michael@0: 
michael@0: #include "SkBlitMask.h"
michael@0: #include "SkColor_opts_neon.h"
michael@0: 
michael@0: static void D32_A8_Black_neon(void* SK_RESTRICT dst, size_t dstRB,
michael@0:                               const void* SK_RESTRICT maskPtr, size_t maskRB,
michael@0:                               SkColor, int width, int height) {
michael@0:     SkPMColor* SK_RESTRICT device = (SkPMColor*)dst;
michael@0:     const uint8_t* SK_RESTRICT mask = (const uint8_t*)maskPtr;
michael@0: 
michael@0:     maskRB -= width;
michael@0:     dstRB -= (width << 2);
michael@0:     do {
michael@0:         int w = width;
michael@0:         while (w >= 8) {
michael@0:             uint8x8_t vmask = vld1_u8(mask);
michael@0:             uint16x8_t vscale = vsubw_u8(vdupq_n_u16(256), vmask);
michael@0:             uint8x8x4_t vdevice = vld4_u8((uint8_t*)device);
michael@0: 
michael@0:             vdevice = SkAlphaMulQ_neon8(vdevice, vscale);
michael@0:             vdevice.val[NEON_A] += vmask;
michael@0: 
michael@0:             vst4_u8((uint8_t*)device, vdevice);
michael@0: 
michael@0:             mask += 8;
michael@0:             device += 8;
michael@0:             w -= 8;
michael@0:         }
michael@0:         while (w-- > 0) {
michael@0:             unsigned aa = *mask++;
michael@0:             *device = (aa << SK_A32_SHIFT)
michael@0:                         + SkAlphaMulQ(*device, SkAlpha255To256(255 - aa));
michael@0:             device += 1;
michael@0:         };
michael@0:         device = (uint32_t*)((char*)device + dstRB);
michael@0:         mask += maskRB;
michael@0:     } while (--height != 0);
michael@0: }
michael@0: 
michael@0: template <bool isColor>
michael@0: static void D32_A8_Opaque_Color_neon(void* SK_RESTRICT dst, size_t dstRB,
michael@0:                                      const void* SK_RESTRICT maskPtr, size_t maskRB,
michael@0:                                      SkColor color, int width, int height) {
michael@0:     SkPMColor pmc = SkPreMultiplyColor(color);
michael@0:     SkPMColor* SK_RESTRICT device = (SkPMColor*)dst;
michael@0:     const uint8_t* SK_RESTRICT mask = (const uint8_t*)maskPtr;
michael@0:     uint8x8x4_t vpmc;
michael@0: 
michael@0:     maskRB -= width;
michael@0:     dstRB -= (width << 2);
michael@0: 
michael@0:     if (width >= 8) {
michael@0:         vpmc.val[NEON_A] = vdup_n_u8(SkGetPackedA32(pmc));
michael@0:         vpmc.val[NEON_R] = vdup_n_u8(SkGetPackedR32(pmc));
michael@0:         vpmc.val[NEON_G] = vdup_n_u8(SkGetPackedG32(pmc));
michael@0:         vpmc.val[NEON_B] = vdup_n_u8(SkGetPackedB32(pmc));
michael@0:     }
michael@0:     do {
michael@0:         int w = width;
michael@0:         while (w >= 8) {
michael@0:             uint8x8_t vmask = vld1_u8(mask);
michael@0:             uint16x8_t vscale, vmask256 = SkAlpha255To256_neon8(vmask);
michael@0:             if (isColor) {
michael@0:                 vscale = vsubw_u8(vdupq_n_u16(256),
michael@0:                             SkAlphaMul_neon8(vpmc.val[NEON_A], vmask256));
michael@0:             } else {
michael@0:                 vscale = vsubw_u8(vdupq_n_u16(256), vmask);
michael@0:             }
michael@0:             uint8x8x4_t vdev = vld4_u8((uint8_t*)device);
michael@0: 
michael@0:             vdev.val[NEON_A] =   SkAlphaMul_neon8(vpmc.val[NEON_A], vmask256)
michael@0:                                + SkAlphaMul_neon8(vdev.val[NEON_A], vscale);
michael@0:             vdev.val[NEON_R] =   SkAlphaMul_neon8(vpmc.val[NEON_R], vmask256)
michael@0:                                + SkAlphaMul_neon8(vdev.val[NEON_R], vscale);
michael@0:             vdev.val[NEON_G] =   SkAlphaMul_neon8(vpmc.val[NEON_G], vmask256)
michael@0:                                + SkAlphaMul_neon8(vdev.val[NEON_G], vscale);
michael@0:             vdev.val[NEON_B] =   SkAlphaMul_neon8(vpmc.val[NEON_B], vmask256)
michael@0:                                + SkAlphaMul_neon8(vdev.val[NEON_B], vscale);
