Index: src/opts/SkXfermode_opts_SSE2.cpp |
diff --git a/src/opts/SkXfermode_opts_SSE2.cpp b/src/opts/SkXfermode_opts_SSE2.cpp |
index fccf66514789950d58da06f9695af0fb90d4409a..b92477094b72cbd8b6ab5b8f884355bf2f969990 100644 |
--- a/src/opts/SkXfermode_opts_SSE2.cpp |
+++ b/src/opts/SkXfermode_opts_SSE2.cpp |
@@ -25,26 +25,6 @@ static inline __m128i SkDiv255Round_SSE2(const __m128i& a) { |
return prod; |
} |
-static inline __m128i saturated_add_SSE2(const __m128i& a, const __m128i& b) { |
- __m128i sum = _mm_add_epi32(a, b); |
- __m128i cmp = _mm_cmpgt_epi32(sum, _mm_set1_epi32(255)); |
- |
- sum = _mm_or_si128(_mm_and_si128(cmp, _mm_set1_epi32(255)), |
- _mm_andnot_si128(cmp, sum)); |
- return sum; |
-} |
- |
-static inline __m128i clamp_signed_byte_SSE2(const __m128i& n) { |
- __m128i cmp1 = _mm_cmplt_epi32(n, _mm_setzero_si128()); |
- __m128i cmp2 = _mm_cmpgt_epi32(n, _mm_set1_epi32(255)); |
- __m128i ret = _mm_and_si128(cmp2, _mm_set1_epi32(255)); |
- |
- __m128i cmp = _mm_or_si128(cmp1, cmp2); |
- ret = _mm_or_si128(_mm_and_si128(cmp, ret), _mm_andnot_si128(cmp, n)); |
- |
- return ret; |
-} |
- |
static inline __m128i clamp_div255round_SSE2(const __m128i& prod) { |
// test if > 0 |
__m128i cmp1 = _mm_cmpgt_epi32(prod, _mm_setzero_si128()); |
@@ -65,131 +45,6 @@ static inline __m128i clamp_div255round_SSE2(const __m128i& prod) { |
return ret; |
} |
- |
-static __m128i srcover_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i isa = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(src)); |
- return _mm_add_epi32(src, SkAlphaMulQ_SSE2(dst, isa)); |
-} |
- |
-static __m128i dstover_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i ida = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(dst)); |
- return _mm_add_epi32(dst, SkAlphaMulQ_SSE2(src, ida)); |
-} |
- |
-static __m128i srcin_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- return SkAlphaMulQ_SSE2(src, SkAlpha255To256_SSE2(da)); |
-} |
- |
-static __m128i dstin_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- return SkAlphaMulQ_SSE2(dst, SkAlpha255To256_SSE2(sa)); |
-} |
- |
-static __m128i srcout_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i ida = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(dst)); |
- return SkAlphaMulQ_SSE2(src, ida); |
-} |
- |
-static __m128i dstout_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i isa = _mm_sub_epi32(_mm_set1_epi32(256), SkGetPackedA32_SSE2(src)); |
- return SkAlphaMulQ_SSE2(dst, isa); |
-} |
- |
-static __m128i srcatop_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa); |
- |
- __m128i a = da; |
- |
- __m128i r1 = SkAlphaMulAlpha_SSE2(da, SkGetPackedR32_SSE2(src)); |
- __m128i r2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedR32_SSE2(dst)); |
- __m128i r = _mm_add_epi32(r1, r2); |
- |
- __m128i g1 = SkAlphaMulAlpha_SSE2(da, SkGetPackedG32_SSE2(src)); |
- __m128i g2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedG32_SSE2(dst)); |
- __m128i g = _mm_add_epi32(g1, g2); |
- |
- __m128i b1 = SkAlphaMulAlpha_SSE2(da, SkGetPackedB32_SSE2(src)); |
- __m128i b2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedB32_SSE2(dst)); |
- __m128i b = _mm_add_epi32(b1, b2); |
- |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
