Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(249)

Issues Search
    My Issues | Starred     Open | Closed | All

Unified Diff: src/codec/SkSwizzler.cpp

Issue 1680743005: NEON Optimized RGBA->PMColor sampling in SkSwizzler (Closed) Base URL: https://skia.googlesource.com/skia.git@master
Patch Set: Share implementations for sampleSize 4 and 8 Created 4 years, 10 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View side-by-side diff with in-line comments
Download patch
« no previous file with comments | « src/codec/SkSwizzler.h ('k') | src/core/SkOpts.h » ('j') | no next file with comments »
Expand Comments ('e') | Collapse Comments ('c') | Show Comments Hide Comments ('s')
Index: src/codec/SkSwizzler.cpp
diff --git a/src/codec/SkSwizzler.cpp b/src/codec/SkSwizzler.cpp
index f9f212072e53e4bb2565e87aadb6fdff0213c1e6..36c3e684eb82a9cd0c8c4e5f31f4e7bcdeae808b 100644
--- a/src/codec/SkSwizzler.cpp
+++ b/src/codec/SkSwizzler.cpp
@@ -20,7 +20,7 @@ static void copy(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc
memcpy(dst, src + offset, width * bpp);
}
-static void sample1(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
+static void sample_1bpp(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
const SkPMColor ctable[]) {
src += offset;
uint8_t* dst8 = (uint8_t*) dst;
@@ -30,7 +30,7 @@ static void sample1(void* dst, const uint8_t* src, int width, int bpp, int delta
}
}
-static void sample2(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
+static void sample_2bpp(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
const SkPMColor ctable[]) {
src += offset;
uint16_t* dst16 = (uint16_t*) dst;
@@ -40,7 +40,7 @@ static void sample2(void* dst, const uint8_t* src, int width, int bpp, int delta
}
}
-static void sample4(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
+static void sample_4bpp(void* dst, const uint8_t* src, int width, int bpp, int deltaSrc, int offset,
const SkPMColor ctable[]) {
src += offset;
uint32_t* dst32 = (uint32_t*) dst;
@@ -503,6 +503,48 @@ static void fast_swizzle_rgba_to_n32_premul(
#endif
}
+static void fast_sample2_rgba_to_n32_premul(
+ void* dst, const uint8_t* src, int width, int bpp, int deltaSrc,
+ int offset, const SkPMColor ctable[]) {
+
+ // This function should only be called if the sampleSize is 2.
+ SkASSERT(deltaSrc == 2*bpp);
+
+#ifdef SK_PMCOLOR_IS_RGBA
+ SkOpts::RGBA_to_rgbA_sample2((uint32_t*) dst, src + offset, width);
+#else
+ SkOpts::RGBA_to_bgrA_sample2((uint32_t*) dst, src + offset, width);
+#endif
+}
+
+static void fast_sample4_rgba_to_n32_premul(
+ void* dst, const uint8_t* src, int width, int bpp, int deltaSrc,
+ int offset, const SkPMColor ctable[]) {
+
+ // This function should only be called if the sampleSize is 4.
+ SkASSERT(deltaSrc == 4*bpp);
+
+#ifdef SK_PMCOLOR_IS_RGBA
+ SkOpts::RGBA_to_rgbA_sample4((uint32_t*) dst, src + offset, width);
+#else
+ SkOpts::RGBA_to_bgrA_sample4((uint32_t*) dst, src + offset, width);
+#endif
+}
+
+static void fast_sample8_rgba_to_n32_premul(
+ void* dst, const uint8_t* src, int width, int bpp, int deltaSrc,
+ int offset, const SkPMColor ctable[]) {
+
+ // This function should only be called if the sampleSize is 8.
