The Tor Browser / file comparison

comparison: gfx/skia/trunk/src/opts/SkBlitRow_opts_arm.cpp

gfx/skia/trunk/src/opts/SkBlitRow_opts_arm.cpp

changeset 0
6474c204b198
equal deleted inserted replaced
-1:000000000000 0:7d86a1e1161d
1 /*
2 * Copyright 2012 The Android Open Source Project
3 *
4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file.
6 */
7
8 #include "SkBlitRow.h"
9 #include "SkColorPriv.h"
10 #include "SkDither.h"
11 #include "SkMathPriv.h"
12 #include "SkUtils.h"
13 #include "SkUtilsArm.h"
14
15 #include "SkCachePreload_arm.h"
16
17 // Define USE_NEON_CODE to indicate that we need to build NEON routines
18 #define USE_NEON_CODE (!SK_ARM_NEON_IS_NONE)
19
20 // Define USE_ARM_CODE to indicate that we need to build ARM routines
21 #define USE_ARM_CODE (!SK_ARM_NEON_IS_ALWAYS)
22
23 #if USE_NEON_CODE
24 #include "SkBlitRow_opts_arm_neon.h"
25 #endif
26
27 #if USE_ARM_CODE
28
29 static void S32A_D565_Opaque(uint16_t* SK_RESTRICT dst,
30 const SkPMColor* SK_RESTRICT src, int count,
31 U8CPU alpha, int /*x*/, int /*y*/) {
32 SkASSERT(255 == alpha);
33
34 asm volatile (
35 "1: \n\t"
36 "ldr r3, [%[src]], #4 \n\t"
37 "cmp r3, #0xff000000 \n\t"
38 "blo 2f \n\t"
39 "and r4, r3, #0x0000f8 \n\t"
40 "and r5, r3, #0x00fc00 \n\t"
41 "and r6, r3, #0xf80000 \n\t"
42 #ifdef SK_ARM_HAS_EDSP
43 "pld [r1, #32] \n\t"
44 #endif
45 "lsl r3, r4, #8 \n\t"
46 "orr r3, r3, r5, lsr #5 \n\t"
47 "orr r3, r3, r6, lsr #19 \n\t"
48 "subs %[count], %[count], #1 \n\t"
49 "strh r3, [%[dst]], #2 \n\t"
50 "bne 1b \n\t"
51 "b 4f \n\t"
52 "2: \n\t"
53 "lsrs r7, r3, #24 \n\t"
54 "beq 3f \n\t"
55 "ldrh r4, [%[dst]] \n\t"
56 "rsb r7, r7, #255 \n\t"
57 "and r6, r4, #0x001f \n\t"
58 #if SK_ARM_ARCH <= 6
59 "lsl r5, r4, #21 \n\t"
60 "lsr r5, r5, #26 \n\t"
61 #else
62 "ubfx r5, r4, #5, #6 \n\t"
63 #endif
64 #ifdef SK_ARM_HAS_EDSP
65 "pld [r0, #16] \n\t"
66 #endif
67 "lsr r4, r4, #11 \n\t"
68 #ifdef SK_ARM_HAS_EDSP
69 "smulbb r6, r6, r7 \n\t"
70 "smulbb r5, r5, r7 \n\t"
71 "smulbb r4, r4, r7 \n\t"
72 #else
73 "mul r6, r6, r7 \n\t"
74 "mul r5, r5, r7 \n\t"
75 "mul r4, r4, r7 \n\t"
76 #endif
77 #if SK_ARM_ARCH >= 6
78 "uxtb r7, r3, ROR #16 \n\t"
79 "uxtb ip, r3, ROR #8 \n\t"
80 #else
81 "mov ip, #0xff \n\t"
82 "and r7, ip, r3, ROR #16 \n\t"
83 "and ip, ip, r3, ROR #8 \n\t"
84 #endif
85 "and r3, r3, #0xff \n\t"
86 "add r6, r6, #16 \n\t"
87 "add r5, r5, #32 \n\t"
88 "add r4, r4, #16 \n\t"
89 "add r6, r6, r6, lsr #5 \n\t"
90 "add r5, r5, r5, lsr #6 \n\t"
91 "add r4, r4, r4, lsr #5 \n\t"
92 "add r6, r7, r6, lsr #5 \n\t"
93 "add r5, ip, r5, lsr #6 \n\t"
94 "add r4, r3, r4, lsr #5 \n\t"
95 "lsr r6, r6, #3 \n\t"
96 "and r5, r5, #0xfc \n\t"
97 "and r4, r4, #0xf8 \n\t"
98 "orr r6, r6, r5, lsl #3 \n\t"
99 "orr r4, r6, r4, lsl #8 \n\t"
100 "strh r4, [%[dst]], #2 \n\t"
101 #ifdef SK_ARM_HAS_EDSP
102 "pld [r1, #32] \n\t"
103 #endif
104 "subs %[count], %[count], #1 \n\t"
105 "bne 1b \n\t"
106 "b 4f \n\t"
107 "3: \n\t"
108 "subs %[count], %[count], #1 \n\t"
109 "add %[dst], %[dst], #2 \n\t"
110 "bne 1b \n\t"
111 "4: \n\t"
112 : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count)
113 :
114 : "memory", "cc", "r3", "r4", "r5", "r6", "r7", "ip"
115 );
116 }
117
118 static void S32A_Opaque_BlitRow32_arm(SkPMColor* SK_RESTRICT dst,
119 const SkPMColor* SK_RESTRICT src,
120 int count, U8CPU alpha) {
121
122 SkASSERT(255 == alpha);
123
124 asm volatile (
125 "cmp %[count], #0 \n\t" /* comparing count with 0 */
126 "beq 3f \n\t" /* if zero exit */
127
128 "mov ip, #0xff \n\t" /* load the 0xff mask in ip */
129 "orr ip, ip, ip, lsl #16 \n\t" /* convert it to 0xff00ff in ip */
