?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Backed out changeset b1aca8e61c5c (bug 958980) for bustage. a=backout
[gecko.git] / gfx / cairo / pixman-8888-over-565.patch
blobd6ab4618f8b67f07726be6106dc55f8086b3c2f9
1 changeset: 96613:3e003f0b8026
2 tag: 2pass
3 tag: qbase
4 tag: qtip
5 tag: tip
6 user: Jeff Muizelaar <jmuizelaar@mozilla.com>
7 date: Thu May 17 19:23:53 2012 -0400
8 summary: Bug 757878. Add a fast path for 8888_over_565 with NEON. r=bgirard,joe
9
10 diff --git a/gfx/cairo/libpixman/src/pixman-arm-common.h b/gfx/cairo/libpixman/src/pixman-arm-common.h
11 --- a/gfx/cairo/libpixman/src/pixman-arm-common.h
12 +++ b/gfx/cairo/libpixman/src/pixman-arm-common.h
13 @@ -355,26 +355,26 @@ scaled_bilinear_scanline_##cputype##_##n
14 if ((flags & SKIP_ZERO_SRC) && zero_src) \
15 return; \
16 pixman_scaled_bilinear_scanline_##name##_##op##_asm_##cputype ( \
17 dst, src_top, src_bottom, wt, wb, vx, unit_x, w); \
18 } \
19 \
20 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_cover_##op, \
21 scaled_bilinear_scanline_##cputype##_##name##_##op, \
22 - src_type, uint32_t, dst_type, COVER, FLAG_NONE) \
23 + NULL, src_type, uint32_t, dst_type, COVER, FLAG_NONE) \
24 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_none_##op, \
25 scaled_bilinear_scanline_##cputype##_##name##_##op, \
26 - src_type, uint32_t, dst_type, NONE, FLAG_NONE) \
27 + NULL, src_type, uint32_t, dst_type, NONE, FLAG_NONE) \
28 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_pad_##op, \
29 scaled_bilinear_scanline_##cputype##_##name##_##op, \
30 - src_type, uint32_t, dst_type, PAD, FLAG_NONE) \
31 + NULL, src_type, uint32_t, dst_type, PAD, FLAG_NONE) \
32 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_normal_##op, \
33 scaled_bilinear_scanline_##cputype##_##name##_##op, \
34 - src_type, uint32_t, dst_type, NORMAL, \
35 + NULL, src_type, uint32_t, dst_type, NORMAL, \
36 FLAG_NONE)
37
38
39 #define PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_A8_DST(flags, cputype, name, op, \
40 src_type, dst_type) \
41 void \
42 pixman_scaled_bilinear_scanline_##name##_##op##_asm_##cputype ( \
43 dst_type * dst, \
44 @@ -404,25 +404,25 @@ scaled_bilinear_scanline_##cputype##_##n
45 if ((flags & SKIP_ZERO_SRC) && zero_src) \
46 return; \
47 pixman_scaled_bilinear_scanline_##name##_##op##_asm_##cputype ( \
48 dst, mask, src_top, src_bottom, wt, wb, vx, unit_x, w); \
49 } \
50 \
51 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_cover_##op, \
52 scaled_bilinear_scanline_##cputype##_##name##_##op, \
53 - src_type, uint8_t, dst_type, COVER, \
54 + NULL, src_type, uint8_t, dst_type, COVER, \
55 FLAG_HAVE_NON_SOLID_MASK) \
56 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_none_##op, \
57 scaled_bilinear_scanline_##cputype##_##name##_##op, \
58 - src_type, uint8_t, dst_type, NONE, \
59 + NULL, src_type, uint8_t, dst_type, NONE, \
60 FLAG_HAVE_NON_SOLID_MASK) \
61 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_pad_##op, \
62 scaled_bilinear_scanline_##cputype##_##name##_##op, \
63 - src_type, uint8_t, dst_type, PAD, \
64 + NULL, src_type, uint8_t, dst_type, PAD, \
65 FLAG_HAVE_NON_SOLID_MASK) \
66 FAST_BILINEAR_MAINLOOP_COMMON (cputype##_##name##_normal_##op, \
67 scaled_bilinear_scanline_##cputype##_##name##_##op, \
68 - src_type, uint8_t, dst_type, NORMAL, \
69 + NULL, src_type, uint8_t, dst_type, NORMAL, \
70 FLAG_HAVE_NON_SOLID_MASK)
71
72
73 #endif
74 diff --git a/gfx/cairo/libpixman/src/pixman-arm-neon.c b/gfx/cairo/libpixman/src/pixman-arm-neon.c
75 --- a/gfx/cairo/libpixman/src/pixman-arm-neon.c
76 +++ b/gfx/cairo/libpixman/src/pixman-arm-neon.c
77 @@ -140,16 +140,33 @@ PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_DST
78 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_DST (0, neon, 8888_0565, SRC,
79 uint32_t, uint16_t)
