1/* 2 * Copyright 2009 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 9#include "SkBitmapScaler.h" 10#include "SkBitmapProcState.h" 11#include "SkColorPriv.h" 12#include "SkPaint.h" 13#include "SkTypes.h" 14#include "SkUtils.h" 15#include "SkUtilsArm.h" 16 17#include "SkConvolver.h" 18 19#if !defined(SK_CPU_ARM64) && SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN) 20void SI8_D16_nofilter_DX_arm( 21 const SkBitmapProcState& s, 22 const uint32_t* SK_RESTRICT xy, 23 int count, 24 uint16_t* SK_RESTRICT colors) SK_ATTRIBUTE_OPTIMIZE_O1; 25 26void SI8_D16_nofilter_DX_arm(const SkBitmapProcState& s, 27 const uint32_t* SK_RESTRICT xy, 28 int count, uint16_t* SK_RESTRICT colors) { 29 SkASSERT(count > 0 && colors != NULL); 30 SkASSERT(s.fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)); 31 SkASSERT(SkPaint::kNone_FilterLevel == s.fFilterLevel); 32 33 const uint16_t* SK_RESTRICT table = s.fBitmap->getColorTable()->lock16BitCache(); 34 const uint8_t* SK_RESTRICT srcAddr = (const uint8_t*)s.fBitmap->getPixels(); 35 36 // buffer is y32, x16, x16, x16, x16, x16 37 // bump srcAddr to the proper row, since we're told Y never changes 38 SkASSERT((unsigned)xy[0] < (unsigned)s.fBitmap->height()); 39 srcAddr = (const uint8_t*)((const char*)srcAddr + 40 xy[0] * s.fBitmap->rowBytes()); 41 42 uint8_t src; 43 44 if (1 == s.fBitmap->width()) { 45 src = srcAddr[0]; 46 uint16_t dstValue = table[src]; 47 sk_memset16(colors, dstValue, count); 48 } else { 49 int i; 50 int count8 = count >> 3; 51 const uint16_t* SK_RESTRICT xx = (const uint16_t*)(xy + 1); 52 53 asm volatile ( 54 "cmp %[count8], #0 \n\t" // compare loop counter with 0 55 "beq 2f \n\t" // if loop counter == 0, exit 56 "1: \n\t" 57 "ldmia %[xx]!, {r5, r7, r9, r11} \n\t" // load ptrs to pixels 0-7 58 "subs %[count8], %[count8], #1 \n\t" // decrement loop counter 59 "uxth r4, r5 \n\t" // extract ptr 0 60 "mov r5, r5, lsr #16 \n\t" // extract ptr 1 61 "uxth r6, r7 \n\t" // extract ptr 2 62 "mov r7, r7, lsr #16 \n\t" // extract ptr 3 63 "ldrb r4, [%[srcAddr], r4] \n\t" // load pixel 0 from image 64 "uxth r8, r9 \n\t" // extract ptr 4 65 "ldrb r5, [%[srcAddr], r5] \n\t" // load pixel 1 from image 66 "mov r9, r9, lsr #16 \n\t" // extract ptr 5 67 "ldrb r6, [%[srcAddr], r6] \n\t" // load pixel 2 from image 68 "uxth r10, r11 \n\t" // extract ptr 6 69 "ldrb r7, [%[srcAddr], r7] \n\t" // load pixel 3 from image 70 "mov r11, r11, lsr #16 \n\t" // extract ptr 7 71 "ldrb r8, [%[srcAddr], r8] \n\t" // load pixel 4 from image 72 "add r4, r4, r4 \n\t" // double pixel 0 for RGB565 lookup 73 "ldrb r9, [%[srcAddr], r9] \n\t" // load pixel 5 from image 74 "add r5, r5, r5 \n\t" // double pixel 1 for RGB565 lookup 75 "ldrb r10, [%[srcAddr], r10] \n\t" // load pixel 6 from image 76 "add r6, r6, r6 \n\t" // double pixel 2 for RGB565 lookup 77 "ldrb r11, [%[srcAddr], r11] \n\t" // load pixel 7 from image 78 "add r7, r7, r7 \n\t" // double pixel 3 for RGB565 lookup 79 "ldrh r4, [%[table], r4] \n\t" // load pixel 0 RGB565 from colmap 80 "add r8, r8, r8 \n\t" // double pixel 4 for RGB565 lookup 81 "ldrh r5, [%[table], r5] \n\t" // load pixel 1 RGB565 from colmap 82 "add r9, r9, r9 \n\t" // double pixel 5 for RGB565 lookup 83 "ldrh r6, [%[table], r6] \n\t" // load pixel 2 RGB565 from colmap 84 "add