Add support for NEON intrinsics to speed up texture compression. We can

src/utils/SkTextureCompressor.cpp

Index: src/utils/SkTextureCompressor.cpp
diff --git a/src/utils/SkTextureCompressor.cpp b/src/utils/SkTextureCompressor.cpp
index c4a6293ed26f6a4407a0f20332e088a8119b8319..a41a8e94d80d9a2dade8a9f36c40a9e0e8c4500e 100644
--- a/src/utils/SkTextureCompressor.cpp
+++ b/src/utils/SkTextureCompressor.cpp
@@ -11,6 +11,8 @@
 #include "SkData.h"
 #include "SkEndian.h"
 
+#include "SkTextureCompression_opts.h"
+
 ////////////////////////////////////////////////////////////////////////////////
 //
 // Utility Functions
@@ -586,15 +588,14 @@ static inline uint64_t interleave6(uint64_t topRows, uint64_t bottomRows) {
 
     // x: b f 00 00 00 a e c g i m 00 00 00 d h j n 00 k o 00 l p
 
-    x |= ((x << 52) & (0x3FULL << 52));
-    x = (x | ((x << 20) & (0x3FULL << 28))) >> 16;
+    x = (x | ((x << 52) & (0x3FULL << 52)) | ((x << 20) & (0x3FULL << 28))) >> 16;
 
-#if defined (SK_CPU_BENDIAN)
     // x: 00 00 00 00 00 00 00 00 b f l p a e c g i m k o d h j n
 
     t = (x ^ (x >> 6)) & 0xFC0000ULL;
     x = x ^ t ^ (t << 6);
 
+#if defined (SK_CPU_BENDIAN)
     // x: 00 00 00 00 00 00 00 00 b f l p a e i m c g k o d h j n
 
     t = (x ^ (x >> 36)) & 0x3FULL;
@@ -610,11 +611,6 @@ static inline uint64_t interleave6(uint64_t topRows, uint64_t bottomRows) {
 #else
     // If our CPU is little endian, then the above logic will
     // produce the following indices:
-    // x: 00 00 00 00 00 00 00 00 c g i m d h b f l p j n a e k o
-
-    t = (x ^ (x >> 6)) & 0xFC0000ULL;

[mtklein, 2014/07/11 15:24:27]

Removing this was intentional?

[krajcevski, 2014/07/11 16:11:47]

Yes, it was the same as line 595 above, which I moved out of the #ifdef.

-    x = x ^ t ^ (t << 6);
-
     // x: 00 00 00 00 00 00 00 00 c g i m d h l p b f j n a e k o
 
     t = (x ^ (x >> 36)) & 0xFC0ULL;
@@ -770,19 +766,31 @@ static inline size_t get_compressed_data_size(Format fmt, int width, int height)
     }
 }
 
-typedef bool (*CompressBitmapProc)(uint8_t* dst, const uint8_t* src,
-                                   int width, int height, int rowBytes);
-
-bool CompressBufferToFormat(uint8_t* dst, const uint8_t* src, SkColorType srcColorType,
-                            int width, int height, int rowBytes, Format format) {
+static class CompressionProcTable 
+{
+public:
+    CompressionProcTable() {

[mtklein, 2014/07/11 15:24:27]

We try to avoid static initializers when possible.  Chrome hates them.

Can you just write this out as a switch or if in CompressBufferToFormat rather
than creating the table ahead of time like this?  Something like,

CompressionProc proc = NULL;
if (kAlpha_8_SkColorType == srcColorType) {
  proc = SkTextureCompressorGetPlatformProc(format)
  if (NULL == proc) {
    switch (format) {
      case kLATC_Format:    proc = compress_a8_to_latc;   break;
      case kR11_EAC_Format: proc = compress_a8_to_r11eac; break;
    }
  }
}
return NULL == proc ? false : proc(dst, src, width, height, rowBytes);

[krajcevski, 2014/07/11 16:11:47]

Done.

+        memset(kProcMap, 0, sizeof(kProcMap));
+
+        kProcMap[kLATC_Format][kAlpha_8_SkColorType] = compress_a8_to_latc;
+        kProcMap[kR11_EAC_Format][kAlpha_8_SkColorType] = compress_a8_to_r11eac;
+        CompressionProc r11opt = SkTextureCompressorGetPlatformProc(kR11_EAC_Format);
+        if (NULL != r11opt) {
+           kProcMap[kR11_EAC_Format][kAlpha_8_SkColorType] = r11opt;
+        }
+    }
 
-    CompressBitmapProc kProcMap[kFormatCnt][kLastEnum_SkColorType + 1];
-    memset(kProcMap, 0, sizeof(kProcMap));
+    CompressionProc getProc(SkColorType colorType, Format fmt) const {
+        return kProcMap[fmt][colorType];
+    }
 
-    kProcMap[kLATC_Format][kAlpha_8_SkColorType] = compress_a8_to_latc;
-    kProcMap[kR11_EAC_Format][kAlpha_8_SkColorType] = compress_a8_to_r11eac;
+private:
+    CompressionProc kProcMap[kFormatCnt][kLastEnum_SkColorType + 1];
+} kCompressionProcTable;
 
-    CompressBitmapProc proc = kProcMap[format][srcColorType];
+bool CompressBufferToFormat(uint8_t* dst, const uint8_t* src, SkColorType srcColorType,
+                            int width, int height, int rowBytes, Format format) {
+    CompressionProc proc = kCompressionProcTable.getProc(srcColorType, format);
     if (NULL != proc) {
         return proc(dst, src, width, height, rowBytes);
     }