Chromium Code Reviews Chromium Code Reviews
Issue 7003082:
Implements RGB to YV12 conversion in YASM. (Closed)
Base URL: svn://chrome-svn/chrome/trunk/src/| OLD | NEW |
|---|---|
| (Empty) | |
| 1 ; Copyright (c) 2011 The Chromium Authors. All rights reserved. | |
|
Alpha Left Google
2011/08/22 14:16:59
Rename this to media/base/yuv_row_rgb2yuv_sse2.inc
| |
| 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 media::simd::ConvertARGBtoYUVRow(const uint8* argb, | |
| 7 ; uint8* y, | |
| 8 ; uint8* u, | |
| 9 ; uint8* v, | |
| 10 ; int width); | |
| 11 ; void media::simd::ConvertRGBtoYUVRow(const uint8* rgb, | |
| 12 ; uint8* y, | |
| 13 ; uint8* u, | |
| 14 ; uint8* v, | |
| 15 ; int width); | |
| 16 ; | |
| 17 ; Converts a row of ARGB (RGB) pixels to the YUV (4:1:1) colorspace. This | |
| 18 ; function reads ARGB (RGB) pixels from right to left, convert their colorspace | |
| 19 ; to YUV, and write the converted pixels to the output buffers. The following | |
| 20 ; code snippetre presents the rough structure of this fucntion. | |
| 21 ; | |
| 22 ; if (width & 1) { | |
| 23 ; --width; | |
| 24 ; convert_one_pixel(argb, y, u, v, width); | |
| 25 ; } | |
| 26 ; if (width & 2) { | |
| 27 ; width -= 2; | |
| 28 ; convert_two_pixels(argb, y, u, v, width); | |
| 29 ; } | |
| 30 ; while (width) { | |
| 31 ; width -= 4; | |
| 32 ; convert_four_pixels(argb, y, u, v, width); | |
| 33 ; } | |
| 34 ; | |
| 35 ; This function has pseudo-code snippets to describe its behaviors. For | |
| 36 ; simplicity, they use the following notations. | |
| 37 ; | |
| 38 ; #define B(n) argb[n * PIXELSIZE + 0]; | |
| 39 ; #define G(n) argb[n * PIXELSIZE + 1]; | |
| 40 ; #define R(n) argb[n * PIXELSIZE + 2]; | |
| 41 ; #define A(n) PIXELSIZE == 4 ? ARGB[n * PIXELSIZE + 3] : 0; | |
| 42 ; #define FIX(n) (n * (1 << SHIFT_BITS) + 0.5) | |
| 43 ; #define ToByte(n) ((n < 0) ? 0 : (n > 255 ? 255 : n)) | |
| 44 ; | |
| 45 ; union { | |
| 46 ; int32 d[4]; | |
| 47 ; int16 w[8]; | |
| 48 ; int8 b[16]; | |
| 49 ; } xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7 | |
| 50 ; | |
| 51 | |
| 52 ; Input Parameters | |
| 53 %ifdef __x86_64__ | |
| 54 %ifdef WIN64 | |
| 55 %define PARAM_ARGB(r) rcx | |
| 56 %define PARAM_Y(r) rdx | |
| 57 %define PARAM_U(r) r8 | |
| 58 %define PARAM_V(r) r9 | |
| 59 %define PARAM_WIDTH(r) INT [(r) + SIZEOF_POINTER * 2] | |
| 60 %else | |
| 61 %define PARAM_ARGB(r) rdi | |