-static __m128i dstatop_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da); |
- |
- __m128i a = sa; |
- |
- __m128i r1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedR32_SSE2(src)); |
- __m128i r2 = SkAlphaMulAlpha_SSE2(sa, SkGetPackedR32_SSE2(dst)); |
- __m128i r = _mm_add_epi32(r1, r2); |
- |
- __m128i g1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedG32_SSE2(src)); |
- __m128i g2 = SkAlphaMulAlpha_SSE2(sa, SkGetPackedG32_SSE2(dst)); |
- __m128i g = _mm_add_epi32(g1, g2); |
- |
- __m128i b1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedB32_SSE2(src)); |
- __m128i b2 = SkAlphaMulAlpha_SSE2(sa, SkGetPackedB32_SSE2(dst)); |
- __m128i b = _mm_add_epi32(b1, b2); |
- |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
-static __m128i xor_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- __m128i isa = _mm_sub_epi32(_mm_set1_epi32(255), sa); |
- __m128i ida = _mm_sub_epi32(_mm_set1_epi32(255), da); |
- |
- __m128i a1 = _mm_add_epi32(sa, da); |
- __m128i a2 = SkAlphaMulAlpha_SSE2(sa, da); |
- a2 = _mm_slli_epi32(a2, 1); |
- __m128i a = _mm_sub_epi32(a1, a2); |
- |
- __m128i r1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedR32_SSE2(src)); |
- __m128i r2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedR32_SSE2(dst)); |
- __m128i r = _mm_add_epi32(r1, r2); |
- |
- __m128i g1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedG32_SSE2(src)); |
- __m128i g2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedG32_SSE2(dst)); |
- __m128i g = _mm_add_epi32(g1, g2); |
- |
- __m128i b1 = SkAlphaMulAlpha_SSE2(ida, SkGetPackedB32_SSE2(src)); |
- __m128i b2 = SkAlphaMulAlpha_SSE2(isa, SkGetPackedB32_SSE2(dst)); |
- __m128i b = _mm_add_epi32(b1, b2); |
- |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
-static __m128i plus_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i b = saturated_add_SSE2(SkGetPackedB32_SSE2(src), |
- SkGetPackedB32_SSE2(dst)); |
- __m128i g = saturated_add_SSE2(SkGetPackedG32_SSE2(src), |
- SkGetPackedG32_SSE2(dst)); |
- __m128i r = saturated_add_SSE2(SkGetPackedR32_SSE2(src), |
- SkGetPackedR32_SSE2(dst)); |
- __m128i a = saturated_add_SSE2(SkGetPackedA32_SSE2(src), |
- SkGetPackedA32_SSE2(dst)); |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
-static __m128i modulate_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i a = SkAlphaMulAlpha_SSE2(SkGetPackedA32_SSE2(src), |
- SkGetPackedA32_SSE2(dst)); |
- __m128i r = SkAlphaMulAlpha_SSE2(SkGetPackedR32_SSE2(src), |
- SkGetPackedR32_SSE2(dst)); |
- __m128i g = SkAlphaMulAlpha_SSE2(SkGetPackedG32_SSE2(src), |
- SkGetPackedG32_SSE2(dst)); |
- __m128i b = SkAlphaMulAlpha_SSE2(SkGetPackedB32_SSE2(src), |
- SkGetPackedB32_SSE2(dst)); |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
static inline __m128i SkMin32_SSE2(const __m128i& a, const __m128i& b) { |
__m128i cmp = _mm_cmplt_epi32(a, b); |
return _mm_or_si128(_mm_and_si128(cmp, a), _mm_andnot_si128(cmp, b)); |
@@ -201,58 +56,6 @@ static inline __m128i srcover_byte_SSE2(const __m128i& a, const __m128i& b) { |
} |
-static inline __m128i blendfunc_multiply_byte_SSE2(const __m128i& sc, const __m128i& dc, |
- const __m128i& sa, const __m128i& da) { |
- // sc * (255 - da) |
