MPEG4_DECORE

/************************************************************************** * * * This code has been developed by John Funnell. This software is an * * implementation of a part of one or more MPEG-4 Video tools as * * specified in ISO/IEC 14496-2 standard. Those intending to use this * * software module in hardware or software products are advised that its * * use may infringe existing patents or copyrights, and any such use * * would be at such party's own risk. The original developer of this * * software module and his/her company, and subsequent editors and their * * companies (including Project Mayo), will have no liability for use of * * this software or modifications or derivatives thereof. * * * * Project Mayo gives users of the Codec a license to this software * * module or modifications thereof for use in hardware or software * * products claiming conformance to the MPEG-4 Video Standard as * * described in the Open DivX license. * * * * The complete Open DivX license can be found at * * http://www.projectmayo.com/opendivx/license.php * * * **************************************************************************/ /** * Copyright (C) 2001 - Project Mayo * * John Funnell * Andrea Graziani * * DivX Advanced Research Center <darc@projectmayo.com> * **/ // yuv2rgb_mmx.c // /* MMX version of colourspace conversion 13 Feb 2001 - John Funnell, added -height feature to invert output, fixed 24-bit overwrite TODO: 1. Extract common part of from 32, 24 and 16 bit conversions into inline func to avoid duplication of the core MMX code. 2. Experiment with using prefetch instructions to improve performance Using matrix of SMPTE 170M This is what we're doing: Step 1. Y -= 16 U -= 128 V -= 128 Step 2. Y /= 219 U /= 224 V /= 224 Step 3. now we want the inverse of this matrix: / 0.299 0.114 0.587 \ | -0.169 0.500 -0.331 | \ 0.500 -0.081 -0.419 / which is, approximately: / 2568 0 3343 \ | 2568 f36e e5e2 | / 65536 * 8 \ 2568 40cf 0 / including the multiplies in Step 2 */ #include <memory.h> // for memset() & memcpy() #include "portab.h" #include "yuv2rgb.h" /** * **/ #define MAXIMUM_Y_WIDTH 800 #define _USE_PREFETCH /* static constants */ /* colourspace conversion matrix values */ static uint64_t mmw_mult_Y = 0x2568256825682568; static uint64_t mmw_mult_U_G = 0xf36ef36ef36ef36e; static uint64_t mmw_mult_U_B = 0x40cf40cf40cf40cf; static uint64_t mmw_mult_V_R = 0x3343334333433343; static uint64_t mmw_mult_V_G = 0xe5e2e5e2e5e2e5e2; /* various masks and other constants */ static uint64_t mmb_0x10 = 0x1010101010101010; static uint64_t mmw_0x0080 = 0x0080008000800080; static uint64_t mmw_0x00ff = 0x00ff00ff00ff00ff; static uint64_t mmw_cut_red = 0x7c007c007c007c00; static uint64_t mmw_cut_green = 0x03e003e003e003e0; static uint64_t mmw_cut_blue = 0x001f001f001f001f; /**** YUV -> RGB conversion, 32-bit output ****/ /* if height_y is negative then the output image will be inverted */ /* note: _stride_out parameter is ignored in yuv to rgb conversion