+ SkASSERT(deltaSrc == 8*bpp);
+
+#ifdef SK_PMCOLOR_IS_RGBA
+ SkOpts::RGBA_to_rgbA_sample8((uint32_t*) dst, src + offset, width);
+#else
+ SkOpts::RGBA_to_bgrA_sample8((uint32_t*) dst, src + offset, width);
+#endif
+}
+
static void swizzle_rgba_to_n32_unpremul(
void* SK_RESTRICT dstRow, const uint8_t* SK_RESTRICT src, int dstWidth,
int bpp, int deltaSrc, int offset, const SkPMColor ctable[]) {
@@ -673,8 +715,9 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
&& nullptr == ctable) {
return nullptr;
}
- RowProc fastProc = nullptr;
+
RowProc proc = nullptr;
+ FastProcs fastProcs;
SkCodec::ZeroInitialized zeroInit = options.fZeroInitialized;
switch (sc) {
case kBit:
@@ -728,8 +771,8 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
proc = &swizzle_index_to_565;
break;
case kIndex_8_SkColorType:
- proc = &sample1;
- fastProc = ©
+ proc = &sample_1bpp;
+ fastProcs.fSampleSize1 = ©
break;
default:
break;
@@ -739,11 +782,11 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
switch (dstInfo.colorType()) {
case kN32_SkColorType:
proc = &swizzle_gray_to_n32;
- fastProc = &fast_swizzle_gray_to_n32;
+ fastProcs.fSampleSize1 = &fast_swizzle_gray_to_n32;
break;
case kGray_8_SkColorType:
- proc = &sample1;
- fastProc = ©
+ proc = &sample_1bpp;
+ fastProcs.fSampleSize1 = ©
break;
case kRGB_565_SkColorType:
proc = &swizzle_gray_to_565;
@@ -759,20 +802,20 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
if (SkCodec::kYes_ZeroInitialized == zeroInit) {
proc = &SkipLeadingGrayAlphaZerosThen
<swizzle_grayalpha_to_n32_unpremul>;
- fastProc = &SkipLeadingGrayAlphaZerosThen
+ fastProcs.fSampleSize1 = &SkipLeadingGrayAlphaZerosThen
<fast_swizzle_grayalpha_to_n32_unpremul>;
} else {
proc = &swizzle_grayalpha_to_n32_unpremul;
- fastProc = &fast_swizzle_grayalpha_to_n32_unpremul;
+ fastProcs.fSampleSize1 = &fast_swizzle_grayalpha_to_n32_unpremul;
}
} else {
if (SkCodec::kYes_ZeroInitialized == zeroInit) {
proc = &SkipLeadingGrayAlphaZerosThen<swizzle_grayalpha_to_n32_premul>;
- fastProc = &SkipLeadingGrayAlphaZerosThen
+ fastProcs.fSampleSize1 = &SkipLeadingGrayAlphaZerosThen
<fast_swizzle_grayalpha_to_n32_premul>;
} else {
proc = &swizzle_grayalpha_to_n32_premul;
- fastProc = &fast_swizzle_grayalpha_to_n32_premul;
+ fastProcs.fSampleSize1 = &fast_swizzle_grayalpha_to_n32_premul;
}
}
break;
@@ -799,18 +842,20 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
if (dstInfo.alphaType() == kUnpremul_SkAlphaType) {
if (SkCodec::kYes_ZeroInitialized == zeroInit) {
proc = &SkipLeading8888ZerosThen<swizzle_bgra_to_n32_unpremul>;
- fastProc = &SkipLeading8888ZerosThen<fast_swizzle_bgra_to_n32_unpremul>;
+ fastProcs.fSampleSize1 = &SkipLeading8888ZerosThen
+ <fast_swizzle_bgra_to_n32_unpremul>;
} else {
proc = &swizzle_bgra_to_n32_unpremul;
- fastProc = &fast_swizzle_bgra_to_n32_unpremul;
+ fastProcs.fSampleSize1 = &fast_swizzle_bgra_to_n32_unpremul;
}
} else {
if (SkCodec::kYes_ZeroInitialized == zeroInit) {
proc = &SkipLeading8888ZerosThen<swizzle_bgra_to_n32_premul>;
- fastProc = &SkipLeading8888ZerosThen<fast_swizzle_bgra_to_n32_premul>;
+ fastProcs.fSampleSize1 = &SkipLeading8888ZerosThen
+ <fast_swizzle_bgra_to_n32_premul>;
} else {
proc = &swizzle_bgra_to_n32_premul;
- fastProc = &fast_swizzle_bgra_to_n32_premul;
+ fastProcs.fSampleSize1 = &fast_swizzle_bgra_to_n32_premul;
}
}
break;
@@ -822,7 +867,7 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
switch (dstInfo.colorType()) {
case kN32_SkColorType:
proc = &swizzle_rgb_to_n32;
- fastProc = &fast_swizzle_rgb_to_n32;
+ fastProcs.fSampleSize1 = &fast_swizzle_rgb_to_n32;
break;
case kRGB_565_SkColorType:
proc = &swizzle_rgb_to_565;
@@ -837,18 +882,29 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
if (dstInfo.alphaType() == kUnpremul_SkAlphaType) {
if (SkCodec::kYes_ZeroInitialized == zeroInit) {
proc = &SkipLeading8888ZerosThen<swizzle_rgba_to_n32_unpremul>;
