1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/skia/trunk/src/opts/SkBlitRow_opts_arm.cpp Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,394 @@
1.4 +/*
1.5 + * Copyright 2012 The Android Open Source Project
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license that can be
1.8 + * found in the LICENSE file.
1.9 + */
1.10 +
1.11 +#include "SkBlitRow.h"
1.12 +#include "SkColorPriv.h"
1.13 +#include "SkDither.h"
1.14 +#include "SkMathPriv.h"
1.15 +#include "SkUtils.h"
1.16 +#include "SkUtilsArm.h"
1.17 +
1.18 +#include "SkCachePreload_arm.h"
1.19 +
1.20 +// Define USE_NEON_CODE to indicate that we need to build NEON routines
1.21 +#define USE_NEON_CODE (!SK_ARM_NEON_IS_NONE)
1.22 +
1.23 +// Define USE_ARM_CODE to indicate that we need to build ARM routines
1.24 +#define USE_ARM_CODE (!SK_ARM_NEON_IS_ALWAYS)
1.25 +
1.26 +#if USE_NEON_CODE
1.27 + #include "SkBlitRow_opts_arm_neon.h"
1.28 +#endif
1.29 +
1.30 +#if USE_ARM_CODE
1.31 +
1.32 +static void S32A_D565_Opaque(uint16_t* SK_RESTRICT dst,
1.33 + const SkPMColor* SK_RESTRICT src, int count,
1.34 + U8CPU alpha, int /*x*/, int /*y*/) {
1.35 + SkASSERT(255 == alpha);
1.36 +
1.37 + asm volatile (
1.38 + "1: \n\t"
1.39 + "ldr r3, [%[src]], #4 \n\t"
1.40 + "cmp r3, #0xff000000 \n\t"
1.41 + "blo 2f \n\t"
1.42 + "and r4, r3, #0x0000f8 \n\t"
1.43 + "and r5, r3, #0x00fc00 \n\t"
1.44 + "and r6, r3, #0xf80000 \n\t"
1.45 +#ifdef SK_ARM_HAS_EDSP
1.46 + "pld [r1, #32] \n\t"
1.47 +#endif
1.48 + "lsl r3, r4, #8 \n\t"
1.49 + "orr r3, r3, r5, lsr #5 \n\t"
1.50 + "orr r3, r3, r6, lsr #19 \n\t"
1.51 + "subs %[count], %[count], #1 \n\t"
1.52 + "strh r3, [%[dst]], #2 \n\t"
1.53 + "bne 1b \n\t"
1.54 + "b 4f \n\t"
1.55 + "2: \n\t"
1.56 + "lsrs r7, r3, #24 \n\t"
1.57 + "beq 3f \n\t"
1.58 + "ldrh r4, [%[dst]] \n\t"
1.59 + "rsb r7, r7, #255 \n\t"
1.60 + "and r6, r4, #0x001f \n\t"
1.61 +#if SK_ARM_ARCH <= 6
1.62 + "lsl r5, r4, #21 \n\t"
1.63 + "lsr r5, r5, #26 \n\t"
1.64 +#else
1.65 + "ubfx r5, r4, #5, #6 \n\t"
1.66 +#endif
1.67 +#ifdef SK_ARM_HAS_EDSP
1.68 + "pld [r0, #16] \n\t"
1.69 +#endif
1.70 + "lsr r4, r4, #11 \n\t"
1.71 +#ifdef SK_ARM_HAS_EDSP
1.72 + "smulbb r6, r6, r7 \n\t"
1.73 + "smulbb r5, r5, r7 \n\t"
1.74 + "smulbb r4, r4, r7 \n\t"
1.75 +#else
1.76 + "mul r6, r6, r7 \n\t"
1.77 + "mul r5, r5, r7 \n\t"
1.78 + "mul r4, r4, r7 \n\t"
1.79 +#endif
1.80 +#if SK_ARM_ARCH >= 6
1.81 + "uxtb r7, r3, ROR #16 \n\t"
1.82 + "uxtb ip, r3, ROR #8 \n\t"
1.83 +#else
1.84 + "mov ip, #0xff \n\t"
1.85 + "and r7, ip, r3, ROR #16 \n\t"
1.86 + "and ip, ip, r3, ROR #8 \n\t"
1.87 +#endif
1.88 + "and r3, r3, #0xff \n\t"
1.89 + "add r6, r6, #16 \n\t"
1.90 + "add r5, r5, #32 \n\t"
1.91 + "add r4, r4, #16 \n\t"
1.92 + "add r6, r6, r6, lsr #5 \n\t"
1.93 + "add r5, r5, r5, lsr #6 \n\t"
1.94 + "add r4, r4, r4, lsr #5 \n\t"
1.95 + "add r6, r7, r6, lsr #5 \n\t"
1.96 + "add r5, ip, r5, lsr #6 \n\t"
1.97 + "add r4, r3, r4, lsr #5 \n\t"
1.98 + "lsr r6, r6, #3 \n\t"
1.99 + "and r5, r5, #0xfc \n\t"
1.100 + "and r4, r4, #0xf8 \n\t"
1.101 + "orr r6, r6, r5, lsl #3 \n\t"
1.102 + "orr r4, r6, r4, lsl #8 \n\t"
1.103 + "strh r4, [%[dst]], #2 \n\t"
1.104 +#ifdef SK_ARM_HAS_EDSP