130
131 "cmp %[count], #2 \n\t" /* compare count with 2 */
132 "blt 2f \n\t" /* if less than 2 -> single loop */
133
134 /* Double Loop */
135 "1: \n\t" /* <double loop> */
136 "ldm %[src]!, {r5,r6} \n\t" /* load the src(s) at r5-r6 */
137 "ldm %[dst], {r7,r8} \n\t" /* loading dst(s) into r7-r8 */
138 "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */
139
140 /* ----------- */
141 "and r9, ip, r7 \n\t" /* r9 = br masked by ip */
142 "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */
143 "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */
144
145 "mul r9, r9, r4 \n\t" /* br = br * scale */
146 "mul r10, r10, r4 \n\t" /* ag = ag * scale */
147 "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */
148
149 "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */
150 "lsr r4, r6, #24 \n\t" /* extracting the alpha from source and storing it to r4 */
151 "orr r7, r9, r10 \n\t" /* br | ag*/
152
153 "add r7, r5, r7 \n\t" /* dst = src + calc dest(r7) */
154 "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 255 -> r4=scale */
155
156 /* ----------- */
157 "and r9, ip, r8 \n\t" /* r9 = br masked by ip */
158
159 "and r10, ip, r8, lsr #8 \n\t" /* r10 = ag masked by ip */
160 "mul r9, r9, r4 \n\t" /* br = br * scale */
161 "sub %[count], %[count], #2 \n\t"
162 "mul r10, r10, r4 \n\t" /* ag = ag * scale */
163
164 "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */
165 "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */
166 "cmp %[count], #1 \n\t" /* comparing count with 1 */
167 "orr r8, r9, r10 \n\t" /* br | ag */
168
169 "add r8, r6, r8 \n\t" /* dst = src + calc dest(r8) */
170
171 /* ----------------- */
172 "stm %[dst]!, {r7,r8} \n\t" /* *dst = r7, increment dst by two (each times 4) */
173 /* ----------------- */
174
175 "bgt 1b \n\t" /* if greater than 1 -> reloop */
176 "blt 3f \n\t" /* if less than 1 -> exit */
177
178 /* Single Loop */
179 "2: \n\t" /* <single loop> */
180 "ldr r5, [%[src]], #4 \n\t" /* load the src pointer into r5 r5=src */
181 "ldr r7, [%[dst]] \n\t" /* loading dst into r7 */
182 "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */
183
184 /* ----------- */
185 "and r9, ip, r7 \n\t" /* r9 = br masked by ip */
186 "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */
187
188 "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */
189 "mul r9, r9, r4 \n\t" /* br = br * scale */
190 "mul r10, r10, r4 \n\t" /* ag = ag * scale */
191 "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */
192
193 "and r10, r10, ip, lsl #8 \n\t" /* mask ag */
194 "orr r7, r9, r10 \n\t" /* br | ag */
195
196 "add r7, r5, r7 \n\t" /* *dst = src + calc dest(r7) */
197
198 /* ----------------- */
199 "str r7, [%[dst]], #4 \n\t" /* *dst = r7, increment dst by one (times 4) */
200 /* ----------------- */
201
202 "3: \n\t" /* <exit> */
203 : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count)
204 :
205 : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "ip", "memory"
206 );
207 }
208
209 /*
210 * ARM asm version of S32A_Blend_BlitRow32
211 */
212 void S32A_Blend_BlitRow32_arm(SkPMColor* SK_RESTRICT dst,
213 const SkPMColor* SK_RESTRICT src,
214 int count, U8CPU alpha) {
215 asm volatile (
216 "cmp %[count], #0 \n\t" /* comparing count with 0 */
217 "beq 3f \n\t" /* if zero exit */
218
219 "mov r12, #0xff \n\t" /* load the 0xff mask in r12 */
220 "orr r12, r12, r12, lsl #16 \n\t" /* convert it to 0xff00ff in r12 */
221
222 /* src1,2_scale */
223 "add %[alpha], %[alpha], #1 \n\t" /* loading %[alpha]=src_scale=alpha+1 */
224
