Changeset 94026 in webkit

Timestamp:

Aug 29, 2011, 3:54:06 PM (15 years ago)

Author:

commit-queue@webkit.org

Message:

Speed up texImage from BGRA
​https://bugs.webkit.org/show_bug.cgi?id=66884

Patch by John Bauman <​jbauman@chromium.org> on 2011-08-29
Reviewed by Kenneth Russell.

BGRA input is common coming from skia, so optimize BGRA->RGBA
conversion and also avoid the pointless RGBA to RGBA conversion.

• platform/graphics/GraphicsContext3D.cpp:

(WebCore::doUnpackingAndPacking):

Location:

trunk/Source/WebCore

Files:

• ChangeLog (modified) (1 diff)
• platform/graphics/GraphicsContext3D.cpp (modified) (5 diffs)

trunk/Source/WebCore/ChangeLog

@@ +1 @@
+2011-08-29  John Bauman  <jbauman@chromium.org>
+
+        Speed up texImage from BGRA
+        https://bugs.webkit.org/show_bug.cgi?id=66884
+
+        Reviewed by Kenneth Russell.
+
+        BGRA input is common coming from skia, so optimize BGRA->RGBA
+        conversion and also avoid the pointless RGBA to RGBA conversion.
+
+        * platform/graphics/GraphicsContext3D.cpp:
+        (WebCore::doUnpackingAndPacking):
+
 2011-08-29  Matthew Delaney  <mdelaney@apple.com>
 

trunk/Source/WebCore/platform/graphics/GraphicsContext3D.cpp

@@ -475 +475 @@
 void unpackOneRowOfBGRA8ToRGBA8(const uint8_t* source, uint8_t* destination, unsigned int pixelsPerRow)
 {
-    for (unsigned int i = 0; i < pixelsPerRow; ++i) {
-        destination[0] = source[2];
-        destination[1] = source[1];
-        destination[2] = source[0];
-        destination[3] = source[3];
-        source += 4;
-        destination += 4;
+    const uint32_t* source32 = reinterpret_cast<const uint32_t*>(source);
+    uint32_t* destination32 = reinterpret_cast<uint32_t*>(destination);
+    for (unsigned int i = 0; i < pixelsPerRow; ++i) {
+        uint32_t bgra = source32[i];
+#if CPU(BIG_ENDIAN)
+        uint32_t brMask = 0xff00ff00;
+        uint32_t gaMask = 0x00ff00ff;
+#else
+        uint32_t brMask = 0x00ff00ff;
+        uint32_t gaMask = 0xff00ff00;
+#endif
+        uint32_t rgba = (((bgra >> 16) | (bgra << 16)) & brMask) | (bgra & gaMask);
+        destination32[i] = rgba;
     }
 }
@@ -869 +875 @@
         source += 4;
         destination += 3;
-    }
-}
-
-// This is only used when the source format is different than SourceFormatRGBA8.
-void packOneRowOfRGBA8ToRGBA8(const uint8_t* source, uint8_t* destination, unsigned int pixelsPerRow)
-{
-    for (unsigned int i = 0; i < pixelsPerRow; ++i) {
-        destination[0] = source[0];
-        destination[1] = source[1];
-        destination[2] = source[2];
-        destination[3] = source[3];
-        source += 4;
-        destination += 4;
     }
 }
@@ -1151 +1144 @@
                                   unsigned int destinationElementsPerPixel)
 {
-    OwnArrayPtr<IntermediateType> temporaryRGBAData = adoptArrayPtr(new IntermediateType[width * 4]);
-    const SourceType* endPointer = sourceData + height * sourceElementsPerRow;
-    unsigned int destinationElementsPerRow = width * destinationElementsPerPixel;
-    while (sourceData < endPointer) {
-        rowUnpackingFunc(sourceData, temporaryRGBAData.get(), width);
-        rowPackingFunc(temporaryRGBAData.get(), destinationData, width);
-        sourceData += sourceElementsPerRow;
-        destinationData += destinationElementsPerRow;
+    if (!rowPackingFunc) {
+        // The row packing is trivial, so don't bother with a temporary buffer.
+        const SourceType* endPointer = sourceData + height * sourceElementsPerRow;
+        unsigned int destinationElementsPerRow = width * destinationElementsPerPixel;
+        while (sourceData < endPointer) {
+            rowUnpackingFunc(sourceData, reinterpret_cast<IntermediateType*>(destinationData), width);
+            sourceData += sourceElementsPerRow;
+            destinationData += destinationElementsPerRow;
+        }
+    } else {
+        OwnArrayPtr<IntermediateType> temporaryRGBAData = adoptArrayPtr(new IntermediateType[width * 4]);
+        const SourceType* endPointer = sourceData + height * sourceElementsPerRow;
+        unsigned int destinationElementsPerRow = width * destinationElementsPerPixel;
+        while (sourceData < endPointer) {
+            rowUnpackingFunc(sourceData, temporaryRGBAData.get(), width);
+            rowPackingFunc(temporaryRGBAData.get(), destinationData, width);
+            sourceData += sourceElementsPerRow;
+            destinationData += destinationElementsPerRow;
+        }
     }
 }
@@ -1198 +1202 @@
         unsigned int destinationElementsPerRow = width * destinationElementsPerPixel;
         while (source < endPointer) {
-            rowPackingFunc(source, destinationData, width);
+            if (rowPackingFunc)
+                rowPackingFunc(source, destinationData, width);
+            else
+                memcpy(destinationData, source, width * 4);
             source += sourceElementsPerRow;
             destinationData += destinationElementsPerRow;
@@ -1436 +1443 @@
             case AlphaDoNothing:
                 ASSERT(sourceDataFormat != SourceFormatRGBA8 || sourceUnpackAlignment > 4); // Handled above with fast case.
-                doPacking<uint8_t>(sourceData, sourceDataFormat, width, height, sourceUnpackAlignment, destination, packOneRowOfRGBA8ToRGBA8, 4);
+                doPacking<uint8_t>(sourceData, sourceDataFormat, width, height, sourceUnpackAlignment, destination, 0, 4);
                 break;
             case AlphaDoPremultiply: