oil 0.1.3 → 0.2.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +5 -5
- data/MIT-LICENSE +1 -1
- data/README.rdoc +3 -0
- data/Rakefile +1 -3
- data/ext/oil/extconf.rb +1 -1
- data/ext/oil/jpeg.c +68 -42
- data/ext/oil/oil.c +19 -2
- data/ext/oil/png.c +52 -31
- data/ext/oil/resample.c +719 -245
- data/ext/oil/resample.h +101 -28
- data/lib/oil.rb +2 -39
- data/test/test_png.rb +1 -2
- metadata +3 -5
- data/ext/oil/yscaler.c +0 -167
- data/ext/oil/yscaler.h +0 -33
data/ext/oil/resample.c
CHANGED
@@ -1,3 +1,24 @@
|
|
1
|
+
/**
|
2
|
+
* Copyright (c) 2014-2016 Timothy Elliott
|
3
|
+
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
4
|
+
* of this software and associated documentation files (the "Software"), to deal
|
5
|
+
* in the Software without restriction, including without limitation the rights
|
6
|
+
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
7
|
+
* copies of the Software, and to permit persons to whom the Software is
|
8
|
+
* furnished to do so, subject to the following conditions:
|
9
|
+
*
|
10
|
+
* The above copyright notice and this permission notice shall be included in
|
11
|
+
* all copies or substantial portions of the Software.
|
12
|
+
*
|
13
|
+
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
14
|
+
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
15
|
+
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
16
|
+
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
17
|
+
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
18
|
+
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
19
|
+
* THE SOFTWARE.
|
20
|
+
*/
|
21
|
+
|
1
22
|
#include "resample.h"
|
2
23
|
#include <stdint.h>
|
3
24
|
#include <math.h>
|
@@ -26,7 +47,7 @@ typedef int64_t fix33_30;
|
|
26
47
|
* The best possible value was determined by comparing to a reference
|
27
48
|
* implementation and comparing values for the minimal number of errors.
|
28
49
|
*/
|
29
|
-
#define TOPOFF
|
50
|
+
#define TOPOFF 8192
|
30
51
|
|
31
52
|
/**
|
32
53
|
* Signed type that uses 1 bit for signedness, 1 bit for the integer, and 30
|
@@ -42,9 +63,9 @@ typedef int32_t fix1_30;
|
|
42
63
|
/**
|
43
64
|
* Calculate the greatest common denominator between a and b.
|
44
65
|
*/
|
45
|
-
static
|
66
|
+
static uint32_t gcd(uint32_t a, uint32_t b)
|
46
67
|
{
|
47
|
-
|
68
|
+
uint32_t c;
|
48
69
|
while (a != 0) {
|
49
70
|
c = a;
|
50
71
|
a = b%a;
|
@@ -56,46 +77,32 @@ static long gcd (long a, long b)
|
|
56
77
|
/**
|
57
78
|
* Round and clamp a fix33_30 value between 0 and 255. Returns an unsigned char.
|
58
79
|
*/
|
59
|
-
static
|
80
|
+
static uint16_t clamp(fix33_30 x)
|
60
81
|
{
|
61
|
-
|
62
|
-
|
63
|
-
}
|
64
|
-
|
65
|
-
/* bump up rounding errors before truncating */
|
66
|
-
x += TOPOFF;
|
67
|
-
|
68
|
-
/* This is safe because we have the < 0 check above and a sample can't
|
69
|
-
* end up with a value over 512 */
|
70
|
-
if (x & (1l<<38)) {
|
71
|
-
return 255;
|
72
|
-
}
|
73
|
-
|
82
|
+
x += 1 << 29;
|
83
|
+
x = x < 0 ? 0 : (x > (65535L << 30) ? (65535L << 30) : x);
|
74
84
|
return x >> 30;
|
75
85
|
}
|
76
86
|
|
77
|
-
|
78
|
-
* Map from a discreet dest coordinate to a continuous source coordinate.
|
79
|
-
* The resulting coordinate can range from -0.5 to the maximum of the
|
80
|
-
* destination image dimension.
|
81
|
-
*/
|
82
|
-
static double map(long pos, double scale)
|
87
|
+
static uint8_t clamp8(fix33_30 x)
|
83
88
|
{
|
84
|
-
return (
|
89
|
+
return (clamp(x) + (1 << 7)) / 257;
|
85
90
|
}
|
86
91
|
|
87
92
|
/**
|
88
93
|
* Given input and output dimensions and an output position, return the
|
89
94
|
* corresponding input position and put the sub-pixel remainder in rest.
|
95
|
+
*
|
96
|
+
* Map from a discreet dest coordinate to a continuous source coordinate.
|
97
|
+
* The resulting coordinate can range from -0.5 to the maximum of the
|
98
|
+
* destination image dimension.
|
90
99
|
*/
|
91
|
-
|
92
|
-
float *rest)
|
100
|
+
int32_t split_map(uint32_t dim_in, uint32_t dim_out, uint32_t pos, float *rest)
|
93
101
|
{
|
94
|
-
double
|
95
|
-
|
102
|
+
double smp;
|
103
|
+
int32_t smp_i;
|
96
104
|
|
97
|
-
|
98
|
-
smp = map(pos, scale);
|
105
|
+
smp = (pos + 0.5) * ((double)dim_in / dim_out) - 0.5;
|
99
106
|
smp_i = smp < 0 ? -1 : smp;
|
100
107
|
*rest = smp - smp_i;
|
101
108
|
return smp_i;
|
@@ -108,42 +115,14 @@ long split_map(unsigned long dim_in, unsigned long dim_out, unsigned long pos,
|
|
108
115
|
* When we reduce an image by a factor of two, we need to scale our resampling
|
109
116
|
* function by two as well in order to avoid aliasing.
|
110
117
|
*/
|
111
|
-
|
118
|
+
uint64_t calc_taps(uint32_t dim_in, uint32_t dim_out)
|
112
119
|
{
|
113
|
-
|
114
|
-
|
120
|
+
uint64_t tmp;
|
115
121
|
if (dim_out > dim_in) {
|
116
122
|
return TAPS;
|
117
123
|
}
|
118
|
-
|
119
|
-
tmp
|
120
|
-
/* Round up to the nearest even integer */
|
121
|
-
return tmp + (tmp&1);
|
122
|
-
}
|
123
|
-
|
124
|
-
/**
|
125
|
-
* Helper macros to extract byte components from an int32_t holding rgba.
|
126
|
-
*/
|
127
|
-
#define rgba_r(x) ((x) & 0x000000FF)
|
128
|
-
#define rgba_g(x) (((x) & 0x0000FF00) >> 8)
|
129
|
-
#define rgba_b(x) (((x) & 0x00FF0000) >> 16)
|
130
|
-
#define rgba_a(x) (((x) & 0xFF000000) >> 24)
|
131
|
-
|
132
|
-
/**
|
133
|
-
* Convert rgb values in fix33_30 types to a uint32_t.
|
134
|
-
*/
|
135
|
-
static uint32_t fix33_30_to_rgbx(fix33_30 r, fix33_30 g, fix33_30 b)
|
136
|
-
{
|
137
|
-
return clamp(r) + ((uint32_t)clamp(g) << 8) +
|
138
|
-
((uint32_t)clamp(b) << 16);
|
139
|
-
}
|
140
|
-
|
141
|
-
/**
|
142
|
-
* Convert rgba values in fix33_30 types to a uint32_t.