80 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_DST (0, neon, 0565_x888, SRC,
81 uint16_t, uint32_t)
82 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_DST (0, neon, 0565_0565, SRC,
83 uint16_t, uint16_t)
84 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_DST (SKIP_ZERO_SRC, neon, 8888_8888, OVER,
85 uint32_t, uint32_t)
86 +static force_inline void
87 +pixman_scaled_bilinear_scanline_8888_8888_SRC (
88 + uint32_t * dst,
89 + const uint32_t * mask,
90 + const uint32_t * src_top,
91 + const uint32_t * src_bottom,
92 + int32_t w,
93 + int wt,
94 + int wb,
95 + pixman_fixed_t vx,
96 + pixman_fixed_t unit_x,
97 + pixman_fixed_t max_vx,
98 + pixman_bool_t zero_src)
99 +{
100 + pixman_scaled_bilinear_scanline_8888_8888_SRC_asm_neon (dst, src_top, src_bottom, wt, wb, vx, unit_x, w);
101 +}
102 +
103 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_DST (SKIP_ZERO_SRC, neon, 8888_8888, ADD,
104 uint32_t, uint32_t)
105
106 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_A8_DST (0, neon, 8888_8_8888, SRC,
107 uint32_t, uint32_t)
108 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_A8_DST (0, neon, 8888_8_0565, SRC,
109 uint32_t, uint16_t)
110 PIXMAN_ARM_BIND_SCALED_BILINEAR_SRC_A8_DST (0, neon, 0565_8_x888, SRC,
111 @@ -261,16 +278,38 @@ pixman_blt_neon (uint32_t *src_bits,
112 (uint32_t *)(((char *) src_bits) +
113 src_y * src_stride * 4 + src_x * 4), src_stride);
114 return TRUE;
115 default:
116 return FALSE;
117 }
118 }
119
120 +static inline void op_bilinear_over_8888_0565(uint16_t *dst, const uint32_t *mask, const uint32_t *src, int width)
121 +{
122 + pixman_composite_over_8888_0565_asm_neon (width, 1, dst, 0, src, 0);
123 +}
124 +
125 +FAST_BILINEAR_MAINLOOP_COMMON (neon_8888_0565_cover_OVER,
126 + pixman_scaled_bilinear_scanline_8888_8888_SRC, op_bilinear_over_8888_0565,
127 + uint32_t, uint32_t, uint16_t,
128 + COVER, FLAG_NONE)
129 +FAST_BILINEAR_MAINLOOP_COMMON (neon_8888_0565_pad_OVER,
130 + pixman_scaled_bilinear_scanline_8888_8888_SRC, op_bilinear_over_8888_0565,
131 + uint32_t, uint32_t, uint16_t,
132 + PAD, FLAG_NONE)
133 +FAST_BILINEAR_MAINLOOP_COMMON (neon_8888_0565_none_OVER,
134 + pixman_scaled_bilinear_scanline_8888_8888_SRC, op_bilinear_over_8888_0565,
135 + uint32_t, uint32_t, uint16_t,
136 + NONE, FLAG_NONE)
137 +FAST_BILINEAR_MAINLOOP_COMMON (neon_8888_0565_normal_OVER,
138 + pixman_scaled_bilinear_scanline_8888_8888_SRC, op_bilinear_over_8888_0565,
139 + uint32_t, uint32_t, uint16_t,
140 + NORMAL, FLAG_NONE)
141 +
142 static const pixman_fast_path_t arm_neon_fast_paths[] =
143 {
144 PIXMAN_STD_FAST_PATH (SRC, r5g6b5, null, r5g6b5, neon_composite_src_0565_0565),
145 PIXMAN_STD_FAST_PATH (SRC, b5g6r5, null, b5g6r5, neon_composite_src_0565_0565),
146 PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, r5g6b5, neon_composite_src_8888_0565),
147 PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, r5g6b5, neon_composite_src_8888_0565),
148 PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, b5g6r5, neon_composite_src_8888_0565),
149 PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, b5g6r5, neon_composite_src_8888_0565),
150 @@ -414,16 +453,18 @@ static const pixman_fast_path_t arm_neon
151 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (SRC, r5g6b5, r5g6b5, neon_0565_8_0565),
152
153 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, neon_8888_8_8888),
154 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, neon_8888_8_8888),
155
156 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (ADD, a8r8g8b8, a8r8g8b8, neon_8888_8_8888),
157 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (ADD, a8r8g8b8, x8r8g8b8, neon_8888_8_8888),
158
159 + SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, r5g6b5, neon_8888_0565),
160 +
161 { PIXMAN_OP_NONE },
162 };
163
