git_clone/thorvg
1
0
Fork 0
mirror of https://github.com/thorvg/thorvg.git synced 2025-06-14 12:04:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

sw_engine: hotfix avx/neon 8 bit rasterizer.

Browse source 
recte/rle 8 bit rasterizer has been broken,
make it work safely with c raster version.

issue: https://github.com/thorvg/thorvg/issues/2146
This commit is contained in:
Hermet Park 2024-06-20 14:10:29 +09:00 • committed by Hermet Park
parent 25370f438e
commit a9592ec3ef
2 changed files with 179 additions and 135 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

48
src/renderer/sw_engine/tvgSwRasterAvx.h
View file

@ -101,16 +101,14 @@ static void avxRasterPixel32(uint32_t *dst, uint32_t val, uint32_t offset, int32
static bool avxRasterTranslucentRect(SwSurface* surface, const SwBBox& region, uint8_t r, uint8_t g, uint8_t b, uint8_t a)
{
if (surface->channelSize != sizeof(uint32_t)) {
TVGERR("SW_ENGINE", "Unsupported Channel Size = %d", surface->channelSize);
return false;
}
auto color = surface->join(r, g, b, a);
auto buffer = surface->buf32 + (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);
//32bits channels
if (surface->channelSize == sizeof(uint32_t)) {
auto color = surface->join(r, g, b, a);
auto buffer = surface->buf32 + (region.min.y * surface->stride) + region.min.x;
uint32_t ialpha = 255 - a;
auto avxColor = _mm_set1_epi32(color);
@ -144,19 +142,29 @@ static bool avxRasterTranslucentRect(SwSurface* surface, const SwBBox& region, u
dst++;
}
}
//8bit grayscale
} else if (surface->channelSize == sizeof(uint8_t)) {
TVGLOG("SW_ENGINE", "Require AVX Optimization, Channel Size = %d", surface->channelSize);
auto buffer = surface->buf8 + (region.min.y * surface->stride) + region.min.x;
auto ialpha = ~a;
for (uint32_t y = 0; y < h; ++y) {
auto dst = &buffer[y * surface->stride];
for (uint32_t x = 0; x < w; ++x, ++dst) {
*dst = a + MULTIPLY(*dst, ialpha);
}
}
}
return true;
}
static bool avxRasterTranslucentRle(SwSurface* surface, const SwRleData* rle, uint8_t r, uint8_t g, uint8_t b, uint8_t a)
{
if (surface->channelSize != sizeof(uint32_t)) {
TVGERR("SW_ENGINE", "Unsupported Channel Size = %d", surface->channelSize);
return false;
}
auto color = surface->join(r, g, b, a);
auto span = rle->spans;
//32bit channels
if (surface->channelSize == sizeof(uint32_t)) {
auto color = surface->join(r, g, b, a);
uint32_t src;
for (uint32_t i = 0; i < rle->size; ++i) {
@ -201,6 +209,20 @@ static bool avxRasterTranslucentRle(SwSurface* surface, const SwRleData* rle, ui
++span;
}
//8bit grayscale
} else if (surface->channelSize == sizeof(uint8_t)) {
TVGLOG("SW_ENGINE", "Require AVX Optimization, Channel Size = %d", surface->channelSize);
uint8_t src;
for (uint32_t i = 0; i < rle->size; ++i, ++span) {
auto dst = &surface->buf8[span->y * surface->stride + span->x];
if (span->coverage < 255) src = MULTIPLY(span->coverage, a);
else src = a;
auto ialpha = ~a;
for (uint32_t x = 0; x < span->len; ++x, ++dst) {
*dst = src + MULTIPLY(*dst, ialpha);
}
}
}
return true;
}

48
src/renderer/sw_engine/tvgSwRasterNeon.h
View file

@ -91,13 +91,11 @@ static void neonRasterPixel32(uint32_t *dst, uint32_t val, uint32_t offset, int3
static bool neonRasterTranslucentRle(SwSurface* surface, const SwRleData* rle, uint8_t r, uint8_t g, uint8_t b, uint8_t a)
{
if (surface->channelSize != sizeof(uint32_t)) {
TVGERR("SW_ENGINE", "Unsupported Channel Size = %d", surface->channelSize);
return false;
}
auto color = surface->join(r, g, b, a);
auto span = rle->spans;
//32bit channels
if (surface->channelSize == sizeof(uint32_t)) {
auto color = surface->join(r, g, b, a);
uint32_t src;
uint8x8_t *vDst = nullptr;
uint16_t align;
@ -130,21 +128,33 @@ static bool neonRasterTranslucentRle(SwSurface* surface, const SwRleData* rle, u
++span;
}
//8bit grayscale
} else if (surface->channelSize == sizeof(uint8_t)) {
TVGLOG("SW_ENGINE", "Require Neon Optimization, Channel Size = %d", surface->channelSize);
uint8_t src;
for (uint32_t i = 0; i < rle->size; ++i, ++span) {
auto dst = &surface->buf8[span->y * surface->stride + span->x];
if (span->coverage < 255) src = MULTIPLY(span->coverage, a);
else src = a;
auto ialpha = ~a;
for (uint32_t x = 0; x < span->len; ++x, ++dst) {
*dst = src + MULTIPLY(*dst, ialpha);
}
}
}
return true;
}
static bool neonRasterTranslucentRect(SwSurface* surface, const SwBBox& region, uint8_t r, uint8_t g, uint8_t b, uint8_t a)
{
if (surface->channelSize != sizeof(uint32_t)) {
TVGERR("SW_ENGINE", "Unsupported Channel Size = %d", surface->channelSize);
return false;
}
auto color = surface->join(r, g, b, a);
auto buffer = surface->buf32 + (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);
//32bits channels
if (surface->channelSize == sizeof(uint32_t)) {
auto color = surface->join(r, g, b, a);
auto buffer = surface->buf32 + (region.min.y * surface->stride) + region.min.x;
auto ialpha = 255 - a;
auto vColor = vdup_n_u32(color);
@ -172,6 +182,18 @@ static bool neonRasterTranslucentRect(SwSurface* surface, const SwBBox& region,
auto leftovers = (w - align) % 2;
if (leftovers > 0) dst[w - 1] = color + ALPHA_BLEND(dst[w - 1], ialpha);
}
//8bit grayscale
} else if (surface->channelSize == sizeof(uint8_t)) {
TVGLOG("SW_ENGINE", "Require Neon Optimization, Channel Size = %d", surface->channelSize);
auto buffer = surface->buf8 + (region.min.y * surface->stride) + region.min.x;
auto ialpha = ~a;
for (uint32_t y = 0; y < h; ++y) {
auto dst = &buffer[y * surface->stride];
for (uint32_t x = 0; x < w; ++x, ++dst) {
*dst = a + MULTIPLY(*dst, ialpha);
}
}
}
return true;
}