sw_engine common: Added neon alpha blending.

Changes:
- Prepare neon verison of ALPHA_BLEND API.
- Use ALPHA_BLEND_NEON in _translucentRle

Notes:
- _translucentRle with neon support reduces execution time of this
function ~ 300 % (measured on uint32_t 400 x 400 buffer).
- API was tested on ARMv7l device with GCC 9.2 based toolchain. Results
  on other devices could be different.

diff --git a/src/lib/sw_engine/tvgSwCommon.h b/src/lib/sw_engine/tvgSwCommon.h
index a0d18dd..6ad2df4 100644 (file)
--- a/src/lib/sw_engine/tvgSwCommon.h
+++ b/src/lib/sw_engine/tvgSwCommon.h
@@ -275,6 +275,14 @@ static inline uint32_t ALPHA_BLEND(uint32_t c, uint32_t a)
             ((((c & 0x00ff00ff) * a + 0x00ff00ff) >> 8) & 0x00ff00ff));
 }
 
+#if defined(THORVG_NEON_VECTOR_SUPPORT)
+static inline uint8x8_t ALPHA_BLEND_NEON(uint8x8_t c, uint8x8_t a)
+{
+       uint16x8_t t = vmull_u8(c, a);
+       return vshrn_n_u16(t, 8);
+}
+#endif
+
 static inline uint32_t COLOR_INTERPOLATE(uint32_t c1, uint32_t a1, uint32_t c2, uint32_t a2)
 {
     auto t = (((c1 & 0xff00ff) * a1 + (c2 & 0xff00ff) * a2) >> 8) & 0xff00ff;

diff --git a/src/lib/sw_engine/tvgSwRaster.cpp b/src/lib/sw_engine/tvgSwRaster.cpp
index e4e86b5..5fcc81a 100644 (file)
--- a/src/lib/sw_engine/tvgSwRaster.cpp
+++ b/src/lib/sw_engine/tvgSwRaster.cpp
@@ -215,12 +215,35 @@ static bool _translucentRle(SwSurface* surface, const SwRleData* rle, uint32_t c
 
     for (uint32_t i = 0; i < rle->size; ++i) {
         auto dst = &surface->buffer[span->y * surface->stride + span->x];
+
+#if defined(THORVG_NEON_VECTOR_SUPPORT)
+        uint8x8_t *vDst = (uint8x8_t*) dst;
+#endif
+
         if (span->coverage < 255) src = ALPHA_BLEND(color, span->coverage);
         else src = color;
         auto ialpha = 255 - surface->blender.alpha(src);
+
+#if defined(THORVG_NEON_VECTOR_SUPPORT)
+        uint8x8_t vSrc = (uint8x8_t) vdup_n_u32(src);
+        uint8x8_t vIalpha = (uint8x8_t) vdup_n_u32(ialpha);
+
+        uint32_t iterations = span->len / 2;
+        uint32_t left = span->len % 2;
+
+        for (uint32_t x = 0; x < iterations; x+=2) {
+            vDst[x] = vadd_u8(vSrc, ALPHA_BLEND_NEON(vDst[x], vIalpha));
+        }
+
+        if (left) {
+            dst[span->len] = src + ALPHA_BLEND(dst[span->len], ialpha);
+        }
+#else
+
         for (uint32_t x = 0; x < span->len; ++x) {
             dst[x] = src + ALPHA_BLEND(dst[x], ialpha);
         }
+#endif
         ++span;
     }
     return true;