r10, r10, r10 \n\t" // double pixel 6 for RGB565 lookup 85 "ldrh r7, [%[table], r7] \n\t" // load pixel 3 RGB565 from colmap 86 "add r11, r11, r11 \n\t" // double pixel 7 for RGB565 lookup 87 "ldrh r8, [%[table], r8] \n\t" // load pixel 4 RGB565 from colmap 88 "ldrh r9, [%[table], r9] \n\t" // load pixel 5 RGB565 from colmap 89 "ldrh r10, [%[table], r10] \n\t" // load pixel 6 RGB565 from colmap 90 "ldrh r11, [%[table], r11] \n\t" // load pixel 7 RGB565 from colmap 91 "pkhbt r5, r4, r5, lsl #16 \n\t" // pack pixels 0 and 1 92 "pkhbt r6, r6, r7, lsl #16 \n\t" // pack pixels 2 and 3 93 "pkhbt r8, r8, r9, lsl #16 \n\t" // pack pixels 4 and 5 94 "pkhbt r10, r10, r11, lsl #16 \n\t" // pack pixels 6 and 7 95 "stmia %[colors]!, {r5, r6, r8, r10} \n\t" // store last 8 pixels 96 "bgt 1b \n\t" // loop if counter > 0 97 "2: \n\t" 98 : [xx] "+r" (xx), [count8] "+r" (count8), [colors] "+r" (colors) 99 : [table] "r" (table), [srcAddr] "r" (srcAddr) 100 : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11" 101 ); 102 103 for (i = (count & 7); i > 0; --i) { 104 src = srcAddr[*xx++]; *colors++ = table[src]; 105 } 106 } 107 108 s.fBitmap->getColorTable()->unlock16BitCache(); 109} 110 111void SI8_opaque_D32_nofilter_DX_arm( 112 const SkBitmapProcState& s, 113 const uint32_t* SK_RESTRICT xy, 114 int count, 115 SkPMColor* SK_RESTRICT colors) SK_ATTRIBUTE_OPTIMIZE_O1; 116 117void SI8_opaque_D32_nofilter_DX_arm(const SkBitmapProcState& s, 118 const uint32_t* SK_RESTRICT xy, 119 int count, SkPMColor* SK_RESTRICT colors) { 120 SkASSERT(count > 0 && colors != NULL); 121 SkASSERT(s.fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)); 122 SkASSERT(SkPaint::kNone_FilterLevel == s.fFilterLevel); 123 124 const SkPMColor* SK_RESTRICT table = s.fBitmap->getColorTable()->lockColors(); 125 const uint8_t* SK_RESTRICT srcAddr = (const uint8_t*)s.fBitmap->getPixels(); 126 127 // buffer is y32, x16, x16, x16, x16, x16 128 // bump srcAddr to the proper row, since we're told Y never changes 129 SkASSERT((unsigned)xy[0] < (unsigned)s.fBitmap->height()); 130 srcAddr = (const uint8_t*)((const char*)srcAddr + xy[0] * s.fBitmap->rowBytes()); 131 132 if (1 == s.fBitmap->width()) { 133 uint8_t src = srcAddr[0]; 134 SkPMColor dstValue = table[src]; 135 sk_memset32(colors, dstValue, count); 136 } else { 137 const uint16_t* xx = (const uint16_t*)(xy + 1); 138 139 asm volatile ( 140 "subs %[count], %[count], #8 \n\t" // decrement count by 8, set flags 141 "blt 2f \n\t" // if count < 0, branch to singles 142 "1: \n\t" // eights loop 143 "ldmia %[xx]!, {r5, r7, r9, r11} \n\t" // load ptrs to pixels 0-7 144 "uxth r4, r5 \n\t" // extract ptr 0 145 "mov r5, r5, lsr #16 \n\t" // extract ptr 1 146 "uxth r6, r7 \n\t" // extract ptr 2 147 "mov r7, r7, lsr #16 \n\t" // extract ptr 3 148 "ldrb r4, [%[srcAddr], r4] \n\t" // load pixel 0 from image 149 "uxth r8, r9 \n\t" // extract ptr 4 150 "ldrb r5, [%[srcAddr], r5] \n\t" // load pixel 1 from image 151 "mov r9, r9, lsr #16 \n\t" // extract ptr 5 152 "ldrb r6, [%[srcAddr], r6] \n\t" // load pixel 2 from image 153 "uxth r10, r11 \n\t" // extract ptr 6 154 "ldrb r7, [%[srcAddr], r7] \n\t" // load pixel 3 from image 155 "mov r11, r11, lsr #16 \n\t" // extract ptr 7 156 "ldrb r8, [%[srcAddr], r8] \n\t" // load pixel 4 from image 157 "ldrb