|
143
|
-
*/
|
144
|
-
static uint32_t fix33_30_to_rgba(fix33_30 r, fix33_30 g, fix33_30 b, fix33_30 a)
|
145
|
-
{
|
146
|
-
return fix33_30_to_rgbx(r, g, b) + ((uint32_t)clamp(a) << 24);
|
124
|
+
tmp = (uint64_t)TAPS * dim_in / dim_out;
|
125
|
+
return tmp - (tmp & 1);
|
147
126
|
}
|
148
127
|
|
149
128
|
/**
|
@@ -152,9 +131,9 @@ static uint32_t fix33_30_to_rgba(fix33_30 r, fix33_30 g, fix33_30 b, fix33_30 a)
|
|
152
131
|
static float catrom(float x)
|
153
132
|
{
|
154
133
|
if (x<1) {
|
155
|
-
return (3*x
|
134
|
+
return ((3*x - 5)*x*x + 2) / 2;
|
156
135
|
}
|
157
|
-
return (-
|
136
|
+
return (((5 - x)*x - 8)*x + 4) / 2;
|
158
137
|
}
|
159
138
|
|
160
139
|
/**
|
@@ -171,127 +150,191 @@ static fix1_30 f_to_fix1_30(float x)
|
|
171
150
|
*
|
172
151
|
* The coefficients are stored as fix1_30 fixed point ints in coeffs.
|
173
152
|
*/
|
174
|
-
static void calc_coeffs(fix1_30 *coeffs, float tx,
|
153
|
+
static void calc_coeffs(fix1_30 *coeffs, float tx, uint32_t taps)
|
175
154
|
{
|
176
|
-
|
177
|
-
float tmp,
|
155
|
+
uint32_t i;
|
156
|
+
float tmp, tap_mult, fudge;
|
178
157
|
fix1_30 tmp_fixed;
|
179
158
|
|
180
|
-
|
181
|
-
|
182
|
-
tap_mult = (double)taps / TAPS;
|
159
|
+
tap_mult = (float)taps / TAPS;
|
183
160
|
tx = 1 - tx - taps / 2;
|
161
|
+
fudge = 1.0;
|
184
162
|
|
185
163
|
for (i=0; i<taps; i++) {
|
186
|
-
tmp = catrom(fabsf(tx) / tap_mult);
|
164
|
+
tmp = catrom(fabsf(tx) / tap_mult) / tap_mult;
|
165
|
+
fudge -= tmp;
|
187
166
|
tmp_fixed = f_to_fix1_30(tmp);
|
188
167
|
coeffs[i] = tmp_fixed;
|
189
|
-
total += tmp;
|
190
168
|
tx += 1;
|
191
169
|
}
|
170
|
+
coeffs[taps / 2] += f_to_fix1_30(fudge);
|
171
|
+
}
|
192
172
|
|
193
|
-
|
194
|
-
|
173
|
+
/* bicubic y-scaler */
|
174
|
+
|
175
|
+
static uint8_t linear_sample_to_srgb(uint16_t in)
|
176
|
+
{
|
177
|
+
double in_f, s1, s2, s3;
|
178
|
+
if (in <= 248) {
|
179
|
+
return (in * 3295 + 32768) >> 16;
|
195
180
|
}
|
181
|
+
in_f = in / 65535.0;
|
182
|
+
s1 = sqrt(in_f);
|
183
|
+
s2 = sqrt(s1);
|
184
|
+
s3 = sqrt(s2);
|
185
|
+
return (0.0427447 + 0.547242 * s1 + 0.928361 * s2 - 0.518123 * s3) * 255 + 0.5;
|
196
186
|
}
|
197
187
|
|
198
|
-
|
199
|
-
*
|
200
|
-
*/
|
201
|
-
static void yscale_gen(long len, long height, fix1_30 *coeffs,
|
202
|
-
unsigned char **in, unsigned char *out)
|
188
|
+
static void strip_scale_rgbx(uint16_t **in, uint32_t strip_height, size_t len,
|
189
|
+
uint8_t *out, fix1_30 *coeffs)
|
203
190
|
{
|
204
|
-
|
205
|
-
|
191
|
+
size_t i;
|
192
|
+
uint32_t j;
|
193
|
+
fix33_30 coeff, sum[3];
|
206
194
|
|
207
|
-
for (i=0; i<len; i
|
208
|
-
|
209
|
-
for (j=0; j<
|
195
|
+
for (i=0; i<len; i+=4) {
|
196
|
+
sum[0] = sum[1] = sum[2] = 0;
|
197
|
+
for (j=0; j<strip_height; j++) {
|
210
198
|
coeff = coeffs[j];
|
211
|
-
|
199
|
+
sum[0] += coeff * in[j][i];
|
200
|
+
sum[1] += coeff * in[j][i + 1];
|
201
|
+
sum[2] += coeff * in[j][i + 2];
|
212
202
|
}
|
213
|
-
out[
|
203
|
+
out[0] = linear_sample_to_srgb(clamp(sum[0]));
|
204
|
+
out[1] = linear_sample_to_srgb(clamp(sum[1]));
|
205
|
+
out[2] = linear_sample_to_srgb(clamp(sum[2]));
|
206
|
+
out[3] = 0;
|
207
|
+
out += 4;
|
214
208
|
}
|
215
209
|
}
|
216
210
|
|
217
|
-
|
218
|
-
*
|
219
|
-
* performance.
