aboutsummaryrefslogtreecommitdiffhomepage
diff options
context:
space:
mode:
-rw-r--r--include/gpu/GrGLConfig.h10
-rw-r--r--include/gpu/GrGLConfig_chrome.h4
-rw-r--r--src/gpu/GrContext.cpp20
-rw-r--r--src/gpu/GrGpu.h8
-rw-r--r--src/gpu/GrGpuGL.cpp85
-rw-r--r--src/gpu/GrGpuGL.h2
6 files changed, 102 insertions, 27 deletions
diff --git a/include/gpu/GrGLConfig.h b/include/gpu/GrGLConfig.h
index ea7bcf341c..cfb5141a18 100644
--- a/include/gpu/GrGLConfig.h
+++ b/include/gpu/GrGLConfig.h
@@ -78,9 +78,9 @@
* The GrGLInterface field fCallback specifies the function ptr and there is an
* additional field fCallbackData of type intptr_t for client data.
*
- * GR_GL_RGBA_8888_READBACK_SLOW: Set this to 1 if it is known that performing
- * glReadPixels with format=GL_RGBA, type=GL_UNISIGNED_BYTE is significantly
- * slower than format=GL_BGRA, type=GL_UNISIGNED_BYTE.
+ * GR_GL_RGBA_8888_PIXEL_OPS_SLOW: Set this to 1 if it is known that performing
+ * glReadPixels / glTex(Sub)Image with format=GL_RGBA, type=GL_UNISIGNED_BYTE is
+ * significantly slower than format=GL_BGRA, type=GL_UNISIGNED_BYTE.
*/
#if !defined(GR_GL_LOG_CALLS)
@@ -115,8 +115,8 @@
#define GR_GL_PER_GL_FUNC_CALLBACK 0
#endif
-#if !defined(GR_GL_RGBA_8888_READBACK_SLOW)
- #define GR_GL_RGBA_8888_READBACK_SLOW 0
+#if !defined(GR_GL_RGBA_8888_PIXEL_OPS_SLOW)
+ #define GR_GL_RGBA_8888_PIXEL_OPS_SLOW 0
#endif
#if(GR_GL_NO_CONSTANT_ATTRIBUTES) && (GR_GL_ATTRIBUTE_MATRICES)
diff --git a/include/gpu/GrGLConfig_chrome.h b/include/gpu/GrGLConfig_chrome.h
index 79324ab511..72d330acc5 100644
--- a/include/gpu/GrGLConfig_chrome.h
+++ b/include/gpu/GrGLConfig_chrome.h
@@ -14,8 +14,8 @@
// ANGLE creates a temp VB for vertex attributes not specified per-vertex.
#define GR_GL_NO_CONSTANT_ATTRIBUTES GR_WIN32_BUILD
-// RGBA Readbacks are a slow path in ANGLE
-#define GR_GL_RGBA_8888_READBACK_SLOW GR_WIN32_BUILD
+// For RGBA teximage/readpixels ANGLE will sw-convert to/from BGRA.
+#define GR_GL_RGBA_8888_PIXEL_OPS_SLOW GR_WIN32_BUILD
// cmd buffer allocates memory and memsets it to zero when it sees glBufferData
// with NULL.
diff --git a/src/gpu/GrContext.cpp b/src/gpu/GrContext.cpp
index fe85b97b92..3a26d2f609 100644
--- a/src/gpu/GrContext.cpp
+++ b/src/gpu/GrContext.cpp
@@ -1803,6 +1803,25 @@ void GrContext::internalWriteRenderTargetPixels(GrRenderTarget* target,
// TODO: when underlying api has a direct way to do this we should use it
// (e.g. glDrawPixels on desktop GL).
+ // If the RT is also a texture and we don't have to do PM/UPM conversion
+ // then take the texture path, which we expect to be at least as fast or
+ // faster since it doesn't use an intermediate texture as we do below.
