|
1 // Copyright (c) 2006-2012 The Chromium Authors. All rights reserved. |
|
2 // |
|
3 // Redistribution and use in source and binary forms, with or without |
|
4 // modification, are permitted provided that the following conditions |
|
5 // are met: |
|
6 // * Redistributions of source code must retain the above copyright |
|
7 // notice, this list of conditions and the following disclaimer. |
|
8 // * Redistributions in binary form must reproduce the above copyright |
|
9 // notice, this list of conditions and the following disclaimer in |
|
10 // the documentation and/or other materials provided with the |
|
11 // distribution. |
|
12 // * Neither the name of Google, Inc. nor the names of its contributors |
|
13 // may be used to endorse or promote products derived from this |
|
14 // software without specific prior written permission. |
|
15 // |
|
16 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
|
17 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
|
18 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
|
19 // FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
|
20 // COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
|
21 // INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
|
22 // BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS |
|
23 // OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED |
|
24 // AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, |
|
25 // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT |
|
26 // OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
|
27 // SUCH DAMAGE. |
|
28 |
|
29 #ifndef SKIA_EXT_CONVOLVER_H_ |
|
30 #define SKIA_EXT_CONVOLVER_H_ |
|
31 |
|
32 #include <cmath> |
|
33 #include <vector> |
|
34 |
|
35 #include "base/basictypes.h" |
|
36 #include "base/cpu.h" |
|
37 #include "mozilla/Assertions.h" |
|
38 #include "skia/SkTypes.h" |
|
39 |
|
40 // avoid confusion with Mac OS X's math library (Carbon) |
|
41 #if defined(__APPLE__) |
|
42 #undef FloatToFixed |
|
43 #undef FixedToFloat |
|
44 #endif |
|
45 |
|
46 namespace skia { |
|
47 |
|
48 // Represents a filter in one dimension. Each output pixel has one entry in this |
|
49 // object for the filter values contributing to it. You build up the filter |
|
50 // list by calling AddFilter for each output pixel (in order). |
|
51 // |
|
52 // We do 2-dimensional convolution by first convolving each row by one |
|
53 // ConvolutionFilter1D, then convolving each column by another one. |
|
54 // |
|
55 // Entries are stored in fixed point, shifted left by kShiftBits. |
|
56 class ConvolutionFilter1D { |
|
57 public: |
|
58 typedef short Fixed; |
|
59 |
|
60 // The number of bits that fixed point values are shifted by. |
|
61 enum { kShiftBits = 14 }; |
|
62 |
|
63 ConvolutionFilter1D(); |
|
64 ~ConvolutionFilter1D(); |
|
65 |
|
66 // Convert between floating point and our fixed point representation. |
|
67 static Fixed FloatToFixed(float f) { |
|
68 return static_cast<Fixed>(f * (1 << kShiftBits)); |
|
69 } |
|
70 static unsigned char FixedToChar(Fixed x) { |
|
71 return static_cast<unsigned char>(x >> kShiftBits); |
|
72 } |
|
73 static float FixedToFloat(Fixed x) { |
|
74 // The cast relies on Fixed being a short, implying that on |
|
75 // the platforms we care about all (16) bits will fit into |
|
76 // the mantissa of a (32-bit) float. |
|
77 static_assert(sizeof(Fixed) == 2, |
|
78 "fixed type should fit in float mantissa"); |
|
79 float raw = static_cast<float>(x); |
|
80 return ldexpf(raw, -kShiftBits); |
|
81 } |
|
82 |
|
83 // Returns the maximum pixel span of a filter. |
|
84 int max_filter() const { return max_filter_; } |
|
85 |
|
86 // Returns the number of filters in this filter. This is the dimension of the |
|
87 // output image. |
|
88 int num_values() const { return static_cast<int>(filters_.size()); } |
|
89 |
|
90 // Appends the given list of scaling values for generating a given output |
|
91 // pixel. |filter_offset| is the distance from the edge of the image to where |
|
92 // the scaling factors start. The scaling factors apply to the source pixels |
|
93 // starting from this position, and going for the next |filter_length| pixels. |
|
94 // |
|
95 // You will probably want to make sure your input is normalized (that is, |
|