|
220
|
-
*/
|
221
|
-
static void yscale_rgba(long width, long height, fix1_30 *coeffs,
|
222
|
-
uint32_t **sl_in, uint32_t *sl_out)
|
211
|
+
static void strip_scale_rgb(uint16_t **in, uint32_t strip_height, size_t len,
|
212
|
+
uint8_t *out, fix1_30 *coeffs)
|
223
213
|
{
|
224
|
-
|
225
|
-
|
226
|
-
|
214
|
+
size_t i;
|
215
|
+
uint32_t j;
|
216
|
+
fix33_30 coeff, sum[3];
|
227
217
|
|
228
|
-
for (i=0; i<
|
229
|
-
|
230
|
-
for (j=0; j<
|
218
|
+
for (i=0; i<len; i+=4) {
|
219
|
+
sum[0] = sum[1] = sum[2] = 0;
|
220
|
+
for (j=0; j<strip_height; j++) {
|
231
221
|
coeff = coeffs[j];
|
232
|
-
|
233
|
-
|
234
|
-
|
235
|
-
b += coeff * rgba_b(sample);
|
236
|
-
a += coeff * rgba_a(sample);
|
222
|
+
sum[0] += coeff * in[j][i];
|
223
|
+
sum[1] += coeff * in[j][i + 1];
|
224
|
+
sum[2] += coeff * in[j][i + 2];
|
237
225
|
}
|
238
|
-
|
226
|
+
out[0] = linear_sample_to_srgb(clamp(sum[0]));
|
227
|
+
out[1] = linear_sample_to_srgb(clamp(sum[1]));
|
228
|
+
out[2] = linear_sample_to_srgb(clamp(sum[2]));
|
229
|
+
out += 3;
|
239
230
|
}
|
240
231
|
}
|
241
232
|
|
242
|
-
|
243
|
-
*
|
244
|
-
|
245
|
-
|
246
|
-
|
247
|
-
|
233
|
+
static void strip_scale_g(uint16_t **in, uint32_t strip_height, size_t len,
|
234
|
+
uint8_t *out, fix1_30 *coeffs)
|
235
|
+
{
|
236
|
+
size_t i;
|
237
|
+
uint32_t j;
|
238
|
+
fix33_30 sum;
|
239
|
+
|
240
|
+
for (i=0; i<len; i++) {
|
241
|
+
sum = 0;
|
242
|
+
for (j=0; j<strip_height; j++) {
|
243
|
+
sum += (fix33_30)coeffs[j] * in[j][i];
|
244
|
+
}
|
245
|
+
out[i] = clamp8(sum);
|
246
|
+
}
|
247
|
+
}
|
248
|
+
|
249
|
+
static void strip_scale_rgba(uint16_t **in, uint32_t strip_height, size_t len,
|
250
|
+
uint8_t *out, fix1_30 *coeffs)
|
248
251
|
{
|
249
|
-
|
250
|
-
|
251
|
-
|
252
|
+
size_t i;
|
253
|
+
uint32_t j;
|
254
|
+
fix33_30 coeff, sum[4];
|
252
255
|
|
253
|
-
for (i=0; i<
|
254
|
-
|
255
|
-
for (j=0; j<
|
256
|
+
for (i=0; i<len; i+=4) {
|
257
|
+
sum[0] = sum[1] = sum[2] = sum[3] = 0;
|
258
|
+
for (j=0; j<strip_height; j++) {
|
256
259
|
coeff = coeffs[j];
|
257
|
-
|
258
|
-
|
259
|
-
|
260
|
-
|
260
|
+
sum[0] += coeff * in[j][i];
|
261
|
+
sum[1] += coeff * in[j][i + 1];
|
262
|
+
sum[2] += coeff * in[j][i + 2];
|
263
|
+
sum[3] += coeff * in[j][i + 3];
|
261
264
|
}
|
262
|
-
|
265
|
+
out[0] = linear_sample_to_srgb(clamp(sum[0]));
|
266
|
+
out[1] = linear_sample_to_srgb(clamp(sum[1]));
|
267
|
+
out[2] = linear_sample_to_srgb(clamp(sum[2]));
|
268
|
+
out[3] = clamp8(sum[3]);
|
269
|
+
out += 4;
|
263
270
|
}
|
264
271
|
}
|
265
272
|
|
266
|
-
void
|
267
|
-
|
273
|
+
static void strip_scale_cmyk(uint16_t **in, uint32_t strip_height, size_t len,
|
274
|
+
uint8_t *out, fix1_30 *coeffs)
|
275
|
+
{
|
276
|
+
size_t i;
|
277
|
+
uint32_t j;
|
278
|
+
fix33_30 coeff, sum[4];
|
279
|
+
|
280
|
+
for (i=0; i<len; i+=4) {
|
281
|
+
sum[0] = sum[1] = sum[2] = sum[3] = 0;
|
282
|
+
for (j=0; j<strip_height; j++) {
|
283
|
+
coeff = coeffs[j];
|
284
|
+
sum[0] += coeff * in[j][i];
|
285
|
+
sum[1] += coeff * in[j][i + 1];
|
286
|
+
sum[2] += coeff * in[j][i + 2];
|
287
|
+
sum[3] += coeff * in[j][i + 3];
|
288
|
+
}
|
289
|
+
out[0] = clamp8(sum[0]);
|
290
|
+
out[1] = clamp8(sum[1]);
|
291
|
+
out[2] = clamp8(sum[2]);
|
292
|
+
out[3] = clamp8(sum[3]);
|
293
|
+
out += 4;
|
294
|
+
}
|
295
|
+
}
|
296
|
+
|
297
|
+
int strip_scale(uint16_t **in, uint32_t strip_height, size_t len, uint8_t *out,
|
298
|
+
float ty, enum oil_colorspace cs)
|
268
299
|
{
|
269
300
|
fix1_30 *coeffs;
|
270
301
|
|
271
302
|
coeffs = malloc(strip_height * sizeof(fix1_30));
|
303
|
+
if (!coeffs) {
|
304
|
+
return -2; // unable to allocate
|
305
|
+
}
|
272
306
|
calc_coeffs(coeffs, ty, strip_height);
|
273
307
|
|
274
|
-
|
275
|
-
|
276
|
-
|
277
|
-
|
278
|
-
|
279
|
-
|
280
|
-
|
281
|
-
|
282
|
-
|
308
|
+
switch(cs) {
|
309
|
+
case OIL_CS_G:
|
310
|
+
case OIL_CS_GA:
|
311
|
+
strip_scale_g(in, strip_height, len, out, coeffs);
|
312
|
+
break;
|
313
|
+
case OIL_CS_RGB:
|
314
|
+
strip_scale_rgb(in, strip_height, len, out, coeffs);
|
315
|
+
break;
|
316
|
+
case OIL_CS_RGBX:
|
317
|
+
strip_scale_rgbx(in, strip_height, len, out, coeffs);
|
318
|
+
break;
|
319
|
+
case OIL_CS_RGBA:
|
320
|
+
strip_scale_rgba(in, strip_height, len, out, coeffs);
|
321
|
+
break;
|
322
|
+
case OIL_CS_CMYK:
|
323
|
+
strip_scale_cmyk(in, strip_height, len, out, coeffs);
|
324
|
+
break;
|
283
325
|
}
|
284
326
|
|
285
327
|
free(coeffs);
|
328
|
+
return 0;
|
286
329
|
}
|
287
330
|
|
288
331
|
/* Bicubic x scaler */
|
289
332
|
|
290
|
-
static void sample_generic(
|
291
|
-
|
333
|
+
