Modify sample buffer size for larger displays.

src/core/SkBitmapProcShader.cpp

Index: src/core/SkBitmapProcShader.cpp
diff --git a/src/core/SkBitmapProcShader.cpp b/src/core/SkBitmapProcShader.cpp
index 5f5eb1866e4282ba2a6e6b12237d5ea5a13deb08..7aa62ba2f2a7c21b817757074eb5d8a5225752fe 100644
--- a/src/core/SkBitmapProcShader.cpp
+++ b/src/core/SkBitmapProcShader.cpp
@@ -205,13 +205,23 @@ SkBitmapProcShader::BitmapProcShaderContext::~BitmapProcShaderContext() {
     fState->~SkBitmapProcState();
 }
 
-#define BUF_MAX     128
+/* Defines the buffer size for sample pixel indexes, used in the sample proc function calls.
+ * If the operation is bigger than the buffer, it's split into multiple calls. This split is bad
+ * for the performance of SIMD optimizations.
+ * A display in portrait mode, with a width of 720 pixels, requires a buffer size of at least 361
+ * to run uninterrupted in the most basic operations. (Formula: Width / 2 + 1)
+ */
+#ifdef ANDROID_LARGE_MEMORY_DEVICE
+    #define BUF_MAX     541 // 1080p phone display
+#else
+    #define BUF_MAX     241 // 480p phone display
+#endif
 
 #define TEST_BUFFER_OVERRITEx
 
 #ifdef TEST_BUFFER_OVERRITE
     #define TEST_BUFFER_EXTRA   32
-    #define TEST_PATTERN    0x88888888
+    #define TEST_PATTERN        0x88888888
 #else
     #define TEST_BUFFER_EXTRA   0
 #endif
@@ -224,7 +234,12 @@ void SkBitmapProcShader::BitmapProcShaderContext::shadeSpan(int x, int y, SkPMCo
         return;
     }
 
-    uint32_t buffer[BUF_MAX + TEST_BUFFER_EXTRA];
+    // Align buffer to 16 bytes to enable more efficient SIMD optimizations.
+#ifdef _MSC_VER
+    __declspec(align(16)) uint32_t buffer[BUF_MAX + TEST_BUFFER_EXTRA];

[scroggo, 2014/04/22 15:18:40]

Is there a way to turn this into a macro? Or does the different order for
align(ed) and uint32_t prevent that?

[henrik.smiding, 2014/04/23 13:34:46]

Done.

+#else
+    uint32_t __attribute__((aligned(16))) buffer[BUF_MAX + TEST_BUFFER_EXTRA];
+#endif
     SkBitmapProcState::MatrixProc   mproc = state.getMatrixProc();
     SkBitmapProcState::SampleProc32 sproc = state.getSampleProc32();
     int max = state.maxCountForBufferSize(sizeof(buffer[0]) * BUF_MAX);