| 62 %define PARAM_Y(r) rsi | |
| 63 %define PARAM_U(r) rdx | |
| 64 %define PARAM_V(r) rcx | |
| 65 %define PARAM_WIDTH(r) r8 | |
| 66 %endif | |
| 67 %else | |
| 68 %define PARAM_ARGB(r) POINTER [(r) + SIZEOF_POINTER * 2] | |
| 69 %define PARAM_Y(r) POINTER [(r) + SIZEOF_POINTER * 2 + SIZEOF_POINTER * 1] | |
| 70 %define PARAM_U(r) POINTER [(r) + SIZEOF_POINTER * 2 + SIZEOF_POINTER * 2] | |
| 71 %define PARAM_V(r) POINTER [(r) + SIZEOF_POINTER * 2 + SIZEOF_POINTER * 3] | |
| 72 %define PARAM_WIDTH(r) INT [(r) + SIZEOF_POINTER * 2 + SIZEOF_POINTER * 4] | |
| 73 %endif ; __x86_64__ | |
| 74 | |
| 75 ; Local variables used in this function. | |
| 76 %define INT_WIDTH32 eax | |
| 77 %define INT_WIDTH8 al | |
| 78 | |
| 79 %define INT_PIXEL32 ebx | |
| 80 %define INT_PIXEL16 bx | |
| 81 %define INT_PIXEL8 bl | |
| 82 | |
| 83 %ifdef __x86_64__ | |
| 84 %define INT_WIDTH rax | |
| 85 %define INT_PIXEL rbx | |
| 86 %define PTR_ARGB PARAM_ARGB(_EBP_) | |
| 87 %define PTR_Y PARAM_Y(_EBP_) | |
| 88 %define PTR_U PARAM_U(_EBP_) | |
| 89 %define PTR_V PARAM_V(_EBP_) | |
| 90 %else | |
| 91 %define INT_WIDTH eax | |
| 92 %define INT_PIXEL ebx | |
| 93 %define PTR_ARGB ecx | |
| 94 %define PTR_Y edx | |
| 95 %define PTR_U esi | |
| 96 %define PTR_V edi | |
| 97 %endif | |
| 98 | |
| 99 %define XMM_TABLE_Y xmm7 | |
| 100 %define XMM_TABLE_0_5 xmm6 | |
| 101 %define XMM_TABLE_UV xmm5 | |
| 102 %define XMM_ZERO xmm4 | |
| 103 | |
| 104 ; SECTION SEG_TEXT | |
| 105 ; CPU PENTIUM4 | |
| 106 | |
| 107 align 16 | |
| 108 global EXTN(SYMBOL_NAME) PRIVATE | |
| 109 | |
| 110 EXTN(SYMBOL_NAME): | |
| 111 push _EBP_ | |
| 112 mov _EBP_, _ESP_ | |
| 113 | |
| 114 ; Save the resisters used in this function. We do not need to save eax, ecx, | |
| 115 ; or edx even though this function uses them. (Callers save them.) | |
| 116 push _EBX_ | |
| 117 push _ESI_ | |
| 118 push _EDI_ | |
| 119 | |
| 120 ; Copy the tables used in this function to xmm5, xmm6, and xmm7. | |
| 121 LOADSYM _EAX_, SIMD_ConvertARGBToYUVRow_kTable | |
| 122 movdqa XMM_TABLE_Y, XMMWORD [_EAX_ + 16 * 0] | |
| 123 movdqa XMM_TABLE_0_5, XMMWORD [_EAX_ + 16 * 1] | |
| 124 movdqa XMM_TABLE_UV, XMMWORD [_EAX_ + 16 * 2] | |
| 125 | |
| 126 ; Copy the input parameters to registers. (We can use the input pointers on | |
| 127 ; x86-64 since we do not change them.) | |
| 128 mov INT_WIDTH, PARAM_WIDTH(_EBP_) | |
| 129 %ifndef __x86_64__ | |