static void sample_generic(uint32_t taps, fix1_30 *coeffs, uint16_t *in,
|
334
|
+
uint16_t *out, uint8_t cmp)
|
292
335
|
{
|
293
|
-
|
294
|
-
|
336
|
+
uint8_t i;
|
337
|
+
uint32_t j;
|
295
338
|
fix33_30 total, coeff;
|
296
339
|
|
297
340
|
for (i=0; i<cmp; i++) {
|
@@ -304,139 +347,113 @@ static void sample_generic(int cmp, long taps, fix1_30 *coeffs,
|
|
304
347
|
}
|
305
348
|
}
|
306
349
|
|
307
|
-
static
|
350
|
+
static void sample_rgba(uint32_t taps, fix1_30 *coeffs, uint16_t *in,
|
351
|
+
uint16_t *out)
|
308
352
|
{
|
309
|
-
|
310
|
-
fix33_30
|
311
|
-
uint32_t sample;
|
353
|
+
uint32_t i;
|
354
|
+
fix33_30 sum[4], coeff;
|
312
355
|
|
313
|
-
|
356
|
+
sum[0] = sum[1] = sum[2] = sum[3] = 0;
|
314
357
|
for (i=0; i<taps; i++) {
|
315
358
|
coeff = coeffs[i];
|
316
|
-
|
317
|
-
|
318
|
-
|
319
|
-
|
320
|
-
|
359
|
+
sum[0] += coeff * in[0];
|
360
|
+
sum[1] += coeff * in[1];
|
361
|
+
sum[2] += coeff * in[2];
|
362
|
+
sum[3] += coeff * in[3];
|
363
|
+
in += 4;
|
321
364
|
}
|
322
|
-
|
365
|
+
out[0] = clamp(sum[0]);
|
366
|
+
out[1] = clamp(sum[1]);
|
367
|
+
out[2] = clamp(sum[2]);
|
368
|
+
out[3] = clamp(sum[3]);
|
323
369
|
}
|
324
370
|
|
325
|
-
static
|
371
|
+
static void sample_rgbx(uint32_t taps, fix1_30 *coeffs, uint16_t *in,
|
372
|
+
uint16_t *out)
|
326
373
|
{
|
327
|
-
|
328
|
-
fix33_30
|
329
|
-
uint32_t sample;
|
374
|
+
uint32_t i;
|
375
|
+
fix33_30 sum[3], coeff;
|
330
376
|
|
331
|
-
|
377
|
+
sum[0] = sum[1] = sum[2] = 0;
|
332
378
|
for (i=0; i<taps; i++) {
|
333
379
|
coeff = coeffs[i];
|
334
|
-
|
335
|
-
|
336
|
-
|
337
|
-
|
380
|
+
sum[0] += coeff * in[0];
|
381
|
+
sum[1] += coeff * in[1];
|
382
|
+
sum[2] += coeff * in[2];
|
383
|
+
in += 4;
|
338
384
|
}
|
339
|
-
|
385
|
+
out[0] = clamp(sum[0]);
|
386
|
+
out[1] = clamp(sum[1]);
|
387
|
+
out[2] = clamp(sum[2]);
|
388
|
+
out[3] = 0;
|
340
389
|
}
|
341
390
|
|
342
|
-
static void xscale_set_sample(
|
343
|
-
|
391
|
+
static void xscale_set_sample(uint32_t taps, fix1_30 *coeffs, uint16_t *in,
|
392
|
+
uint16_t *out, enum oil_colorspace cs)
|
344
393
|
{
|
345
|
-
|
346
|
-
|
347
|
-
|
348
|
-
|
349
|
-
|
350
|
-
|
394
|
+
switch(cs) {
|
395
|
+
case OIL_CS_G:
|
396
|
+
case OIL_CS_GA:
|
397
|
+
case OIL_CS_RGB:
|
398
|
+
sample_generic(taps, coeffs, in, out, CS_TO_CMP(cs));
|
399
|
+
break;
|
400
|
+
case OIL_CS_RGBX:
|
401
|
+
sample_rgbx(taps, coeffs, in, out);
|
402
|
+
break;
|
403
|
+
case OIL_CS_RGBA:
|
404
|
+
case OIL_CS_CMYK:
|
405
|
+
sample_rgba(taps, coeffs, in, out);
|
406
|
+
break;
|
351
407
|
}
|
352
408
|
}
|
353
409
|
|
354
|
-
|
355
|
-
|
356
|
-
/**
|
357
|
-
* Scanline with extra space at the beginning and end. This allows us to extend
|
358
|
-
* a scanline to the left and right. This in turn allows resizing functions
|
359
|
-
* to operate past the edges of the scanline without having to check for
|
360
|
-
* boundaries.
|
361
|
-
*/
|
362
|
-
struct padded_sl {
|
363
|
-
unsigned char *buf;
|
364
|
-
unsigned char *pad_left;
|
365
|
-
long inner_width;
|
366
|
-
long pad_width;
|
367
|
-
int cmp;
|
368
|
-
};
|
369
|
-
|
370
|
-
void padded_sl_init(struct padded_sl *psl, long inner_width, long pad_width,
|
371
|
-
int cmp)
|
372
|
-
{
|
373
|
-
psl->inner_width = inner_width;
|
374
|
-
psl->pad_width = pad_width;
|
375
|
-
psl->cmp = cmp;
|
376
|
-
psl->pad_left = malloc((inner_width + 2 * pad_width) * cmp);
|
377
|
-
psl->buf = psl->pad_left + pad_width * cmp;
|
378
|
-
}
|
379
|
-
|
380
|
-
void padded_sl_free(struct padded_sl *psl)
|
381
|
-
{
|
382
|
-
free(psl->pad_left);
|
383
|
-
}
|
384
|
-
|
385
|
-
/**
|
386
|
-
* pad points to the first byte in the pad area.
|
387
|
-
* src points to the sample that will be replicated in the pad area.
|
388
|
-
* width is the number of samples in the pad area.
|
389
|
-
* cmp is the number of components per sample.
|
390
|
-
*/
|
391
|
-
static void padded_sl_pad(unsigned char *pad, unsigned char *src, int width,
|
392
|
-
int cmp)
|
410
|
+
void padded_sl_extend_edges(uint16_t *buf, uint32_t width, size_t pad_len,
|
411
|
+
uint8_t cmp)
|
393
412
|
{
|
394
|
-
|
395
|
-
|
396
|
-
|
397
|
-
|
398
|
-
|
413
|
+
uint16_t *pad_right;
|
414
|
+
size_t i;
|
415
|
+
pad_right = buf + pad_len + (size_t)width * cmp;
|
416
|
+
for (i=0; i<pad_len; i++) {
|
417
|
+
buf[i] = (buf + pad_len)[i % cmp];
|
418
|
+
pad_right[i] = (pad_right - cmp)[i % cmp];
|
419
|
+
}
|
399
420
|
}
|
400
421
|
|
401
|
-
|
422
|
+
size_t padded_sl_len_offset(uint32_t in_width, uint32_t out_width,
|
423
|
+
uint8_t cmp, size_t *offset)
|
402
424
|
{
|
403
|
-
|
404
|
-
|
405
|
-
|
406
|
-
|
407
|
-