225 "cmp %[count], #2 \n\t" /* comparing count with 2 */
226 "blt 2f \n\t" /* if less than 2 -> single loop */
227
228 /* Double Loop */
229 "1: \n\t" /* <double loop> */
230 "ldm %[src]!, {r5, r6} \n\t" /* loading src pointers into r5 and r6 */
231 "ldm %[dst], {r7, r8} \n\t" /* loading dst pointers into r7 and r8 */
232
233 /* dst1_scale and dst2_scale*/
234 "lsr r9, r5, #24 \n\t" /* src >> 24 */
235 "lsr r10, r6, #24 \n\t" /* src >> 24 */
236 #ifdef SK_ARM_HAS_EDSP
237 "smulbb r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
238 "smulbb r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
239 #else
240 "mul r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
241 "mul r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
242 #endif
243 "lsr r9, r9, #8 \n\t" /* r9 >> 8 */
244 "lsr r10, r10, #8 \n\t" /* r10 >> 8 */
245 "rsb r9, r9, #256 \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */
246 "rsb r10, r10, #256 \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */
247
248 /* ---------------------- */
249
250 /* src1, src1_scale */
251 "and r11, r12, r5, lsr #8 \n\t" /* ag = r11 = r5 masked by r12 lsr by #8 */
252 "and r4, r12, r5 \n\t" /* rb = r4 = r5 masked by r12 */
253 "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */
254 "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */
255 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
256 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
257 "orr r5, r11, r4 \n\t" /* r5 = (src1, src_scale) */
258
259 /* dst1, dst1_scale */
260 "and r11, r12, r7, lsr #8 \n\t" /* ag = r11 = r7 masked by r12 lsr by #8 */
261 "and r4, r12, r7 \n\t" /* rb = r4 = r7 masked by r12 */
262 "mul r11, r11, r9 \n\t" /* ag = r11 times dst_scale (r9) */
263 "mul r4, r4, r9 \n\t" /* rb = r4 times dst_scale (r9) */
264 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
265 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
266 "orr r9, r11, r4 \n\t" /* r9 = (dst1, dst_scale) */
267
268 /* ---------------------- */
269 "add r9, r5, r9 \n\t" /* *dst = src plus dst both scaled */
270 /* ---------------------- */
271
272 /* ====================== */
273
274 /* src2, src2_scale */
275 "and r11, r12, r6, lsr #8 \n\t" /* ag = r11 = r6 masked by r12 lsr by #8 */
276 "and r4, r12, r6 \n\t" /* rb = r4 = r6 masked by r12 */
277 "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */
278 "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */
279 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
280 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
281 "orr r6, r11, r4 \n\t" /* r6 = (src2, src_scale) */
282
283 /* dst2, dst2_scale */
284 "and r11, r12, r8, lsr #8 \n\t" /* ag = r11 = r8 masked by r12 lsr by #8 */
285 "and r4, r12, r8 \n\t" /* rb = r4 = r8 masked by r12 */
286 "mul r11, r11, r10 \n\t" /* ag = r11 times dst_scale (r10) */
287 "mul r4, r4, r10 \n\t" /* rb = r4 times dst_scale (r6) */
288 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
289 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
290 "orr r10, r11, r4 \n\t" /* r10 = (dst2, dst_scale) */
291
292 "sub %[count], %[count], #2 \n\t" /* decrease count by 2 */
293 /* ---------------------- */
294 "add r10, r6, r10 \n\t" /* *dst = src plus dst both scaled */
295 /* ---------------------- */
296 "cmp %[count], #1 \n\t" /* compare count with 1 */
297 /* ----------------- */
298 "stm %[dst]!, {r9, r10} \n\t" /* copy r9 and r10 to r7 and r8 respectively */
299 /* ----------------- */
300
301 "bgt 1b \n\t" /* if %[count] greater than 1 reloop */
302 "blt 3f \n\t" /* if %[count] less than 1 exit */