+ if (NULL != target->asTexture() &&
+ GrPixelConfigIsUnpremultiplied(target->config()) ==
+ GrPixelConfigIsUnpremultiplied(config)) {
+
+ this->internalWriteTexturePixels(target->asTexture(),
+ left, top, width, height,
+ config, buffer, rowBytes, flags);
+ return;
+ }
+
+ bool swapRAndB = fGpu->preferredReadPixelsConfig(config) ==
+ GrPixelConfigSwapRAndB(config);
+ if (swapRAndB) {
+ config = GrPixelConfigSwapRAndB(config);
+ }
+
const GrTextureDesc desc = {
kNone_GrTextureFlags, kNone_GrAALevel, width, height, { config }
};
@@ -1827,6 +1846,7 @@ void GrContext::internalWriteRenderTargetPixels(GrRenderTarget* target,
sampler.setClampNoFilter();
matrix.setIDiv(texture->width(), texture->height());
sampler.setMatrix(matrix);
+ sampler.setRAndBSwap(swapRAndB);
fGpu->setSamplerState(0, sampler);
GrVertexLayout layout = GrDrawTarget::StagePosAsTexCoordVertexLayoutBit(0);
diff --git a/src/gpu/GrGpu.h b/src/gpu/GrGpu.h
index 9b39319f4f..6741aec513 100644
--- a/src/gpu/GrGpu.h
+++ b/src/gpu/GrGpu.h
@@ -195,6 +195,13 @@ public:
}
/**
+ * Same as above but applies to writeTexturePixels
+ */
+ virtual GrPixelConfig preferredWritePixelsConfig(GrPixelConfig config) {
+ return config;
+ }
+
+ /**
* OpenGL's readPixels returns the result bottom-to-top while the skia
* API is top-to-bottom. Thus we have to do a y-axis flip. The obvious
* solution is to have the subclass do the flip using either the CPU or GPU.
@@ -247,6 +254,7 @@ public:
/**
* Updates the pixels in a rectangle of a texture.
+ *
* @param left left edge of the rectangle to write (inclusive)
* @param top top edge of the rectangle to write (inclusive)
* @param width width of rectangle to write in pixels.
diff --git a/src/gpu/GrGpuGL.cpp b/src/gpu/GrGpuGL.cpp
index facfac914a..cd37a54e9a 100644
--- a/src/gpu/GrGpuGL.cpp
+++ b/src/gpu/GrGpuGL.cpp
@@ -439,7 +439,15 @@ void GrGpuGL::initStencilFormats() {
}
GrPixelConfig GrGpuGL::preferredReadPixelsConfig(GrPixelConfig config) {
- if (GR_GL_RGBA_8888_READBACK_SLOW && GrPixelConfigIsRGBA8888(config)) {
+ if (GR_GL_RGBA_8888_PIXEL_OPS_SLOW && GrPixelConfigIsRGBA8888(config)) {
+ return GrPixelConfigSwapRAndB(config);
+ } else {
+ return config;
+ }
+}
+
+GrPixelConfig GrGpuGL::preferredWritePixelsConfig(GrPixelConfig config) {
+ if (GR_GL_RGBA_8888_PIXEL_OPS_SLOW && GrPixelConfigIsRGBA8888(config)) {
return GrPixelConfigSwapRAndB(config);
} else {
return config;
@@ -693,21 +701,46 @@ void GrGpuGL::onWriteTexturePixels(GrTexture* texture,
this->uploadTexData(desc, left, top, width, height, config, buffer, rowBytes);
}
+namespace {
+bool adjust_pixel_ops_params(int surfaceWidth,
+ int surfaceHeight,
+ size_t bpp,
+ int* left, int* top, int* width, int* height,
+ const void** data,
+ size_t* rowBytes) {
+ if (!*rowBytes) {
+ *rowBytes = *width * bpp;
+ }
+
+ GrIRect subRect = GrIRect::MakeXYWH(*left, *top, *width, *height);
+ GrIRect bounds = GrIRect::MakeWH(surfaceWidth, surfaceHeight);
+
+ if (!subRect.intersect(bounds)) {
+ return false;
+ }
+ *data = reinterpret_cast<const void*>(reinterpret_cast<intptr_t>(*data) +
+ (subRect.fTop - *top) * *rowBytes + (subRect.fLeft - *left) * bpp);
+
+ *left = subRect.fLeft;
+ *top = subRect.fTop;
+ *width = subRect.width();
+ *height = subRect.height();
+ return true;
+}
+}
+
void GrGpuGL::uploadTexData(const GrGLTexture::Desc& desc,
int left, int top, int width, int height,
GrPixelConfig dataConfig,
const void* data,
size_t rowBytes) {
- GrIRect bounds = GrIRect::MakeWH(desc.fWidth, desc.fHeight);
