1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/skia/trunk/src/opts/SkBlitMask_opts_arm_neon.cpp Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,254 @@
1.4 +
1.5 +#include "SkBlitMask.h"
1.6 +#include "SkColor_opts_neon.h"
1.7 +
1.8 +static void D32_A8_Black_neon(void* SK_RESTRICT dst, size_t dstRB,
1.9 + const void* SK_RESTRICT maskPtr, size_t maskRB,
1.10 + SkColor, int width, int height) {
1.11 + SkPMColor* SK_RESTRICT device = (SkPMColor*)dst;
1.12 + const uint8_t* SK_RESTRICT mask = (const uint8_t*)maskPtr;
1.13 +
1.14 + maskRB -= width;
1.15 + dstRB -= (width << 2);
1.16 + do {
1.17 + int w = width;
1.18 + while (w >= 8) {
1.19 + uint8x8_t vmask = vld1_u8(mask);
1.20 + uint16x8_t vscale = vsubw_u8(vdupq_n_u16(256), vmask);
1.21 + uint8x8x4_t vdevice = vld4_u8((uint8_t*)device);
1.22 +
1.23 + vdevice = SkAlphaMulQ_neon8(vdevice, vscale);
1.24 + vdevice.val[NEON_A] += vmask;
1.25 +
1.26 + vst4_u8((uint8_t*)device, vdevice);
1.27 +
1.28 + mask += 8;
1.29 + device += 8;
1.30 + w -= 8;
1.31 + }
1.32 + while (w-- > 0) {
1.33 + unsigned aa = *mask++;
1.34 + *device = (aa << SK_A32_SHIFT)
1.35 + + SkAlphaMulQ(*device, SkAlpha255To256(255 - aa));
1.36 + device += 1;
1.37 + };
1.38 + device = (uint32_t*)((char*)device + dstRB);
1.39 + mask += maskRB;
1.40 + } while (--height != 0);
1.41 +}
1.42 +
1.43 +template <bool isColor>
1.44 +static void D32_A8_Opaque_Color_neon(void* SK_RESTRICT dst, size_t dstRB,
1.45 + const void* SK_RESTRICT maskPtr, size_t maskRB,
1.46 + SkColor color, int width, int height) {
1.47 + SkPMColor pmc = SkPreMultiplyColor(color);
1.48 + SkPMColor* SK_RESTRICT device = (SkPMColor*)dst;
1.49 + const uint8_t* SK_RESTRICT mask = (const uint8_t*)maskPtr;
1.50 + uint8x8x4_t vpmc;
1.51 +
1.52 + maskRB -= width;
1.53 + dstRB -= (width << 2);
1.54 +
1.55 + if (width >= 8) {
1.56 + vpmc.val[NEON_A] = vdup_n_u8(SkGetPackedA32(pmc));
1.57 + vpmc.val[NEON_R] = vdup_n_u8(SkGetPackedR32(pmc));
1.58 + vpmc.val[NEON_G] = vdup_n_u8(SkGetPackedG32(pmc));
1.59 + vpmc.val[NEON_B] = vdup_n_u8(SkGetPackedB32(pmc));
1.60 + }
1.61 + do {
1.62 + int w = width;
1.63 + while (w >= 8) {
1.64 + uint8x8_t vmask = vld1_u8(mask);
1.65 + uint16x8_t vscale, vmask256 = SkAlpha255To256_neon8(vmask);
1.66 + if (isColor) {
1.67 + vscale = vsubw_u8(vdupq_n_u16(256),
1.68 + SkAlphaMul_neon8(vpmc.val[NEON_A], vmask256));
1.69 + } else {
1.70 + vscale = vsubw_u8(vdupq_n_u16(256), vmask);
1.71 + }
1.72 + uint8x8x4_t vdev = vld4_u8((uint8_t*)device);
1.73 +
1.74 + vdev.val[NEON_A] = SkAlphaMul_neon8(vpmc.val[NEON_A], vmask256)
1.75 + + SkAlphaMul_neon8(vdev.val[NEON_A], vscale);
1.76 + vdev.val[NEON_R] = SkAlphaMul_neon8(vpmc.val[NEON_R], vmask256)
1.77 + + SkAlphaMul_neon8(vdev.val[NEON_R], vscale);
1.78 + vdev.val[NEON_G] = SkAlphaMul_neon8(vpmc.val[NEON_G], vmask256)
1.79 + + SkAlphaMul_neon8(vdev.val[NEON_G], vscale);
1.80 + vdev.val[NEON_B] = SkAlphaMul_neon8(vpmc.val[NEON_B], vmask256)
1.81 + + SkAlphaMul_neon8(vdev.val[NEON_B], vscale);
1.82 +