michael@0: 
michael@0:             vst4_u8((uint8_t*)device, vdev);
michael@0: 
michael@0:             mask += 8;
michael@0:             device += 8;
michael@0:             w -= 8;
michael@0:         }
michael@0: 
michael@0:         while (w--) {
michael@0:             unsigned aa = *mask++;
michael@0:             if (isColor) {
michael@0:                 *device = SkBlendARGB32(pmc, *device, aa);
michael@0:             } else {
michael@0:                 *device = SkAlphaMulQ(pmc, SkAlpha255To256(aa))
michael@0:                             + SkAlphaMulQ(*device, SkAlpha255To256(255 - aa));
michael@0:             }
michael@0:             device += 1;
michael@0:         };
michael@0: 
michael@0:         device = (uint32_t*)((char*)device + dstRB);
michael@0:         mask += maskRB;
michael@0: 
michael@0:     } while (--height != 0);
michael@0: }
michael@0: 
michael@0: static void D32_A8_Opaque_neon(void* SK_RESTRICT dst, size_t dstRB,
michael@0:                                const void* SK_RESTRICT maskPtr, size_t maskRB,
michael@0:                                SkColor color, int width, int height) {
michael@0:     D32_A8_Opaque_Color_neon<false>(dst, dstRB, maskPtr, maskRB, color, width, height);
michael@0: }
michael@0: 
michael@0: static void D32_A8_Color_neon(void* SK_RESTRICT dst, size_t dstRB,
michael@0:                               const void* SK_RESTRICT maskPtr, size_t maskRB,
michael@0:                               SkColor color, int width, int height) {
michael@0:     D32_A8_Opaque_Color_neon<true>(dst, dstRB, maskPtr, maskRB, color, width, height);
michael@0: }
michael@0: 
michael@0: SkBlitMask::ColorProc D32_A8_Factory_neon(SkColor color) {
michael@0:     if (SK_ColorBLACK == color) {
michael@0:         return D32_A8_Black_neon;
michael@0:     } else if (0xFF == SkColorGetA(color)) {
michael@0:         return D32_A8_Opaque_neon;
michael@0:     } else {
michael@0:         return D32_A8_Color_neon;
michael@0:     }
michael@0: }
michael@0: 
michael@0: ////////////////////////////////////////////////////////////////////////////////
michael@0: 
michael@0: void SkBlitLCD16OpaqueRow_neon(SkPMColor dst[], const uint16_t src[],
michael@0:                                         SkColor color, int width,
michael@0:                                         SkPMColor opaqueDst) {
michael@0:     int colR = SkColorGetR(color);
michael@0:     int colG = SkColorGetG(color);
michael@0:     int colB = SkColorGetB(color);
michael@0: 
michael@0:     uint8x8_t vcolR, vcolG, vcolB;
michael@0:     uint8x8_t vopqDstA, vopqDstR, vopqDstG, vopqDstB;
michael@0: 
michael@0:     if (width >= 8) {
michael@0:         vcolR = vdup_n_u8(colR);
michael@0:         vcolG = vdup_n_u8(colG);
michael@0:         vcolB = vdup_n_u8(colB);
michael@0:         vopqDstA = vdup_n_u8(SkGetPackedA32(opaqueDst));
michael@0:         vopqDstR = vdup_n_u8(SkGetPackedR32(opaqueDst));
michael@0:         vopqDstG = vdup_n_u8(SkGetPackedG32(opaqueDst));
michael@0:         vopqDstB = vdup_n_u8(SkGetPackedB32(opaqueDst));
michael@0:     }
michael@0: 
michael@0:     while (width >= 8) {
michael@0:         uint8x8x4_t vdst;
michael@0:         uint16x8_t vmask;
michael@0:         uint16x8_t vmaskR, vmaskG, vmaskB;
michael@0:         uint8x8_t vsel_trans, vsel_opq;
michael@0: 
michael@0:         vdst = vld4_u8((uint8_t*)dst);
michael@0:         vmask = vld1q_u16(src);
michael@0: 
michael@0:         // Prepare compare masks
michael@0:         vsel_trans = vmovn_u16(vceqq_u16(vmask, vdupq_n_u16(0)));
michael@0:         vsel_opq = vmovn_u16(vceqq_u16(vmask, vdupq_n_u16(0xFFFF)));
michael@0: 
michael@0:         // Get all the color masks on 5 bits
michael@0:         vmaskR = vshrq_n_u16(vmask, SK_R16_SHIFT);
michael@0:         vmaskG = vshrq_n_u16(vshlq_n_u16(vmask, SK_R16_BITS),
michael@0:                              SK_B16_BITS + SK_R16_BITS + 1);
