| OLD | NEW |
| (Empty) | |
| 1 ; Copyright (c) 2011 The Chromium Authors. All rights reserved. |
| 2 ; Use of this source code is governed by a BSD-style license that can be |
| 3 ; found in the LICENSE file. |
| 4 |
| 5 ; |
| 6 ; void SYMBOL(const uint8* argb, uint8* y, uint8* u, uint8* v, int width); |
| 7 ; |
| 8 ; The main code that converts RGB pixels to YUV pixels. This function roughly |
| 9 ; consists of three parts: converting one ARGB pixel to YUV pixels, converting |
| 10 ; two ARGB pixels to YUV pixels, and converting four ARGB pixels to YUV pixels. |
| 11 ; To write the structure of this function in C, it becomes the snippet listed |
| 12 ; below. |
| 13 ; |
| 14 ; if (width & 1) { |
| 15 ; --width; |
| 16 ; // Convert one ARGB pixel to one Y pixel, one U pixel, and one V pixel. |
| 17 ; } |
| 18 ; |
| 19 ; if (width & 2) { |
| 20 ; width -= 2; |
| 21 ; // Convert two ARGB pixels to two Y pixels, one U pixel, and one V pixel. |
| 22 ; } |
| 23 ; |
| 24 ; while (width) { |
| 25 ; width -= 4; |
| 26 ; // Convert four ARGB pixels to four Y pixels, two U pixels, and two V |
| 27 ; // pixels. |
| 28 ; } |
| 29 ; |
| 30 global mangle(SYMBOL) PRIVATE |
| 31 align function_align |
| 32 |
| 33 mangle(SYMBOL): |
| 34 %assign stack_offset 0 |
| 35 PROLOGUE 5, 6, 8, ARGB, Y, U, V, WIDTH, TEMP |
| 36 |
| 37 ; Initialize constants used in this function. (We use immediates to avoid |
| 38 ; dependency onto GOT.) |
| 39 LOAD_XMM XMM_CONST_Y0, 0x00420219 |
| 40 LOAD_XMM XMM_CONST_Y1, 0x00007F00 |
| 41 LOAD_XMM XMM_CONST_U, 0x00DAB670 |
| 42 LOAD_XMM XMM_CONST_V, 0x0070A2EE |
| 43 LOAD_XMM XMM_CONST_128, 0x00800080 |
| 44 |
| 45 .convert_one_pixel: |
| 46 ; Divide the input width by two so it represents the offsets for u[] and v[]. |
| 47 ; When the width is odd, We read the rightmost ARGB pixel and convert its |
| 48 ; colorspace to YUV. This code stores one Y pixel, one U pixel, and one V |
| 49 ; pixel. |
| 50 sar WIDTHq, 1 |
| 51 jnc .convert_two_pixels |
| 52 |
| 53 ; Read one ARGB (or RGB) pixel. |
| 54 READ_ARGB xmm0, 1 |
| 55 |
| 56 ; Calculate y[0] from one RGB pixel read above. |
| 57 CALC_Y xmm1, xmm0 |
| 58 movd TEMPd, xmm1 |
| 59 mov BYTE [Yq + WIDTHq * 2], TEMPb |
| 60 |
| 61 ; Calculate u[0] from one RGB pixel read above. If this is an odd line, the |
| 62 ; output pixel contains the U value calculated in the previous call. We also |
| 63 ; read this pixel and calculate their average. |
| 64 INIT_UV TEMPd, Uq, 4 |
| 65 CALC_UV xmm1, xmm0, XMM_CONST_U, TEMPd |
| 66 movd TEMPd, xmm1 |
| 67 mov BYTE [Uq + WIDTHq], TEMPb |
| 68 |
| 69 ; Calculate v[0] from one RGB pixel. Same as u[0], we read the result of the |
| 70 ; previous call and get their average. |
| 71 INIT_UV TEMPd, Uq, 4 |
| 72 CALC_UV xmm1, xmm0, XMM_CONST_V, TEMPd |
| 73 movd TEMPd, xmm1 |
| 74 mov BYTE [Vq + WIDTHq], TEMPb |
| 75 |
| 76 .convert_two_pixels: |
| 77 ; If the input width is not a multiple of four, read the rightmost two ARGB |
| 78 ; pixels and convert their colorspace to YUV. This code stores two Y pixels, |
| 79 ; one U pixel, and one V pixel. |
| 80 test WIDTHb, 2 / 2 |
| 81 jz .convert_four_pixels |
| 82 sub WIDTHb, 2 / 2 |
| 83 |
| 84 ; Read two ARGB (or RGB) pixels. |
| 85 READ_ARGB xmm0, 2 |
| 86 |
| 87 ; Calculate r[0] and r[1] from two RGB pixels read above. |
| 88 CALC_Y xmm1, xmm0 |
| 89 movd TEMPd, xmm1 |
| 90 mov WORD [Yq + WIDTHq * 2], TEMPw |
| 91 |
| 92 ; Skip calculating u and v if the output buffer is NULL. |
| 93 test Uq, Uq |
| 94 jz .convert_four_pixels |
| 95 |