164 static pixman_bool_t
165 arm_neon_blt (pixman_implementation_t *imp,
166 uint32_t * src_bits,
167 uint32_t * dst_bits,
168 int src_stride,
169 diff --git a/gfx/cairo/libpixman/src/pixman-fast-path.c b/gfx/cairo/libpixman/src/pixman-fast-path.c
170 --- a/gfx/cairo/libpixman/src/pixman-fast-path.c
171 +++ b/gfx/cairo/libpixman/src/pixman-fast-path.c
172 @@ -1356,63 +1356,63 @@ scaled_bilinear_scanline_565_565_SRC (ui
173 vx += unit_x;
174 *dst++ = d;
175 }
176 }
177
178 #endif
179
180 FAST_BILINEAR_MAINLOOP_COMMON (565_565_cover_SRC,
181 - scaled_bilinear_scanline_565_565_SRC,
182 + scaled_bilinear_scanline_565_565_SRC, NULL,
183 uint16_t, uint32_t, uint16_t,
184 COVER, FLAG_NONE)
185 FAST_BILINEAR_MAINLOOP_COMMON (565_565_pad_SRC,
186 - scaled_bilinear_scanline_565_565_SRC,
187 + scaled_bilinear_scanline_565_565_SRC, NULL,
188 uint16_t, uint32_t, uint16_t,
189 PAD, FLAG_NONE)
190 FAST_BILINEAR_MAINLOOP_COMMON (565_565_none_SRC,
191 - scaled_bilinear_scanline_565_565_SRC,
192 + scaled_bilinear_scanline_565_565_SRC, NULL,
193 uint16_t, uint32_t, uint16_t,
194 NONE, FLAG_NONE)
195 FAST_BILINEAR_MAINLOOP_COMMON (565_565_normal_SRC,
196 - scaled_bilinear_scanline_565_565_SRC,
197 + scaled_bilinear_scanline_565_565_SRC, NULL,
198 uint16_t, uint32_t, uint16_t,
199 NORMAL, FLAG_NONE)
200
201 FAST_BILINEAR_MAINLOOP_COMMON (8888_565_cover_OVER,
202 - scaled_bilinear_scanline_8888_565_OVER,
203 + scaled_bilinear_scanline_8888_565_OVER, NULL,
204 uint32_t, uint32_t, uint16_t,
205 COVER, FLAG_NONE)
206 FAST_BILINEAR_MAINLOOP_COMMON (8888_565_pad_OVER,
207 - scaled_bilinear_scanline_8888_565_OVER,
208 + scaled_bilinear_scanline_8888_565_OVER, NULL,
209 uint32_t, uint32_t, uint16_t,
210 PAD, FLAG_NONE)
211 FAST_BILINEAR_MAINLOOP_COMMON (8888_565_none_OVER,
212 - scaled_bilinear_scanline_8888_565_OVER,
213 + scaled_bilinear_scanline_8888_565_OVER, NULL,
214 uint32_t, uint32_t, uint16_t,
215 NONE, FLAG_NONE)
216 FAST_BILINEAR_MAINLOOP_COMMON (8888_565_normal_OVER,
217 - scaled_bilinear_scanline_8888_565_OVER,
218 + scaled_bilinear_scanline_8888_565_OVER, NULL,
219 uint32_t, uint32_t, uint16_t,
220 NORMAL, FLAG_NONE)
221
222 FAST_BILINEAR_MAINLOOP_COMMON (8888_8888_cover_OVER,
223 - scaled_bilinear_scanline_8888_8888_OVER,
224 + scaled_bilinear_scanline_8888_8888_OVER, NULL,
225 uint32_t, uint32_t, uint32_t,
226 COVER, FLAG_NONE)
227 FAST_BILINEAR_MAINLOOP_COMMON (8888_8888_pad_OVER,
228 - scaled_bilinear_scanline_8888_8888_OVER,
229 + scaled_bilinear_scanline_8888_8888_OVER, NULL,
230 uint32_t, uint32_t, uint32_t,
231 PAD, FLAG_NONE)
232 FAST_BILINEAR_MAINLOOP_COMMON (8888_8888_none_OVER,
233 - scaled_bilinear_scanline_8888_8888_OVER,
234 + scaled_bilinear_scanline_8888_8888_OVER, NULL,
235 uint32_t, uint32_t, uint32_t,
236 NONE, FLAG_NONE)
237 FAST_BILINEAR_MAINLOOP_COMMON (8888_8888_normal_OVER,
238 - scaled_bilinear_scanline_8888_8888_OVER,
239 + scaled_bilinear_scanline_8888_8888_OVER, NULL,
240 uint32_t, uint32_t, uint32_t,
241 NORMAL, FLAG_NONE)
242
243 #define REPEAT_MIN_WIDTH 32
244
245 static void
246 fast_composite_tiled_repeat (pixman_implementation_t *imp,
247 pixman_composite_info_t *info)
248 diff --git a/gfx/cairo/libpixman/src/pixman-inlines.h b/gfx/cairo/libpixman/src/pixman-inlines.h
249 --- a/gfx/cairo/libpixman/src/pixman-inlines.h
250 +++ b/gfx/cairo/libpixman/src/pixman-inlines.h
251 @@ -816,18 +816,48 @@ bilinear_pad_repeat_get_scanline_bounds
252 *
253 * Note: normally the sum of 'weight_top' and 'weight_bottom' is equal to 256,
254 * but sometimes it may be less than that for NONE repeat when handling
255 * fuzzy antialiased top or bottom image edges. Also both top and
256 * bottom weight variables are guaranteed to have value in 0-255
257 * range and can fit into unsigned byte or be used with 8-bit SIMD
258 * multiplication instructions.