| 130 mov PTR_ARGB, PARAM_ARGB(_EBP_) | |
| 131 mov PTR_Y, PARAM_Y(_EBP_) | |
| 132 mov PTR_U, PARAM_U(_EBP_) | |
| 133 mov PTR_V, PARAM_V(_EBP_) | |
| 134 %endif | |
| 135 | |
| 136 .convert_one_pixel: | |
| 137 ; Divide the input width by two so it represents the offsets for u[] and v[]. | |
| 138 ; When the width is odd. Wee read the rightmost ARGB pixel and convert its | |
| 139 ; colorspace to YUV. This code stores one Y pixel, one U pixel, and one V | |
| 140 ; pixel. | |
| 141 sar INT_WIDTH, 1 | |
| 142 jnc .convert_two_pixels | |
| 143 | |
| 144 ; xmm0.w[0] = B(0); xmm0.w[1] = G(0); xmm0.w[2] = R(0); xmm0.w[3] = A(0); | |
| 145 %if PIXELSIZE == 4 | |
| 146 movd xmm0, DWORD [PTR_ARGB + INT_WIDTH * 4 * 2] | |
| 147 %elif PIXELSIZE == 3 | |
| 148 movd xmm1, INT_WIDTH | |
| 149 lea INT_WIDTH, [INT_WIDTH + INT_WIDTH * 2] | |
| 150 movzx INT_PIXEL32, BYTE [PTR_ARGB + INT_WIDTH * 2 + 2] | |
| 151 shl INT_PIXEL32, 16 | |
| 152 mov INT_PIXEL16, WORD [PTR_ARGB + INT_WIDTH * 2] | |
| 153 movd xmm0, INT_PIXEL32 | |
| 154 movd INT_WIDTH, xmm1 | |
| 155 %else | |
| 156 %error unsupported pixel size. | |
| 157 %endif | |
| 158 pxor XMM_ZERO, XMM_ZERO | |
| 159 punpcklbw xmm0, XMM_ZERO | |
| 160 | |
| 161 ; xmm2.d[0] = FIX(0.098) * B(0) + FIX(0.504) * G(0); | |
| 162 ; xmm2.d[1] = FIX(0.257) * R(0) + FIX(0.000) * A(0); | |
| 163 movdqa xmm2, xmm0 | |
| 164 pmaddwd xmm2, XMM_TABLE_Y | |
| 165 | |
| 166 ; xmm2.d[0] = (xmm2.d[0] + xmm2.d[1] + FIX(0.5)) / FIX(1) + 16; | |
| 167 pshufd xmm4, xmm2, 10001000B | |
| 168 pshufd xmm2, xmm2, 11011101B | |
| 169 paddd xmm2, xmm4 | |
| 170 | |
| 171 paddd xmm2, XMM_TABLE_0_5 | |
| 172 psrad xmm2, SHIFT_BITS | |
| 173 movdqa xmm3, XMM_TABLE_0_5 | |
| 174 psrad xmm3, SHIFT_BITS - 5 | |
| 175 paddd xmm2, xmm3 | |
| 176 | |
| 177 ; y[0] = ToByte(xmm2.d[0]); | |
| 178 packssdw xmm2, xmm2 | |
| 179 packuswb xmm2, xmm2 | |
| 180 movd INT_PIXEL32, xmm2 | |
| 181 mov BYTE [PTR_Y + INT_WIDTH * 2], INT_PIXEL8 | |
| 182 | |
| 183 ; Calculate u[0] from xmm0. | |
| 184 test PTR_U, PTR_U | |
| 185 jz .convert_two_pixels | |
| 186 | |
| 187 ; xmm1.d[0] = FIX(0.439) * B(0) + -FIX(0.291) * G(0); | |
| 188 ; xmm1.d[1] = -FIX(0.148) * R(0) + FIX(0.000) * A(0); | |
| 189 movdqa xmm1, xmm0 | |
| 190 pshufd xmm2, XMM_TABLE_UV, 01000100B | |
| 191 pmaddwd xmm1, xmm2 | |
| 192 | |
| 193 ; xmm1.d[0] = (xmm1.d[0] + xmm1.d[1] + FIX(0.5)) / FIX(1) + 128; | |
