jcsample-altivec.c (5617B)
1 /* 2 * AltiVec optimizations for libjpeg-turbo 3 * 4 * Copyright (C) 2015, D. R. Commander. All Rights Reserved. 5 * 6 * This software is provided 'as-is', without any express or implied 7 * warranty. In no event will the authors be held liable for any damages 8 * arising from the use of this software. 9 * 10 * Permission is granted to anyone to use this software for any purpose, 11 * including commercial applications, and to alter it and redistribute it 12 * freely, subject to the following restrictions: 13 * 14 * 1. The origin of this software must not be misrepresented; you must not 15 * claim that you wrote the original software. If you use this software 16 * in a product, an acknowledgment in the product documentation would be 17 * appreciated but is not required. 18 * 2. Altered source versions must be plainly marked as such, and must not be 19 * misrepresented as being the original software. 20 * 3. This notice may not be removed or altered from any source distribution. 21 */ 22 23 /* CHROMA DOWNSAMPLING */ 24 25 #include "jsimd_altivec.h" 26 #include "jcsample.h" 27 28 29 void jsimd_h2v1_downsample_altivec(JDIMENSION image_width, 30 int max_v_samp_factor, 31 JDIMENSION v_samp_factor, 32 JDIMENSION width_in_blocks, 33 JSAMPARRAY input_data, 34 JSAMPARRAY output_data) 35 { 36 int outrow, outcol; 37 JDIMENSION output_cols = width_in_blocks * DCTSIZE; 38 JSAMPROW inptr, outptr; 39 40 __vector unsigned char this0, next0, out; 41 __vector unsigned short this0e, this0o, next0e, next0o, outl, outh; 42 43 /* Constants */ 44 __vector unsigned short pw_bias = { __4X2(0, 1) }, 45 pw_one = { __8X(1) }; 46 __vector unsigned char even_odd_index = 47 { 0, 2, 4, 6, 8, 10, 12, 14, 1, 3, 5, 7, 9, 11, 13, 15 }, 48 pb_zero = { __16X(0) }; 49 50 expand_right_edge(input_data, max_v_samp_factor, image_width, 51 output_cols * 2); 52 53 for (outrow = 0; outrow < v_samp_factor; outrow++) { 54 outptr = output_data[outrow]; 55 inptr = input_data[outrow]; 56 57 for (outcol = output_cols; outcol > 0; 58 outcol -= 16, inptr += 32, outptr += 16) { 59 60 this0 = vec_ld(0, inptr); 61 this0 = vec_perm(this0, this0, even_odd_index); 62 this0e = (__vector unsigned short)VEC_UNPACKHU(this0); 63 this0o = (__vector unsigned short)VEC_UNPACKLU(this0); 64 outl = vec_add(this0e, this0o); 65 outl = vec_add(outl, pw_bias); 66 outl = vec_sr(outl, pw_one); 67 68 if (outcol > 8) { 69 next0 = vec_ld(16, inptr); 70 next0 = vec_perm(next0, next0, even_odd_index); 71 next0e = (__vector unsigned short)VEC_UNPACKHU(next0); 72 next0o = (__vector unsigned short)VEC_UNPACKLU(next0); 73 outh = vec_add(next0e, next0o); 74 outh = vec_add(outh, pw_bias); 75 outh = vec_sr(outh, pw_one); 76 } else 77 outh = vec_splat_u16(0); 78 79 out = vec_pack(outl, outh); 80 vec_st(out, 0, outptr); 81 } 82 } 83 } 84 85 86 void 87 jsimd_h2v2_downsample_altivec(JDIMENSION image_width, int max_v_samp_factor, 88 JDIMENSION v_samp_factor, 89 JDIMENSION width_in_blocks, 90 JSAMPARRAY input_data, JSAMPARRAY output_data) 91 { 92 int inrow, outrow, outcol; 93 JDIMENSION output_cols = width_in_blocks * DCTSIZE; 94 JSAMPROW inptr0, inptr1, outptr; 95 96 __vector unsigned char this0, next0, this1, next1, out; 97 __vector unsigned short this0e, this0o, next0e, next0o, this1e, this1o, 98 next1e, next1o, out0l, out0h, out1l, out1h, outl, outh; 99 100 /* Constants */ 101 __vector unsigned short pw_bias = { __4X2(1, 2) }, 102 pw_two = { __8X(2) }; 103 __vector unsigned char even_odd_index = 104 { 0, 2, 4, 6, 8, 10, 12, 14, 1, 3, 5, 7, 9, 11, 13, 15 }, 105 pb_zero = { __16X(0) }; 106 107 expand_right_edge(input_data, max_v_samp_factor, image_width, 108 output_cols * 2); 109 110 for (inrow = 0, outrow = 0; outrow < v_samp_factor; 111 inrow += 2, outrow++) { 112 113 inptr0 = input_data[inrow]; 114 inptr1 = input_data[inrow + 1]; 115 outptr = output_data[outrow]; 116 117 for (outcol = output_cols; outcol > 0; 118 outcol -= 16, inptr0 += 32, inptr1 += 32, outptr += 16) { 119 120 this0 = vec_ld(0, inptr0); 121 this0 = vec_perm(this0, this0, even_odd_index); 122 this0e = (__vector unsigned short)VEC_UNPACKHU(this0); 123 this0o = (__vector unsigned short)VEC_UNPACKLU(this0); 124 out0l = vec_add(this0e, this0o); 125 126 this1 = vec_ld(0, inptr1); 127 this1 = vec_perm(this1, this1, even_odd_index); 128 this1e = (__vector unsigned short)VEC_UNPACKHU(this1); 129 this1o = (__vector unsigned short)VEC_UNPACKLU(this1); 130 out1l = vec_add(this1e, this1o); 131 132 outl = vec_add(out0l, out1l); 133 outl = vec_add(outl, pw_bias); 134 outl = vec_sr(outl, pw_two); 135 136 if (outcol > 8) { 137 next0 = vec_ld(16, inptr0); 138 next0 = vec_perm(next0, next0, even_odd_index); 139 next0e = (__vector unsigned short)VEC_UNPACKHU(next0); 140 next0o = (__vector unsigned short)VEC_UNPACKLU(next0); 141 out0h = vec_add(next0e, next0o); 142 143 next1 = vec_ld(16, inptr1); 144 next1 = vec_perm(next1, next1, even_odd_index); 145 next1e = (__vector unsigned short)VEC_UNPACKHU(next1); 146 next1o = (__vector unsigned short)VEC_UNPACKLU(next1); 147 out1h = vec_add(next1e, next1o); 148 149 outh = vec_add(out0h, out1h); 150 outh = vec_add(outh, pw_bias); 151 outh = vec_sr(outh, pw_two); 152 } else 153 outh = vec_splat_u16(0); 154 155 out = vec_pack(outl, outh); 156 vec_st(out, 0, outptr); 157 } 158 } 159 }