259 */
260 -#define FAST_BILINEAR_MAINLOOP_INT(scale_func_name, scanline_func, src_type_t, mask_type_t, \
261 - dst_type_t, repeat_mode, flags) \
262 +
263 +/* Replace a single "scanline_func" with "fetch_func" & "op_func" to allow optional
264 + * two stage processing (bilinear fetch to a temp buffer, followed by unscaled
265 + * combine), "op_func" may be NULL, in this case we keep old behavior.
266 + * This is ugly and gcc issues some warnings, but works.
267 + *
268 + * An advice: clang has much better error reporting than gcc for deeply nested macros.
269 + */
270 +
271 +#define scanline_func(dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
272 + scanline_buf, mask, src_top, src_bottom, width, \
273 + weight_top, weight_bottom, vx, unit_x, max_vx, zero_src) \
274 + do { \
275 + if (op_func != NULL) \
276 + { \
277 + fetch_func ((void *)scanline_buf, (mask), (src_top), (src_bottom), (width), \
278 + (weight_top), (weight_bottom), (vx), (unit_x), (max_vx), (zero_src)); \
279 + ((void (*)(dst_type_t *, const mask_type_t *, const src_type_t *, int)) op_func)\
280 + ((dst), (mask), (src_type_t *)scanline_buf, (width)); \
281 + } \
282 + else \
283 + { \
284 + fetch_func ((void*)(dst), (mask), (src_top), (src_bottom), (width), (weight_top), \
285 + (weight_bottom), (vx), (unit_x), (max_vx), (zero_src)); \
286 + } \
287 + } while (0)
288 +
289 +
290 +#define SCANLINE_BUFFER_LENGTH 3072
291 +
292 +#define FAST_BILINEAR_MAINLOOP_INT(scale_func_name, fetch_func, op_func, src_type_t, \
293 + mask_type_t, dst_type_t, repeat_mode, flags) \
294 static void \
295 fast_composite_scaled_bilinear ## scale_func_name (pixman_implementation_t *imp, \
296 pixman_composite_info_t *info) \
297 { \
298 PIXMAN_COMPOSITE_ARGS (info); \
299 dst_type_t *dst_line; \
300 mask_type_t *mask_line; \
301 src_type_t *src_first_line; \
302 @@ -842,16 +872,19 @@ fast_composite_scaled_bilinear ## scale_
303 mask_type_t solid_mask; \
304 const mask_type_t *mask = &solid_mask; \
305 int src_stride, mask_stride, dst_stride; \
306 \
307 int src_width; \
308 pixman_fixed_t src_width_fixed; \
309 int max_x; \
310 pixman_bool_t need_src_extension; \
311 + \
312 + uint64_t stack_scanline_buffer[SCANLINE_BUFFER_LENGTH]; \
313 + uint8_t *scanline_buffer = (uint8_t *) stack_scanline_buffer; \
314 \
315 PIXMAN_IMAGE_GET_LINE (dest_image, dest_x, dest_y, dst_type_t, dst_stride, dst_line, 1); \
316 if (flags & FLAG_HAVE_SOLID_MASK) \
317 { \
318 solid_mask = _pixman_image_get_solid (imp, mask_image, dest_image->bits.format); \
319 mask_stride = 0; \
320 } \
321 else if (flags & FLAG_HAVE_NON_SOLID_MASK) \
322 @@ -914,16 +947,24 @@ fast_composite_scaled_bilinear ## scale_
323 else \
324 { \
325 src_width = src_image->bits.width; \
326 need_src_extension = FALSE; \
327 } \
328 \
329 src_width_fixed = pixman_int_to_fixed (src_width); \
330 } \
331 + \
332 + if (op_func != NULL && width * sizeof(src_type_t) > sizeof(stack_scanline_buffer)) \
333 + { \
334 + scanline_buffer = pixman_malloc_ab (width, sizeof(src_type_t)); \
335 + \
336 + if (!scanline_buffer) \
337 + return; \
338 + } \
339 \
340 while (--height >= 0) \
341 { \
342 int weight1, weight2; \
343 dst = dst_line; \
344 dst_line += dst_stride; \
345 vx = v.vector[0]; \
346 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
347 @@ -956,36 +997,39 @@ fast_composite_scaled_bilinear ## scale_
348 repeat (PIXMAN_REPEAT_PAD, &y2, src_image->bits.height); \
349 src1 = src_first_line + src_stride * y1; \