| 194 pshufd xmm2, xmm1, 11011101B | |
| 195 pshufd xmm1, xmm1, 10001000B | |
| 196 paddd xmm1, xmm2 | |
| 197 paddd xmm1, XMM_TABLE_0_5 | |
| 198 psrad xmm1, SHIFT_BITS | |
| 199 %if SHIFT_BITS == 8 | |
| 200 paddd xmm1, XMM_TABLE_0_5 | |
| 201 %else | |
| 202 movdqa xmm3, XMM_TABLE_0_5 | |
| 203 psrad xmm3, SHIFT_BITS - 8 | |
| 204 paddd xmm1, xmm3 | |
| 205 %endif | |
| 206 | |
| 207 ; u[0] = ToByte(xmm1.d[0]); | |
| 208 packssdw xmm1, xmm1 | |
| 209 packuswb xmm1, xmm1 | |
| 210 movd INT_PIXEL32, xmm1 | |
| 211 mov BYTE [PTR_U + INT_WIDTH], INT_PIXEL8 | |
| 212 | |
| 213 ; Calculate v[0] from xmm0. | |
| 214 ; xmm0.d[0] = -FIX(0.071) * B(0) + -FIX(0.368) * G(0); | |
| 215 ; xmm0.d[1] = FIX(0.439) * R(0) + FIX(0.000) * A(0); | |
| 216 pshufd xmm2, XMM_TABLE_UV, 11101110B | |
| 217 pmaddwd xmm0, xmm2 | |
| 218 | |
| 219 ; xmm0.d[0] = (xmm0.d[0] + xmm0.d[1] + FIX(0.5)) / FIX(1) + 128; | |
| 220 pshufd xmm2, xmm0, 11011101B | |
| 221 pshufd xmm1, xmm0, 10001000B | |
| 222 paddd xmm0, xmm2 | |
| 223 paddd xmm0, XMM_TABLE_0_5 | |
| 224 psrad xmm0, SHIFT_BITS | |
| 225 %if SHIFT_BITS == 8 | |
| 226 paddd xmm0, XMM_TABLE_0_5 | |
| 227 %else | |
| 228 movdqa xmm3, XMM_TABLE_0_5 | |
| 229 psrad xmm3, SHIFT_BITS - 8 | |
| 230 paddd xmm0, xmm3 | |
| 231 %endif | |
| 232 | |
| 233 ; v[0] = ToByte(xmm0.d[0]); | |
| 234 packssdw xmm0, xmm0 | |
| 235 packuswb xmm0, xmm0 | |
| 236 movd INT_PIXEL32, xmm0 | |
| 237 mov BYTE [PTR_V + INT_WIDTH], INT_PIXEL8 | |
| 238 | |
| 239 .convert_two_pixels: | |
| 240 ; If the input width is not a multiple of four, read the rightmost two ARGB | |
| 241 ; pixels and convert their colorspace to YUV. This code stores two Y pixels, | |
| 242 ; one U pixel, and one V pixel. | |
| 243 test INT_WIDTH8, 2 / 2 | |
| 244 jz .convert_four_pixels | |
| 245 sub INT_WIDTH8, 2 / 2 | |
| 246 | |
| 247 ; xmm0.w[0] = B(0); xmm0.w[1] = G(0); xmm0.w[2] = R(0); xmm0.w[3] = A(0); | |
| 248 ; xmm0.w[4] = B(1); xmm0.w[5] = G(1); xmm0.w[6] = R(1); xmm0.w[7] = A(1); | |
| 249 %if PIXELSIZE == 4 | |
| 250 movq xmm0, MMWORD [PTR_ARGB + INT_WIDTH * 4 * 2] | |
| 251 %elif PIXELSIZE == 3 | |
| 252 mov _EBX_, INT_WIDTH | |
| 253 lea _EBX_, [_EBX_ + _EBX_ * 2] | |
| 254 movd xmm0, DWORD [PTR_ARGB + _EBX_ * 2] | |
| 255 pinsrw xmm0, WORD [PTR_ARGB + _EBX_ * 2 + 4], 3 | |
| 256 movdqa xmm1, xmm0 | |
| 257 psrldq xmm1, 3 | |
| 258 pslldq xmm1, 3 | |
| 259 pxor xmm0, xmm1 | |