padded_sl_pad(pad_right, pad_right - psl->cmp, psl->pad_width, psl->cmp);
|
425
|
+
uint64_t taps;
|
426
|
+
taps = calc_taps(in_width, out_width);
|
427
|
+
*offset = (taps / 2 + 1) * cmp;
|
428
|
+
return ((size_t)in_width * cmp + *offset * 2) * sizeof(uint16_t);
|
408
429
|
}
|
409
430
|
|
410
|
-
|
411
|
-
|
431
|
+
int xscale_padded(uint16_t *in, uint32_t in_width, uint16_t *out,
|
432
|
+
uint32_t out_width, enum oil_colorspace cs)
|
412
433
|
{
|
413
434
|
float tx;
|
414
435
|
fix1_30 *coeffs;
|
415
|
-
|
416
|
-
|
417
|
-
|
436
|
+
uint32_t i, j, in_chunk, out_chunk, scale_gcd;
|
437
|
+
int32_t xsmp_i;
|
438
|
+
uint64_t taps;
|
439
|
+
uint16_t *out_pos, *tmp;
|
440
|
+
uint8_t cmp;
|
441
|
+
|
442
|
+
if (!in_width || !out_width) {
|
443
|
+
return -1; // bad input parameter
|
444
|
+
}
|
418
445
|
|
446
|
+
cmp = CS_TO_CMP(cs);
|
419
447
|
taps = calc_taps(in_width, out_width);
|
420
448
|
coeffs = malloc(taps * sizeof(fix1_30));
|
449
|
+
if (!coeffs) {
|
450
|
+
return -2; // unable to allocate space for coefficients
|
451
|
+
}
|
421
452
|
|
422
453
|
scale_gcd = gcd(in_width, out_width);
|
423
454
|
in_chunk = in_width / scale_gcd;
|
424
455
|
out_chunk = out_width / scale_gcd;
|
425
456
|
|
426
|
-
if (in_width < taps * 2) {
|
427
|
-
padded_sl_init(&psl, in_width, taps / 2 + 1, cmp);
|
428
|
-
memcpy(psl.buf, in, in_width * cmp);
|
429
|
-
rpadv = psl.buf;
|
430
|
-
} else {
|
431
|
-
/* just the ends of the scanline with edges extended */
|
432
|
-
padded_sl_init(&psl, 2 * taps - 2, taps / 2 + 1, cmp);
|
433
|
-
memcpy(psl.buf, in, (taps - 1) * cmp);
|
434
|
-
memcpy(psl.buf + (taps - 1) * cmp, in + (in_width - taps + 1) * cmp, (taps - 1) * cmp);
|
435
|
-
rpadv = psl.buf + (2 * taps - 2 - in_width) * cmp;
|
436
|
-
}
|
437
|
-
|
438
|
-
padded_sl_extend_edges(&psl);
|
439
|
-
|
440
457
|
for (i=0; i<out_chunk; i++) {
|
441
458
|
xsmp_i = split_map(in_width, out_width, i, &tx);
|
442
459
|
calc_coeffs(coeffs, tx, taps);
|
@@ -444,21 +461,478 @@ void xscale(unsigned char *in, long in_width, unsigned char *out,
|
|
444
461
|
xsmp_i += 1 - taps / 2;
|
445
462
|
out_pos = out + i * cmp;
|
446
463
|
for (j=0; j<scale_gcd; j++) {
|
447
|
-
|
448
|
-
|
449
|
-
} else if (xsmp_i > in_width - taps) {
|
450
|
-
tmp = rpadv;
|
451
|
-
} else {
|
452
|
-
tmp = in;
|
453
|
-
}
|
454
|
-
tmp += xsmp_i * cmp;
|
455
|
-
xscale_set_sample(taps, coeffs, tmp, out_pos, cmp, opts);
|
464
|
+
tmp = in + xsmp_i * cmp;
|
465
|
+
xscale_set_sample(taps, coeffs, tmp, out_pos, cs);
|
456
466
|
out_pos += out_chunk * cmp;
|
457
|
-
|
458
467
|
xsmp_i += in_chunk;
|
459
468
|
}
|
460
469
|
}
|
461
470
|
|
462
|
-
padded_sl_free(&psl);
|
463
471
|
free(coeffs);
|
472
|
+
return 0;
|
473
|
+
}
|
474
|
+
|
475
|
+
/* scanline ring buffer */
|
476
|
+
|
477
|
+
int sl_rbuf_init(struct sl_rbuf *rb, uint32_t height, size_t sl_len)
|
478
|
+
{
|
479
|
+
rb->height = height;
|
480
|
+
rb->count = 0;
|
481
|
+
rb->length = sl_len;
|
482
|
+
rb->buf = malloc(sl_len * height * sizeof(uint16_t));
|
483
|
+
if (!rb->buf) {
|
484
|
+
return -2;
|
485
|
+
}
|
486
|
+
rb->virt = malloc(sizeof(uint8_t *) * height);
|
487
|
+
if (!rb->virt) {
|
488
|
+
free(rb->buf);
|
489
|
+
return -2;
|
490
|
+
}
|
491
|
+
return 0;
|
492
|
+
}
|
493
|
+
|
494
|
+
void sl_rbuf_free(struct sl_rbuf *rb)
|
495
|
+
{
|
496
|
+
free(rb->buf);
|
497
|
+
free(rb->virt);
|
498
|
+
}
|
499
|
+
|
500
|
+
uint16_t *sl_rbuf_next(struct sl_rbuf *rb)
|
501
|
+
{
|
502
|
+
return rb->buf + (rb->count++ % rb->height) * rb->length;
|
503
|
+
}
|
504
|
+
|
505
|
+
uint16_t **sl_rbuf_virt(struct sl_rbuf *rb, uint32_t last_target)
|
506
|
+
{
|
507
|
+
uint32_t i, safe, height, last_idx;
|
508
|
+
height = rb->height;
|
509
|
+
last_idx = rb->count - 1;
|
510
|
+
|
511
|
+
// Make sure we have the 1st scanline if extending upwards
|
512
|
+
if (last_target < last_idx && last_idx > height - 1) {
|
513
|
+
return 0;
|
514
|
+
}
|
515
|
+
|
516
|
+
for (i=0; i<height; i++) {
|
517
|
+
safe = last_target < i ? 0 : last_target - i;
|
518
|
+
safe = safe > last_idx ? last_idx : safe;
|
519
|
+
rb->virt[height - i - 1] = rb->buf + (safe % height) * rb->length;
|
520
|
+
}
|
521
|
+
return rb->virt;
|
522
|
+
}
|
523
|
+
|
524
|
+
/* xscaler */
|
525
|
+
|
526
|
+
int xscaler_init(struct xscaler *xs, uint32_t width_in, uint32_t width_out,
|
527
|
+
enum oil_colorspace cs)
|
528
|
+
{
|
529
|
+
size_t psl_len, psl_offset;
|
530
|
+
uint16_t *psl_buf;
|
531
|
+
|
532
|
+
psl_len = padded_sl_len_offset(width_in, width_out, CS_TO_CMP(cs), &psl_offset);
|
533
|
+
psl_buf = malloc(psl_len);
|
534
|
+
if (!psl_buf) {
|
535
|
+
return -2;
|
536
|
+
}
|
537
|