350 src2 = src_first_line + src_stride * y2; \
351 \
352 if (left_pad > 0) \
353 { \
354 buf1[0] = buf1[1] = src1[0]; \
355 buf2[0] = buf2[1] = src2[0]; \
356 - scanline_func (dst, mask, \
357 - buf1, buf2, left_pad, weight1, weight2, 0, 0, 0, FALSE); \
358 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
359 + scanline_buffer, mask, buf1, buf2, left_pad, weight1, weight2, \
360 + 0, 0, 0, FALSE); \
361 dst += left_pad; \
362 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
363 mask += left_pad; \
364 } \
365 if (width > 0) \
366 { \
367 - scanline_func (dst, mask, \
368 - src1, src2, width, weight1, weight2, vx, unit_x, 0, FALSE); \
369 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
370 + scanline_buffer, mask, src1, src2, width, weight1, weight2, \
371 + vx, unit_x, 0, FALSE); \
372 dst += width; \
373 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
374 mask += width; \
375 } \
376 if (right_pad > 0) \
377 { \
378 buf1[0] = buf1[1] = src1[src_image->bits.width - 1]; \
379 buf2[0] = buf2[1] = src2[src_image->bits.width - 1]; \
380 - scanline_func (dst, mask, \
381 - buf1, buf2, right_pad, weight1, weight2, 0, 0, 0, FALSE); \
382 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
383 + scanline_buffer, mask, buf1, buf2, right_pad, weight1, weight2, \
384 + 0, 0, 0, FALSE); \
385 } \
386 } \
387 else if (PIXMAN_REPEAT_ ## repeat_mode == PIXMAN_REPEAT_NONE) \
388 { \
389 src_type_t *src1, *src2; \
390 src_type_t buf1[2]; \
391 src_type_t buf2[2]; \
392 /* handle top/bottom zero padding by just setting weights to 0 if needed */ \
393 @@ -1011,64 +1055,67 @@ fast_composite_scaled_bilinear ## scale_
394 } \
395 src1 = src_first_line + src_stride * y1; \
396 src2 = src_first_line + src_stride * y2; \
397 \
398 if (left_pad > 0) \
399 { \
400 buf1[0] = buf1[1] = 0; \
401 buf2[0] = buf2[1] = 0; \
402 - scanline_func (dst, mask, \
403 - buf1, buf2, left_pad, weight1, weight2, 0, 0, 0, TRUE); \
404 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
405 + scanline_buffer, mask, buf1, buf2, left_pad, weight1, weight2, \
406 + 0, 0, 0, TRUE); \
407 dst += left_pad; \
408 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
409 mask += left_pad; \
410 } \
411 if (left_tz > 0) \
412 { \
413 buf1[0] = 0; \
414 buf1[1] = src1[0]; \
415 buf2[0] = 0; \
416 buf2[1] = src2[0]; \
417 - scanline_func (dst, mask, \
418 - buf1, buf2, left_tz, weight1, weight2, \
419 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
420 + scanline_buffer, mask, buf1, buf2, left_tz, weight1, weight2, \
421 pixman_fixed_frac (vx), unit_x, 0, FALSE); \
422 dst += left_tz; \
423 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
424 mask += left_tz; \
425 vx += left_tz * unit_x; \
426 } \
427 if (width > 0) \
428 { \
429 - scanline_func (dst, mask, \
430 - src1, src2, width, weight1, weight2, vx, unit_x, 0, FALSE); \
431 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
432 + scanline_buffer, mask, src1, src2, width, weight1, weight2, \
433 + vx, unit_x, 0, FALSE); \
434 dst += width; \
435 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
436 mask += width; \
437 vx += width * unit_x; \
438 } \
439 if (right_tz > 0) \
440 { \
441 buf1[0] = src1[src_image->bits.width - 1]; \
442 buf1[1] = 0; \
443 buf2[0] = src2[src_image->bits.width - 1]; \
444 buf2[1] = 0; \
445 - scanline_func (dst, mask, \
446 - buf1, buf2, right_tz, weight1, weight2, \
447 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