| 260 pslldq xmm1, 1 | |
| 261 por xmm0, xmm1 | |
| 262 %else | |
| 263 %error unsupported pixel size. | |
| 264 %endif | |
| 265 pxor XMM_ZERO, XMM_ZERO | |
| 266 punpcklbw xmm0, XMM_ZERO | |
| 267 | |
| 268 ; xmm2.d[0] = FIX(0.098) * B(0) + FIX(0.504) * G(0); | |
| 269 ; xmm2.d[1] = FIX(0.257) * R(0) + FIX(0.000) * A(0) ; | |
| 270 ; xmm2.d[2] = FIX(0.098) * B(1) + FIX(0.504) * G(1); | |
| 271 ; xmm2.d[3] = FIX(0.257) * R(1) + FIX(0.000) * A(1); | |
| 272 movdqa xmm2, xmm0 | |
| 273 pmaddwd xmm2, XMM_TABLE_Y | |
| 274 | |
| 275 ; xmm2.d[0] = (xmm2.d[0] + xmm2.d[1] + FIX(0.5)) / FIX(1) + 16; | |
| 276 ; xmm2.d[1] = (xmm2.d[2] + xmm2.d[3] + FIX(0.5)) / FIX(1) + 16; | |
| 277 pshufd xmm4, xmm2, 10001000B | |
| 278 pshufd xmm2, xmm2, 11011101B | |
| 279 paddd xmm2, xmm4 | |
| 280 | |
| 281 paddd xmm2, XMM_TABLE_0_5 | |
| 282 psrld xmm2, SHIFT_BITS | |
| 283 movdqa xmm3, XMM_TABLE_0_5 | |
| 284 psrld xmm3, SHIFT_BITS - 5 | |
| 285 paddd xmm2, xmm3 | |
| 286 | |
| 287 ; y[0] = ToByte(xmm2.d[0]); y[1] = ToByte(xmm2.d[1]); | |
| 288 packssdw xmm2, xmm2 | |
| 289 packuswb xmm2, xmm2 | |
| 290 movd INT_PIXEL32, xmm2 | |
| 291 mov WORD [PTR_Y + INT_WIDTH * 2], INT_PIXEL16 | |
| 292 | |
| 293 ; Calculate u[0] from xmm0. | |
| 294 test PTR_U, PTR_U | |
| 295 jz .convert_four_pixels | |
| 296 | |
| 297 ; xmm1.d[0] = FIX(0.439) * B(0) + -FIX(0.291) * G(0); | |
| 298 ; xmm1.d[1] = -FIX(0.148) * R(0) + FIX(0.000) * A(0); | |
| 299 movdqa xmm1, xmm0 | |
| 300 pshufd xmm2, XMM_TABLE_UV, 01000100B | |
| 301 pmaddwd xmm1, xmm2 | |
| 302 | |
| 303 ; xmm1.d[0] = (xmm1.d[0] + xmm1.d[1] + FIX(0.5)) / FIX(1) + 128; | |
| 304 pshufd xmm2, xmm1, 11011101B | |
| 305 pshufd xmm1, xmm1, 10001000B | |
| 306 paddd xmm1, xmm2 | |
| 307 paddd xmm1, XMM_TABLE_0_5 | |
| 308 psrad xmm1, SHIFT_BITS | |
| 309 %if SHIFT_BITS == 8 | |
| 310 paddd xmm1, XMM_TABLE_0_5 | |
| 311 %else | |
| 312 movdqa xmm3, XMM_TABLE_0_5 | |
| 313 psrad xmm3, SHIFT_BITS - 8 | |
| 314 paddd xmm1, xmm3 | |
| 315 %endif | |
| 316 | |
| 317 ; u[0] = ToByte(xmm1.d[0]); | |
| 318 packssdw xmm1, xmm1 | |
| 319 packuswb xmm1, xmm1 | |
| 320 movd INT_PIXEL32, xmm1 | |
| 321 mov BYTE [PTR_U + INT_WIDTH], INT_PIXEL8 | |
| 322 | |
| 323 ; Calculate v[0] from xmm0. | |
| 324 ; xmm0.d[0] = -FIX(0.071) * B(0) + -FIX(0.368) * G(0); | |
| 325 ; xmm0.d[1] = FIX(0.439) * R(0) + FIX(0.000) * A(0); | |
| 326 pshufd xmm2, XMM_TABLE_UV, 11101110B | |
