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 // Define USE_NEON_CODE to indicate that we need to build NEON routines 16 #define USE_NEON_CODE (!SK_ARM_NEON_IS_NONE) 17 18 // Define USE_ARM_CODE to indicate that we need to build ARM routines 19 #define USE_ARM_CODE (!SK_ARM_NEON_IS_ALWAYS) 20 21 #if USE_NEON_CODE 22 #include "SkBlitRow_opts_arm_neon.h" 23 #endif 24 25 #if USE_ARM_CODE 26 27 static void S32A_D565_Opaque(uint16_t* SK_RESTRICT dst, 28 const SkPMColor* SK_RESTRICT src, int count, 29 U8CPU alpha, int /*x*/, int /*y*/) { 30 SkASSERT(255 == alpha); 31 32 asm volatile ( 33 "1: \n\t" 34 "ldr r3, [%[src]], #4 \n\t" 35 "cmp r3, #0xff000000 \n\t" 36 "blo 2f \n\t" 37 "and r4, r3, #0x0000f8 \n\t" 38 "and r5, r3, #0x00fc00 \n\t" 39 "and r6, r3, #0xf80000 \n\t" 40 #ifdef SK_ARM_HAS_EDSP 41 "pld [r1, #32] \n\t" 42 #endif 43 "lsl r3, r4, #8 \n\t" 44 "orr r3, r3, r5, lsr #5 \n\t" 45 "orr r3, r3, r6, lsr #19 \n\t" 46 "subs %[count], %[count], #1 \n\t" 47 "strh r3, [%[dst]], #2 \n\t" 48 "bne 1b \n\t" 49 "b 4f \n\t" 50 "2: \n\t" 51 "lsrs r7, r3, #24 \n\t" 52 "beq 3f \n\t" 53 "ldrh r4, [%[dst]] \n\t" 54 "rsb r7, r7, #255 \n\t" 55 "and r6, r4, #0x001f \n\t" 56 #if SK_ARM_ARCH <= 6 57 "lsl r5, r4, #21 \n\t" 58 "lsr r5, r5, #26 \n\t" 59 #else 60 "ubfx r5, r4, #5, #6 \n\t" 61 #endif 62 #ifdef SK_ARM_HAS_EDSP 63 "pld [r0, #16] \n\t" 64 #endif 65 "lsr r4, r4, #11 \n\t" 66 #ifdef SK_ARM_HAS_EDSP 67 "smulbb r6, r6, r7 \n\t" 68 "smulbb r5, r5, r7 \n\t" 69 "smulbb r4, r4, r7 \n\t" 70 #else 71 "mul r6, r6, r7 \n\t" 72 "mul r5, r5, r7 \n\t" 73 "mul r4, r4, r7 \n\t" 74 #endif 75 #if SK_ARM_ARCH >= 6 76 "uxtb r7, r3, ROR #16 \n\t" 77 "uxtb ip, r3, ROR #8 \n\t" 78 #else 79 "mov ip, #0xff \n\t" 80 "and r7, ip, r3, ROR #16 \n\t" 81 "and ip, ip, r3, ROR #8 \n\t" 82 #endif 83 "and r3, r3, #0xff \n\t" 84 "add r6, r6, #16 \n\t" 85 "add r5, r5, #32 \n\t" 86 "add r4, r4, #16 \n\t" 87 "add r6, r6, r6, lsr #5 \n\t" 88 "add r5, r5, r5, lsr #6 \n\t" 89 "add r4, r4, r4, lsr #5 \n\t" 90 "add r6, r7, r6, lsr #5 \n\t" 91 "add r5, ip, r5, lsr #6 \n\t" 92 "add r4, r3, r4, lsr #5 \n\t" 93 "lsr r6, r6, #3 \n\t" 94 "and r5, r5, #0xfc \n\t" 95 "and r4, r4, #0xf8 \n\t" 96 "orr r6, r6, r5, lsl #3 \n\t" 97 "orr r4, r6, r4, lsl #8 \n\t" 98 "strh r4, [%[dst]], #2 \n\t" 99 #ifdef SK_ARM_HAS_EDSP 100 "pld [r1, #32] \n\t" 101 #endif 102 "subs %[count], %[count], #1 \n\t" 103 "bne 1b \n\t" 104 "b 4f \n\t" 105 "3: \n\t" 106 "subs %[count], %[count], #1 \n\t" 107 "add %[dst], %[dst], #2 \n\t" 108 "bne 1b \n\t" 109 "4: \n\t" 110 : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) 111 : 112 : "memory", "cc", "r3", "r4", "r5", "r6", "r7", "ip" 113 ); 114 } 115 116 static void S32A_Opaque_BlitRow32_arm(SkPMColor* SK_RESTRICT dst, 117 const SkPMColor* SK_RESTRICT src, 118 int count, U8CPU alpha) { 119 120 SkASSERT(255 == alpha); 121 122 asm volatile ( 123 "cmp %[count], #0 \n\t" /* comparing count with 0 */ 124 "beq 3f \n\t" /* if zero exit */ 125 126 "mov ip, #0xff \n\t" /* load the 0xff mask in ip */ 127 "orr ip, ip, ip, lsl #16 \n\t" /* convert it to 0xff00ff in ip */ 128 129 "cmp %[count], #2 \n\t" /* compare count with 2 */ 130 "blt 2f \n\t" /* if less than 2 -> single loop */ 131 132 /* Double Loop */ 133 "1: \n\t" /* <double loop> */ 134 "ldm %[src]!, {r5,r6} \n\t" /* load the src(s) at r5-r6 */ 135 "ldm %[dst], {r7,r8} \n\t" /* loading dst(s) into r7-r8 */ 136 "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */ 137 138 /* ----------- */ 139 "and r9, ip, r7 \n\t" /* r9 = br masked by ip */ 140 "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */ 141 "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */ 142 143 "mul r9, r9, r4 \n\t" /* br = br * scale */ 144 "mul r10, r10, r4 \n\t" /* ag = ag * scale */ 145 "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */ 146 147 "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */ 148 "lsr r4, r6, #24 \n\t" /* extracting the alpha from source and storing it to r4 */ 149 "orr r7, r9, r10 \n\t" /* br | ag*/ 150 151 "add r7, r5, r7 \n\t" /* dst = src + calc dest(r7) */ 152 "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 255 -> r4=scale */ 153 154 /* ----------- */ 155 "and r9, ip, r8 \n\t" /* r9 = br masked by ip */ 156 157 "and r10, ip, r8, lsr #8 \n\t" /* r10 = ag masked by ip */ 158 "mul r9, r9, r4 \n\t" /* br = br * scale */ 159 "sub %[count], %[count], #2 \n\t" 160 "mul r10, r10, r4 \n\t" /* ag = ag * scale */ 161 162 "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */ 163 "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */ 164 "cmp %[count], #1 \n\t" /* comparing count with 1 */ 165 "orr r8, r9, r10 \n\t" /* br | ag */ 166 167 "add r8, r6, r8 \n\t" /* dst = src + calc dest(r8) */ 168 169 /* ----------------- */ 170 "stm %[dst]!, {r7,r8} \n\t" /* *dst = r7, increment dst by two (each times 4) */ 171 /* ----------------- */ 172 173 "bgt 1b \n\t" /* if greater than 1 -> reloop */ 174 "blt 3f \n\t" /* if less than 1 -> exit */ 175 176 /* Single Loop */ 177 "2: \n\t" /* <single loop> */ 178 "ldr r5, [%[src]], #4 \n\t" /* load the src pointer into r5 r5=src */ 179 "ldr r7, [%[dst]] \n\t" /* loading dst into r7 */ 180 "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */ 181 182 /* ----------- */ 183 "and r9, ip, r7 \n\t" /* r9 = br masked by ip */ 184 "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */ 185 186 "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */ 187 "mul r9, r9, r4 \n\t" /* br = br * scale */ 188 "mul r10, r10, r4 \n\t" /* ag = ag * scale */ 189 "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */ 190 191 "and r10, r10, ip, lsl #8 \n\t" /* mask ag */ 192 "orr r7, r9, r10 \n\t" /* br | ag */ 193 194 "add r7, r5, r7 \n\t" /* *dst = src + calc dest(r7) */ 195 196 /* ----------------- */ 197 "str r7, [%[dst]], #4 \n\t" /* *dst = r7, increment dst by one (times 4) */ 198 /* ----------------- */ 199 200 "3: \n\t" /* <exit> */ 201 : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count) 202 : 203 : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "ip", "memory" 204 ); 205 } 206 207 /* 208 * ARM asm version of S32A_Blend_BlitRow32 209 */ 210 void S32A_Blend_BlitRow32_arm(SkPMColor* SK_RESTRICT dst, 211 const SkPMColor* SK_RESTRICT src, 212 int count, U8CPU alpha) { 213 asm volatile ( 214 "cmp %[count], #0 \n\t" /* comparing count with 0 */ 215 "beq 3f \n\t" /* if zero exit */ 216 217 "mov r12, #0xff \n\t" /* load the 0xff mask in r12 */ 218 "orr r12, r12, r12, lsl #16 \n\t" /* convert it to 0xff00ff in r12 */ 219 220 /* src1,2_scale */ 221 "add %[alpha], %[alpha], #1 \n\t" /* loading %[alpha]=src_scale=alpha+1 */ 222 223 "cmp %[count], #2 \n\t" /* comparing count with 2 */ 224 "blt 2f \n\t" /* if less than 2 -> single loop */ 225 226 /* Double Loop */ 227 "1: \n\t" /* <double loop> */ 228 "ldm %[src]!, {r5, r6} \n\t" /* loading src pointers into r5 