--- xvid.c 2002/12/30 10:49:17 1.33.2.19 +++ xvid.c 2004/04/12 14:05:08 1.51 @@ -3,14 +3,7 @@ * XVID MPEG-4 VIDEO CODEC * - Native API implementation - * - * This program 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, will have no liability for use of this - * software or modifications or derivatives thereof. + * Copyright(C) 2001-2004 Peter Ross * * This program is free software ; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -26,21 +19,15 @@ * along with this program ; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - ****************************************************************************/ - -/***************************************************************************** - * - * History - * - * - 23.06.2002 added XVID_CPU_CHKONLY - * - 17.03.2002 Added interpolate8x8_halfpel_hv_xmm - * - 22.12.2001 API change: added xvid_init() - Isibaar - * - 16.12.2001 inital version; (c)2001 peter ross - * - * $Id: xvid.c,v 1.33.2.19 2002/12/30 10:49:17 suxen_drol Exp $ + * $Id: xvid.c,v 1.51 2004/04/12 14:05:08 edgomez Exp $ * ****************************************************************************/ +#include +#include +#include +#include + #include "xvid.h" #include "decoder.h" #include "encoder.h" @@ -52,81 +39,117 @@ #include "image/reduced.h" #include "utils/mem_transfer.h" #include "utils/mbfunctions.h" -#include "quant/quant_h263.h" -#include "quant/quant_mpeg4.h" +#include "quant/quant.h" #include "motion/motion.h" #include "motion/sad.h" #include "utils/emms.h" #include "utils/timer.h" #include "bitstream/mbcoding.h" +#include "image/qpel.h" +#include "image/postprocessing.h" -#if defined(ARCH_X86) && defined(EXPERIMENTAL_SSE2_CODE) - -#ifdef WIN32 -#include -#else -#include -#include +#if defined(_DEBUG) +unsigned int xvid_debug = 0; /* xvid debug mask */ #endif +#if defined(ARCH_IS_IA32) && defined(_MSC_VER) +# include +#elif defined(ARCH_IS_IA32) || defined(ARCH_IS_PPC) +# include +# include -#ifndef WIN32 - -static jmp_buf mark; + static jmp_buf mark; -static void -sigill_handler(int signal) -{ - longjmp(mark, 1); -} + static void + sigill_handler(int signal) + { + longjmp(mark, 1); + } #endif /* -calls the funcptr, and returns whether SIGILL (illegal instruction) was signalled -return values: --1 : could not determine -0 : SIGILL was *not* signalled -1 : SIGILL was signalled -*/ - -int + * Calls the funcptr, and returns whether SIGILL (illegal instruction) was + * signalled + * + * Return values: + * -1 : could not determine + * 0 : SIGILL was *not* signalled + * 1 : SIGILL was signalled + */ +#if defined(ARCH_IS_IA32) && defined(_MSC_VER) +static int sigill_check(void (*func)()) { -#ifdef WIN32 _try { func(); - } - _except(EXCEPTION_EXECUTE_HANDLER) { + } _except(EXCEPTION_EXECUTE_HANDLER) { if (_exception_code() == STATUS_ILLEGAL_INSTRUCTION) - return 1; + return(1); } - return 0; -#else - void * old_handler; + return(0); +} +#elif defined(ARCH_IS_IA32) || defined(ARCH_IS_PPC) +static int +sigill_check(void (*func)()) +{ + void *old_handler; int jmpret; - + /* Set our SIGILL handler */ old_handler = signal(SIGILL, sigill_handler); - if (old_handler == SIG_ERR) - { - return -1; + + /* Check for error */ + if (old_handler == SIG_ERR) { + return(-1); } - jmpret = setjmp(mark); - if (jmpret == 0) - { + /* Save stack context, so if func triggers a SIGILL, we can still roll + * back to a valid CPU state */ + jmpret = setjmp(mark); + + /* If setjmp returned directly, then its returned value is 0, and we