+
|
538
|
+
xs->psl_buf = psl_buf;
|
539
|
+
xs->psl_offset = psl_offset;
|
540
|
+
xs->psl_pos0 = psl_buf + psl_offset;
|
541
|
+
xs->width_in = width_in;
|
542
|
+
xs->width_out = width_out;
|
543
|
+
xs->cs = cs;
|
544
|
+
|
545
|
+
return 0;
|
546
|
+
}
|
547
|
+
|
548
|
+
void xscaler_free(struct xscaler *xs)
|
549
|
+
{
|
550
|
+
free(xs->psl_buf);
|
551
|
+
}
|
552
|
+
|
553
|
+
void xscaler_scale(struct xscaler *xs, uint16_t *out_buf)
|
554
|
+
{
|
555
|
+
padded_sl_extend_edges(xs->psl_buf, xs->width_in, xs->psl_offset, CS_TO_CMP(xs->cs));
|
556
|
+
xscale_padded(xs->psl_pos0, xs->width_in, out_buf, xs->width_out, xs->cs);
|
557
|
+
}
|
558
|
+
|
559
|
+
/* yscaler */
|
560
|
+
|
561
|
+
static void yscaler_map_pos(struct yscaler *ys, uint32_t pos)
|
562
|
+
{
|
563
|
+
long target;
|
564
|
+
target = split_map(ys->in_height, ys->out_height, pos, &ys->ty);
|
565
|
+
ys->target = target + ys->rb.height / 2;
|
566
|
+
}
|
567
|
+
|
568
|
+
int yscaler_init(struct yscaler *ys, uint32_t in_height, uint32_t out_height,
|
569
|
+
uint32_t width, enum oil_colorspace cs)
|
570
|
+
{
|
571
|
+
uint8_t cmp;
|
572
|
+
int ret;
|
573
|
+
uint32_t taps;
|
574
|
+
taps = calc_taps(in_height, out_height);
|
575
|
+
ys->in_height = in_height;
|
576
|
+
ys->out_height = out_height;
|
577
|
+
ys->width = width;
|
578
|
+
ys->cs = cs;
|
579
|
+
cmp = CS_TO_CMP(cs);
|
580
|
+
if (cs == OIL_CS_RGB) {
|
581
|
+
cmp = 4;
|
582
|
+
}
|
583
|
+
ret = sl_rbuf_init(&ys->rb, taps, width * cmp);
|
584
|
+
yscaler_map_pos(ys, 0);
|
585
|
+
return ret;
|
586
|
+
}
|
587
|
+
|
588
|
+
void yscaler_free(struct yscaler *ys)
|
589
|
+
{
|
590
|
+
sl_rbuf_free(&ys->rb);
|
591
|
+
}
|
592
|
+
|
593
|
+
uint16_t *yscaler_next(struct yscaler *ys)
|
594
|
+
{
|
595
|
+
if (ys->rb.count == ys->in_height || ys->rb.count > ys->target) {
|
596
|
+
return 0;
|
597
|
+
}
|
598
|
+
return sl_rbuf_next(&ys->rb);
|
599
|
+
}
|
600
|
+
|
601
|
+
int yscaler_scale(struct yscaler *ys, uint8_t *out, uint32_t pos)
|
602
|
+
{
|
603
|
+
int ret;
|
604
|
+
uint16_t **virt;
|
605
|
+
virt = sl_rbuf_virt(&ys->rb, ys->target);
|
606
|
+
ret = strip_scale(virt, ys->rb.height, ys->rb.length, out, ys->ty, ys->cs);
|
607
|
+
yscaler_map_pos(ys, pos + 1);
|
608
|
+
return ret;
|
609
|
+
}
|
610
|
+
|
611
|
+
/* Color Space Helpers */
|
612
|
+
|
613
|
+
#define EXPAND8(X) ((X<<8) + X)
|
614
|
+
|
615
|
+
static uint16_t srgb_sample_to_linear(uint8_t x)
|
616
|
+
{
|
617
|
+
static const uint16_t s2l_map[256] = {
|
618
|
+
0x0000, 0x0014, 0x0028, 0x003c, 0x0050, 0x0063, 0x0077, 0x008b,
|
619
|
+
0x009f, 0x00b3, 0x00c7, 0x00db, 0x00f1, 0x0108, 0x0120, 0x0139,
|
620
|
+
0x0154, 0x016f, 0x018c, 0x01ab, 0x01ca, 0x01eb, 0x020e, 0x0232,
|
621
|
+
0x0257, 0x027d, 0x02a5, 0x02ce, 0x02f9, 0x0325, 0x0353, 0x0382,
|
622
|
+
0x03b3, 0x03e5, 0x0418, 0x044d, 0x0484, 0x04bc, 0x04f6, 0x0532,
|
623
|
+
0x056f, 0x05ad, 0x05ed, 0x062f, 0x0673, 0x06b8, 0x06fe, 0x0747,
|
624
|
+
0x0791, 0x07dd, 0x082a, 0x087a, 0x08ca, 0x091d, 0x0972, 0x09c8,
|
625
|
+
0x0a20, 0x0a79, 0x0ad5, 0x0b32, 0x0b91, 0x0bf2, 0x0c55, 0x0cba,
|
626
|
+
0x0d20, 0x0d88, 0x0df2, 0x0e5e, 0x0ecc, 0x0f3c, 0x0fae, 0x1021,
|
627
|
+
0x1097, 0x110e, 0x1188, 0x1203, 0x1280, 0x1300, 0x1381, 0x1404,
|
628
|
+
0x1489, 0x1510, 0x159a, 0x1625, 0x16b2, 0x1741, 0x17d3, 0x1866,
|
629
|
+
0x18fb, 0x1993, 0x1a2c, 0x1ac8, 0x1b66, 0x1c06, 0x1ca7, 0x1d4c,
|
630
|
+
0x1df2, 0x1e9a, 0x1f44, 0x1ff1, 0x20a0, 0x2150, 0x2204, 0x22b9,
|
631
|
+
0x2370, 0x242a, 0x24e5, 0x25a3, 0x2664, 0x2726, 0x27eb, 0x28b1,
|
632
|
+
0x297b, 0x2a46, 0x2b14, 0x2be3, 0x2cb6, 0x2d8a, 0x2e61, 0x2f3a,
|
633
|
+
0x3015, 0x30f2, 0x31d2, 0x32b4, 0x3399, 0x3480, 0x3569, 0x3655,
|
634
|
+
0x3742, 0x3833, 0x3925, 0x3a1a, 0x3b12, 0x3c0b, 0x3d07, 0x3e06,
|
635
|
+
0x3f07, 0x400a, 0x4110, 0x4218, 0x4323, 0x4430, 0x453f, 0x4651,
|
636
|
+
0x4765, 0x487c, 0x4995, 0x4ab1, 0x4bcf, 0x4cf0, 0x4e13, 0x4f39,
|
637
|
+
0x5061, 0x518c, 0x52b9, 0x53e9, 0x551b, 0x5650, 0x5787, 0x58c1,
|
638
|
+
0x59fe, 0x5b3d, 0x5c7e, 0x5dc2, 0x5f09, 0x6052, 0x619e, 0x62ed,
|
639
|
+
0x643e, 0x6591, 0x66e8, 0x6840, 0x699c, 0x6afa, 0x6c5b, 0x6dbe,
|
640
|
+
0x6f24, 0x708d, 0x71f8, 0x7366, 0x74d7, 0x764a, 0x77c0, 0x7939,
|
641
|
+
0x7ab4, 0x7c32, 0x7db3, 0x7f37, 0x80bd, 0x8246, 0x83d1, 0x855f,
|
642
|
+
0x86f0, 0x8884, 0x8a1b, 0x8bb4, 0x8d50, 0x8eef, 0x9090, 0x9235,
|
643
|
+
0x93dc, 0x9586, 0x9732, 0x98e2, 0x9a94, 0x9c49, 0x9e01, 0x9fbb,
|
644
|
+
0xa179, 0xa339, 0xa4fc, 0xa6c2, 0xa88b, 0xaa56, 0xac25, 0xadf6,
|
645
|
+
0xafca, 0xb1a1, 0xb37b, 0xb557, 0xb737, 0xb919, 0xbaff, 0xbce7,
|
646
|
+
0xbed2, 0xc0c0, 0xc2b1, 0xc4a5, 0xc69c, 0xc895, 0xca92, 0xcc91,