448 + scanline_buffer, mask, buf1, buf2, right_tz, weight1, weight2, \
449 pixman_fixed_frac (vx), unit_x, 0, FALSE); \
450 dst += right_tz; \
451 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
452 mask += right_tz; \
453 } \
454 if (right_pad > 0) \
455 { \
456 buf1[0] = buf1[1] = 0; \
457 buf2[0] = buf2[1] = 0; \
458 - scanline_func (dst, mask, \
459 - buf1, buf2, right_pad, weight1, weight2, 0, 0, 0, TRUE); \
460 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
461 + scanline_buffer, mask, buf1, buf2, right_pad, weight1, weight2, \
462 + 0, 0, 0, TRUE); \
463 } \
464 } \
465 else if (PIXMAN_REPEAT_ ## repeat_mode == PIXMAN_REPEAT_NORMAL) \
466 { \
467 int32_t num_pixels; \
468 int32_t width_remain; \
469 src_type_t * src_line_top; \
470 src_type_t * src_line_bottom; \
471 @@ -1120,17 +1167,18 @@ fast_composite_scaled_bilinear ## scale_
472 * vx is in range [0, src_width_fixed - pixman_fixed_e] \
473 * So we are safe from overflow. \
474 */ \
475 num_pixels = ((src_width_fixed - vx - pixman_fixed_e) / unit_x) + 1; \
476 \
477 if (num_pixels > width_remain) \
478 num_pixels = width_remain; \
479 \
480 - scanline_func (dst, mask, buf1, buf2, num_pixels, \
481 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, \
482 + dst, scanline_buffer, mask, buf1, buf2, num_pixels, \
483 weight1, weight2, pixman_fixed_frac(vx), \
484 unit_x, src_width_fixed, FALSE); \
485 \
486 width_remain -= num_pixels; \
487 vx += num_pixels * unit_x; \
488 dst += num_pixels; \
489 \
490 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
491 @@ -1149,41 +1197,47 @@ fast_composite_scaled_bilinear ## scale_
492 * So we are safe from overflow here. \
493 */ \
494 num_pixels = ((src_width_fixed - pixman_fixed_1 - vx - pixman_fixed_e) \
495 / unit_x) + 1; \
496 \
497 if (num_pixels > width_remain) \
498 num_pixels = width_remain; \
499 \
500 - scanline_func (dst, mask, src_line_top, src_line_bottom, num_pixels, \
501 - weight1, weight2, vx, unit_x, src_width_fixed, FALSE); \
502 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, \
503 + dst, scanline_buffer, mask, src_line_top, src_line_bottom, \
504 + num_pixels, weight1, weight2, vx, unit_x, src_width_fixed, \
505 + FALSE); \
506 \
507 width_remain -= num_pixels; \
508 vx += num_pixels * unit_x; \
509 dst += num_pixels; \
510 \
511 if (flags & FLAG_HAVE_NON_SOLID_MASK) \
512 mask += num_pixels; \
513 } \
514 } \
515 } \
516 else \
517 { \
518 - scanline_func (dst, mask, src_first_line + src_stride * y1, \
519 + scanline_func (dst_type_t, mask_type_t, src_type_t, fetch_func, op_func, dst, \
520 + scanline_buffer, mask, \
521 + src_first_line + src_stride * y1, \
522 src_first_line + src_stride * y2, width, \
523 weight1, weight2, vx, unit_x, max_vx, FALSE); \
524 } \
525 } \
526 + if (scanline_buffer != (uint8_t *) stack_scanline_buffer) \
527 + free (scanline_buffer); \
528 }
529
530 /* A workaround for old sun studio, see: https://bugs.freedesktop.org/show_bug.cgi?id=32764 */
531 -#define FAST_BILINEAR_MAINLOOP_COMMON(scale_func_name, scanline_func, src_type_t, mask_type_t, \
532 +#define FAST_BILINEAR_MAINLOOP_COMMON(scale_func_name, fetch_func, op_func, src_type_t, mask_type_t,\
533 dst_type_t, repeat_mode, flags) \
534 - FAST_BILINEAR_MAINLOOP_INT(_ ## scale_func_name, scanline_func, src_type_t, mask_type_t,\
535 + FAST_BILINEAR_MAINLOOP_INT(_ ## scale_func_name, fetch_func, op_func, src_type_t, mask_type_t,\
536 dst_type_t, repeat_mode, flags)
537