still + * have to test the passed func. Otherwise it means the stack context has + * been restored by a longjmp() call, which in our case happens only in the + * signal handler */ + if (jmpret == 0) { func(); } + /* Restore old signal handler */ signal(SIGILL, old_handler); - return jmpret; -#endif + return(jmpret); } #endif + +/* detect cpu flags */ +static unsigned int +detect_cpu_flags() +{ + /* enable native assembly optimizations by default */ + unsigned int cpu_flags = XVID_CPU_ASM; + +#if defined(ARCH_IS_IA32) + cpu_flags |= check_cpu_features(); + if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) + cpu_flags &= ~XVID_CPU_SSE; + + if ((cpu_flags & XVID_CPU_SSE2) && sigill_check(sse2_os_trigger)) + cpu_flags &= ~XVID_CPU_SSE2; +#endif + +#if defined(ARCH_IS_PPC) + if (!sigill_check(altivec_trigger)) + cpu_flags |= XVID_CPU_ALTIVEC; +#endif + + return cpu_flags; +} + + /***************************************************************************** * XviD Init Entry point * @@ -134,7 +157,7 @@ * to the CPU features forced by the library client or autodetected (depending * on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all * image colorspace transformation tables. - * + * * Returned value : XVID_ERR_OK * + API_VERSION in the input XVID_INIT_PARAM structure * + core build " " " " " @@ -142,43 +165,15 @@ ****************************************************************************/ -static -int xvid_init_init(XVID_INIT_PARAM * init_param) +static +int xvid_gbl_init(xvid_gbl_init_t * init) { - int cpu_flags; - - /* Inform the client the API version */ - init_param->api_version = API_VERSION; - - /* Inform the client the core build - unused because we're still alpha */ - init_param->core_build = 1000; - - /* Do we have to force CPU features ? */ - if ((init_param->cpu_flags & XVID_CPU_FORCE)) { - - cpu_flags = init_param->cpu_flags; + unsigned int cpu_flags; - } else { - - cpu_flags = check_cpu_features(); - -#if defined(ARCH_X86) && defined(EXPERIMENTAL_SSE2_CODE) - if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) - cpu_flags &= ~XVID_CPU_SSE; - - if ((cpu_flags & XVID_CPU_SSE2) && sigill_check(sse2_os_trigger)) - cpu_flags &= ~XVID_CPU_SSE2; -#endif - } - - if ((init_param->cpu_flags & XVID_CPU_CHKONLY)) - { - init_param->cpu_flags = cpu_flags; - return XVID_ERR_OK; - } - - init_param->cpu_flags = cpu_flags; + if (XVID_VERSION_MAJOR(init->version) != 1) /* v1.x.x */ + return XVID_ERR_VERSION; + cpu_flags = (init->cpu_flags & XVID_CPU_FORCE) ? init->cpu_flags : detect_cpu_flags(); /* Initialize the function pointers */ idct_int32_init(); @@ -194,21 +189,27 @@ /* Restore FPU context : emms_c is a nop functions */ emms = emms_c; + /* Qpel stuff */ + xvid_QP_Funcs = &xvid_QP_Funcs_C; + xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_C; + xvid_Init_QP(); + /* Quantization functions */ - quant_intra = quant_intra_c; - dequant_intra = dequant_intra_c; - quant_inter = quant_inter_c; - dequant_inter = dequant_inter_c; - - quant4_intra = quant4_intra_c; - dequant4_intra = dequant4_intra_c; - quant4_inter = quant4_inter_c; - dequant4_inter = dequant4_inter_c; + quant_h263_intra = quant_h263_intra_c; + quant_h263_inter = quant_h263_inter_c; + dequant_h263_intra = dequant_h263_intra_c; + dequant_h263_inter = dequant_h263_inter_c; + + quant_mpeg_intra = quant_mpeg_intra_c; + quant_mpeg_inter = quant_mpeg_inter_c; + dequant_mpeg_intra = dequant_mpeg_intra_c; + dequant_mpeg_inter = dequant_mpeg_inter_c; /* Block