|
647
|
+
0xce94, 0xd099, 0xd2a1, 0xd4ad, 0xd6bb, 0xd8cc, 0xdae0, 0xdcf7,
|
648
|
+
0xdf11, 0xe12e, 0xe34e, 0xe571, 0xe797, 0xe9c0, 0xebec, 0xee1b,
|
649
|
+
0xf04d, 0xf282, 0xf4ba, 0xf6f5, 0xf933, 0xfb74, 0xfdb8, 0xffff,
|
650
|
+
};
|
651
|
+
return s2l_map[x];
|
652
|
+
}
|
653
|
+
|
654
|
+
static void srgbx_preprocess_nx(uint8_t *in, uint16_t *out, uint32_t in_width, uint32_t n)
|
655
|
+
{
|
656
|
+
uint32_t i, j;
|
657
|
+
uint32_t sums[3];
|
658
|
+
for (i=0; i<in_width/n; i++) {
|
659
|
+
sums[0] = sums[1] = sums[2] = 0;
|
660
|
+
for (j=0; j<n; j++) {
|
661
|
+
sums[0] += srgb_sample_to_linear(in[0]);
|
662
|
+
sums[1] += srgb_sample_to_linear(in[1]);
|
663
|
+
sums[2] += srgb_sample_to_linear(in[2]);
|
664
|
+
in += 4;
|
665
|
+
}
|
666
|
+
out[0] = sums[0] / n;
|
667
|
+
out[1] = sums[1] / n;
|
668
|
+
out[2] = sums[2] / n;
|
669
|
+
out[3] = 0;
|
670
|
+
out += 4;
|
671
|
+
}
|
672
|
+
}
|
673
|
+
|
674
|
+
static void srgba_preprocess_nx(uint8_t *in, uint16_t *out, uint32_t in_width, uint32_t n)
|
675
|
+
{
|
676
|
+
uint32_t i, j;
|
677
|
+
uint32_t sums[4];
|
678
|
+
for (i=0; i<in_width/n; i++) {
|
679
|
+
sums[0] = sums[1] = sums[2] = sums[3] = 0;
|
680
|
+
for (j=0; j<n; j++) {
|
681
|
+
sums[0] += srgb_sample_to_linear(in[0]);
|
682
|
+
sums[1] += srgb_sample_to_linear(in[1]);
|
683
|
+
sums[2] += srgb_sample_to_linear(in[2]);
|
684
|
+
sums[3] += EXPAND8(in[3]);
|
685
|
+
in += 4;
|
686
|
+
}
|
687
|
+
out[0] = sums[0] / n;
|
688
|
+
out[1] = sums[1] / n;
|
689
|
+
out[2] = sums[2] / n;
|
690
|
+
out[3] = sums[3] / n;
|
691
|
+
out += 4;
|
692
|
+
}
|
693
|
+
}
|
694
|
+
|
695
|
+
static void srgb_preprocess_nx(uint8_t *in, uint16_t *out, uint32_t in_width, uint32_t n)
|
696
|
+
{
|
697
|
+
uint32_t i, j;
|
698
|
+
uint32_t sums[3];
|
699
|
+
for (i=0; i<in_width/n; i++) {
|
700
|
+
sums[0] = sums[1] = sums[2] = 0;
|
701
|
+
for (j=0; j<n; j++) {
|
702
|
+
sums[0] += srgb_sample_to_linear(in[0]);
|
703
|
+
sums[1] += srgb_sample_to_linear(in[1]);
|
704
|
+
sums[2] += srgb_sample_to_linear(in[2]);
|
705
|
+
in += 3;
|
706
|
+
}
|
707
|
+
out[0] = sums[0] / n;
|
708
|
+
out[1] = sums[1] / n;
|
709
|
+
out[2] = sums[2] / n;
|
710
|
+
out[3] = 0;
|
711
|
+
out += 4;
|
712
|
+
}
|
713
|
+
}
|
714
|
+
|
715
|
+
static void g_preprocess_nx(uint8_t *in, uint16_t *out, uint32_t in_width, uint32_t n)
|
716
|
+
{
|
717
|
+
uint32_t i, j;
|
718
|
+
uint32_t sum;
|
719
|
+
for (i=0; i<in_width/n; i++) {
|
720
|
+
sum = 0;
|
721
|
+
for (j=0; j<n; j++) {
|
722
|
+
sum += EXPAND8(in[0]);
|
723
|
+
in++;
|
724
|
+
}
|
725
|
+
out[0] = sum / n;
|
726
|
+
out++;
|
727
|
+
}
|
728
|
+
}
|
729
|
+
|
730
|
+
static void ga_preprocess_nx(uint8_t *in, uint16_t *out, uint32_t in_width, uint32_t n)
|
731
|
+
{
|
732
|
+
uint32_t i, j;
|
733
|
+
uint32_t sums[2];
|
734
|
+
for (i=0; i<in_width/n; i++) {
|
735
|
+
sums[0] = sums[1] = 0;
|
736
|
+
for (j=0; j<n; j++) {
|
737
|
+
sums[0] += EXPAND8(in[0]);
|
738
|
+
sums[1] += EXPAND8(in[1]);
|
739
|
+
in += 2;
|
740
|
+
}
|
741
|
+
out[0] = sums[0] / n;
|
742
|
+
out[1] = sums[1] / n;
|
743
|
+
out += 2;
|
744
|
+
}
|
745
|
+
}
|
746
|
+
|
747
|
+
static void cmyk_preprocess_nx(uint8_t *in, uint16_t *out, uint32_t in_width, uint32_t n)
|
748
|
+
{
|
749
|
+
uint32_t i, j;
|
750
|
+
uint32_t sums[4];
|
751
|
+
for (i=0; i<in_width/n; i++) {
|
752
|
+
sums[0] = sums[1] = sums[2] = sums[3] = 0;
|
753
|
+
for (j=0; j<n; j++) {
|
754
|
+
sums[0] += EXPAND8(in[0]);
|
755
|
+
sums[1] += EXPAND8(in[1]);
|
756
|
+
sums[2] += EXPAND8(in[2]);
|
757
|
+
sums[3] += EXPAND8(in[3]);
|
758
|
+
in += 4;
|
759
|
+
}
|
760
|
+
out[0] = sums[0] / n;
|
761
|
+
out[1] = sums[1] / n;
|
762
|
+
out[2] = sums[2] / n;
|
763
|
+
out[3] = sums[3] / n;
|
764
|
+
out += 4;
|
765
|
+
}
|
766
|
+
}
|
767
|
+
|
768
|
+
static uint32_t calc_pre_shrink(uint32_t dim_in, uint32_t dim_out)
|
769
|
+
{
|
770
|
+
uint32_t max;
|
771
|
+
max = (2 * dim_in) / dim_out / 3;
|
772
|
+
if (max >= 4 && (dim_in % 4) == 0) {
|
773
|
+
return 4;
|
774
|
+
}
|
775
|
+
if (max >= 2 && (dim_in % 2) == 0) {
|
776
|
+
return 2;
|
777
|
+
}
|
778
|
+
return 1;
|
779
|
+
}
|
780
|
+
|
781
|
+
int preprocess_xscaler_init(struct preprocess_xscaler *pxs, uint32_t width_in,
|
782
|
+
uint32_t width_out, enum oil_colorspace cs_in)
|
783
|
+
{
|
784
|
+
enum oil_colorspace cs_out;
|
785
|
+
|
786
|
+
pxs->width_in = width_in;
|
787
|
+
pxs->cs_in = cs_in;
|
788
|
+
pxs->scale_factor = calc_pre_shrink(width_in, width_out);
|
789
|
+
|
790
|
+
/* Auto promote rgb components to rgbx for performance */
|
791
|
+
cs_out = cs_in == OIL_CS_RGB ? OIL_CS_RGBX : cs_in;