- fastProc = &SkipLeading8888ZerosThen<fast_swizzle_rgba_to_n32_unpremul>;
+ fastProcs.fSampleSize1 = &SkipLeading8888ZerosThen
+ <fast_swizzle_rgba_to_n32_unpremul>;
} else {
proc = &swizzle_rgba_to_n32_unpremul;
- fastProc = &fast_swizzle_rgba_to_n32_unpremul;
+ fastProcs.fSampleSize1 = &fast_swizzle_rgba_to_n32_unpremul;
}
} else {
if (SkCodec::kYes_ZeroInitialized == zeroInit) {
proc = &SkipLeading8888ZerosThen<swizzle_rgba_to_n32_premul>;
- fastProc = &SkipLeading8888ZerosThen<fast_swizzle_rgba_to_n32_premul>;
+ fastProcs.fSampleSize1 = &SkipLeading8888ZerosThen
+ <fast_swizzle_rgba_to_n32_premul>;
+ fastProcs.fSampleSize2 = &SkipLeading8888ZerosThen
+ <fast_sample2_rgba_to_n32_premul>;
+ fastProcs.fSampleSize4 = &SkipLeading8888ZerosThen
+ <fast_sample4_rgba_to_n32_premul>;
+ fastProcs.fSampleSize8 = &SkipLeading8888ZerosThen
+ <fast_sample8_rgba_to_n32_premul>;
} else {
proc = &swizzle_rgba_to_n32_premul;
- fastProc = &fast_swizzle_rgba_to_n32_premul;
+ fastProcs.fSampleSize1 = &fast_swizzle_rgba_to_n32_premul;
+ fastProcs.fSampleSize2 = &fast_sample2_rgba_to_n32_premul;
+ fastProcs.fSampleSize4 = &fast_sample4_rgba_to_n32_premul;
+ fastProcs.fSampleSize8 = &fast_sample8_rgba_to_n32_premul;
}
}
break;
@@ -860,7 +916,7 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
switch (dstInfo.colorType()) {
case kN32_SkColorType:
proc = &swizzle_cmyk_to_n32;
- fastProc = &fast_swizzle_cmyk_to_n32;
+ fastProcs.fSampleSize1 = &fast_swizzle_cmyk_to_n32;
break;
case kRGB_565_SkColorType:
proc = &swizzle_cmyk_to_565;
@@ -870,16 +926,16 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
}
break;
case kNoOp8:
- proc = &sample1;
- fastProc = &copy;
+ proc = &sample_1bpp;
+ fastProcs.fSampleSize1 = &copy;
break;
case kNoOp16:
- proc = sample2;
- fastProc = &copy;
+ proc = sample_2bpp;
+ fastProcs.fSampleSize1 = &copy;
break;
case kNoOp32:
- proc = &sample4;
- fastProc = &copy;
+ proc = &sample_4bpp;
+ fastProcs.fSampleSize1 = &copy;
break;
default:
break;
@@ -905,15 +961,15 @@ SkSwizzler* SkSwizzler::CreateSwizzler(SkSwizzler::SrcConfig sc,
srcWidth = frame->width();
}
- return new SkSwizzler(fastProc, proc, ctable, srcOffset, srcWidth, dstOffset, dstWidth,
+ return new SkSwizzler(proc, fastProcs, ctable, srcOffset, srcWidth, dstOffset, dstWidth,
srcBPP, dstBPP);
}
-SkSwizzler::SkSwizzler(RowProc fastProc, RowProc proc, const SkPMColor* ctable, int srcOffset,
- int srcWidth, int dstOffset, int dstWidth, int srcBPP, int dstBPP)
- : fFastProc(fastProc)
- , fSlowProc(proc)
- , fActualProc(fFastProc ? fFastProc : fSlowProc)
+SkSwizzler::SkSwizzler(RowProc proc, const FastProcs& fastProcs, const SkPMColor* ctable,
+ int srcOffset, int srcWidth, int dstOffset, int dstWidth, int srcBPP, int dstBPP)
+ : fSlowProc(proc)
+ , fFastProcs(fastProcs)
+ , fActualProc(fFastProcs.fSampleSize1 ? fFastProcs.fSampleSize1 : fSlowProc)
, fColorTable(ctable)
, fSrcOffset(srcOffset)
, fDstOffset(dstOffset)
@@ -937,10 +993,27 @@ int SkSwizzler::onSetSampleX(int sampleX) {
fSwizzleWidth = get_scaled_dimension(fSrcWidth, sampleX);
fAllocatedWidth = get_scaled_dimension(fDstWidth, sampleX);
- // The optimized swizzler routines do not (yet) support sampling.
- if (1 == fSampleX && fFastProc) {
- fActualProc = fFastProc;
- } else {
+ // Choose an optimized swizzler function (if it exists).
+ fActualProc = nullptr;
+ switch (fSampleX) {
+ case 1:
+ fActualProc = fFastProcs.fSampleSize1;
+ break;
+ case 2:
+ fActualProc = fFastProcs.fSampleSize2;
+ break;
+ case 4:
+ fActualProc = fFastProcs.fSampleSize4;
+ break;
+ case 8:
+ fActualProc = fFastProcs.fSampleSize8;
+ break;
+ default:
+ break;
+ }
+
+ // Otherwise, use fSlowProc.
+ if (!fActualProc) {
fActualProc = fSlowProc;
}
« no previous file with comments | « src/codec/SkSwizzler.h ('k') | src/core/SkOpts.h » ('j') | no next file with comments »

Powered by Google App Engine
This is Rietveld 408576698