transfer related functions */ transfer_8to16copy = transfer_8to16copy_c; transfer_16to8copy = transfer_16to8copy_c; transfer_8to16sub = transfer_8to16sub_c; + transfer_8to16subro = transfer_8to16subro_c; transfer_8to16sub2 = transfer_8to16sub2_c; transfer_16to8add = transfer_16to8add_c; transfer8x8_copy = transfer8x8_copy_c; @@ -235,17 +236,14 @@ interpolate8x8_avg2 = interpolate8x8_avg2_c; interpolate8x8_avg4 = interpolate8x8_avg4_c; - /* reduced resoltuion */ + /* postprocessing */ + image_brightness = image_brightness_c; + /* reduced resolution */ copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C; add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C; -#ifdef ARCH_X86 - vfilter_31 = xvid_VFilter_31_x86; - hfilter_31 = xvid_HFilter_31_x86; -#else vfilter_31 = xvid_VFilter_31_C; hfilter_31 = xvid_HFilter_31_C; -#endif filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_C; filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_C; @@ -260,6 +258,7 @@ bgra_to_yv12 = bgra_to_yv12_c; abgr_to_yv12 = abgr_to_yv12_c; rgba_to_yv12 = rgba_to_yv12_c; + argb_to_yv12 = argb_to_yv12_c; yuyv_to_yv12 = yuyv_to_yv12_c; uyvy_to_yv12 = uyvy_to_yv12_c; @@ -269,10 +268,10 @@ bgrai_to_yv12 = bgrai_to_yv12_c; abgri_to_yv12 = abgri_to_yv12_c; rgbai_to_yv12 = rgbai_to_yv12_c; + argbi_to_yv12 = argbi_to_yv12_c; yuyvi_to_yv12 = yuyvi_to_yv12_c; uyvyi_to_yv12 = uyvyi_to_yv12_c; - /* All colorspace transformation functions YV12->User format */ yv12_to_rgb555 = yv12_to_rgb555_c; yv12_to_rgb565 = yv12_to_rgb565_c; @@ -280,30 +279,36 @@ yv12_to_bgra = yv12_to_bgra_c; yv12_to_abgr = yv12_to_abgr_c; yv12_to_rgba = yv12_to_rgba_c; + yv12_to_argb = yv12_to_argb_c; yv12_to_yuyv = yv12_to_yuyv_c; yv12_to_uyvy = yv12_to_uyvy_c; - + yv12_to_rgb555i = yv12_to_rgb555i_c; yv12_to_rgb565i = yv12_to_rgb565i_c; yv12_to_bgri = yv12_to_bgri_c; yv12_to_bgrai = yv12_to_bgrai_c; yv12_to_abgri = yv12_to_abgri_c; yv12_to_rgbai = yv12_to_rgbai_c; + yv12_to_argbi = yv12_to_argbi_c; yv12_to_yuyvi = yv12_to_yuyvi_c; yv12_to_uyvyi = yv12_to_uyvyi_c; /* Functions used in motion estimation algorithms */ - calc_cbp = calc_cbp_c; - sad16 = sad16_c; - sad8 = sad8_c; - sad16bi = sad16bi_c; - sad8bi = sad8bi_c; - dev16 = dev16_c; - sad16v = sad16v_c; - -// Halfpel8_Refine = Halfpel8_Refine_c; - -#ifdef ARCH_X86 + calc_cbp = calc_cbp_c; + sad16 = sad16_c; + sad8 = sad8_c; + sad16bi = sad16bi_c; + sad8bi = sad8bi_c; + dev16 = dev16_c; + sad16v = sad16v_c; + sse8_16bit = sse8_16bit_c; + +#if defined(ARCH_IS_IA32) + + if ((cpu_flags & XVID_CPU_ASM)) { + vfilter_31 = xvid_VFilter_31_x86; + hfilter_31 = xvid_HFilter_31_x86; + } if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) || (cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) || @@ -313,27 +318,32 @@ emms = emms_mmx; } - if ((cpu_flags & XVID_CPU_MMX) > 0) { + if ((cpu_flags & XVID_CPU_MMX)) { /* Forward and Inverse Discrete Cosine Transformation functions */ - fdct = fdct_mmx; + fdct = fdct_mmx_skal; idct = idct_mmx; + /* Qpel stuff */ + xvid_QP_Funcs = &xvid_QP_Funcs_mmx; + xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_mmx; + /* Quantization related functions */ - quant_intra = quant_intra_mmx; - dequant_intra = dequant_intra_mmx; - quant_inter = quant_inter_mmx; - dequant_inter = dequant_inter_mmx; - - quant4_intra = quant4_intra_mmx; - dequant4_intra = dequant4_intra_mmx; - quant4_inter = quant4_inter_mmx; - dequant4_inter = dequant4_inter_mmx; + quant_h263_intra = quant_h263_intra_mmx; + quant_h263_inter = quant_h263_inter_mmx; + dequant_h263_intra = dequant_h263_intra_mmx; + dequant_h263_inter = dequant_h263_inter_mmx; + + quant_mpeg_intra = quant_mpeg_intra_mmx; + quant_mpeg_inter = quant_mpeg_inter_mmx; + dequant_mpeg_intra = dequant_mpeg_intra_mmx; + dequant_mpeg_inter = dequant_mpeg_inter_mmx; /* Block related functions */ transfer_8to16copy = transfer_8to16copy_mmx; transfer_16to8copy = transfer_16to8copy_mmx; transfer_8to16sub = transfer_8to16sub_mmx; + transfer_8to16subro = transfer_8to16subro_mmx; transfer_8to16sub2 = transfer_8to16sub2_mmx; transfer_16to8add = transfer_16to8add_mmx; transfer8x8_copy = transfer8x8_copy_mmx; @@ -352,6 +362,9 @@ interpolate8x8_avg2 = interpolate8x8_avg2_mmx; interpolate8x8_avg4 = interpolate8x8_avg4_mmx; + /* postprocessing */ + image_brightness = image_brightness_mmx; + /* reduced resolution */ copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_mmx; add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_mmx; @@ -371,7 +384,7 @@ yv12_to_bgra = yv12_to_bgra_mmx; yv12_to_yuyv = yv12_to_yuyv_mmx; yv12_to_uyvy = yv12_to_uyvy_mmx; - + yv12_to_yuyvi = yv12_to_yuyvi_mmx; yv12_to_uyvyi = yv12_to_uyvyi_mmx; @@ -383,11 +396,13 @@ sad8bi = sad8bi_mmx; dev16 = dev16_mmx; sad16v = sad16v_mmx; - + sse8_16bit = sse8_16bit_mmx; } /* these 3dnow functions are faster than mmx, but slower than xmm. */ - if ((cpu_flags & XVID_CPU_3DNOW) > 0) { + if ((cpu_flags & XVID_CPU_3DNOW)) { + + emms = emms_3dn; /* ME functions */ sad16bi = sad16bi_3dn; @@ -398,9 +413,10 @@ } - if ((cpu_flags & XVID_CPU_MMXEXT) > 0) { + if ((cpu_flags & XVID_CPU_MMXEXT)) { - /* Inverse DCT */ + /* DCT */ + fdct = fdct_xmm_skal; idct = idct_xmm; /* Interpolation */ @@ -413,11 +429,11 @@ add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm; /* Quantization */ - quant4_intra = quant4_intra_xmm; - quant4_inter = quant4_inter_xmm; + quant_mpeg_intra = quant_mpeg_intra_xmm; + quant_mpeg_inter = quant_mpeg_inter_xmm; - dequant_intra = dequant_intra_xmm; - dequant_inter = dequant_inter_xmm; + dequant_h263_intra = dequant_h263_intra_xmm; + dequant_h263_inter = dequant_h263_inter_xmm; /* Buffer transfer */ transfer_8to16sub2 = transfer_8to16sub2_xmm; @@ -436,7 +452,7 @@ sad16v = sad16v_xmm; } - if ((cpu_flags & XVID_CPU_3DNOW) > 0) { + if ((cpu_flags & XVID_CPU_3DNOW)) { /* Interpolation */ interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn; @@ -444,70 +460,72 @@ interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn; } - if ((cpu_flags & XVID_CPU_3DNOWEXT) > 0) { - - /* Inverse DCT */ - idct = idct_3dne; + if ((cpu_flags & XVID_CPU_3DNOWEXT)) { /* Buffer transfer */ transfer_8to16copy = transfer_8to16copy_3dne; transfer_16to8copy = transfer_16to8copy_3dne; transfer_8to16sub = transfer_8to16sub_3dne; - transfer_8to16sub2 = transfer_8to16sub2_3dne; + transfer_8to16subro = transfer_8to16subro_3dne; transfer_16to8add = transfer_16to8add_3dne; transfer8x8_copy = transfer8x8_copy_3dne; - /* Quantization */ - dequant4_intra = dequant4_intra_3dne; - dequant4_inter = dequant4_inter_3dne; - quant_intra = quant_intra_3dne; - quant_inter = quant_inter_3dne; - dequant_intra = dequant_intra_3dne; - dequant_inter = dequant_inter_3dne; - - /* ME functions */ - calc_cbp = calc_cbp_3dne; - sad16 = sad16_3dne; - sad8 = sad8_3dne; - sad16bi = sad16bi_3dne; - sad8bi = sad8bi_3dne; - dev16 = dev16_3dne; - - /* Interpolation */ - interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dne; - interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne; - interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne; + if ((cpu_flags & XVID_CPU_MMXEXT)) { + /* Inverse DCT */ + idct = idct_3dne; + + /* Buffer transfer */ + transfer_8to16sub2 = transfer_8to16sub2_3dne; + + /* Interpolation */ + interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dne; + interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne; + interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne; + + /* Quantization */ + quant_h263_intra = quant_h263_intra_3dne; /* cmov only */ + quant_h263_inter = quant_h263_inter_3dne; + dequant_mpeg_intra = dequant_mpeg_intra_3dne; /* cmov only */ + dequant_mpeg_inter = dequant_mpeg_inter_3dne; + dequant_h263_intra = dequant_h263_intra_3dne; + dequant_h263_inter = dequant_h263_inter_3dne; + + /* ME functions */ + calc_cbp = calc_cbp_3dne; + + sad16 = sad16_3dne; + sad8 = sad8_3dne; + sad16bi = sad16bi_3dne; + sad8bi = sad8bi_3dne; + dev16 = dev16_3dne; + } } - - if ((cpu_flags & XVID_CPU_SSE2) > 0) { -#ifdef EXPERIMENTAL_SSE2_CODE + if ((cpu_flags & XVID_CPU_SSE2)) { calc_cbp = calc_cbp_sse2; /* Quantization */ - quant_intra = quant_intra_sse2; - dequant_intra = dequant_intra_sse2; - quant_inter = quant_inter_sse2; - dequant_inter = dequant_inter_sse2; + quant_h263_intra = quant_h263_intra_sse2; + quant_h263_inter = quant_h263_inter_sse2; + dequant_h263_intra = dequant_h263_intra_sse2; + dequant_h263_inter = dequant_h263_inter_sse2; - /* ME */ + /* SAD operators */ sad16 = sad16_sse2; dev16 = dev16_sse2; - /* Forward and Inverse DCT */ - idct = idct_sse2; - fdct = fdct_sse2; -#endif + /* DCT operators + * no iDCT because it's not "Walken matching" */ + fdct = fdct_sse2_skal; } +#endif /* ARCH_IS_IA32 */ -#endif - -#ifdef ARCH_IA64 - if ((cpu_flags & XVID_CPU_IA64) > 0) { //use assembler routines? +#if defined(ARCH_IS_IA64) + if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */ idct_ia64_init(); fdct = fdct_ia64; - idct = idct_ia64; //not yet working, crashes + idct = idct_ia64; /*not yet working, crashes */ interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_ia64; interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_ia64; interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_ia64; @@ -515,305 +533,167 @@ sad16bi = sad16bi_ia64; sad8 = sad8_ia64; dev16 = dev16_ia64; -// Halfpel8_Refine = Halfpel8_Refine_ia64; - quant_intra = quant_intra_ia64; - dequant_intra = dequant_intra_ia64; - quant_inter = quant_inter_ia64; - dequant_inter = dequant_inter_ia64; +/* Halfpel8_Refine = Halfpel8_Refine_ia64; */ + quant_h263_intra = quant_h263_intra_ia64; + quant_h263_inter = quant_h263_inter_ia64; + dequant_h263_intra = dequant_h263_intra_ia64; + dequant_h263_inter = dequant_h263_inter_ia64; transfer_8to16copy = transfer_8to16copy_ia64; transfer_16to8copy = transfer_16to8copy_ia64; transfer_8to16sub = transfer_8to16sub_ia64; transfer_8to16sub2 = transfer_8to16sub2_ia64; transfer_16to8add = transfer_16to8add_ia64; transfer8x8_copy = transfer8x8_copy_ia64; - DEBUG("Using IA-64 assembler routines.