michael@0:         vmaskB = vmask & vdupq_n_u16(SK_B16_MASK);
michael@0: 
michael@0:         // Upscale to 0..32
michael@0:         vmaskR = vmaskR + vshrq_n_u16(vmaskR, 4);
michael@0:         vmaskG = vmaskG + vshrq_n_u16(vmaskG, 4);
michael@0:         vmaskB = vmaskB + vshrq_n_u16(vmaskB, 4);
michael@0: 
michael@0:         vdst.val[NEON_A] = vbsl_u8(vsel_trans, vdst.val[NEON_A], vdup_n_u8(0xFF));
michael@0:         vdst.val[NEON_A] = vbsl_u8(vsel_opq, vopqDstA, vdst.val[NEON_A]);
michael@0: 
michael@0:         vdst.val[NEON_R] = SkBlend32_neon8(vcolR, vdst.val[NEON_R], vmaskR);
michael@0:         vdst.val[NEON_G] = SkBlend32_neon8(vcolG, vdst.val[NEON_G], vmaskG);
michael@0:         vdst.val[NEON_B] = SkBlend32_neon8(vcolB, vdst.val[NEON_B], vmaskB);
michael@0: 
michael@0:         vdst.val[NEON_R] = vbsl_u8(vsel_opq, vopqDstR, vdst.val[NEON_R]);
michael@0:         vdst.val[NEON_G] = vbsl_u8(vsel_opq, vopqDstG, vdst.val[NEON_G]);
michael@0:         vdst.val[NEON_B] = vbsl_u8(vsel_opq, vopqDstB, vdst.val[NEON_B]);
michael@0: 
michael@0:         vst4_u8((uint8_t*)dst, vdst);
michael@0: 
michael@0:         dst += 8;
michael@0:         src += 8;
michael@0:         width -= 8;
michael@0:     }
michael@0: 
michael@0:     // Leftovers
michael@0:     for (int i = 0; i < width; i++) {
michael@0:         dst[i] = SkBlendLCD16Opaque(colR, colG, colB, dst[i], src[i],
michael@0:                                     opaqueDst);
michael@0:     }
michael@0: }
michael@0: 
michael@0: void SkBlitLCD16Row_neon(SkPMColor dst[], const uint16_t src[],
michael@0:                                    SkColor color, int width, SkPMColor) {
michael@0:     int colA = SkColorGetA(color);
michael@0:     int colR = SkColorGetR(color);
michael@0:     int colG = SkColorGetG(color);
michael@0:     int colB = SkColorGetB(color);
michael@0: 
michael@0:     colA = SkAlpha255To256(colA);
michael@0: 
michael@0:     uint8x8_t vcolR, vcolG, vcolB;
michael@0:     uint16x8_t vcolA;
michael@0: 
michael@0:     if (width >= 8) {
michael@0:         vcolA = vdupq_n_u16(colA);
michael@0:         vcolR = vdup_n_u8(colR);
michael@0:         vcolG = vdup_n_u8(colG);
michael@0:         vcolB = vdup_n_u8(colB);
michael@0:     }
michael@0: 
michael@0:     while (width >= 8) {
michael@0:         uint8x8x4_t vdst;
michael@0:         uint16x8_t vmask;
michael@0:         uint16x8_t vmaskR, vmaskG, vmaskB;
michael@0: 
michael@0:         vdst = vld4_u8((uint8_t*)dst);
michael@0:         vmask = vld1q_u16(src);
michael@0: 
michael@0:         // Get all the color masks on 5 bits
michael@0:         vmaskR = vshrq_n_u16(vmask, SK_R16_SHIFT);
michael@0:         vmaskG = vshrq_n_u16(vshlq_n_u16(vmask, SK_R16_BITS),
michael@0:                              SK_B16_BITS + SK_R16_BITS + 1);
michael@0:         vmaskB = vmask & vdupq_n_u16(SK_B16_MASK);
michael@0: 
michael@0:         // Upscale to 0..32
michael@0:         vmaskR = vmaskR + vshrq_n_u16(vmaskR, 4);
michael@0:         vmaskG = vmaskG + vshrq_n_u16(vmaskG, 4);
michael@0:         vmaskB = vmaskB + vshrq_n_u16(vmaskB, 4);
michael@0: 
michael@0:         vmaskR = vshrq_n_u16(vmaskR * vcolA, 8);
michael@0:         vmaskG = vshrq_n_u16(vmaskG * vcolA, 8);
michael@0:         vmaskB = vshrq_n_u16(vmaskB * vcolA, 8);
michael@0: 
michael@0:         vdst.val[NEON_A] = vdup_n_u8(0xFF);
michael@0:         vdst.val[NEON_R] = SkBlend32_neon8(vcolR, vdst.val[NEON_R], vmaskR);
michael@0:         vdst.val[NEON_G] = SkBlend32_neon8(vcolG, vdst.val[NEON_G], vmaskG);
michael@0:         vdst.val[NEON_B] = SkBlend32_neon8(vcolB, vdst.val[NEON_B], vmaskB);
michael@0: 
michael@0:         vst4_u8((uint8_t*)dst, vdst);
michael@0: 
michael@0:         dst += 8;
michael@0:         src += 8;
michael@0:         width -= 8;
michael@0:     }
michael@0: 
michael@0:     for (int i = 0; i < width; i++) {
michael@0:         dst[i] = SkBlendLCD16(colA, colR, colG, colB, dst[i], src[i]);
michael@0:     }
michael@0: }