|
792
|
+
return xscaler_init(&pxs->xs, width_in / pxs->scale_factor, width_out, cs_out);
|
793
|
+
}
|
794
|
+
|
795
|
+
void preprocess_xscaler_free(struct preprocess_xscaler *pxs)
|
796
|
+
{
|
797
|
+
xscaler_free(&pxs->xs);
|
798
|
+
}
|
799
|
+
|
800
|
+
static void pre_convert_g(uint8_t *in, uint16_t *out,
|
801
|
+
uint32_t width_in, uint32_t scale_factor)
|
802
|
+
{
|
803
|
+
switch(scale_factor) {
|
804
|
+
case 1:
|
805
|
+
g_preprocess_nx(in, out, width_in, 1);
|
806
|
+
break;
|
807
|
+
case 2:
|
808
|
+
g_preprocess_nx(in, out, width_in, 2);
|
809
|
+
break;
|
810
|
+
case 4:
|
811
|
+
g_preprocess_nx(in, out, width_in, 4);
|
812
|
+
break;
|
813
|
+
}
|
814
|
+
}
|
815
|
+
|
816
|
+
static void pre_convert_ga(uint8_t *in, uint16_t *out,
|
817
|
+
uint32_t width_in, uint32_t scale_factor)
|
818
|
+
{
|
819
|
+
switch(scale_factor) {
|
820
|
+
case 1:
|
821
|
+
ga_preprocess_nx(in, out, width_in, 1);
|
822
|
+
break;
|
823
|
+
case 2:
|
824
|
+
ga_preprocess_nx(in, out, width_in, 2);
|
825
|
+
break;
|
826
|
+
case 4:
|
827
|
+
ga_preprocess_nx(in, out, width_in, 4);
|
828
|
+
break;
|
829
|
+
}
|
830
|
+
}
|
831
|
+
|
832
|
+
static void pre_convert_cmyk(uint8_t *in, uint16_t *out,
|
833
|
+
uint32_t width_in, uint32_t scale_factor)
|
834
|
+
{
|
835
|
+
switch(scale_factor) {
|
836
|
+
case 1:
|
837
|
+
cmyk_preprocess_nx(in, out, width_in, 1);
|
838
|
+
break;
|
839
|
+
case 2:
|
840
|
+
cmyk_preprocess_nx(in, out, width_in, 2);
|
841
|
+
break;
|
842
|
+
case 4:
|
843
|
+
cmyk_preprocess_nx(in, out, width_in, 4);
|
844
|
+
break;
|
845
|
+
}
|
846
|
+
}
|
847
|
+
|
848
|
+
static void pre_convert_rgbx(uint8_t *in, uint16_t *out,
|
849
|
+
uint32_t width_in, uint32_t scale_factor)
|
850
|
+
{
|
851
|
+
switch(scale_factor) {
|
852
|
+
case 1:
|
853
|
+
srgbx_preprocess_nx(in, out, width_in, 1);
|
854
|
+
break;
|
855
|
+
case 2:
|
856
|
+
srgbx_preprocess_nx(in, out, width_in, 2);
|
857
|
+
break;
|
858
|
+
case 4:
|
859
|
+
srgbx_preprocess_nx(in, out, width_in, 4);
|
860
|
+
break;
|
861
|
+
}
|
862
|
+
}
|
863
|
+
|
864
|
+
static void pre_convert_rgba(uint8_t *in, uint16_t *out,
|
865
|
+
uint32_t width_in, uint32_t scale_factor)
|
866
|
+
{
|
867
|
+
switch(scale_factor) {
|
868
|
+
case 1:
|
869
|
+
srgba_preprocess_nx(in, out, width_in, 1);
|
870
|
+
break;
|
871
|
+
case 2:
|
872
|
+
srgba_preprocess_nx(in, out, width_in, 2);
|
873
|
+
break;
|
874
|
+
case 4:
|
875
|
+
srgba_preprocess_nx(in, out, width_in, 4);
|
876
|
+
break;
|
877
|
+
}
|
878
|
+
}
|
879
|
+
|
880
|
+
static void pre_convert_rgb(uint8_t *in, uint16_t *out,
|
881
|
+
uint32_t width_in, uint32_t scale_factor)
|
882
|
+
{
|
883
|
+
switch(scale_factor) {
|
884
|
+
case 1:
|
885
|
+
srgb_preprocess_nx(in, out, width_in, 1);
|
886
|
+
break;
|
887
|
+
case 2:
|
888
|
+
srgb_preprocess_nx(in, out, width_in, 2);
|
889
|
+
break;
|
890
|
+
case 4:
|
891
|
+
srgb_preprocess_nx(in, out, width_in, 4);
|
892
|
+
break;
|
893
|
+
}
|
894
|
+
}
|
895
|
+
|
896
|
+
void preprocess_xscaler_scale(struct preprocess_xscaler *pxs, uint8_t *in,
|
897
|
+
uint16_t *out)
|
898
|
+
{
|
899
|
+
switch(pxs->cs_in) {
|
900
|
+
case OIL_CS_G:
|
901
|
+
pre_convert_g(in, pxs->xs.psl_pos0, pxs->width_in, pxs->scale_factor);
|
902
|
+
break;
|
903
|
+
case OIL_CS_GA:
|
904
|
+
pre_convert_ga(in, pxs->xs.psl_pos0, pxs->width_in, pxs->scale_factor);
|
905
|
+
break;
|
906
|
+
case OIL_CS_RGB:
|
907
|
+
pre_convert_rgb(in, pxs->xs.psl_pos0, pxs->width_in, pxs->scale_factor);
|
908
|
+
break;
|
909
|
+
case OIL_CS_RGBX:
|
910
|
+
pre_convert_rgbx(in, pxs->xs.psl_pos0, pxs->width_in, pxs->scale_factor);
|
911
|
+
break;
|
912
|
+
case OIL_CS_RGBA:
|
913
|
+
pre_convert_rgba(in, pxs->xs.psl_pos0, pxs->width_in, pxs->scale_factor);
|
914
|
+
break;
|
915
|
+
case OIL_CS_CMYK:
|
916
|
+
pre_convert_cmyk(in, pxs->xs.psl_pos0, pxs->width_in, pxs->scale_factor);
|
917
|
+
break;
|
918
|
+
}
|
919
|
+
xscaler_scale(&pxs->xs, out);
|
920
|
+
}
|
921
|
+
|
922
|
+
|
923
|
+
/* Utility helpers */
|
924
|
+
void fix_ratio(uint32_t src_width, uint32_t src_height, uint32_t *out_width,
|
925
|
+
uint32_t *out_height)
|
926
|
+
{
|
927
|
+
double width_ratio, height_ratio;
|
928
|
+
|
929
|
+
width_ratio = *out_width / (double)src_width;
|
930
|
+
height_ratio = *out_height / (double)src_height;
|
931
|
+
if (width_ratio < height_ratio) {
|
932
|
+
*out_height = round(width_ratio * src_height);
|
933
|
+
*out_height = *out_height ? *out_height : 1;
|
934
|
+
} else {
|
935
|
+
*out_width = round(height_ratio * src_width);
|
936
|
+
*out_width = *out_width ? *out_width : 1;
|
937
|
+
}
|
464
938
|
}
|