it's assumed that stride_out = 4 * width_y for the 32 bit color bitmap */ void yuv2rgb_32(uint8_t *puc_y, int stride_y, uint8_t *puc_u, uint8_t *puc_v, int stride_uv, uint8_t *puc_out, int width_y, int height_y, unsigned int _stride_out) { int y, horiz_count; int stride_out = width_y * 4; if (height_y < 0) { /* we are flipping our output upside-down */ height_y = -height_y; puc_y += (height_y - 1) * stride_y ; puc_u += (height_y/2 - 1) * stride_uv; puc_v += (height_y/2 - 1) * stride_uv; stride_y = -stride_y; stride_uv = -stride_uv; } horiz_count = -(width_y >> 3); for (y=0; y<height_y; y++) { _asm { push eax push ebx push ecx push edx push edi mov eax, puc_out mov ebx, puc_y mov ecx, puc_u mov edx, puc_v mov edi, horiz_count horiz_loop: movd mm2, [ecx] pxor mm7, mm7 movd mm3, [edx] punpcklbw mm2, mm7 ; mm2 = __u3__u2__u1__u0 movq mm0, [ebx] ; mm0 = y7y6y5y4y3y2y1y0 punpcklbw mm3, mm7 ; mm3 = __v3__v2__v1__v0 movq mm1, mmw_0x00ff ; mm1 = 00ff00ff00ff00ff psubusb mm0, mmb_0x10 ; mm0 -= 16 psubw mm2, mmw_0x0080 ; mm2 -= 128 pand mm1, mm0 ; mm1 = __y6__y4__y2__y0 psubw mm3, mmw_0x0080 ; mm3 -= 128 psllw mm1, 3 ; mm1 *= 8 psrlw mm0, 8 ; mm0 = __y7__y5__y3__y1 psllw mm2, 3 ; mm2 *= 8 pmulhw mm1, mmw_mult_Y ; mm1 *= luma coeff psllw mm0, 3 ; mm0 *= 8 psllw mm3, 3 ; mm3 *= 8 movq mm5, mm3 ; mm5 = mm3 = v pmulhw mm5, mmw_mult_V_R ; mm5 = red chroma movq mm4, mm2 ; mm4 = mm2 = u pmulhw mm0, mmw_mult_Y ; mm0 *= luma coeff movq mm7, mm1 ; even luma part pmulhw mm2, mmw_mult_U_G ; mm2 *= u green coeff paddsw mm7, mm5 ; mm7 = luma + chroma __r6__r4__r2__r0 pmulhw mm3, mmw_mult_V_G ; mm3 *= v green coeff packuswb mm7, mm7 ; mm7 = r6r4r2r0r6r4r2r0 pmulhw mm4, mmw_mult_U_B ; mm4 = blue chroma paddsw mm5, mm0 ; mm5 = luma + chroma __r7__r5__r3__r1 packuswb mm5, mm5 ; mm6 = r7r5r3r1r7r5r3r1 paddsw mm2, mm3 ; mm2 = green chroma movq mm3, mm1 ; mm3 = __y6__y4__y2__y0 movq mm6, mm1 ; mm6 = __y6__y4__y2__y0 paddsw mm3, mm4 ; mm3 = luma + chroma __b6__b4__b2__b0 paddsw mm6, mm2 ; mm6 = luma + chroma __g6__g4__g2__g0 punpcklbw mm7, mm5 ; mm7 = r7r6r5r4r3r2r1r0 paddsw mm2, mm0 ; odd luma part plus chroma part __g7__g5__g3__g1 packuswb mm6, mm6 ; mm2 = g6g4g2g0g6g4g2g0 packuswb mm2, mm2 ; mm2 = g7g5g3g1g7g5g3g1 packuswb mm3, mm3 ; mm3 = b6b4b2b0b6b4b2b0 paddsw mm4, mm0 ; odd luma part plus chroma part __b7__b5__b3__b1 packuswb mm4, mm4 ; mm4 = b7b5b3b1b7b5b3b1 punpcklbw mm6, mm2 ; mm6 = g7g6g5g4g3g2g1g0 punpcklbw mm3, mm4 ; mm3 = b7b6b5b4b3b2b1b0 /* 32-bit shuffle.... */ pxor mm0, mm0 ; is this needed? movq mm1, mm6 ; mm1 = g7g6g5g4g3g2g1g0 punpcklbw mm1, mm0 ; mm1 = __g3__g2__g1__g0 movq mm0, mm3 ; mm0 = b7b6b5b4b3b2b1b0 punpcklbw mm0, mm7 ; mm0 = r3b3r2b2r1b1r0b0 movq mm2, mm0 ; mm2 = r3b3r2b2r1b1r0b0 punpcklbw mm0, mm1 ; mm0 = __r1g1b1__r0g0b0 punpckhbw mm2, mm1 ; mm2 = __r3g3b3__r2g2b2 /* 32-bit save... */ movq [eax], mm0 ; eax[0] = __r1g1b1__r0g0b0 movq mm1, mm6 ; mm1 = g7g6g5g4g3g2g1g0 movq 8[eax], mm2 ; eax[8] = __r3g3b3__r2g2b2 /* 32-bit shuffle.... */ pxor mm0, mm0