1.83 + vst4_u8((uint8_t*)device, vdev);
1.84 +
1.85 + mask += 8;
1.86 + device += 8;
1.87 + w -= 8;
1.88 + }
1.89 +
1.90 + while (w--) {
1.91 + unsigned aa = *mask++;
1.92 + if (isColor) {
1.93 + *device = SkBlendARGB32(pmc, *device, aa);
1.94 + } else {
1.95 + *device = SkAlphaMulQ(pmc, SkAlpha255To256(aa))
1.96 + + SkAlphaMulQ(*device, SkAlpha255To256(255 - aa));
1.97 + }
1.98 + device += 1;
1.99 + };
1.100 +
1.101 + device = (uint32_t*)((char*)device + dstRB);
1.102 + mask += maskRB;
1.103 +
1.104 + } while (--height != 0);
1.105 +}
1.106 +
1.107 +static void D32_A8_Opaque_neon(void* SK_RESTRICT dst, size_t dstRB,
1.108 + const void* SK_RESTRICT maskPtr, size_t maskRB,
1.109 + SkColor color, int width, int height) {
1.110 + D32_A8_Opaque_Color_neon<false>(dst, dstRB, maskPtr, maskRB, color, width, height);
1.111 +}
1.112 +
1.113 +static void D32_A8_Color_neon(void* SK_RESTRICT dst, size_t dstRB,
1.114 + const void* SK_RESTRICT maskPtr, size_t maskRB,
1.115 + SkColor color, int width, int height) {
1.116 + D32_A8_Opaque_Color_neon<true>(dst, dstRB, maskPtr, maskRB, color, width, height);
1.117 +}
1.118 +
1.119 +SkBlitMask::ColorProc D32_A8_Factory_neon(SkColor color) {
1.120 + if (SK_ColorBLACK == color) {
1.121 + return D32_A8_Black_neon;
1.122 + } else if (0xFF == SkColorGetA(color)) {
1.123 + return D32_A8_Opaque_neon;
1.124 + } else {
1.125 + return D32_A8_Color_neon;
1.126 + }
1.127 +}
1.128 +
1.129 +////////////////////////////////////////////////////////////////////////////////
1.130 +
1.131 +void SkBlitLCD16OpaqueRow_neon(SkPMColor dst[], const uint16_t src[],
1.132 + SkColor color, int width,
1.133 + SkPMColor opaqueDst) {
1.134 + int colR = SkColorGetR(color);
1.135 + int colG = SkColorGetG(color);
1.136 + int colB = SkColorGetB(color);
1.137 +
1.138 + uint8x8_t vcolR, vcolG, vcolB;
1.139 + uint8x8_t vopqDstA, vopqDstR, vopqDstG, vopqDstB;
1.140 +
1.141 + if (width >= 8) {
1.142 + vcolR = vdup_n_u8(colR);
1.143 + vcolG = vdup_n_u8(colG);
1.144 + vcolB = vdup_n_u8(colB);
1.145 + vopqDstA = vdup_n_u8(SkGetPackedA32(opaqueDst));
1.146 + vopqDstR = vdup_n_u8(SkGetPackedR32(opaqueDst));
1.147 + vopqDstG = vdup_n_u8(SkGetPackedG32(opaqueDst));
1.148 + vopqDstB = vdup_n_u8(SkGetPackedB32(opaqueDst));
1.149 + }
1.150 +
1.151 + while (width >= 8) {
1.152 + uint8x8x4_t vdst;
1.153 + uint16x8_t vmask;
1.154 + uint16x8_t vmaskR, vmaskG, vmaskB;
1.155 + uint8x8_t vsel_trans, vsel_opq;
1.156 +
1.157 + vdst = vld4_u8((uint8_t*)dst);
1.158 + vmask = vld1q_u16(src);
1.159 +
1.160 + // Prepare compare masks
1.161 + vsel_trans = vmovn_u16(vceqq_u16(vmask, vdupq_n_u16(0)));
1.162 + vsel_opq = vmovn_u16(vceqq_u16(vmask, vdupq_n_u16(0xFFFF)));
1.163 +
1.164 + // Get all the color masks on 5 bits
1.165 + vmaskR = vshrq_n_u16(vmask, SK_R16_SHIFT);
1.166 + vmaskG = vshrq_n_u16(vshlq_n_u16(vmask, SK_R16_BITS),
1.167 + SK_B16_BITS + SK_R16_BITS + 1);
1.168 + vmaskB = vmask & vdupq_n_u16(SK_B16_MASK);
1.169 +
1.170 + // Upscale to 0..32
