sw common: Added neon translucent rect API implementation.

Changes:
Added neonRasterTranslucentRect implementation. Rendering was tested on
32 lottie fiels. Without neon ~ 18.1 FPS was measured. With neon ~ 20.1
FPS was measured.

diff --git a/src/lib/sw_engine/tvgSwRaster.cpp b/src/lib/sw_engine/tvgSwRaster.cpp
index 8498561..d42a6d3 100644 (file)
--- a/src/lib/sw_engine/tvgSwRaster.cpp
+++ b/src/lib/sw_engine/tvgSwRaster.cpp
@@ -120,18 +120,13 @@ static uint32_t _applyBilinearInterpolation(const uint32_t *img, uint32_t w, uin
 
 static bool _translucentRect(SwSurface* surface, const SwBBox& region, uint32_t color)
 {
-    auto buffer = surface->buffer + (region.min.y * surface->stride) + region.min.x;
-    auto h = static_cast<uint32_t>(region.max.y - region.min.y);
-    auto w = static_cast<uint32_t>(region.max.x - region.min.x);
-    auto ialpha = 255 - surface->blender.alpha(color);
-
-    for (uint32_t y = 0; y < h; ++y) {
-        auto dst = &buffer[y * surface->stride];
-        for (uint32_t x = 0; x < w; ++x) {
-            dst[x] = color + ALPHA_BLEND(dst[x], ialpha);
-        }
-    }
-    return true;
+    #if defined(THORVG_AVX_VECTOR_SUPPORT)
+        return cRasterTranslucentRect(surface, region, color);
+    #elif defined(THORVG_NEON_VECTOR_SUPPORT)
+        return neonRasterTranslucentRect(surface, region, color);
+    #else
+        return cRasterTranslucentRect(surface, region, color);
+    #endif
 }
 
 

diff --git a/src/lib/sw_engine/tvgSwRasterC.h b/src/lib/sw_engine/tvgSwRasterC.h
index c638a19..b634edf 100644 (file)
--- a/src/lib/sw_engine/tvgSwRasterC.h
+++ b/src/lib/sw_engine/tvgSwRasterC.h
@@ -47,4 +47,20 @@ static inline bool cRasterTranslucentRle(SwSurface* surface, const SwRleData* rl
         ++span;
     }
     return true;
+}
+
+static inline bool cRasterTranslucentRect(SwSurface* surface, const SwBBox& region, uint32_t color)
+{
+    auto buffer = surface->buffer + (region.min.y * surface->stride) + region.min.x;
+    auto h = static_cast<uint32_t>(region.max.y - region.min.y);
+    auto w = static_cast<uint32_t>(region.max.x - region.min.x);
+    auto ialpha = 255 - surface->blender.alpha(color);
+
+    for (uint32_t y = 0; y < h; ++y) {
+        auto dst = &buffer[y * surface->stride];
+        for (uint32_t x = 0; x < w; ++x) {
+            dst[x] = color + ALPHA_BLEND(dst[x], ialpha);
+        }
+    }
+    return true;
 }
\ No newline at end of file

diff --git a/src/lib/sw_engine/tvgSwRasterNeon.h b/src/lib/sw_engine/tvgSwRasterNeon.h
index 2d32caa..a1c1b0f 100644 (file)
--- a/src/lib/sw_engine/tvgSwRasterNeon.h
+++ b/src/lib/sw_engine/tvgSwRasterNeon.h
@@ -88,4 +88,41 @@ static inline bool neonRasterTranslucentRle(SwSurface* surface, const SwRleData*
     return true;
 }
 
+
+static inline bool neonRasterTranslucentRect(SwSurface* surface, const SwBBox& region, uint32_t color)
+{
+    uint32_t *buffer = surface->buffer + (region.min.y * surface->stride) + region.min.x;
+    uint32_t h = (uint32_t)(region.max.y - region.min.y);
+    uint32_t w = (uint32_t)(region.max.x - region.min.x);
+
+    uint32_t ialpha = 255 - surface->blender.alpha(color);
+
+    uint8x8_t vColor = (uint8x8_t) vdup_n_u32(color);
+    uint8x8_t vIalpha = (uint8x8_t) vdup_n_u8((uint8_t) ialpha);
+
+    uint8x8_t *vDst = NULL;
+
+    for (uint32_t y = 0; y < h; ++y) {
+        uint32_t align = 0;
+        uint32_t *dst = &buffer[y * surface->stride];
+
+        if ((((uint32_t) dst) & 0x7) != 0) {
+            vDst = (uint8x8_t*)(dst+1);
+            align = 1;
+        } else {
+            vDst = (uint8x8_t*) dst;
+        }
+
+        uint32_t iterations = w / 2;
+        uint32_t left = w % 2;
+
+        if (align) *dst = color + ALPHA_BLEND(*dst, ialpha);
+
+        for (uint32_t x = 0; x < iterations; ++x) vDst[x] = vadd_u8(vColor, ALPHA_BLEND_NEON(vDst[x], vIalpha));
+        
+        if (left) dst[w] = color + ALPHA_BLEND(dst[w], ialpha);
+    }
+    return true;
+}
+
 #endif
\ No newline at end of file