- __m128i ret1 = _mm_sub_epi32(_mm_set1_epi32(255), da); |
- ret1 = _mm_mullo_epi16(sc, ret1); |
- |
- // dc * (255 - sa) |
- __m128i ret2 = _mm_sub_epi32(_mm_set1_epi32(255), sa); |
- ret2 = _mm_mullo_epi16(dc, ret2); |
- |
- // sc * dc |
- __m128i ret3 = _mm_mullo_epi16(sc, dc); |
- |
- __m128i ret = _mm_add_epi32(ret1, ret2); |
- ret = _mm_add_epi32(ret, ret3); |
- |
- return clamp_div255round_SSE2(ret); |
-} |
- |
-static __m128i multiply_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- __m128i a = srcover_byte_SSE2(sa, da); |
- |
- __m128i sr = SkGetPackedR32_SSE2(src); |
- __m128i dr = SkGetPackedR32_SSE2(dst); |
- __m128i r = blendfunc_multiply_byte_SSE2(sr, dr, sa, da); |
- |
- __m128i sg = SkGetPackedG32_SSE2(src); |
- __m128i dg = SkGetPackedG32_SSE2(dst); |
- __m128i g = blendfunc_multiply_byte_SSE2(sg, dg, sa, da); |
- |
- |
- __m128i sb = SkGetPackedB32_SSE2(src); |
- __m128i db = SkGetPackedB32_SSE2(dst); |
- __m128i b = blendfunc_multiply_byte_SSE2(sb, db, sa, da); |
- |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
-static __m128i screen_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i a = srcover_byte_SSE2(SkGetPackedA32_SSE2(src), |
- SkGetPackedA32_SSE2(dst)); |
- __m128i r = srcover_byte_SSE2(SkGetPackedR32_SSE2(src), |
- SkGetPackedR32_SSE2(dst)); |
- __m128i g = srcover_byte_SSE2(SkGetPackedG32_SSE2(src), |
- SkGetPackedG32_SSE2(dst)); |
- __m128i b = srcover_byte_SSE2(SkGetPackedB32_SSE2(src), |
- SkGetPackedB32_SSE2(dst)); |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
// Portable version overlay_byte() is in SkXfermode.cpp. |
static inline __m128i overlay_byte_SSE2(const __m128i& sc, const __m128i& dc, |
const __m128i& sa, const __m128i& da) { |
@@ -580,67 +383,11 @@ static __m128i softlight_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
return SkPackARGB32_SSE2(a, r, g, b); |
} |
-static inline __m128i difference_byte_SSE2(const __m128i& sc, const __m128i& dc, |
- const __m128i& sa, const __m128i& da) { |
- __m128i tmp1 = _mm_mullo_epi16(sc, da); |
- __m128i tmp2 = _mm_mullo_epi16(dc, sa); |
- __m128i tmp = SkMin32_SSE2(tmp1, tmp2); |
- |
- __m128i ret1 = _mm_add_epi32(sc, dc); |
- __m128i ret2 = _mm_slli_epi32(SkDiv255Round_SSE2(tmp), 1); |
- __m128i ret = _mm_sub_epi32(ret1, ret2); |
- |
- ret = clamp_signed_byte_SSE2(ret); |
- return ret; |
-} |
- |
-static __m128i difference_modeproc_SSE2(const __m128i& src, |
- const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- |
- __m128i a = srcover_byte_SSE2(sa, da); |
- __m128i r = difference_byte_SSE2(SkGetPackedR32_SSE2(src), |
- SkGetPackedR32_SSE2(dst), sa, da); |
- __m128i g = difference_byte_SSE2(SkGetPackedG32_SSE2(src), |
- SkGetPackedG32_SSE2(dst), sa, da); |
- __m128i b = difference_byte_SSE2(SkGetPackedB32_SSE2(src), |
- SkGetPackedB32_SSE2(dst), sa, da); |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
- |
-static inline __m128i exclusion_byte_SSE2(const __m128i& sc, const __m128i& dc, |
- const __m128i&, __m128i&) { |
- __m128i tmp1 = _mm_mullo_epi16(_mm_set1_epi32(255), sc); // 255 * sc |
- __m128i tmp2 = _mm_mullo_epi16(_mm_set1_epi32(255), dc); // 255 * dc |
- tmp1 = _mm_add_epi32(tmp1, tmp2); |
- tmp2 = _mm_mullo_epi16(sc, dc); // sc * dc |
- tmp2 = _mm_slli_epi32(tmp2, 1); // 2 * sc * dc |