538 #define SCALED_BILINEAR_FLAGS \
539 (FAST_PATH_SCALE_TRANSFORM | \
540 FAST_PATH_NO_ALPHA_MAP | \
541 FAST_PATH_BILINEAR_FILTER | \
542 FAST_PATH_NO_ACCESSORS | \
543 FAST_PATH_NARROW_FORMAT)
544 diff --git a/gfx/cairo/libpixman/src/pixman-sse2.c b/gfx/cairo/libpixman/src/pixman-sse2.c
545 --- a/gfx/cairo/libpixman/src/pixman-sse2.c
546 +++ b/gfx/cairo/libpixman/src/pixman-sse2.c
547 @@ -5404,30 +5404,33 @@ scaled_bilinear_scanline_sse2_8888_8888_
548 if (w & 1)
549 {
550 BILINEAR_INTERPOLATE_ONE_PIXEL (pix1);
551 *dst = pix1;
552 }
553
554 }
555
556 +/* Add extra NULL argument to the existing bilinear fast paths to indicate
557 + * that we don't need two-pass processing */
558 +
559 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_cover_SRC,
560 - scaled_bilinear_scanline_sse2_8888_8888_SRC,
561 + scaled_bilinear_scanline_sse2_8888_8888_SRC, NULL,
562 uint32_t, uint32_t, uint32_t,
563 COVER, FLAG_NONE)
564 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_pad_SRC,
565 - scaled_bilinear_scanline_sse2_8888_8888_SRC,
566 + scaled_bilinear_scanline_sse2_8888_8888_SRC, NULL,
567 uint32_t, uint32_t, uint32_t,
568 PAD, FLAG_NONE)
569 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_none_SRC,
570 - scaled_bilinear_scanline_sse2_8888_8888_SRC,
571 + scaled_bilinear_scanline_sse2_8888_8888_SRC, NULL,
572 uint32_t, uint32_t, uint32_t,
573 NONE, FLAG_NONE)
574 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_normal_SRC,
575 - scaled_bilinear_scanline_sse2_8888_8888_SRC,
576 + scaled_bilinear_scanline_sse2_8888_8888_SRC, NULL,
577 uint32_t, uint32_t, uint32_t,
578 NORMAL, FLAG_NONE)
579
580 static force_inline void
581 scaled_bilinear_scanline_sse2_8888_8888_OVER (uint32_t * dst,
582 const uint32_t * mask,
583 const uint32_t * src_top,
584 const uint32_t * src_bottom,
585 @@ -5505,32 +5508,66 @@ scaled_bilinear_scanline_sse2_8888_8888_
586 }
587
588 w--;
589 dst++;
590 }
591 }
592
593 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_cover_OVER,
594 - scaled_bilinear_scanline_sse2_8888_8888_OVER,
595 + scaled_bilinear_scanline_sse2_8888_8888_OVER, NULL,
596 uint32_t, uint32_t, uint32_t,
597 COVER, FLAG_NONE)
598 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_pad_OVER,
599 - scaled_bilinear_scanline_sse2_8888_8888_OVER,
600 + scaled_bilinear_scanline_sse2_8888_8888_OVER, NULL,
601 uint32_t, uint32_t, uint32_t,
602 PAD, FLAG_NONE)
603 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_none_OVER,
604 - scaled_bilinear_scanline_sse2_8888_8888_OVER,
605 + scaled_bilinear_scanline_sse2_8888_8888_OVER, NULL,
606 uint32_t, uint32_t, uint32_t,
607 NONE, FLAG_NONE)
608 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8888_normal_OVER,
609 - scaled_bilinear_scanline_sse2_8888_8888_OVER,
610 + scaled_bilinear_scanline_sse2_8888_8888_OVER, NULL,
611 uint32_t, uint32_t, uint32_t,
612 NORMAL, FLAG_NONE)
613
614 +
615 +/* An example of SSE2 two-stage bilinear_over_8888_0565 fast path, which is implemented
616 + as scaled_bilinear_scanline_sse2_8888_8888_SRC + op_bilinear_over_8888_0565 */
617 +
618 +void op_bilinear_over_8888_0565(uint16_t *dst, const uint32_t *mask, const uint32_t *src, int width)
619 +{
620 + /* Note: this is not really fast and should be based on 8 pixel loop from sse2_composite_over_8888_0565 */
621 + while (--width >= 0)
622 + {
623 + *dst = composite_over_8888_0565pixel (*src, *dst);
624 + src++;
625 + dst++;
626 + }
627 +}
628 +
629 +FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_0565_cover_OVER,
