blend_neon.h (4809B)
1 /* 2 * Copyright (c) 2023, Alliance for Open Media. All rights reserved. 3 * 4 * This source code is subject to the terms of the BSD 2 Clause License and 5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License 6 * was not distributed with this source code in the LICENSE file, you can 7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open 8 * Media Patent License 1.0 was not distributed with this source code in the 9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent. 10 */ 11 12 #ifndef AOM_AOM_DSP_ARM_BLEND_NEON_H_ 13 #define AOM_AOM_DSP_ARM_BLEND_NEON_H_ 14 15 #include <arm_neon.h> 16 17 #include "aom_dsp/blend.h" 18 19 static inline uint8x16_t alpha_blend_a64_u8x16(uint8x16_t m, uint8x16_t a, 20 uint8x16_t b) { 21 const uint8x16_t m_inv = vsubq_u8(vdupq_n_u8(AOM_BLEND_A64_MAX_ALPHA), m); 22 23 uint16x8_t blend_u16_lo = vmull_u8(vget_low_u8(m), vget_low_u8(a)); 24 uint16x8_t blend_u16_hi = vmull_u8(vget_high_u8(m), vget_high_u8(a)); 25 26 blend_u16_lo = vmlal_u8(blend_u16_lo, vget_low_u8(m_inv), vget_low_u8(b)); 27 blend_u16_hi = vmlal_u8(blend_u16_hi, vget_high_u8(m_inv), vget_high_u8(b)); 28 29 uint8x8_t blend_u8_lo = vrshrn_n_u16(blend_u16_lo, AOM_BLEND_A64_ROUND_BITS); 30 uint8x8_t blend_u8_hi = vrshrn_n_u16(blend_u16_hi, AOM_BLEND_A64_ROUND_BITS); 31 32 return vcombine_u8(blend_u8_lo, blend_u8_hi); 33 } 34 35 static inline uint8x8_t alpha_blend_a64_u8x8(uint8x8_t m, uint8x8_t a, 36 uint8x8_t b) { 37 const uint8x8_t m_inv = vsub_u8(vdup_n_u8(AOM_BLEND_A64_MAX_ALPHA), m); 38 39 uint16x8_t blend_u16 = vmull_u8(m, a); 40 41 blend_u16 = vmlal_u8(blend_u16, m_inv, b); 42 43 return vrshrn_n_u16(blend_u16, AOM_BLEND_A64_ROUND_BITS); 44 } 45 46 #if CONFIG_AV1_HIGHBITDEPTH 47 static inline uint16x8_t alpha_blend_a64_u16x8(uint16x8_t m, uint16x8_t a, 48 uint16x8_t b) { 49 uint16x8_t m_inv = vsubq_u16(vdupq_n_u16(AOM_BLEND_A64_MAX_ALPHA), m); 50 51 uint32x4_t blend_u32_lo = vmull_u16(vget_low_u16(a), vget_low_u16(m)); 52 uint32x4_t blend_u32_hi = vmull_u16(vget_high_u16(a), vget_high_u16(m)); 53 54 blend_u32_lo = vmlal_u16(blend_u32_lo, vget_low_u16(b), vget_low_u16(m_inv)); 55 blend_u32_hi = 56 vmlal_u16(blend_u32_hi, vget_high_u16(b), vget_high_u16(m_inv)); 57 58 uint16x4_t blend_u16_lo = 59 vrshrn_n_u32(blend_u32_lo, AOM_BLEND_A64_ROUND_BITS); 60 uint16x4_t blend_u16_hi = 61 vrshrn_n_u32(blend_u32_hi, AOM_BLEND_A64_ROUND_BITS); 62 63 return vcombine_u16(blend_u16_lo, blend_u16_hi); 64 } 65 66 static inline uint16x4_t alpha_blend_a64_u16x4(uint16x4_t m, uint16x4_t a, 67 uint16x4_t b) { 68 const uint16x4_t m_inv = vsub_u16(vdup_n_u16(AOM_BLEND_A64_MAX_ALPHA), m); 69 70 uint32x4_t blend_u16 = vmull_u16(m, a); 71 72 blend_u16 = vmlal_u16(blend_u16, m_inv, b); 73 74 return vrshrn_n_u32(blend_u16, AOM_BLEND_A64_ROUND_BITS); 75 } 76 #endif // CONFIG_AV1_HIGHBITDEPTH 77 78 static inline uint8x8_t avg_blend_u8x8(uint8x8_t a, uint8x8_t b) { 79 return vrhadd_u8(a, b); 80 } 81 82 static inline uint8x16_t avg_blend_u8x16(uint8x16_t a, uint8x16_t b) { 83 return vrhaddq_u8(a, b); 84 } 85 86 static inline uint8x8_t avg_blend_pairwise_u8x8(uint8x8_t a, uint8x8_t b) { 87 return vrshr_n_u8(vpadd_u8(a, b), 1); 88 } 89 90 static inline uint8x16_t avg_blend_pairwise_u8x16(uint8x16_t a, uint8x16_t b) { 91 #if AOM_ARCH_AARCH64 92 return vrshrq_n_u8(vpaddq_u8(a, b), 1); 93 #else 94 uint8x8_t sum_pairwise_a = vpadd_u8(vget_low_u8(a), vget_high_u8(a)); 95 uint8x8_t sum_pairwise_b = vpadd_u8(vget_low_u8(b), vget_high_u8(b)); 96 return vrshrq_n_u8(vcombine_u8(sum_pairwise_a, sum_pairwise_b), 1); 97 #endif // AOM_ARCH_AARCH64 98 } 99 100 static inline uint8x8_t avg_blend_pairwise_u8x8_4(uint8x8_t a, uint8x8_t b, 101 uint8x8_t c, uint8x8_t d) { 102 uint8x8_t a_c = vpadd_u8(a, c); 103 uint8x8_t b_d = vpadd_u8(b, d); 104 return vrshr_n_u8(vqadd_u8(a_c, b_d), 2); 105 } 106 107 static inline uint8x16_t avg_blend_pairwise_u8x16_4(uint8x16_t a, uint8x16_t b, 108 uint8x16_t c, 109 uint8x16_t d) { 110 #if AOM_ARCH_AARCH64 111 uint8x16_t a_c = vpaddq_u8(a, c); 112 uint8x16_t b_d = vpaddq_u8(b, d); 113 return vrshrq_n_u8(vqaddq_u8(a_c, b_d), 2); 114 #else 115 uint8x8_t sum_pairwise_a = vpadd_u8(vget_low_u8(a), vget_high_u8(a)); 116 uint8x8_t sum_pairwise_b = vpadd_u8(vget_low_u8(b), vget_high_u8(b)); 117 uint8x8_t sum_pairwise_c = vpadd_u8(vget_low_u8(c), vget_high_u8(c)); 118 uint8x8_t sum_pairwise_d = vpadd_u8(vget_low_u8(d), vget_high_u8(d)); 119 uint8x16_t a_c = vcombine_u8(sum_pairwise_a, sum_pairwise_c); 120 uint8x16_t b_d = vcombine_u8(sum_pairwise_b, sum_pairwise_d); 121 return vrshrq_n_u8(vqaddq_u8(a_c, b_d), 2); 122 #endif // AOM_ARCH_AARCH64 123 } 124 125 #endif // AOM_AOM_DSP_ARM_BLEND_NEON_H_