1.105 + "pld [r1, #32] \n\t"
1.106 +#endif
1.107 + "subs %[count], %[count], #1 \n\t"
1.108 + "bne 1b \n\t"
1.109 + "b 4f \n\t"
1.110 + "3: \n\t"
1.111 + "subs %[count], %[count], #1 \n\t"
1.112 + "add %[dst], %[dst], #2 \n\t"
1.113 + "bne 1b \n\t"
1.114 + "4: \n\t"
1.115 + : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count)
1.116 + :
1.117 + : "memory", "cc", "r3", "r4", "r5", "r6", "r7", "ip"
1.118 + );
1.119 +}
1.120 +
1.121 +static void S32A_Opaque_BlitRow32_arm(SkPMColor* SK_RESTRICT dst,
1.122 + const SkPMColor* SK_RESTRICT src,
1.123 + int count, U8CPU alpha) {
1.124 +
1.125 + SkASSERT(255 == alpha);
1.126 +
1.127 + asm volatile (
1.128 + "cmp %[count], #0 \n\t" /* comparing count with 0 */
1.129 + "beq 3f \n\t" /* if zero exit */
1.130 +
1.131 + "mov ip, #0xff \n\t" /* load the 0xff mask in ip */
1.132 + "orr ip, ip, ip, lsl #16 \n\t" /* convert it to 0xff00ff in ip */
1.133 +
1.134 + "cmp %[count], #2 \n\t" /* compare count with 2 */
1.135 + "blt 2f \n\t" /* if less than 2 -> single loop */
1.136 +
1.137 + /* Double Loop */
1.138 + "1: \n\t" /* <double loop> */
1.139 + "ldm %[src]!, {r5,r6} \n\t" /* load the src(s) at r5-r6 */
1.140 + "ldm %[dst], {r7,r8} \n\t" /* loading dst(s) into r7-r8 */
1.141 + "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */
1.142 +
1.143 + /* ----------- */
1.144 + "and r9, ip, r7 \n\t" /* r9 = br masked by ip */
1.145 + "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */
1.146 + "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */
1.147 +
1.148 + "mul r9, r9, r4 \n\t" /* br = br * scale */
1.149 + "mul r10, r10, r4 \n\t" /* ag = ag * scale */
1.150 + "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */
1.151 +
1.152 + "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */
1.153 + "lsr r4, r6, #24 \n\t" /* extracting the alpha from source and storing it to r4 */
1.154 + "orr r7, r9, r10 \n\t" /* br | ag*/
1.155 +
1.156 + "add r7, r5, r7 \n\t" /* dst = src + calc dest(r7) */
1.157 + "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 255 -> r4=scale */
1.158 +
1.159 + /* ----------- */
1.160 + "and r9, ip, r8 \n\t" /* r9 = br masked by ip */
1.161 +
1.162 + "and r10, ip, r8, lsr #8 \n\t" /* r10 = ag masked by ip */
1.163 + "mul r9, r9, r4 \n\t" /* br = br * scale */
1.164 + "sub %[count], %[count], #2 \n\t"
1.165 + "mul r10, r10, r4 \n\t" /* ag = ag * scale */
1.166 +
1.167 + "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */
1.168 + "and r10, r10, ip, lsl #8 \n\t" /* mask ag with reverse mask */
1.169 + "cmp %[count], #1 \n\t" /* comparing count with 1 */
1.170 + "orr r8, r9, r10 \n\t" /* br | ag */
1.171 +
1.172 + "add r8, r6, r8 \n\t" /* dst = src + calc dest(r8) */
1.173 +
1.174 + /* ----------------- */
1.175 + "stm %[dst]!, {r7,r8} \n\t" /* *dst = r7, increment dst by two (each times 4) */
1.176 + /* ----------------- */
1.177 +
1.178 + "bgt 1b \n\t" /* if greater than 1 -> reloop */
1.179 + "blt 3f \n\t" /* if less than 1 -> exit */
1.180 +
1.181 + /* Single Loop */
1.182 + "2: \n\t" /* <single loop> */
1.183 + "ldr r5, [%[src]], #4 \n\t" /* load the src pointer into r5 r5=src */
1.184 + "ldr r7, [%[dst]] \n\t" /* loading dst into r7 */