1.171 + vmaskR = vmaskR + vshrq_n_u16(vmaskR, 4);
1.172 + vmaskG = vmaskG + vshrq_n_u16(vmaskG, 4);
1.173 + vmaskB = vmaskB + vshrq_n_u16(vmaskB, 4);
1.174 +
1.175 + vdst.val[NEON_A] = vbsl_u8(vsel_trans, vdst.val[NEON_A], vdup_n_u8(0xFF));
1.176 + vdst.val[NEON_A] = vbsl_u8(vsel_opq, vopqDstA, vdst.val[NEON_A]);
1.177 +
1.178 + vdst.val[NEON_R] = SkBlend32_neon8(vcolR, vdst.val[NEON_R], vmaskR);
1.179 + vdst.val[NEON_G] = SkBlend32_neon8(vcolG, vdst.val[NEON_G], vmaskG);
1.180 + vdst.val[NEON_B] = SkBlend32_neon8(vcolB, vdst.val[NEON_B], vmaskB);
1.181 +
1.182 + vdst.val[NEON_R] = vbsl_u8(vsel_opq, vopqDstR, vdst.val[NEON_R]);
1.183 + vdst.val[NEON_G] = vbsl_u8(vsel_opq, vopqDstG, vdst.val[NEON_G]);
1.184 + vdst.val[NEON_B] = vbsl_u8(vsel_opq, vopqDstB, vdst.val[NEON_B]);
1.185 +
1.186 + vst4_u8((uint8_t*)dst, vdst);
1.187 +
1.188 + dst += 8;
1.189 + src += 8;
1.190 + width -= 8;
1.191 + }
1.192 +
1.193 + // Leftovers
1.194 + for (int i = 0; i < width; i++) {
1.195 + dst[i] = SkBlendLCD16Opaque(colR, colG, colB, dst[i], src[i],
1.196 + opaqueDst);
1.197 + }
1.198 +}
1.199 +
1.200 +void SkBlitLCD16Row_neon(SkPMColor dst[], const uint16_t src[],
1.201 + SkColor color, int width, SkPMColor) {
1.202 + int colA = SkColorGetA(color);
1.203 + int colR = SkColorGetR(color);
1.204 + int colG = SkColorGetG(color);
1.205 + int colB = SkColorGetB(color);
1.206 +
1.207 + colA = SkAlpha255To256(colA);
1.208 +
1.209 + uint8x8_t vcolR, vcolG, vcolB;
1.210 + uint16x8_t vcolA;
1.211 +
1.212 + if (width >= 8) {
1.213 + vcolA = vdupq_n_u16(colA);
1.214 + vcolR = vdup_n_u8(colR);
1.215 + vcolG = vdup_n_u8(colG);
1.216 + vcolB = vdup_n_u8(colB);
1.217 + }
1.218 +
1.219 + while (width >= 8) {
1.220 + uint8x8x4_t vdst;
1.221 + uint16x8_t vmask;
1.222 + uint16x8_t vmaskR, vmaskG, vmaskB;
1.223 +
1.224 + vdst = vld4_u8((uint8_t*)dst);
1.225 + vmask = vld1q_u16(src);
1.226 +
1.227 + // Get all the color masks on 5 bits
1.228 + vmaskR = vshrq_n_u16(vmask, SK_R16_SHIFT);
1.229 + vmaskG = vshrq_n_u16(vshlq_n_u16(vmask, SK_R16_BITS),
1.230 + SK_B16_BITS + SK_R16_BITS + 1);
1.231 + vmaskB = vmask & vdupq_n_u16(SK_B16_MASK);
1.232 +
1.233 + // Upscale to 0..32
1.234 + vmaskR = vmaskR + vshrq_n_u16(vmaskR, 4);
1.235 + vmaskG = vmaskG + vshrq_n_u16(vmaskG, 4);
1.236 + vmaskB = vmaskB + vshrq_n_u16(vmaskB, 4);
1.237 +
1.238 + vmaskR = vshrq_n_u16(vmaskR * vcolA, 8);
1.239 + vmaskG = vshrq_n_u16(vmaskG * vcolA, 8);
1.240 + vmaskB = vshrq_n_u16(vmaskB * vcolA, 8);
1.241 +
1.242 + vdst.val[NEON_A] = vdup_n_u8(0xFF);
1.243 + vdst.val[NEON_R] = SkBlend32_neon8(vcolR, vdst.val[NEON_R], vmaskR);
1.244 + vdst.val[NEON_G] = SkBlend32_neon8(vcolG, vdst.val[NEON_G], vmaskG);
1.245 + vdst.val[NEON_B] = SkBlend32_neon8(vcolB, vdst.val[NEON_B], vmaskB);
1.246 +
1.247 + vst4_u8((uint8_t*)dst, vdst);
1.248 +
1.249 + dst += 8;
1.250 + src += 8;
1.251 + width -= 8;
1.252 + }
1.253 +
1.254 + for (int i = 0; i < width; i++) {
1.255 + dst[i] = SkBlendLCD16(colA, colR, colG, colB, dst[i], src[i]);
1.256 + }
1.257 +}
