OLD | NEW |
1 // Copyright (c) 2011 The Chromium Authors. All rights reserved. | 1 // Copyright (c) 2011 The Chromium Authors. All rights reserved. |
2 // Use of this source code is governed by a BSD-style license that can be | 2 // Use of this source code is governed by a BSD-style license that can be |
3 // found in the LICENSE file. | 3 // found in the LICENSE file. |
4 | 4 |
5 #ifndef SKIA_EXT_CONVOLVER_H_ | 5 #ifndef SKIA_EXT_CONVOLVER_H_ |
6 #define SKIA_EXT_CONVOLVER_H_ | 6 #define SKIA_EXT_CONVOLVER_H_ |
7 #pragma once | 7 #pragma once |
8 | 8 |
9 #include <cmath> | 9 #include <cmath> |
10 #include <vector> | 10 #include <vector> |
11 | 11 |
12 #include "base/basictypes.h" | 12 #include "base/basictypes.h" |
13 #include "base/cpu.h" | 13 #include "base/cpu.h" |
| 14 #include "third_party/skia/include/core/SkTypes.h" |
14 | 15 |
15 #if defined(ARCH_CPU_X86_FAMILY) | 16 #if defined(ARCH_CPU_X86_FAMILY) |
16 #if defined(__x86_64__) || defined(_M_X64) || defined(__SSE2__) || _M_IX86_FP==2 | 17 #if defined(__x86_64__) || defined(_M_X64) || defined(__SSE2__) || _M_IX86_FP==2 |
17 // This is where we had compiler support for SSE2 instructions. | 18 // This is where we had compiler support for SSE2 instructions. |
18 #define SIMD_SSE2 1 | 19 #define SIMD_SSE2 1 |
19 #endif | 20 #endif |
20 #endif | 21 #endif |
21 | 22 |
22 // avoid confusion with Mac OS X's math library (Carbon) | 23 // avoid confusion with Mac OS X's math library (Carbon) |
23 #if defined(__APPLE__) | 24 #if defined(__APPLE__) |
(...skipping 11 matching lines...) Expand all Loading... |
35 // ConvolutionFilter1D, then convolving each column by another one. | 36 // ConvolutionFilter1D, then convolving each column by another one. |
36 // | 37 // |
37 // Entries are stored in fixed point, shifted left by kShiftBits. | 38 // Entries are stored in fixed point, shifted left by kShiftBits. |
38 class ConvolutionFilter1D { | 39 class ConvolutionFilter1D { |
39 public: | 40 public: |
40 typedef short Fixed; | 41 typedef short Fixed; |
41 | 42 |
42 // The number of bits that fixed point values are shifted by. | 43 // The number of bits that fixed point values are shifted by. |
43 enum { kShiftBits = 14 }; | 44 enum { kShiftBits = 14 }; |
44 | 45 |
45 ConvolutionFilter1D(); | 46 SK_API ConvolutionFilter1D(); |
46 ~ConvolutionFilter1D(); | 47 SK_API ~ConvolutionFilter1D(); |
47 | 48 |
48 // Convert between floating point and our fixed point representation. | 49 // Convert between floating point and our fixed point representation. |
49 static Fixed FloatToFixed(float f) { | 50 static Fixed FloatToFixed(float f) { |
50 return static_cast<Fixed>(f * (1 << kShiftBits)); | 51 return static_cast<Fixed>(f * (1 << kShiftBits)); |
51 } | 52 } |
52 static unsigned char FixedToChar(Fixed x) { | 53 static unsigned char FixedToChar(Fixed x) { |
53 return static_cast<unsigned char>(x >> kShiftBits); | 54 return static_cast<unsigned char>(x >> kShiftBits); |
54 } | 55 } |
55 static float FixedToFloat(Fixed x) { | 56 static float FixedToFloat(Fixed x) { |
56 // The cast relies on Fixed being a short, implying that on | 57 // The cast relies on Fixed being a short, implying that on |
(...skipping 16 matching lines...) Expand all Loading... |
73 // the scaling factors start. The scaling factors apply to the source pixels | 74 // the scaling factors start. The scaling factors apply to the source pixels |
74 // starting from this position, and going for the next |filter_length| pixels. | 75 // starting from this position, and going for the next |filter_length| pixels. |
75 // | 76 // |
76 // You will probably want to make sure your input is normalized (that is, | 77 // You will probably want to make sure your input is normalized (that is, |
77 // all entries in |filter_values| sub to one) to prevent affecting the overall | 78 // all entries in |filter_values| sub to one) to prevent affecting the overall |
78 // brighness of the image. | 79 // brighness of the image. |
79 // | 80 // |
80 // The filter_length must be > 0. | 81 // The filter_length must be > 0. |
81 // | 82 // |
82 // This version will automatically convert your input to fixed point. | 83 // This version will automatically convert your input to fixed point. |
83 void AddFilter(int filter_offset, | 84 SK_API void AddFilter(int filter_offset, |
84 const float* filter_values, | 85 const float* filter_values, |
85 int filter_length); | 86 int filter_length); |
86 | 87 |
87 // Same as the above version, but the input is already fixed point. | 88 // Same as the above version, but the input is already fixed point. |
