| Index: src/opts/SkBitmapFilter_opts_SSE2.cpp
|
| diff --git a/src/opts/SkBitmapFilter_opts_SSE2.cpp b/src/opts/SkBitmapFilter_opts_SSE2.cpp
|
| index 2996f535d97abe1413cb7ed810f631668b29033d..de3dd3b8e9e04161b642c12d27412719a600d90c 100644
|
| --- a/src/opts/SkBitmapFilter_opts_SSE2.cpp
|
| +++ b/src/opts/SkBitmapFilter_opts_SSE2.cpp
|
| @@ -40,78 +40,6 @@ static inline void print128f(__m128 value) {
|
| }
|
| #endif
|
|
|
| -// because the border is handled specially, this is guaranteed to have all 16 pixels
|
| -// available to it without running off the bitmap's edge.
|
| -
|
| -void highQualityFilter_SSE2(const SkBitmapProcState& s, int x, int y,
|
| - SkPMColor* SK_RESTRICT colors, int count) {
|
| -
|
| - const int maxX = s.fBitmap->width();
|
| - const int maxY = s.fBitmap->height();
|
| - SkAutoTMalloc<SkScalar> xWeights(maxX);
|
| - const SkBitmapFilter* filter = s.getBitmapFilter();
|
| -
|
| - while (count-- > 0) {
|
| - SkPoint srcPt;
|
| - s.fInvProc(s.fInvMatrix, x + 0.5f, y + 0.5f, &srcPt);
|
| - srcPt.fX -= SK_ScalarHalf;
|
| - srcPt.fY -= SK_ScalarHalf;
|
| -
|
| - __m128 weight = _mm_setzero_ps();
|
| - __m128 accum = _mm_setzero_ps();
|
| -
|
| - int y0 = SkClampMax(SkScalarCeilToInt(srcPt.fY - filter->width()), maxY);
|
| - int y1 = SkClampMax(SkScalarFloorToInt(srcPt.fY + filter->width() + 1), maxY);
|
| - int x0 = SkClampMax(SkScalarCeilToInt(srcPt.fX - filter->width()), maxX);
|
| - int x1 = SkClampMax(SkScalarFloorToInt(srcPt.fX + filter->width()) + 1, maxX);
|
| -
|
| - for (int srcX = x0; srcX < x1 ; srcX++) {
|
| - // Looking these up once instead of each loop is a ~15% speedup.
|
| - xWeights[srcX - x0] = filter->lookupScalar((srcPt.fX - srcX));
|
| - }
|
| -
|
| - for (int srcY = y0; srcY < y1; srcY++) {
|
| - SkScalar yWeight = filter->lookupScalar((srcPt.fY - srcY));
|
| -
|
| - for (int srcX = x0; srcX < x1 ; srcX++) {
|
| - SkScalar xWeight = xWeights[srcX - x0];
|
| -
|
| - SkScalar combined_weight = SkScalarMul(xWeight, yWeight);
|
| - __m128 weightVector = _mm_set1_ps(combined_weight);
|
| - weight = _mm_add_ps( weight, weightVector );
|
| -
|
| - SkPMColor color = *s.fBitmap->getAddr32(srcX, srcY);
|
| - if (!color) {
|
| - continue;
|
| - }
|
| -
|
| - __m128i c = _mm_cvtsi32_si128(color);
|
| - c = _mm_unpacklo_epi8(c, _mm_setzero_si128());
|
| - c = _mm_unpacklo_epi16(c, _mm_setzero_si128());
|
| - __m128 cfloat = _mm_cvtepi32_ps(c);
|
| -
|
| - accum = _mm_add_ps(accum, _mm_mul_ps(cfloat, weightVector));
|
| - }
|
| - }
|
| -
|
| - accum = _mm_div_ps(accum, weight);
|
| - accum = _mm_add_ps(accum, _mm_set1_ps(0.5f));
|
| - __m128i accumInt = _mm_cvttps_epi32(accum);
|
| - accumInt = _mm_packs_epi32(accumInt, _mm_setzero_si128());
|
| - accumInt = _mm_packus_epi16(accumInt, _mm_setzero_si128());
|
| - SkPMColor c = _mm_cvtsi128_si32(accumInt);
|
| -
|
| - int a = SkClampMax(SkGetPackedA32(c), 255);
|
| - int r = SkClampMax(SkGetPackedR32(c), a);
|
| - int g = SkClampMax(SkGetPackedG32(c), a);
|
| - int b = SkClampMax(SkGetPackedB32(c), a);
|
| -
|
| - *colors++ = SkPackARGB32(a, r, g, b);
|
| -
|
| - x++;
|
| - }
|
| -}
|
| -
|
| // Convolves horizontally along a single row. The row data is given in
|
| // |src_data| and continues for the num_values() of the filter.
|
| void convolveHorizontally_SSE2(const unsigned char* src_data,
|
|
|