- |
- __m128i r = _mm_sub_epi32(tmp1, tmp2); |
- return clamp_div255round_SSE2(r); |
-} |
- |
-static __m128i exclusion_modeproc_SSE2(const __m128i& src, const __m128i& dst) { |
- __m128i sa = SkGetPackedA32_SSE2(src); |
- __m128i da = SkGetPackedA32_SSE2(dst); |
- |
- __m128i a = srcover_byte_SSE2(sa, da); |
- __m128i r = exclusion_byte_SSE2(SkGetPackedR32_SSE2(src), |
- SkGetPackedR32_SSE2(dst), sa, da); |
- __m128i g = exclusion_byte_SSE2(SkGetPackedG32_SSE2(src), |
- SkGetPackedG32_SSE2(dst), sa, da); |
- __m128i b = exclusion_byte_SSE2(SkGetPackedB32_SSE2(src), |
- SkGetPackedB32_SSE2(dst), sa, da); |
- return SkPackARGB32_SSE2(a, r, g, b); |
-} |
//////////////////////////////////////////////////////////////////////////////// |
typedef __m128i (*SkXfermodeProcSIMD)(const __m128i& src, const __m128i& dst); |
-extern SkXfermodeProcSIMD gSSE2XfermodeProcs[]; |
- |
void SkSSE2ProcCoeffXfermode::xfer32(SkPMColor dst[], const SkPMColor src[], |
int count, const SkAlpha aa[]) const { |
SkASSERT(dst && src && count >= 0); |
@@ -765,49 +512,19 @@ void SkSSE2ProcCoeffXfermode::toString(SkString* str) const { |
} |
#endif |
-//////////////////////////////////////////////////////////////////////////////// |
- |
-// 4 pixels modeprocs with SSE2 |
-SkXfermodeProcSIMD gSSE2XfermodeProcs[] = { |
- NULL, // kClear_Mode |
- NULL, // kSrc_Mode |
- NULL, // kDst_Mode |
- srcover_modeproc_SSE2, |
- dstover_modeproc_SSE2, |
- srcin_modeproc_SSE2, |
- dstin_modeproc_SSE2, |
- srcout_modeproc_SSE2, |
- dstout_modeproc_SSE2, |
- srcatop_modeproc_SSE2, |
- dstatop_modeproc_SSE2, |
- xor_modeproc_SSE2, |
- plus_modeproc_SSE2, |
- modulate_modeproc_SSE2, |
- screen_modeproc_SSE2, |
- |
- overlay_modeproc_SSE2, |
- darken_modeproc_SSE2, |
- lighten_modeproc_SSE2, |
- colordodge_modeproc_SSE2, |
- colorburn_modeproc_SSE2, |
- hardlight_modeproc_SSE2, |
- softlight_modeproc_SSE2, |
- difference_modeproc_SSE2, |
- exclusion_modeproc_SSE2, |
- multiply_modeproc_SSE2, |
- |
- NULL, // kHue_Mode |
- NULL, // kSaturation_Mode |
- NULL, // kColor_Mode |
- NULL, // kLuminosity_Mode |
-}; |
- |
SkProcCoeffXfermode* SkPlatformXfermodeFactory_impl_SSE2(const ProcCoeff& rec, |
SkXfermode::Mode mode) { |
- void* procSIMD = reinterpret_cast<void*>(gSSE2XfermodeProcs[mode]); |
- |
- if (procSIMD != NULL) { |
- return SkNEW_ARGS(SkSSE2ProcCoeffXfermode, (rec, mode, procSIMD)); |
+ SkXfermodeProcSIMD proc = nullptr; |
+ // TODO(mtklein): implement these Sk4px. |
+ switch (mode) { |
+ case SkProcCoeffXfermode::kOverlay_Mode: proc = overlay_modeproc_SSE2; break; |
+ case SkProcCoeffXfermode::kDarken_Mode: proc = darken_modeproc_SSE2; break; |
+ case SkProcCoeffXfermode::kLighten_Mode: proc = lighten_modeproc_SSE2; break; |
+ case SkProcCoeffXfermode::kColorDodge_Mode: proc = colordodge_modeproc_SSE2; break; |
+ case SkProcCoeffXfermode::kColorBurn_Mode: proc = colorburn_modeproc_SSE2; break; |
+ case SkProcCoeffXfermode::kHardLight_Mode: proc = hardlight_modeproc_SSE2; break; |
+ case SkProcCoeffXfermode::kSoftLight_Mode: proc = softlight_modeproc_SSE2; break; |
+ default: break; |
} |
- return NULL; |
+ return proc ? SkNEW_ARGS(SkSSE2ProcCoeffXfermode, (rec, mode, (void*)proc)) : nullptr; |
} |