303 /* else get into the single loop */
304 /* Single Loop */
305 "2: \n\t" /* <single loop> */
306 "ldr r5, [%[src]], #4 \n\t" /* loading src pointer into r5: r5=src */
307 "ldr r7, [%[dst]] \n\t" /* loading dst pointer into r7: r7=dst */
308
309 "lsr r6, r5, #24 \n\t" /* src >> 24 */
310 "and r8, r12, r5, lsr #8 \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */
311 #ifdef SK_ARM_HAS_EDSP
312 "smulbb r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
313 #else
314 "mul r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
315 #endif
316 "and r9, r12, r5 \n\t" /* rb = r9 = r5 masked by r12 */
317 "lsr r6, r6, #8 \n\t" /* r6 >> 8 */
318 "mul r8, r8, %[alpha] \n\t" /* ag = r8 times scale */
319 "rsb r6, r6, #256 \n\t" /* r6 = 255 - r6 + 1 */
320
321 /* src, src_scale */
322 "mul r9, r9, %[alpha] \n\t" /* rb = r9 times scale */
323 "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
324 "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */
325 "orr r10, r8, r9 \n\t" /* r10 = (scr, src_scale) */
326
327 /* dst, dst_scale */
328 "and r8, r12, r7, lsr #8 \n\t" /* ag = r8 = r7 masked by r12 lsr by #8 */
329 "and r9, r12, r7 \n\t" /* rb = r9 = r7 masked by r12 */
330 "mul r8, r8, r6 \n\t" /* ag = r8 times scale (r6) */
331 "mul r9, r9, r6 \n\t" /* rb = r9 times scale (r6) */
332 "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
333 "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */
334 "orr r7, r8, r9 \n\t" /* r7 = (dst, dst_scale) */
335
336 "add r10, r7, r10 \n\t" /* *dst = src plus dst both scaled */
337
338 /* ----------------- */
339 "str r10, [%[dst]], #4 \n\t" /* *dst = r10, postincrement dst by one (times 4) */
340 /* ----------------- */
341
342 "3: \n\t" /* <exit> */
343 : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count), [alpha] "+r" (alpha)
344 :
345 : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12", "memory"
346 );
347
348 }
349
350 ///////////////////////////////////////////////////////////////////////////////
351
352 static const SkBlitRow::Proc sk_blitrow_platform_565_procs_arm[] = {
353 // no dither
354 // NOTE: For the functions below, we don't have a special version
355 // that assumes that each source pixel is opaque. But our S32A is
356 // still faster than the default, so use it.
357 S32A_D565_Opaque, // S32_D565_Opaque
358 NULL, // S32_D565_Blend
359 S32A_D565_Opaque, // S32A_D565_Opaque
360 NULL, // S32A_D565_Blend
361
362 // dither
363 NULL, // S32_D565_Opaque_Dither
364 NULL, // S32_D565_Blend_Dither
365 NULL, // S32A_D565_Opaque_Dither
366 NULL, // S32A_D565_Blend_Dither
367 };
368
369 static const SkBlitRow::Proc32 sk_blitrow_platform_32_procs_arm[] = {
370 NULL, // S32_Opaque,
371 NULL, // S32_Blend,
372 S32A_Opaque_BlitRow32_arm, // S32A_Opaque,
373 S32A_Blend_BlitRow32_arm // S32A_Blend
374 };
375
376 #endif // USE_ARM_CODE
377
378 SkBlitRow::Proc SkBlitRow::PlatformProcs565(unsigned flags) {
379 return SK_ARM_NEON_WRAP(sk_blitrow_platform_565_procs_arm)[flags];
380 }
381
382 SkBlitRow::Proc32 SkBlitRow::PlatformProcs32(unsigned flags) {
383 return SK_ARM_NEON_WRAP(sk_blitrow_platform_32_procs_arm)[flags];
384 }
385
386 ///////////////////////////////////////////////////////////////////////////////
387 #define Color32_arm NULL
388 SkBlitRow::ColorProc SkBlitRow::PlatformColorProc() {
389 return SK_ARM_NEON_WRAP(Color32_arm);
390 }
391
392 SkBlitRow::ColorRectProc PlatformColorRectProcFactory() {
393 return NULL;
394 }

Mercurial Repository: The Tor Browser

Europalab SCM Repository Server

mercurial