blend_a64_vmask_neon.c (3429B)
1 /* 2 * 3 * Copyright (c) 2018, Alliance for Open Media. All rights reserved. 4 * 5 * This source code is subject to the terms of the BSD 2 Clause License and 6 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 7 * was not distributed with this source code in the LICENSE file, you can 8 * obtain it at www.aomedia.org/license/software. If the Alliance for Open 9 * Media Patent License 1.0 was not distributed with this source code in the 10 * PATENTS file, you can obtain it at www.aomedia.org/license/patent. 11 */ 12 13 #include <arm_neon.h> 14 #include <assert.h> 15 16 #include "aom/aom_integer.h" 17 #include "aom_dsp/aom_dsp_common.h" 18 #include "aom_dsp/blend.h" 19 #include "aom_dsp/arm/blend_neon.h" 20 #include "aom_dsp/arm/mem_neon.h" 21 #include "aom_ports/mem.h" 22 #include "config/aom_dsp_rtcd.h" 23 24 void aom_blend_a64_vmask_neon(uint8_t *dst, uint32_t dst_stride, 25 const uint8_t *src0, uint32_t src0_stride, 26 const uint8_t *src1, uint32_t src1_stride, 27 const uint8_t *mask, int w, int h) { 28 assert(IMPLIES(src0 == dst, src0_stride == dst_stride)); 29 assert(IMPLIES(src1 == dst, src1_stride == dst_stride)); 30 31 assert(h >= 2); 32 assert(w >= 2); 33 assert(IS_POWER_OF_TWO(h)); 34 assert(IS_POWER_OF_TWO(w)); 35 36 if (w > 8) { 37 do { 38 uint8x16_t m0 = vdupq_n_u8(mask[0]); 39 int i = 0; 40 do { 41 uint8x16_t s0 = vld1q_u8(src0 + i); 42 uint8x16_t s1 = vld1q_u8(src1 + i); 43 44 uint8x16_t blend = alpha_blend_a64_u8x16(m0, s0, s1); 45 46 vst1q_u8(dst + i, blend); 47 48 i += 16; 49 } while (i < w); 50 51 mask += 1; 52 src0 += src0_stride; 53 src1 += src1_stride; 54 dst += dst_stride; 55 } while (--h != 0); 56 } else if (w == 8) { 57 do { 58 uint8x8_t m0 = vdup_n_u8(mask[0]); 59 uint8x8_t s0 = vld1_u8(src0); 60 uint8x8_t s1 = vld1_u8(src1); 61 62 uint8x8_t blend = alpha_blend_a64_u8x8(m0, s0, s1); 63 64 vst1_u8(dst, blend); 65 66 mask += 1; 67 src0 += src0_stride; 68 src1 += src1_stride; 69 dst += dst_stride; 70 } while (--h != 0); 71 } else if (w == 4) { 72 do { 73 const uint16x4_t m0 = vdup_n_u16((uint16_t)mask[0]); 74 const uint16x4_t m1 = vdup_n_u16((uint16_t)mask[1]); 75 const uint8x8_t m = vmovn_u16(vcombine_u16(m0, m1)); 76 uint8x8_t s0 = load_unaligned_u8_4x2(src0, src0_stride); 77 uint8x8_t s1 = load_unaligned_u8_4x2(src1, src1_stride); 78 79 uint8x8_t blend = alpha_blend_a64_u8x8(m, s0, s1); 80 81 store_u8x4_strided_x2(dst, dst_stride, blend); 82 83 mask += 2; 84 src0 += 2 * src0_stride; 85 src1 += 2 * src1_stride; 86 dst += 2 * dst_stride; 87 h -= 2; 88 } while (h != 0); 89 } else if (w == 2 && h >= 16) { 90 do { 91 uint16x4_t m0 = vdup_n_u16(0); 92 m0 = vld1_lane_u16((uint16_t *)mask, m0, 0); 93 uint8x8_t m = 94 vzip_u8(vreinterpret_u8_u16(m0), vreinterpret_u8_u16(m0)).val[0]; 95 uint8x8_t s0 = load_unaligned_u8_2x2(src0, src0_stride); 96 uint8x8_t s1 = load_unaligned_u8_2x2(src1, src1_stride); 97 98 uint8x8_t blend = alpha_blend_a64_u8x8(m, s0, s1); 99 100 store_u8x2_strided_x2(dst, dst_stride, blend); 101 102 mask += 2; 103 src0 += 2 * src0_stride; 104 src1 += 2 * src1_stride; 105 dst += 2 * dst_stride; 106 h -= 2; 107 } while (h != 0); 108 } else { 109 aom_blend_a64_vmask_c(dst, dst_stride, src0, src0_stride, src1, src1_stride, 110 mask, w, h); 111 } 112 }