96 // all entries in |filter_values| sub to one) to prevent affecting the overall |
|
97 // brighness of the image. |
|
98 // |
|
99 // The filter_length must be > 0. |
|
100 // |
|
101 // This version will automatically convert your input to fixed point. |
|
102 void AddFilter(int filter_offset, |
|
103 const float* filter_values, |
|
104 int filter_length); |
|
105 |
|
106 // Same as the above version, but the input is already fixed point. |
|
107 void AddFilter(int filter_offset, |
|
108 const Fixed* filter_values, |
|
109 int filter_length); |
|
110 |
|
111 // Retrieves a filter for the given |value_offset|, a position in the output |
|
112 // image in the direction we're convolving. The offset and length of the |
|
113 // filter values are put into the corresponding out arguments (see AddFilter |
|
114 // above for what these mean), and a pointer to the first scaling factor is |
|
115 // returned. There will be |filter_length| values in this array. |
|
116 inline const Fixed* FilterForValue(int value_offset, |
|
117 int* filter_offset, |
|
118 int* filter_length) const { |
|
119 const FilterInstance& filter = filters_[value_offset]; |
|
120 *filter_offset = filter.offset; |
|
121 *filter_length = filter.length; |
|
122 if (filter.length == 0) { |
|
123 return NULL; |
|
124 } |
|
125 return &filter_values_[filter.data_location]; |
|
126 } |
|
127 |
|
128 |
|
129 inline void PaddingForSIMD(int padding_count) { |
|
130 // Padding |padding_count| of more dummy coefficients after the coefficients |
|
131 // of last filter to prevent SIMD instructions which load 8 or 16 bytes |
|
132 // together to access invalid memory areas. We are not trying to align the |
|
133 // coefficients right now due to the opaqueness of <vector> implementation. |
|
134 // This has to be done after all |AddFilter| calls. |
|
135 for (int i = 0; i < padding_count; ++i) |
|
136 filter_values_.push_back(static_cast<Fixed>(0)); |
|
137 } |
|
138 |
|
139 private: |
|
140 struct FilterInstance { |
|
141 // Offset within filter_values for this instance of the filter. |
|
142 int data_location; |
|
143 |
|
144 // Distance from the left of the filter to the center. IN PIXELS |
|
145 int offset; |
|
146 |
|
147 // Number of values in this filter instance. |
|
148 int length; |
|
149 }; |
|
150 |
|
151 // Stores the information for each filter added to this class. |
|
152 std::vector<FilterInstance> filters_; |
|
153 |
|
154 // We store all the filter values in this flat list, indexed by |
|
155 // |FilterInstance.data_location| to avoid the mallocs required for storing |
|
156 // each one separately. |
|
157 std::vector<Fixed> filter_values_; |
|
158 |
|
159 // The maximum size of any filter we've added. |
|
160 int max_filter_; |
|
161 }; |
|
162 |
|
163 // Does a two-dimensional convolution on the given source image. |
|
164 // |
|
165 // It is assumed the source pixel offsets referenced in the input filters |
|
166 // reference only valid pixels, so the source image size is not required. Each |
|
167 // row of the source image starts |source_byte_row_stride| after the previous |
|
168 // one (this allows you to have rows with some padding at the end). |
|
169 // |
|
170 // The result will be put into the given output buffer. The destination image |
|
171 // size will be xfilter.num_values() * yfilter.num_values() pixels. It will be |
|
172 // in rows of exactly xfilter.num_values() * 4 bytes. |
|
173 // |
|
174 // |source_has_alpha| is a hint that allows us to avoid doing computations on |
|
175 // the alpha channel if the image is opaque. If you don't know, set this to |
|
176 // true and it will work properly, but setting this to false will be a few |
|
177 // percent faster if you know the image is opaque. |
|
178 // |
|
179 // The layout in memory is assumed to be 4-bytes per pixel in B-G-R-A order |
|
180 // (this is ARGB when loaded into 32-bit words on a little-endian machine). |
|
181 void BGRAConvolve2D(const unsigned char* source_data, |
|
182 int source_byte_row_stride, |
|
183 bool source_has_alpha, |
|
184 const ConvolutionFilter1D& xfilter, |
|
185 const ConvolutionFilter1D& yfilter, |
|
186 int output_byte_row_stride, |
|
187 unsigned char* output, |
|
188 bool use_sse2); |
|
189 } // namespace skia |
|
190 |
|
191 #endif // SKIA_EXT_CONVOLVER_H_ |