVal);
+    }
+
+    for (; i < len; ++i) {
+        dst[i] = val;
+    }
+}
+
+
 static void avxRasterPixel32(uint32_t *dst, uint32_t val, uint32_t offset, int32_t len)
 {
     //1. calculate how many iterations we need to cover the length
diff --git a/src/renderer/sw_engine/tvgSwRasterNeon.h b/src/renderer/sw_engine/tvgSwRasterNeon.h
index 2bb52837..30d8813c 100644
--- a/src/renderer/sw_engine/tvgSwRasterNeon.h
+++ b/src/renderer/sw_engine/tvgSwRasterNeon.h
@@ -31,6 +31,23 @@ static inline uint8x8_t ALPHA_BLEND(uint8x8_t c, uint8x8_t a)
 }
 
 
+static void neonRasterGrayscale8(uint8_t* dst, uint8_t val, uint32_t offset, int32_t len)
+{
+    dst += offset;
+
+    int32_t i = 0;
+    uint8x16_t valVec = vdupq_n_u8(val);
+
+    for (; i <= len - 16; i += 16) {
+        vst1q_u8(dst + i, valVec);
+    }
+
+    for (; i < len; i++) {
+        dst[i] = val;
+    }
+}
+
+
 static void neonRasterPixel32(uint32_t *dst, uint32_t val, uint32_t offset, int32_t len)
 {
     uint32_t iterations = len / 4;