| 327 pmaddwd xmm0, xmm2 | |
| 328 | |
| 329 ; xmm0.d[0] = (xmm0.d[0] + xmm0.d[1] + FIX(0.5)) / FIX(1) + 128; | |
| 330 pshufd xmm2, xmm0, 11011101B | |
| 331 pshufd xmm0, xmm0, 10001000B | |
| 332 paddd xmm0, xmm2 | |
| 333 paddd xmm0, XMM_TABLE_0_5 | |
| 334 psrad xmm0, SHIFT_BITS | |
| 335 %if SHIFT_BITS == 8 | |
| 336 paddd xmm0, XMM_TABLE_0_5 | |
| 337 %else | |
| 338 movdqa xmm3, XMM_TABLE_0_5 | |
| 339 psrad xmm3, SHIFT_BITS - 8 | |
| 340 paddd xmm0, xmm3 | |
| 341 %endif | |
| 342 | |
| 343 ; v[0] = ToByte(xmm0.d[0]); | |
| 344 packssdw xmm0, xmm0 | |
| 345 packuswb xmm0, xmm0 | |
| 346 movd INT_PIXEL32, xmm0 | |
| 347 mov BYTE [PTR_V + INT_WIDTH], INT_PIXEL8 | |
| 348 | |
| 349 .convert_four_pixels: | |
| 350 ; Read four ARGB pixels and convert their colorspace to YUV. This code stores | |
| 351 ; four Y pixels, two U pixels, and two V pixels. | |
| 352 test INT_WIDTH32, INT_WIDTH32 | |
| 353 jz .convert_finish | |
| 354 | |
| 355 .convert_four_pixels_main: | |
| 356 sub INT_WIDTH32, 4 / 2 | |
| 357 | |
| 358 ; xmm0.w[0] = B(0); xmm0.w[1] = G(0); xmm0.w[2] = R(0); xmm0.w[3] = A(0); | |
| 359 ; xmm0.w[4] = B(1); xmm0.w[5] = G(1); xmm0.w[6] = R(1); xmm0.w[7] = A(1); | |
| 360 ; xmm1.w[0] = B(2); xmm1.w[1] = G(2); xmm1.w[2] = R(2); xmm1.w[3] = A(2); | |
| 361 ; xmm1.w[4] = B(3); xmm1.w[5] = G(3); xmm1.w[6] = R(3); xmm1.w[7] = A(3); | |
| 362 %if PIXELSIZE == 4 | |
| 363 movdqa xmm0, XMMWORD [PTR_ARGB + INT_WIDTH * 4 * 2] | |
| 364 %elif PIXELSIZE == 3 | |
| 365 mov _EBX_, INT_WIDTH | |
| 366 lea _EBX_, [_EBX_ + _EBX_ * 2] | |
| 367 movq xmm0, MMWORD [PTR_ARGB + _EBX_ * 2] | |
| 368 movd xmm1, DWORD [PTR_ARGB + _EBX_ * 2 + 8] | |
| 369 shufps xmm0, xmm1, 01000100B | |
| 370 movdqa xmm1, xmm0 | |
| 371 psrldq xmm1, 6 | |
| 372 pslldq xmm1, 6 | |
| 373 pxor xmm0, xmm1 | |
| 374 | |
| 375 movdqa xmm2, xmm0 | |
| 376 psrldq xmm2, 3 | |
| 377 pslldq xmm2, 3 | |
| 378 pxor xmm0, xmm2 | |
| 379 pslldq xmm2, 1 | |
| 380 por xmm0, xmm2 | |
| 381 | |
| 382 movdqa xmm3, xmm1 | |
| 383 psrldq xmm3, 6 + 3 | |
| 384 pslldq xmm3, 6 + 3 | |
| 385 pxor xmm1, xmm3 | |
| 386 pslldq xmm3, 1 | |
| 387 por xmm1, xmm3 | |
| 388 | |
| 389 pslldq xmm1, 2 | |
| 390 por xmm0, xmm1 | |
| 391 %else | |
| 392 %error unsupported pixel size. | |
| 393 %endif | |
| 394 movdqa xmm1, xmm0 | |
| 395 pxor XMM_ZERO, XMM_ZERO | |
| 396 punpckhbw xmm1, XMM_ZERO | |
| 397 punpcklbw xmm0, XMM_ZERO | |
| 398 | |
