SIMD implementation of Convolver for Lanczos filter etc.

skia/ext/image_operations.cc

Index: skia/ext/image_operations.cc
diff --git a/skia/ext/image_operations.cc b/skia/ext/image_operations.cc
index 51d2e4e252c26a690affc0a57232983df2228d72..e58f79d9d00604bbfead19d510a9cabdbea71467 100644
--- a/skia/ext/image_operations.cc
+++ b/skia/ext/image_operations.cc
@@ -316,6 +316,8 @@ void ResizeFilter::ComputeFilters(int src_size,
     output->AddFilter(src_begin, &fixed_filter_values[0],
                       static_cast<int>(fixed_filter_values->size()));
   }
+
+  output->PaddingForSIMD(8);

[evannier, 2011/02/14 23:45:13]

At the API layer, this seems wrong to be visible from the outside.
For example one could conceive that an optimization (that I discussed in my
comments) that the layer below would be responsible for detecting cycles in the
values, at which point, when you push the values in 'AddFilter' you have no
guarantee that any value was actually added.

[fbarchard, 2011/02/15 23:16:45]

should be 16?

[jiesun, 2011/02/17 20:17:58]

I agree this is ugly, but there is no "finalizefilters". padding only happened
at the end of all filters.  filter_values is also private. ...

[jiesun, 2011/02/17 20:17:58]

16 byte = 8 coefficient because Fixed is short(2bytes)

 }
 
 ImageOperations::ResizeMethod ResizeMethodToAlgorithmMethod(
@@ -461,6 +463,32 @@ SkBitmap ImageOperations::ResizeSubpixel(const SkBitmap& source,
 #endif  // OS_POSIX && !OS_MACOSX
 }
 
+#ifdef ARCH_CPU_X86_FAMILY
+#ifdef _MSC_VER
+#include <intrin.h>
+#endif
+#endif
+
+bool ImageOperations::hasSSE2() {
+#ifdef ARCH_CPU_X86_FAMILY
+#ifdef _MSC_VER
+  int cpu_info[4] = {-1};
+  __cpuid(cpu_info, 0);
+  int num_ids = cpu_info[0];

[fbarchard, 2011/02/15 23:16:45]

is this necessary?  i thought everyone that supports cpuid at all, support
function 1?

[jiesun, 2011/02/17 20:17:58]

Done.

+  if (num_ids > 0) {
+    __cpuid(cpu_info, 1);
+    return (cpu_info[3] & (1<<26)) != 0;
+  }
+  return false;
+#else
+  // TODO(jiesun): This has to be resolved. We could not use skia implementation.
+  return true;
+#endif
+#else
+  return false;
+#endif
+}
+
 // static
 SkBitmap ImageOperations::ResizeBasic(const SkBitmap& source,
                                       ResizeMethod method,
@@ -509,7 +537,7 @@ SkBitmap ImageOperations::ResizeBasic(const SkBitmap& source,
   BGRAConvolve2D(source_subset, static_cast<int>(source.rowBytes()),
                  !source.isOpaque(), filter.x_filter(), filter.y_filter(),
                  static_cast<int>(result.rowBytes()),
-                 static_cast<unsigned char*>(result.getPixels()));
+                 static_cast<unsigned char*>(result.getPixels()), hasSSE2());
 
   // Preserve the "opaque" flag for use as an optimization later.
   result.setIsOpaque(source.isOpaque());