1.185 + "lsr r4, r5, #24 \n\t" /* extracting the alpha from source and storing it to r4 */
1.186 +
1.187 + /* ----------- */
1.188 + "and r9, ip, r7 \n\t" /* r9 = br masked by ip */
1.189 + "rsb r4, r4, #256 \n\t" /* subtracting the alpha from 256 -> r4=scale */
1.190 +
1.191 + "and r10, ip, r7, lsr #8 \n\t" /* r10 = ag masked by ip */
1.192 + "mul r9, r9, r4 \n\t" /* br = br * scale */
1.193 + "mul r10, r10, r4 \n\t" /* ag = ag * scale */
1.194 + "and r9, ip, r9, lsr #8 \n\t" /* lsr br by 8 and mask it */
1.195 +
1.196 + "and r10, r10, ip, lsl #8 \n\t" /* mask ag */
1.197 + "orr r7, r9, r10 \n\t" /* br | ag */
1.198 +
1.199 + "add r7, r5, r7 \n\t" /* *dst = src + calc dest(r7) */
1.200 +
1.201 + /* ----------------- */
1.202 + "str r7, [%[dst]], #4 \n\t" /* *dst = r7, increment dst by one (times 4) */
1.203 + /* ----------------- */
1.204 +
1.205 + "3: \n\t" /* <exit> */
1.206 + : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count)
1.207 + :
1.208 + : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "ip", "memory"
1.209 + );
1.210 +}
1.211 +
1.212 +/*
1.213 + * ARM asm version of S32A_Blend_BlitRow32
1.214 + */
1.215 +void S32A_Blend_BlitRow32_arm(SkPMColor* SK_RESTRICT dst,
1.216 + const SkPMColor* SK_RESTRICT src,
1.217 + int count, U8CPU alpha) {
1.218 + asm volatile (
1.219 + "cmp %[count], #0 \n\t" /* comparing count with 0 */
1.220 + "beq 3f \n\t" /* if zero exit */
1.221 +
1.222 + "mov r12, #0xff \n\t" /* load the 0xff mask in r12 */
1.223 + "orr r12, r12, r12, lsl #16 \n\t" /* convert it to 0xff00ff in r12 */
1.224 +
1.225 + /* src1,2_scale */
1.226 + "add %[alpha], %[alpha], #1 \n\t" /* loading %[alpha]=src_scale=alpha+1 */
1.227 +
1.228 + "cmp %[count], #2 \n\t" /* comparing count with 2 */
1.229 + "blt 2f \n\t" /* if less than 2 -> single loop */
1.230 +
1.231 + /* Double Loop */
1.232 + "1: \n\t" /* <double loop> */
1.233 + "ldm %[src]!, {r5, r6} \n\t" /* loading src pointers into r5 and r6 */
1.234 + "ldm %[dst], {r7, r8} \n\t" /* loading dst pointers into r7 and r8 */
1.235 +
1.236 + /* dst1_scale and dst2_scale*/
1.237 + "lsr r9, r5, #24 \n\t" /* src >> 24 */
1.238 + "lsr r10, r6, #24 \n\t" /* src >> 24 */
1.239 +#ifdef SK_ARM_HAS_EDSP
1.240 + "smulbb r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
1.241 + "smulbb r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
1.242 +#else
1.243 + "mul r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
1.244 + "mul r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
1.245 +#endif
1.246 + "lsr r9, r9, #8 \n\t" /* r9 >> 8 */
1.247 + "lsr r10, r10, #8 \n\t" /* r10 >> 8 */
1.248 + "rsb r9, r9, #256 \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */
1.249 + "rsb r10, r10, #256 \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */
1.250 +
1.251 + /* ---------------------- */
1.252 +
1.253 + /* src1, src1_scale */
1.254 + "and r11, r12, r5, lsr #8 \n\t" /* ag = r11 = r5 masked by r12 lsr by #8 */
1.255 + "and r4, r12, r5 \n\t" /* rb = r4 = r5 masked by r12 */
1.256 + "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */
1.257 + "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */
1.258 + "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
1.259 + "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
1.260 + "orr r5, r11, r4 \n\t" /* r5 = (src1, src_scale) */
1.261 +
1.262 + /* dst1, dst1_scale */
1.263 + "and r11, r12, r7, lsr #8 \n\t" /* ag = r11 = r7 masked by r12 lsr by #8 */
1.264 + "and r4, r12, r7 \n\t" /* rb = r4 = r7 masked by r12 */
1.265 + "mul r11, r11, r9 \n\t" /* ag = r11 times dst_scale (r9) */
1.266 + "mul r4, r4, r9 \n\t" /* rb = r4 times dst_scale (r9) */
1.267 + "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
1.268 + "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
1.269 + "orr r9, r11, r4 \n\t" /* r9 = (dst1, dst_scale) */
1.270 +
1.271 + /* ---------------------- */
1.272 + "add r9, r5, r9 \n\t" /* *dst = src plus dst both scaled */
1.273 + /* ---------------------- */
1.274 +
1.275 + /* ====================== */
1.276 +
1.277 + /* src2, src2_scale */
1.278 + "and r11, r12, r6, lsr #8 \n\t" /* ag = r11 = r6 masked by r12 lsr by #8 */
1.279 + "and r4, r12, r6 \n\t" /* rb = r4 = r6 masked by r12 */
1.280 + "mul r11, r11, %[alpha] \n\t" /* ag = r11 times src_scale */
1.281 + "mul r4, r4, %[alpha] \n\t" /* rb = r4 times src_scale */
1.282 + "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
1.283 + "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
1.284 + "orr r6, r11, r4 \n\t" /* r6 = (src2, src_scale) */
1.285 +
1.286 + /* dst2, dst2_scale */
1.287 + "and r11, r12, r8, lsr #8 \n\t" /* ag = r11 = r8 masked by r12 lsr by #8 */
1.288 + "and r4, r12, r8 \n\t" /* rb = r4 = r8 masked by r12 */
1.289 + "mul r11, r11, r10 \n\t" /* ag = r11 times dst_scale (r10) */
1.290 + "mul r4, r4, r10 \n\t" /* rb = r4 times dst_scale (r6) */
1.291 + "and r11, r11, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
1.292 + "and r4, r12, r4, lsr #8 \n\t" /* rb masked by mask (r12) */
1.293 + "orr r10, r11, r4 \n\t" /* r10 = (dst2, dst_scale) */
1.294 +
1.295 + "sub %[count], %[count], #2 \n\t" /* decrease count by 2 */
1.296 + /* ---------------------- */
1.297 + "add r10, r6, r10 \n\t" /* *dst = src plus dst both scaled */
1.298 + /* ---------------------- */
1.299 + "cmp %[count], #1 \n\t" /* compare count with 1 */
1.300 + /* ----------------- */
1.301 + "stm %[dst]!, {r9, r10} \n\t" /* copy r9 and r10 to r7 and r8 respectively */
1.302 + /* ----------------- */
1.303 +
1.304 + "bgt 1b \n\t" /* if %[count] greater than 1 reloop */
1.305 + "blt 3f \n\t" /* if %[count] less than 1 exit */
1.306 + /* else get into the single loop */
1.307 + /* Single Loop */
1.308 + "2: \n\t" /* <single loop> */
1.309 + "ldr r5, [%[src]], #4 \n\t" /* loading src pointer into r5: r5=src */
1.310 + "ldr r7, [%[dst]] \n\t" /* loading dst pointer into r7: r7=dst */
1.311 +
1.312 + "lsr r6, r5, #24 \n\t" /* src >> 24 */
1.313 + "and r8, r12, r5, lsr #8 \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */
1.314 +#ifdef SK_ARM_HAS_EDSP
1.315 + "smulbb r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
1.316 +#else
1.317 + "mul r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
1.318 +#endif
1.319 + "and r9, r12, r5 \n\t" /* rb = r9 = r5 masked by r12 */