r9, [%[srcAddr], r9] \n\t" // load pixel 5 from image 158 "ldrb r10, [%[srcAddr], r10] \n\t" // load pixel 6 from image 159 "ldrb r11, [%[srcAddr], r11] \n\t" // load pixel 7 from image 160 "ldr r4, [%[table], r4, lsl #2] \n\t" // load pixel 0 SkPMColor from colmap 161 "ldr r5, [%[table], r5, lsl #2] \n\t" // load pixel 1 SkPMColor from colmap 162 "ldr r6, [%[table], r6, lsl #2] \n\t" // load pixel 2 SkPMColor from colmap 163 "ldr r7, [%[table], r7, lsl #2] \n\t" // load pixel 3 SkPMColor from colmap 164 "ldr r8, [%[table], r8, lsl #2] \n\t" // load pixel 4 SkPMColor from colmap 165 "ldr r9, [%[table], r9, lsl #2] \n\t" // load pixel 5 SkPMColor from colmap 166 "ldr r10, [%[table], r10, lsl #2] \n\t" // load pixel 6 SkPMColor from colmap 167 "ldr r11, [%[table], r11, lsl #2] \n\t" // load pixel 7 SkPMColor from colmap 168 "subs %[count], %[count], #8 \n\t" // decrement loop counter 169 "stmia %[colors]!, {r4-r11} \n\t" // store 8 pixels 170 "bge 1b \n\t" // loop if counter >= 0 171 "2: \n\t" 172 "adds %[count], %[count], #8 \n\t" // fix up counter, set flags 173 "beq 4f \n\t" // if count == 0, branch to exit 174 "3: \n\t" // singles loop 175 "ldrh r4, [%[xx]], #2 \n\t" // load pixel ptr 176 "subs %[count], %[count], #1 \n\t" // decrement loop counter 177 "ldrb r5, [%[srcAddr], r4] \n\t" // load pixel from image 178 "ldr r6, [%[table], r5, lsl #2] \n\t" // load SkPMColor from colmap 179 "str r6, [%[colors]], #4 \n\t" // store pixel, update ptr 180 "bne 3b \n\t" // loop if counter != 0 181 "4: \n\t" // exit 182 : [xx] "+r" (xx), [count] "+r" (count), [colors] "+r" (colors) 183 : [table] "r" (table), [srcAddr] "r" (srcAddr) 184 : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11" 185 ); 186 } 187 188 s.fBitmap->getColorTable()->unlockColors(); 189} 190#endif // !defined(SK_CPU_ARM64) && SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN) 191 192/////////////////////////////////////////////////////////////////////////////// 193 194/* If we replace a sampleproc, then we null-out the associated shaderproc, 195 otherwise the shader won't even look at the matrix/sampler 196 */ 197void SkBitmapProcState::platformProcs() { 198#if !defined(SK_CPU_ARM64) && SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN) 199 bool isOpaque = 256 == fAlphaScale; 200 bool justDx = false; 201 202 if (fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)) { 203 justDx = true; 204 } 205 206 switch (fBitmap->colorType()) { 207 case kIndex_8_SkColorType: 208 if (justDx && SkPaint::kNone_FilterLevel == fFilterLevel) { 209#if 0 /* crashing on android device */ 210 fSampleProc16 = SI8_D16_nofilter_DX_arm; 211 fShaderProc16 = NULL; 212#endif 213 if (isOpaque) { 214 // this one is only very slighty faster than the C version 215 fSampleProc32 = SI8_opaque_D32_nofilter_DX_arm; 216 fShaderProc32 = NULL; 217 } 218 } 219 break; 220 default: 221 break; 222 } 223#endif 224} 225 226/////////////////////////////////////////////////////////////////////////////// 227 228extern void platformConvolutionProcs_arm_neon(SkConvolutionProcs* procs); 229 230void platformConvolutionProcs_arm(SkConvolutionProcs* procs) { 231} 232 233void SkBitmapScaler::PlatformConvolutionProcs(SkConvolutionProcs* procs) { 234 SK_ARM_NEON_WRAP(platformConvolutionProcs_arm)(procs); 235} 236