| 399 ; xmm2.d[0] = FIX(0.098) * B(0) + FIX(0.504) * G(0); | |
| 400 ; xmm2.d[1] = FIX(0.257) * R(0) + FIX(0.000) * A(0); | |
| 401 ; xmm2.d[2] = FIX(0.098) * B(1) + FIX(0.504) * G(1); | |
| 402 ; xmm2.d[3] = FIX(0.257) * R(1) + FIX(0.000) * A(1); | |
| 403 ; xmm3.d[0] = FIX(0.098) * B(2) + FIX(0.504) * G(2); | |
| 404 ; xmm3.d[1] = FIX(0.257) * R(2) + FIX(0.000) * A(2); | |
| 405 ; xmm3.d[2] = FIX(0.098) * B(3) + FIX(0.504) * G(3); | |
| 406 ; xmm3.d[3] = FIX(0.257) * R(3) + FIX(0.000) * A(3); | |
| 407 movdqa xmm2, xmm0 | |
| 408 movdqa xmm3, xmm1 | |
| 409 pmaddwd xmm2, XMM_TABLE_Y | |
| 410 pmaddwd xmm3, XMM_TABLE_Y | |
| 411 | |
| 412 ; xmm2.d[0] = (xmm2.d[0] + xmm2.d[1] + FIX(0.5)) / FIX(1) + 16; | |
| 413 ; xmm2.d[1] = (xmm2.d[2] + xmm2.d[3] + FIX(0.5)) / FIX(1) + 16; | |
| 414 ; xmm2.d[2] = (xmm3.d[0] + xmm3.d[1] + FIX(0.5)) / FIX(1) + 16; | |
| 415 ; xmm2.d[3] = (xmm3.d[2] + xmm3.d[3] + FIX(0.5)) / FIX(1) + 16; | |
| 416 movdqa xmm4, xmm2 | |
| 417 shufps xmm4, xmm3, 10001000B | |
| 418 shufps xmm2, xmm3, 11011101B | |
| 419 paddd xmm2, xmm4 | |
| 420 | |
| 421 paddd xmm2, XMM_TABLE_0_5 | |
| 422 psrad xmm2, SHIFT_BITS | |
| 423 movdqa xmm3, XMM_TABLE_0_5 | |
| 424 psrad xmm3, SHIFT_BITS - 5 | |
| 425 paddd xmm2, xmm3 | |
| 426 | |
| 427 ; y[0] = ToByte(xmm2.d[0]); y[1] = ToByte(xmm2.d[1]); | |
| 428 ; y[2] = ToByte(xmm2.d[2]); y[3] = ToByte(xmm2.d[3]); | |
| 429 packssdw xmm2, xmm2 | |
| 430 packuswb xmm2, xmm2 | |
| 431 movd DWORD [PTR_Y + INT_WIDTH * 2], xmm2 | |
| 432 | |
| 433 ; Retrieve (R(0), G(0), B(0), A(0)) and (R(2), G(2), B(2), A(2)) from xmm1 | |
| 434 ; and xmm1. | |
| 435 ; xmm0.w[0] = B(0); xmm0.w[1] = G(0); xmm0.w[2] = R(0); xmm0.w[3] = A(0); | |
| 436 ; xmm0.w[4] = B(2); xmm0.w[5] = G(2); xmm0.w[6] = R(2); xmm0.w[7] = A(2); | |
| 437 shufps xmm0, xmm1, 01000100B | |
| 438 | |
| 439 ; Calculate u[0] and u[2] from xmm0. | |
| 440 test PTR_U, PTR_U | |
| 441 jz .convert_four_pixels_next | |
| 442 | |
| 443 ; xmm1.d[0] = FIX(0.439) * B(0) + -FIX(0.291) * G(0); | |
| 444 ; xmm1.d[1] = -FIX(0.148) * R(0) + FIX(0.000) * A(0); | |
| 445 ; xmm1.d[2] = FIX(0.439) * B(2) + -FIX(0.291) * G(2); | |
| 446 ; xmm1.d[3] = -FIX(0.148) * R(2) + FIX(0.000) * A(2); | |
| 447 movdqa xmm1, xmm0 | |
| 448 pshufd xmm2, XMM_TABLE_UV, 01000100B | |
| 449 pmaddwd xmm1, xmm2 | |
| 450 | |
| 451 ; xmm1.d[0] = (xmm1.d[0] + xmm1.d[1] + FIX(0.5)) / FIX(1) + 128; | |