630 + scaled_bilinear_scanline_sse2_8888_8888_SRC, op_bilinear_over_8888_0565,
631 + uint32_t, uint32_t, uint16_t,
632 + COVER, FLAG_NONE)
633 +FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_0565_pad_OVER,
634 + scaled_bilinear_scanline_sse2_8888_8888_SRC, op_bilinear_over_8888_0565,
635 + uint32_t, uint32_t, uint16_t,
636 + PAD, FLAG_NONE)
637 +FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_0565_none_OVER,
638 + scaled_bilinear_scanline_sse2_8888_8888_SRC, op_bilinear_over_8888_0565,
639 + uint32_t, uint32_t, uint16_t,
640 + NONE, FLAG_NONE)
641 +FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_0565_normal_OVER,
642 + scaled_bilinear_scanline_sse2_8888_8888_SRC, op_bilinear_over_8888_0565,
643 + uint32_t, uint32_t, uint16_t,
644 + NORMAL, FLAG_NONE)
645 +
646 +/*****************************/
647 +
648 static force_inline void
649 scaled_bilinear_scanline_sse2_8888_8_8888_OVER (uint32_t * dst,
650 const uint8_t * mask,
651 const uint32_t * src_top,
652 const uint32_t * src_bottom,
653 int32_t w,
654 int wt,
655 int wb,
656 @@ -5669,29 +5706,29 @@ scaled_bilinear_scanline_sse2_8888_8_888
657 }
658
659 w--;
660 dst++;
661 }
662 }
663
664 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8_8888_cover_OVER,
665 - scaled_bilinear_scanline_sse2_8888_8_8888_OVER,
666 + scaled_bilinear_scanline_sse2_8888_8_8888_OVER, NULL,
667 uint32_t, uint8_t, uint32_t,
668 COVER, FLAG_HAVE_NON_SOLID_MASK)
669 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8_8888_pad_OVER,
670 - scaled_bilinear_scanline_sse2_8888_8_8888_OVER,
671 + scaled_bilinear_scanline_sse2_8888_8_8888_OVER, NULL,
672 uint32_t, uint8_t, uint32_t,
673 PAD, FLAG_HAVE_NON_SOLID_MASK)
674 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8_8888_none_OVER,
675 - scaled_bilinear_scanline_sse2_8888_8_8888_OVER,
676 + scaled_bilinear_scanline_sse2_8888_8_8888_OVER, NULL,
677 uint32_t, uint8_t, uint32_t,
678 NONE, FLAG_HAVE_NON_SOLID_MASK)
679 FAST_BILINEAR_MAINLOOP_COMMON (sse2_8888_8_8888_normal_OVER,
680 - scaled_bilinear_scanline_sse2_8888_8_8888_OVER,
681 + scaled_bilinear_scanline_sse2_8888_8_8888_OVER, NULL,
682 uint32_t, uint8_t, uint32_t,
683 NORMAL, FLAG_HAVE_NON_SOLID_MASK)
684
685 static const pixman_fast_path_t sse2_fast_paths[] =
686 {
687 /* PIXMAN_OP_OVER */
688 PIXMAN_STD_FAST_PATH (OVER, solid, a8, r5g6b5, sse2_composite_over_n_8_0565),
689 PIXMAN_STD_FAST_PATH (OVER, solid, a8, b5g6r5, sse2_composite_over_n_8_0565),
690 @@ -5808,16 +5845,21 @@ static const pixman_fast_path_t sse2_fas
691 SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, sse2_8888_8888),
692 SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, sse2_8888_8888),
693
694 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8, sse2_8888_8_8888),
695 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8, sse2_8888_8_8888),
696 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8, sse2_8888_8_8888),
697 SIMPLE_BILINEAR_A8_MASK_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8, sse2_8888_8_8888),
698
699 + /* and here the needed entries are added to the fast path table */
700 +
701 + SIMPLE_BILINEAR_FAST_PATH (OVER, a8r8g8b8, r5g6b5, sse2_8888_0565),
702 + SIMPLE_BILINEAR_FAST_PATH (OVER, a8b8g8r8, b5g6r5, sse2_8888_0565),
703 +
704 { PIXMAN_OP_NONE },
705 };
706
707 static pixman_bool_t
708 sse2_blt (pixman_implementation_t *imp,
709 uint32_t * src_bits,
710 uint32_t * dst_bits,
711 int src_stride,
712