| 96 ; Calculate u[0] from two RGB pixels read above. (For details, read the above |
| 97 ; comment in .convert_one_pixel). |
| 98 INIT_UV TEMPd, Uq, 2 |
| 99 CALC_UV xmm1, xmm0, XMM_CONST_U, TEMPd |
| 100 movd TEMPd, xmm1 |
| 101 mov BYTE [Uq + WIDTHq], TEMPb |
| 102 |
| 103 ; Calculate v[0] from two RGB pixels read above. |
| 104 INIT_UV TEMPd, Vq, 2 |
| 105 CALC_UV xmm1, xmm0, XMM_CONST_V, TEMPd |
| 106 movd TEMPd, xmm1 |
| 107 mov BYTE [Vq + WIDTHq], TEMPb |
| 108 |
| 109 .convert_four_pixels: |
| 110 ; Read four ARGB pixels and convert their colorspace to YUV. This code stores |
| 111 ; four Y pixels, two U pixels, and two V pixels. |
| 112 test WIDTHq, WIDTHq |
| 113 jz .convert_finish |
| 114 |
| 115 %if PIXELSIZE == 4 |
| 116 ; Check if the input buffer is aligned to a 16-byte boundary and use movdqa |
| 117 ; for reading the ARGB pixels. |
| 118 test ARGBw, 15 |
| 119 jnz .convert_four_pixels_unaligned |
| 120 |
| 121 .convert_four_pixels_aligned: |
| 122 sub WIDTHq, 4 / 2 |
| 123 |
| 124 ; Read four ARGB pixels. (We can use movdqa here since we have checked if the |
| 125 ; source address is aligned.) |
| 126 movdqa xmm0, DQWORD [ARGBq + WIDTHq * 4 * 2] |
| 127 |
| 128 ; Calculate y[0], y[1], y[2],and, y[3] from the input ARGB pixels. |
| 129 CALC_Y xmm1, xmm0 |
| 130 movd DWORD [Yq + WIDTHq * 2], xmm1 |
| 131 |
| 132 %if SUBSAMPLING == 0 |
| 133 ; Skip calculating u and v if the output buffer is NULL, which means we are |
| 134 ; converting an odd line. (When we enable subsampling, these buffers must |
| 135 ; contain the u and v values for the previous call, i.e. these variables must |
| 136 ; not be NULL.) |
| 137 test Uq, Uq |
| 138 jz .convert_four_pixels_aligned_next |
| 139 %endif |
| 140 |
| 141 ; Calculate u[0] and u[1] from four ARGB pixels read above. |
| 142 INIT_UV TEMPd, Uq, 4 |
| 143 CALC_UV xmm1, xmm0, XMM_CONST_U, TEMPd |
| 144 movd TEMPd, xmm1 |
| 145 mov WORD [Uq + WIDTHq], TEMPw |
| 146 |
| 147 ; Calculate v[0] and v[1] from four ARGB pixels read above. |
| 148 INIT_UV TEMPd, Vq, 4 |
| 149 CALC_UV xmm1, xmm0, XMM_CONST_V, TEMPd |
| 150 movd TEMPd, xmm1 |
| 151 mov WORD [Vq + WIDTHq], TEMPw |
| 152 |
| 153 %if SUBSAMPLING == 0 |
| 154 .convert_four_pixels_aligned_next: |
| 155 %endif |
| 156 |
| 157 test WIDTHq, WIDTHq |
| 158 jnz .convert_four_pixels_aligned |
| 159 |
| 160 jmp .convert_finish |
| 161 %endif |
| 162 |
| 163 .convert_four_pixels_unaligned: |
| 164 sub WIDTHq, 4 / 2 |
| 165 |
| 166 ; Read four ARGB (or RGB) pixels. |
| 167 READ_ARGB xmm0, 4 |
| 168 |
| 169 ; Calculate y[0], y[1], y[2],and, y[3] from the input ARGB pixels. |
| 170 CALC_Y xmm1, xmm0 |
| 171 movd DWORD [Yq + WIDTHq * 2], xmm1 |
| 172 |
| 173 %if SUBSAMPLING == 0 |
| 174 ; Skip calculating u and v if the output buffer is NULL. |
| 175 test Uq, Uq |
| 176 jz .convert_four_pixels_unaligned_next |
| 177 %endif |
| 178 |
| 179 ; Calculate u[0] and u[1] from the input ARGB pixels. |
| 180 INIT_UV TEMPd, Uq, 4 |
| 181 CALC_UV xmm1, xmm0, XMM_CONST_U, TEMPd |
| 182 movd TEMPd, xmm1 |
| 183 mov WORD [Uq + WIDTHq], TEMPw |
| 184 |
| 185 ; Calculate v[0] and v[1] from the input ARGB pixels. |
| 186 INIT_UV TEMPd, Vq, 4 |
| 187 CALC_UV xmm1, xmm0, XMM_CONST_V, TEMPd |
| 188 movd TEMPd, xmm1 |
| 189 mov WORD [Vq + WIDTHq], TEMPw |
| 190 |
| 191 %if SUBSAMPLING == 0 |
| 192 .convert_four_pixels_unaligned_next: |
| 193 %endif |
| 194 |
| 195 test WIDTHq, WIDTHq |
| 196 jnz .convert_four_pixels_unaligned |
| 197 |
| 198 .convert_finish: |
| 199 ; Just exit this function since this is a void function. |
| 200 RET |
| OLD | NEW |