1.320 + "lsr r6, r6, #8 \n\t" /* r6 >> 8 */
1.321 + "mul r8, r8, %[alpha] \n\t" /* ag = r8 times scale */
1.322 + "rsb r6, r6, #256 \n\t" /* r6 = 255 - r6 + 1 */
1.323 +
1.324 + /* src, src_scale */
1.325 + "mul r9, r9, %[alpha] \n\t" /* rb = r9 times scale */
1.326 + "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
1.327 + "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */
1.328 + "orr r10, r8, r9 \n\t" /* r10 = (scr, src_scale) */
1.329 +
1.330 + /* dst, dst_scale */
1.331 + "and r8, r12, r7, lsr #8 \n\t" /* ag = r8 = r7 masked by r12 lsr by #8 */
1.332 + "and r9, r12, r7 \n\t" /* rb = r9 = r7 masked by r12 */
1.333 + "mul r8, r8, r6 \n\t" /* ag = r8 times scale (r6) */
1.334 + "mul r9, r9, r6 \n\t" /* rb = r9 times scale (r6) */
1.335 + "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
1.336 + "and r9, r12, r9, lsr #8 \n\t" /* rb masked by mask (r12) */
1.337 + "orr r7, r8, r9 \n\t" /* r7 = (dst, dst_scale) */
1.338 +
1.339 + "add r10, r7, r10 \n\t" /* *dst = src plus dst both scaled */
1.340 +
1.341 + /* ----------------- */
1.342 + "str r10, [%[dst]], #4 \n\t" /* *dst = r10, postincrement dst by one (times 4) */
1.343 + /* ----------------- */
1.344 +
1.345 + "3: \n\t" /* <exit> */
1.346 + : [dst] "+r" (dst), [src] "+r" (src), [count] "+r" (count), [alpha] "+r" (alpha)
1.347 + :
1.348 + : "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12", "memory"
1.349 + );
1.350 +
1.351 +}
1.352 +
1.353 +///////////////////////////////////////////////////////////////////////////////
1.354 +
1.355 +static const SkBlitRow::Proc sk_blitrow_platform_565_procs_arm[] = {
1.356 + // no dither
1.357 + // NOTE: For the functions below, we don't have a special version
1.358 + // that assumes that each source pixel is opaque. But our S32A is
1.359 + // still faster than the default, so use it.
1.360 + S32A_D565_Opaque, // S32_D565_Opaque
1.361 + NULL, // S32_D565_Blend
1.362 + S32A_D565_Opaque, // S32A_D565_Opaque
1.363 + NULL, // S32A_D565_Blend
1.364 +
1.365 + // dither
1.366 + NULL, // S32_D565_Opaque_Dither
1.367 + NULL, // S32_D565_Blend_Dither
1.368 + NULL, // S32A_D565_Opaque_Dither
1.369 + NULL, // S32A_D565_Blend_Dither
1.370 +};
1.371 +
1.372 +static const SkBlitRow::Proc32 sk_blitrow_platform_32_procs_arm[] = {
1.373 + NULL, // S32_Opaque,
1.374 + NULL, // S32_Blend,
1.375 + S32A_Opaque_BlitRow32_arm, // S32A_Opaque,
1.376 + S32A_Blend_BlitRow32_arm // S32A_Blend
1.377 +};
1.378 +
1.379 +#endif // USE_ARM_CODE
1.380 +
1.381 +SkBlitRow::Proc SkBlitRow::PlatformProcs565(unsigned flags) {
1.382 + return SK_ARM_NEON_WRAP(sk_blitrow_platform_565_procs_arm)[flags];
1.383 +}
1.384 +
1.385 +SkBlitRow::Proc32 SkBlitRow::PlatformProcs32(unsigned flags) {
1.386 + return SK_ARM_NEON_WRAP(sk_blitrow_platform_32_procs_arm)[flags];
1.387 +}
1.388 +
1.389 +///////////////////////////////////////////////////////////////////////////////
1.390 +#define Color32_arm NULL
1.391 +SkBlitRow::ColorProc SkBlitRow::PlatformColorProc() {
1.392 + return SK_ARM_NEON_WRAP(Color32_arm);
1.393 +}
1.394 +
1.395 +SkBlitRow::ColorRectProc PlatformColorRectProcFactory() {
1.396 + return NULL;
1.397 +}
