SIMD implementation of Convolver for Lanczos filter etc.

skia/ext/convolver.h

 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef SKIA_EXT_CONVOLVER_H_
 #define SKIA_EXT_CONVOLVER_H_
 #pragma once
 
 #include <cmath>
 #include <vector>
 
 #include "base/basictypes.h"
+#include "base/cpu.h"
 
 // avoid confusion with Mac OS X's math library (Carbon)
 #if defined(__APPLE__)
 #undef FloatToFixed
 #undef FixedToFloat
 #endif
 
 namespace skia {
 
 // Represents a filter in one dimension. Each output pixel has one entry in this
@@ skipping 68 matching lines @@
                                      int* filter_length) const {
     const FilterInstance& filter = filters_[value_offset];
     *filter_offset = filter.offset;
     *filter_length = filter.length;
     if (filter.length == 0) {
       return NULL;
     }
     return &filter_values_[filter.data_location];
   }
 
+  inline void PaddingForSIMD(int align) {

[brettw, 2011/02/21 04:45:45]

I would have thought that you'd want to pad up to the next multiple of the
alignment. If you have 9 things in the filter, do you really just want to leave
it at that, or do you want to pad to 16?

[jiesun, 2011/02/22 21:37:03]

padding in SIMD serve two purpose:
1. make element align to element size boundary, such as 16 or 8.
2. prevent SIMD load and store instruction to access violation.
here I only try to solve problem (2). therefore only padding once after all
filters are added.
solving (1) need to padding for each filters ( not a big deal). but because of
current implementation use <vector> and it is tricky to align at the beginning.

+    // Padding more coefficients to allow SIMD to load after the last one.
+    // This has to be done after all |AddFilter| calls.
+    for (int i = 0; i < align; ++i)
+      filter_values_.push_back(static_cast<Fixed>(0));
+  }
+
  private:
   struct FilterInstance {
     // Offset within filter_values for this instance of the filter.
     int data_location;
 
     // Distance from the left of the filter to the center. IN PIXELS
     int offset;
 
     // Number of values in this filter instance.
     int length;
@@ skipping 30 matching lines @@
 // The layout in memory is assumed to be 4-bytes per pixel in B-G-R-A order
 // (this is ARGB when loaded into 32-bit words on a little-endian machine).
 void BGRAConvolve2D(const unsigned char* source_data,
                     int source_byte_row_stride,
                     bool source_has_alpha,
                     const ConvolutionFilter1D& xfilter,
                     const ConvolutionFilter1D& yfilter,
                     int output_byte_row_stride,
                     unsigned char* output);
 
+void BGRAConvolve2D_C(const unsigned char* source_data,
+                    int source_byte_row_stride,

[brettw, 2011/02/21 04:45:45]

Style: be sure to align the args in these two functions to after the (

[jiesun, 2011/02/22 21:37:03]

Done.

+                    bool source_has_alpha,
+                    const ConvolutionFilter1D& filter_x,
+                    const ConvolutionFilter1D& filter_y,
+                    int output_byte_row_stride,
+                    unsigned char* output);
+
+void BGRAConvolve2D_SSE2(const unsigned char* source_data,
+                    int source_byte_row_stride,
+                    bool source_has_alpha,
+                    const ConvolutionFilter1D& filter_x,
+                    const ConvolutionFilter1D& filter_y,
+                    int output_byte_row_stride,
+                    unsigned char* output);
 }  // namespace skia
 
 #endif  // SKIA_EXT_CONVOLVER_H_