88 void AddFilter(int filter_offset, | 89 void AddFilter(int filter_offset, |
89 const Fixed* filter_values, | 90 const Fixed* filter_values, |
90 int filter_length); | 91 int filter_length); |
91 | 92 |
92 // Retrieves a filter for the given |value_offset|, a position in the output | 93 // Retrieves a filter for the given |value_offset|, a position in the output |
93 // image in the direction we're convolving. The offset and length of the | 94 // image in the direction we're convolving. The offset and length of the |
94 // filter values are put into the corresponding out arguments (see AddFilter | 95 // filter values are put into the corresponding out arguments (see AddFilter |
95 // above for what these mean), and a pointer to the first scaling factor is | 96 // above for what these mean), and a pointer to the first scaling factor is |
(...skipping 15 matching lines...) Expand all Loading... |
111 // Padding |padding_count| of more dummy coefficients after the coefficients | 112 // Padding |padding_count| of more dummy coefficients after the coefficients |
112 // of last filter to prevent SIMD instructions which load 8 or 16 bytes | 113 // of last filter to prevent SIMD instructions which load 8 or 16 bytes |
113 // together to access invalid memory areas. We are not trying to align the | 114 // together to access invalid memory areas. We are not trying to align the |
114 // coefficients right now due to the opaqueness of <vector> implementation. | 115 // coefficients right now due to the opaqueness of <vector> implementation. |
115 // This has to be done after all |AddFilter| calls. | 116 // This has to be done after all |AddFilter| calls. |
116 for (int i = 0; i < padding_count; ++i) | 117 for (int i = 0; i < padding_count; ++i) |
117 filter_values_.push_back(static_cast<Fixed>(0)); | 118 filter_values_.push_back(static_cast<Fixed>(0)); |
118 } | 119 } |
119 | 120 |
120 private: | 121 private: |
121 struct FilterInstance { | 122 struct SK_API FilterInstance { |
122 // Offset within filter_values for this instance of the filter. | 123 // Offset within filter_values for this instance of the filter. |
123 int data_location; | 124 int data_location; |
124 | 125 |
125 // Distance from the left of the filter to the center. IN PIXELS | 126 // Distance from the left of the filter to the center. IN PIXELS |
126 int offset; | 127 int offset; |
127 | 128 |
128 // Number of values in this filter instance. | 129 // Number of values in this filter instance. |
129 int length; | 130 int length; |
130 }; | 131 }; |
131 | 132 |
(...skipping 20 matching lines...) Expand all Loading... |
152 // size will be xfilter.num_values() * yfilter.num_values() pixels. It will be | 153 // size will be xfilter.num_values() * yfilter.num_values() pixels. It will be |
153 // in rows of exactly xfilter.num_values() * 4 bytes. | 154 // in rows of exactly xfilter.num_values() * 4 bytes. |
154 // | 155 // |
155 // |source_has_alpha| is a hint that allows us to avoid doing computations on | 156 // |source_has_alpha| is a hint that allows us to avoid doing computations on |
156 // the alpha channel if the image is opaque. If you don't know, set this to | 157 // the alpha channel if the image is opaque. If you don't know, set this to |
157 // true and it will work properly, but setting this to false will be a few | 158 // true and it will work properly, but setting this to false will be a few |
158 // percent faster if you know the image is opaque. | 159 // percent faster if you know the image is opaque. |
159 // | 160 // |
160 // The layout in memory is assumed to be 4-bytes per pixel in B-G-R-A order | 161 // The layout in memory is assumed to be 4-bytes per pixel in B-G-R-A order |
161 // (this is ARGB when loaded into 32-bit words on a little-endian machine). | 162 // (this is ARGB when loaded into 32-bit words on a little-endian machine). |
162 void BGRAConvolve2D(const unsigned char* source_data, | 163 SK_API void BGRAConvolve2D(const unsigned char* source_data, |
163 int source_byte_row_stride, | 164 int source_byte_row_stride, |
164 bool source_has_alpha, | 165 bool source_has_alpha, |
165 const ConvolutionFilter1D& xfilter, | 166 const ConvolutionFilter1D& xfilter, |
166 const ConvolutionFilter1D& yfilter, | 167 const ConvolutionFilter1D& yfilter, |
167 int output_byte_row_stride, | 168 int output_byte_row_stride, |
168 unsigned char* output, | 169 unsigned char* output, |
169 bool use_sse2); | 170 bool use_sse2); |
170 } // namespace skia | 171 } // namespace skia |
171 | 172 |
172 #endif // SKIA_EXT_CONVOLVER_H_ | 173 #endif // SKIA_EXT_CONVOLVER_H_ |
OLD | NEW |