- GrIRect subrect = GrIRect::MakeXYWH(left, top, width, height);
- if (!bounds.contains(subrect)) {
+
+ size_t bpp = GrBytesPerPixel(dataConfig);
+ if (!adjust_pixel_ops_params(desc.fWidth, desc.fHeight, bpp, &left, &top,
+ &width, &height, &data, &rowBytes)) {
return;
}
-
- // ES2 glCompressedTexSubImage2D doesn't support any formats
- // (at least without extensions)
- GrAssert(desc.fInternalFormat != GR_GL_PALETTE8_RGBA8 ||
- bounds == subrect);
+ size_t trimRowBytes = width * bpp;
// in case we need a temporary, trimmed copy of the src pixels
SkAutoSMalloc<128 * 128> tempStorage;
@@ -720,11 +753,6 @@ void GrGpuGL::uploadTexData(const GrGLTexture::Desc& desc,
return;
}
- size_t bpp = GrBytesPerPixel(dataConfig);
- size_t trimRowBytes = width * bpp;
- if (!rowBytes) {
- rowBytes = trimRowBytes;
- }
/*
* check whether to allocate a temporary buffer for flipping y or
* because our srcData has extra bytes past each row. If so, we need
@@ -758,13 +786,14 @@ void GrGpuGL::uploadTexData(const GrGLTexture::Desc& desc,
}
dst += trimRowBytes;
}
- // now point dat to our copied version
+ // now point data to our copied version
data = tempStorage.get();
}
}
GL_CALL(PixelStorei(GR_GL_UNPACK_ALIGNMENT, static_cast<GrGLint>(bpp)));
- if (bounds == subrect) {
+ if (0 == left && 0 == top &&
+ desc.fWidth == width && desc.fHeight == height) {
GL_CALL(TexImage2D(GR_GL_TEXTURE_2D, 0, desc.fInternalFormat,
desc.fWidth, desc.fHeight, 0,
externalFormat, externalType, data));
@@ -1320,9 +1349,18 @@ bool GrGpuGL::readPixelsWillPayForYFlip(GrRenderTarget* renderTarget,
size_t rowBytes) {
// if we have to do memcpy to handle non-trim rowBytes then we
// get the flip for free. Otherwise it costs.
- return this->glCaps().fPackRowLengthSupport ||
- 0 == rowBytes ||
- GrBytesPerPixel(config) * width == rowBytes;
+ if (this->glCaps().fPackRowLengthSupport) {
+ return true;
+ }
+ // If we have to do memcpys to handle rowBytes then y-flip is free
+ // Note the rowBytes might be tight to the passed in data, but if data
+ // gets clipped in x to the target the rowBytes will no longer be tight.
+ if (left >= 0 && (left + width) < renderTarget->width()) {
+ return 0 == rowBytes ||
+ GrBytesPerPixel(config) * width == rowBytes;
+ } else {
+ return false;
+ }
}
bool GrGpuGL::onReadPixels(GrRenderTarget* target,
@@ -1338,6 +1376,13 @@ bool GrGpuGL::onReadPixels(GrRenderTarget* target,
if (!this->canBeTexture(config, &internalFormat, &format, &type)) {
return false;
}
+ size_t bpp = GrBytesPerPixel(config);
+ if (!adjust_pixel_ops_params(target->width(), target->height(), bpp,
+ &left, &top, &width, &height,
+ const_cast<const void**>(&buffer),
+ &rowBytes)) {
+ return false;
+ }
// resolve the render target if necessary
GrGLRenderTarget* tgt = static_cast<GrGLRenderTarget*>(target);
@@ -1366,7 +1411,7 @@ bool GrGpuGL::onReadPixels(GrRenderTarget* target,
GrGLIRect readRect;
readRect.setRelativeTo(glvp, left, top, width, height);
- size_t tightRowBytes = GrBytesPerPixel(config) * width;
+ size_t tightRowBytes = bpp * width;
if (0 == rowBytes) {
rowBytes = tightRowBytes;
}
diff --git a/src/gpu/GrGpuGL.h b/src/gpu/GrGpuGL.h
index 18baedf968..3b8e6d74af 100644
--- a/src/gpu/GrGpuGL.h
+++ b/src/gpu/GrGpuGL.h
@@ -31,6 +31,8 @@ public:
virtual GrPixelConfig preferredReadPixelsConfig(GrPixelConfig config)
SK_OVERRIDE;
+ virtual GrPixelConfig preferredWritePixelsConfig(GrPixelConfig config)
+ SK_OVERRIDE;
virtual bool readPixelsWillPayForYFlip(
GrRenderTarget* renderTarget,