and r6 */ 229 "ldm %[dst], {r7, r8} \n\t" /* loading dst pointers into r7 and r8 */ 230 231 /* dst1_scale and dst2_scale*/ 232 "lsr r9, r5, #24 \n\t" /* src >> 24 */ 233 "lsr r10, r6, #24 \n\t" /* src >> 24 */ 234 #ifdef SK_ARM_HAS_EDSP 235 "smulbb r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */ 236 "smulbb r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */ 237 #else 238 "mul r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */ 239 "mul r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */ 240 #endif 241 "lsr r9, r9, #8 \n\t" /* r9 >> 8 */ 242 "lsr r10, r10, #8 \n\t" /* r10 >> 8 */ 243 "rsb r9, r9, #256 \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */ 244 "rsb r10, r10, #256 \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */ 245 246 /* ---------------------- */ 247 248 /* src1, src1_scale */ 249 "and r11, r12, r5, lsr #8 \n\t" /* ag = r11 = r5 masked by r12 lsr by #8 */ 250 "and r4, r12, r5 \n\t" /* rb = r4 = r5 masked by r12 */ 251 "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */ 252 "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */ 253 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ 254 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ 255 "orr r5, r11, r4 \n\t" /* r5 = (src1, src_scale) */ 256 257 /* dst1, dst1_scale */ 258 "and r11, r12, r7, lsr #8 \n\t" /* ag = r11 = r7 masked by r12 lsr by #8 */ 259 "and r4, r12, r7 \n\t" /* rb = r4 = r7 masked by r12 */ 260 "mul r11, r11, r9 \n\t" /* ag = r11 times dst_scale (r9) */ 261 "mul r4, r4, r9 \n\t" /* rb = r4 times dst_scale (r9) */ 262 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ 263 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ 264 "orr r9, r11, r4 \n\t" /* r9 = (dst1, dst_scale) */ 265 266 /* ---------------------- */ 267 "add r9, r5, r9 \n\t" /* *dst = src plus dst both scaled */ 268 /* ---------------------- */ 269 270 /* ====================== */ 271 272 /* src2, src2_scale */ 273 "and r11, r12, r6, lsr #8 \n\t" /* ag = r11 = r6 masked by r12 lsr by #8 */ 274 "and r4, r12, r6 \n\t" /* rb = r4 = r6 masked by r12 */ 275 "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */ 276 "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */ 277 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ 278 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ 279 "orr r6, r11, r4 \n\t" /* r6 = (src2, src_scale) */ 280 281 /* dst2, dst2_scale */ 282 "and r11, r12, r8, lsr #8 \n\t" /* ag = r11 = r8 masked by r12 lsr by #8 */ 283 "and r4, r12, r8 \n\t" /* rb = r4 = r8 masked by r12 */ 284 "mul r11, r11, r10 \n\t" /* ag = r11 times dst_scale (r10) */ 285 "mul r4, r4, r10 \n\t" /* rb = r4 times dst_scale (r6) */ 286 "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ 287 "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */ 288 "orr r10, r11, r4 \n\t" /* r10 = (dst2, dst_scale) */ 289 290 "sub %[count], %[count], #2 \n\t" /* decrease count by 2 */ 291 /* ---------------------- */ 292 "add r10, r6, r10 \n\t" /* *dst = src plus dst both scaled */ 293 /* ---------------------- */ 294 "cmp %[count], #1 \n\t" /* compare count with 1 */ 295 /* ----------------- */ 296 "stm %[dst]!, {r9, r10} \n\t" /* copy r9 and r10 to r7 and r8 respectively */ 297 /* ----------------- */ 298 299 "bgt 1b \n\t" /* if %[count] greater than 1 reloop */ 300 "blt 3f \n\t" /* if %[count] less than 1 exit */ 301 /* else get into the single loop */ 302 /* Single Loop */ 303 "2: \n\t" /* <single loop> */ 304 "ldr r5, [%[src]], #4 \n\t" /* loading src pointer into r5: r5=src */ 305 "ldr r7, [%[dst]] \n\t" /* loading dst pointer into r7: r7=dst */ 306 307 "lsr r6, r5, #24 \n\t" /* src >> 24 */ 308 "and r8, r12, r5, lsr #8 \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */ 309 #ifdef SK_ARM_HAS_EDSP 310 "smulbb r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */ 311 #else 312 "mul r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */ 313 #endif 314 "and r9, r12, r5 \n\t" /* rb = r9 = r5 masked by r12 */ 315 "lsr r6, r6, #8 \n\t" /* r6 >> 8 */ 316 "mul r8, r8, %[alpha] \n\t" /* ag = r8 times scale */ 317 "rsb r6, r6, #256 \n\t" /* r6 = 255 - r6 + 1 */ 318 319 /* src, src_scale */ 320 "mul r9, r9, %[alpha] \n\t" /* rb = r9 times scale */ 321 "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ 322 "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */ 323 "orr r10, r8, r9 \n\t" /* r10 = (scr, src_scale) */ 324 325 /* dst, dst_scale */ 326 "and r8, r12, r7, lsr #8 \n\t" /* ag = r8 = r7 masked by r12 lsr by #8 */ 327 "and r9, r12, r7 \n\t" /* rb = r9 = r7 masked by r12 */ 328 "mul r8, r8, r6 \n\t" /* ag = r8 times scale (r6) */ 329 "mul r9, r9, r6 \n\t" /* rb = r9 times scale (r6) */ 330 "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */ 331 "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */ 332 "orr r7, r8, r9 \n\t" /* r7 = (dst, dst_scale) */ 333 334 "add r10, r7, r10 \n\t" /* *dst = src plus dst both scaled */ 335 336 /* ----------------- */ 337 "str r10, [%[dst]], #4 \n\t" /* *dst = r10, postincrement dst by one (times 4) */ 338 /* ----------------- */ 339 340 "3: \n\t" /* <exit> */ 341 : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count), [alpha] "+r" (alpha) 342 : 343 : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12", "memory" 344 ); 345 346 } 347 348 /////////////////////////////////////////////////////////////////////////////// 349 350 static const SkBlitRow::Proc16 sk_blitrow_platform_565_procs_arm[] = { 351 // no dither 352 // NOTE: For the functions below, we don't have a special version 353 // that assumes that each source pixel is opaque. But our S32A is 354 // still faster than the default, so use it. 355 S32A_D565_Opaque, // S32_D565_Opaque 356 NULL, // S32_D565_Blend 357 S32A_D565_Opaque, // S32A_D565_Opaque 358 NULL, // S32A_D565_Blend 359 360 // dither 361 NULL, // S32_D565_Opaque_Dither 362 NULL, // S32_D565_Blend_Dither 363 NULL, // S32A_D565_Opaque_Dither 364 NULL, // S32A_D565_Blend_Dither 365 }; 366 367 static const SkBlitRow::ColorProc16 sk_blitrow_platform_565_colorprocs_arm[] = { 368 NULL, // Color32A_D565, 369 NULL, // Color32A_D565_Dither 370 }; 371 372 static const SkBlitRow::Proc32 sk_blitrow_platform_32_procs_arm[] = { 373 NULL, // S32_Opaque, 374 NULL, // S32_Blend, 375 S32A_Opaque_BlitRow32_arm, // S32A_Opaque, 376 S32A_Blend_BlitRow32_arm // S32A_Blend 377 }; 378 379 #endif // USE_ARM_CODE 380 381 SkBlitRow::Proc16 SkBlitRow::PlatformFactory565(unsigned flags) { 382 return SK_ARM_NEON_WRAP(sk_blitrow_platform_565_procs_arm)[flags]; 383 } 384 385 SkBlitRow::ColorProc16 SkBlitRow::PlatformColorFactory565(unsigned flags) { 386 return SK_ARM_NEON_WRAP(sk_blitrow_platform_565_colorprocs_arm)[flags]; 387 } 388 389 SkBlitRow::Proc32 SkBlitRow::PlatformProcs32(unsigned flags) { 390 return SK_ARM_NEON_WRAP(sk_blitrow_platform_32_procs_arm)[flags]; 391 } 392 393