| 452 ; xmm1.d[1] = (xmm1.d[2] + xmm1.d[3] + FIX(0.5)) / FIX(1) + 128; | |
| 453 pshufd xmm2, xmm1, 11011101B | |
| 454 pshufd xmm1, xmm1, 10001000B | |
| 455 paddd xmm1, xmm2 | |
| 456 paddd xmm1, XMM_TABLE_0_5 | |
| 457 psrad xmm1, SHIFT_BITS | |
| 458 %if SHIFT_BITS == 8 | |
| 459 paddd xmm1, XMM_TABLE_0_5 | |
| 460 %else | |
| 461 movdqa xmm3, XMM_TABLE_0_5 | |
| 462 psrad xmm3, SHIFT_BITS - 8 | |
| 463 paddd xmm1, xmm3 | |
| 464 %endif | |
| 465 | |
| 466 ; u[0] = ToByte(xmm1.d[0]); u[1] = ToByte(xmm1.d[1]); | |
| 467 packssdw xmm1, xmm1 | |
| 468 packuswb xmm1, xmm1 | |
| 469 movd INT_PIXEL32, xmm1 | |
| 470 mov WORD [PTR_U + INT_WIDTH], INT_PIXEL16 | |
| 471 | |
| 472 ; Calculate v[0] and v[2] from xmm0. | |
| 473 ; xmm1.d[0] = -FIX(0.071) * B(0) + -FIX(0.368) * G(0); | |
| 474 ; xmm1.d[1] = FIX(0.439) * R(0) + FIX(0.000) * A(0); | |
| 475 ; xmm1.d[2] = -FIX(0.071) * B(2) + -FIX(0.368) * G(2); | |
| 476 ; xmm1.d[3] = FIX(0.439) * R(2) + FIX(0.000) * A(2); | |
| 477 movdqa xmm1, xmm0 | |
| 478 pshufd xmm2, XMM_TABLE_UV, 11101110B | |
| 479 pmaddwd xmm1, xmm2 | |
| 480 | |
| 481 ; xmm1.d[0] = (xmm1.d[0] + xmm1.d[1] + FIX(0.5)) / FIX(1) + 128; | |
| 482 ; xmm1.d[1] = (xmm1.d[2] + xmm1.d[3] + FIX(0.5)) / FIX(1) + 128; | |
| 483 pshufd xmm2, xmm1, 11011101B | |
| 484 pshufd xmm1, xmm1, 10001000B | |
| 485 paddd xmm1, xmm2 | |
| 486 paddd xmm1, XMM_TABLE_0_5 | |
| 487 psrad xmm1, SHIFT_BITS | |
| 488 %if SHIFT_BITS == 8 | |
| 489 paddd xmm1, XMM_TABLE_0_5 | |
| 490 %else | |
| 491 movdqa xmm3, XMM_TABLE_0_5 | |
| 492 psrad xmm3, SHIFT_BITS - 8 | |
| 493 paddd xmm1, xmm3 | |
| 494 %endif | |
| 495 | |
| 496 ; v[0] = ToByte(xmm1.d[0]); v[1] = ToByte(xmm1.d[1]); | |
| 497 packssdw xmm1, xmm1 | |
| 498 packuswb xmm1, xmm1 | |
| 499 movd INT_PIXEL32, xmm1 | |
| 500 mov WORD [PTR_V + INT_WIDTH], INT_PIXEL16 | |
| 501 | |
| 502 .convert_four_pixels_next: | |
| 503 test INT_WIDTH32, INT_WIDTH32 | |
| 504 jnz .convert_four_pixels_main | |
| 505 | |
| 506 .convert_finish: | |
| 507 ; Restore the registers used in this function, the stack frame, and the stack | |
| 508 ; pointer. | |
| 509 pop _EDI_ | |
| 510 pop _ESI_ | |
| 511 pop _EBX_ | |
| 512 | |
| 513 mov _ESP_, _EBP_ | |
| 514 pop _EBP_ | |
| 515 | |
| 516 ret | |
| 517 | |
| 518 ; For some reason, the OS X linker does not honor the request to align the | |
| 519 ; segment unless we do this. | |
| 520 align 16 | |
| OLD | NEW |
