OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2012 The Android Open Source Project | 2 * Copyright 2012 The Android Open Source Project |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 #include "SkBlitRow_opts_arm_neon.h" | 8 #include "SkBlitRow_opts_arm_neon.h" |
9 | 9 |
10 #include "SkBlitMask.h" | 10 #include "SkBlitMask.h" |
(...skipping 1661 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
1672 SkPMColorAssert(c); | 1672 SkPMColorAssert(c); |
1673 SkASSERT(SkGetPackedA32(c) == 255); | 1673 SkASSERT(SkGetPackedA32(c) == 255); |
1674 | 1674 |
1675 unsigned dither = DITHER_VALUE(x); | 1675 unsigned dither = DITHER_VALUE(x); |
1676 *dst++ = SkDitherRGB32To565(c, dither); | 1676 *dst++ = SkDitherRGB32To565(c, dither); |
1677 DITHER_INC_X(x); | 1677 DITHER_INC_X(x); |
1678 } while (--count != 0); | 1678 } while (--count != 0); |
1679 } | 1679 } |
1680 } | 1680 } |
1681 | 1681 |
| 1682 #define SK_SUPPORT_LEGACY_COLOR32_MATHx |
| 1683 |
| 1684 // Color32 and its SIMD specializations use the blend_256_round_alt algorithm |
| 1685 // from tests/BlendTest.cpp. It's not quite perfect, but it's never wrong in th
e |
| 1686 // interesting edge cases, and it's quite a bit faster than blend_perfect. |
| 1687 // |
| 1688 // blend_256_round_alt is our currently blessed algorithm. Please use it or an
analogous one. |
| 1689 void Color32_arm_neon(SkPMColor* dst, const SkPMColor* src, int count, SkPMColor
color) { |
| 1690 switch (SkGetPackedA32(color)) { |
| 1691 case 0: memmove(dst, src, count * sizeof(SkPMColor)); return; |
| 1692 case 255: sk_memset32(dst, color, count); return; |
| 1693 } |
| 1694 |
| 1695 uint16x8_t colorHigh = vshll_n_u8((uint8x8_t)vdup_n_u32(color), 8); |
| 1696 #ifdef SK_SUPPORT_LEGACY_COLOR32_MATH // blend_256_plus1_trunc, busted |
| 1697 uint16x8_t colorAndRound = colorHigh; |
| 1698 #else // blend_256_round_alt, good |
| 1699 uint16x8_t colorAndRound = vaddq_u16(colorHigh, vdupq_n_u16(128)); |
| 1700 #endif |
| 1701 |
| 1702 unsigned invA = 255 - SkGetPackedA32(color); |
| 1703 #ifdef SK_SUPPORT_LEGACY_COLOR32_MATH // blend_256_plus1_trunc, busted |
| 1704 uint8x8_t invA8 = vdup_n_u8(invA); |
| 1705 #else // blend_256_round_alt, good |
| 1706 SkASSERT(invA + (invA >> 7) < 256); // This next part only works if alpha i
s not 0. |
| 1707 uint8x8_t invA8 = vdup_n_u8(invA + (invA >> 7)); |
| 1708 #endif |
| 1709 |
| 1710 // Does the core work of blending color onto 4 pixels, returning the resulti
ng 4 pixels. |
| 1711 auto kernel = [&](const uint32x4_t& src4) -> uint32x4_t { |
| 1712 uint16x8_t lo = vmull_u8(vget_low_u8( (uint8x16_t)src4), invA8), |
| 1713 hi = vmull_u8(vget_high_u8((uint8x16_t)src4), invA8); |
| 1714 return (uint32x4_t) |
| 1715 vcombine_u8(vaddhn_u16(colorAndRound, lo), vaddhn_u16(colorAndRound,
hi)); |
| 1716 }; |
| 1717 |
| 1718 while (count >= 8) { |
| 1719 uint32x4_t dst0 = kernel(vld1q_u32(src+0)), |
| 1720 dst4 = kernel(vld1q_u32(src+4)); |
| 1721 vst1q_u32(dst+0, dst0); |
| 1722 vst1q_u32(dst+4, dst4); |
| 1723 src += 8; |
| 1724 dst += 8; |
| 1725 count -= 8; |
| 1726 } |
| 1727 if (count >= 4) { |
| 1728 vst1q_u32(dst, kernel(vld1q_u32(src))); |
| 1729 src += 4; |
| 1730 dst += 4; |
| 1731 count -= 4; |
| 1732 } |
| 1733 if (count >= 2) { |
| 1734 uint32x2_t src2 = vld1_u32(src); |
| 1735 vst1_u32(dst, vget_low_u32(kernel(vcombine_u32(src2, src2)))); |
| 1736 src += 2; |
| 1737 dst += 2; |
| 1738 count -= 2; |
| 1739 } |
| 1740 if (count >= 1) { |
| 1741 vst1q_lane_u32(dst, kernel(vdupq_n_u32(*src)), 0); |
| 1742 } |
| 1743 } |
| 1744 |
1682 /////////////////////////////////////////////////////////////////////////////// | 1745 /////////////////////////////////////////////////////////////////////////////// |
1683 | 1746 |
1684 const SkBlitRow::Proc16 sk_blitrow_platform_565_procs_arm_neon[] = { | 1747 const SkBlitRow::Proc16 sk_blitrow_platform_565_procs_arm_neon[] = { |
1685 // no dither | 1748 // no dither |
1686 S32_D565_Opaque_neon, | 1749 S32_D565_Opaque_neon, |
1687 S32_D565_Blend_neon, | 1750 S32_D565_Blend_neon, |
1688 S32A_D565_Opaque_neon, | 1751 S32A_D565_Opaque_neon, |
1689 #if 0 | 1752 #if 0 |
1690 S32A_D565_Blend_neon, | 1753 S32A_D565_Blend_neon, |
1691 #else | 1754 #else |
(...skipping 31 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
1723 S32A_Opaque_BlitRow32_neon_src_alpha, // S32A_Opaque, | 1786 S32A_Opaque_BlitRow32_neon_src_alpha, // S32A_Opaque, |
1724 #else | 1787 #else |
1725 S32A_Opaque_BlitRow32_neon, // S32A_Opaque, | 1788 S32A_Opaque_BlitRow32_neon, // S32A_Opaque, |
1726 #endif | 1789 #endif |
1727 #ifdef SK_CPU_ARM32 | 1790 #ifdef SK_CPU_ARM32 |
1728 S32A_Blend_BlitRow32_neon // S32A_Blend | 1791 S32A_Blend_BlitRow32_neon // S32A_Blend |
1729 #else | 1792 #else |
1730 NULL | 1793 NULL |
1731 #endif | 1794 #endif |
1732 }; | 1795 }; |
OLD | NEW |