\n"); } -#endif +#endif -#ifdef ARCH_PPC -#ifdef ARCH_PPC_ALTIVEC - calc_cbp = calc_cbp_altivec; - fdct = fdct_altivec; - idct = idct_altivec; - sadInit = sadInit_altivec; - sad16 = sad16_altivec; - sad8 = sad8_altivec; - dev16 = dev16_altivec; -#else - calc_cbp = calc_cbp_ppc; +#if defined(ARCH_IS_PPC) + if ((cpu_flags & XVID_CPU_ALTIVEC)) { + /* sad operators */ + sad16 = sad16_altivec_c; + sad16bi = sad16bi_altivec_c; + sad8 = sad8_altivec_c; + dev16 = dev16_altivec_c; + + sse8_16bit = sse8_16bit_altivec_c; + + /* mem transfer */ + transfer_8to16copy = transfer_8to16copy_altivec_c; + transfer_16to8copy = transfer_16to8copy_altivec_c; + transfer_8to16sub = transfer_8to16sub_altivec_c; + transfer_8to16subro = transfer_8to16subro_altivec_c; + transfer_8to16sub2 = transfer_8to16sub2_altivec_c; + transfer_16to8add = transfer_16to8add_altivec_c; + transfer8x8_copy = transfer8x8_copy_altivec_c; + + /* Inverse DCT */ + idct = idct_altivec_c; + + /* Interpolation */ + interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_altivec_c; + interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_altivec_c; + interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_altivec_c; + + interpolate8x8_avg2 = interpolate8x8_avg2_altivec_c; + interpolate8x8_avg4 = interpolate8x8_avg4_altivec_c; + + interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_altivec_c; + + /* Colorspace conversion */ + bgra_to_yv12 = bgra_to_yv12_altivec_c; + abgr_to_yv12 = abgr_to_yv12_altivec_c; + rgba_to_yv12 = rgba_to_yv12_altivec_c; + argb_to_yv12 = argb_to_yv12_altivec_c; + + yuyv_to_yv12 = yuyv_to_yv12_altivec_c; + uyvy_to_yv12 = uyvy_to_yv12_altivec_c; + + yv12_to_yuyv = yv12_to_yuyv_altivec_c; + yv12_to_uyvy = yv12_to_uyvy_altivec_c; + + /* Quantization */ + quant_h263_intra = quant_h263_intra_altivec_c; + quant_h263_inter = quant_h263_inter_altivec_c; + dequant_h263_intra = dequant_h263_intra_altivec_c; + dequant_h263_inter = dequant_h263_inter_altivec_c; + } #endif + +#if defined(_DEBUG) + xvid_debug = init->debug; #endif - return XVID_ERR_OK; + return(0); } - static int -xvid_init_convert(XVID_INIT_CONVERTINFO* convert) -{ - // const int flip1 = (convert->input.colorspace & XVID_CSP_VFLIP) ^ (convert->output.colorspace & XVID_CSP_VFLIP); - const int width = convert->width; - const int height = convert->height; - const int width2 = convert->width/2; - const int height2 = convert->height/2; - IMAGE img; - - switch (convert->input.colorspace & ~XVID_CSP_VFLIP) - { - case XVID_CSP_YV12 : - img.y = convert->input.y; - img.v = (uint8_t*)convert->input.y + width*height; - img.u = (uint8_t*)convert->input.y + width*height + width2*height2; - image_output(&img, width, height, width, - convert->output.y, convert->output.y_stride, - convert->output.colorspace, convert->interlacing); - break; - - default : - return XVID_ERR_FORMAT; - } - - - emms(); - return XVID_ERR_OK; -} - - - -void fill8(uint8_t * block, int size, int value) -{ - int i; - for (i = 0; i < size; i++) - block[i] = value; -} - -void fill16(int16_t * block, int size, int value) +xvid_gbl_info(xvid_gbl_info_t * info) { - int i; - for (i = 0; i < size; i++) - block[i] = value; -} + if (XVID_VERSION_MAJOR(info->version) != 1) /* v1.x.x */ + return XVID_ERR_VERSION; -#define RANDOM(min,max) min + (rand() % (max-min)) + info->actual_version = XVID_VERSION; + info->build = "xvid-1.0.0"; + info->cpu_flags = detect_cpu_flags(); -void random8(uint8_t * block, int size, int min, int max) -{ - int i; - for (i = 0; i < size; i++) - block[i] = RANDOM(min,max); -} - -void random16(int16_t * block, int size, int min, int max) -{ - int i; - for (i = 0; i < size; i++) - block[i] = RANDOM(min,max); -} - -int compare16(const int16_t * blockA, const int16_t * blockB, int size) -{ - int i; - for (i = 0; i < size; i++) - if (blockA[i] != blockB[i]) - return 1; +#if defined(_SMP) && defined(WIN32) + info->num_threads = pthread_num_processors_np();; +#else + info->num_threads = 0; +#endif return 0; } - -int test_h263_intra(quanth263_intraFunc * funcA, quanth263_intraFunc * funcB, - const char * nameA, const char * nameB, - int min, int max) +static int +xvid_gbl_convert(xvid_gbl_convert_t* convert) { - int q,i; - int64_t timeSTART; - int64_t timeA = 0; - int64_t timeB = 0; - DECLARE_ALIGNED_MATRIX(arrayX, 1, 64, int16_t, CACHE_LINE); - DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); - DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); - - for (q = 1; q <= 31; q++) /* quantizer */ - { - for (i = min; i < max; i++) /* input coeff */ - { - fill16(arrayX, 64, i); - - timeSTART = read_counter(); - funcA(arrayA, arrayX, q, q); - timeA += read_counter() - timeSTART; - - timeSTART = read_counter(); - funcB(arrayB, arrayX, q, q); - timeB += read_counter() - timeSTART; - - if (compare16(arrayA, arrayB, 64)) - { - printf("%s/%s error: q=%i, i=%i\n", nameA?nameA:"?", nameB?nameB:"?", q, i); - return 0; - } - } - } + int width; + int height; + int width2; + int height2; + IMAGE img; - if (nameA) printf("%s:\t%I64i\n", nameA, timeA); - if (nameB) printf("%s:\t%I64i\n", nameB, timeB); + if (XVID_VERSION_MAJOR(convert->version) != 1) /* v1.x.x */ + return XVID_ERR_VERSION; - return 0; -} +#if 0 + const int flip1 = (convert->input.colorspace & XVID_CSP_VFLIP) ^ (convert->output.colorspace & XVID_CSP_VFLIP); +#endif + width = convert->width; + height = convert->height; + width2 = convert->width/2; + height2 = convert->height/2; -int test_h263_inter(quanth263_interFunc * funcA, quanth263_interFunc * funcB, - const char * nameA, const char * nameB, - int min, int max) -{ - int q,i; - int64_t timeSTART; - int64_t timeA = 0; - int64_t timeB = 0; - DECLARE_ALIGNED_MATRIX(arrayX, 1, 64, int16_t, CACHE_LINE); - DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); - DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); - - for (q = 1; q <= 31; q++) /* quantizer */ + switch (convert->input.csp & ~XVID_CSP_VFLIP) { - for (i = min; i < max; i++) /* input coeff */ - { - fill16(arrayX, 64, i); - - timeSTART = read_counter(); - funcA(arrayA, arrayX, q); - timeA += read_counter() - timeSTART; - - timeSTART = read_counter(); - funcB(arrayB, arrayX, q); - timeB += read_counter() - timeSTART; - - if (compare16(arrayA, arrayB, 64)) - { - printf("%s/%s error: q=%i, i=%i\n", nameA?nameA:"?", nameB?nameB:"?", q, i); - return 0; - } - } + case XVID_CSP_YV12 : + img.y = convert->input.plane[0]; + img.v = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height; + img.u = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height + (convert->input.stride[0]/2)*height2; + image_output(&img, width, height, width, + (uint8_t**)convert->output.plane, convert->output.stride, + convert->output.csp, convert->interlacing); + break; + + default : + return XVID_ERR_FORMAT; } - if (nameA) printf("%s:\t%I64i\n", nameA, timeA); - if (nameB) printf("%s:\t%I64i\n", nameB, timeB); + emms(); return 0; } - - -int xvid_init_test() -{ - int cpu_flags; - - printf("xvid_init_test\n"); - -#if defined(ARCH_X86) - cpu_flags = check_cpu_features(); - - emms_mmx(); - - printf("--- quant intra ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_intra(quant_intra_c, quant_intra_mmx, "c", "mmx", -2048, 2047); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_h263_intra(quant_intra_c, quant_intra_3dne, NULL, "3dne", -2048, 2047); - if (cpu_flags & XVID_CPU_SSE2) - test_h263_intra(quant_intra_c, quant_intra_sse2, NULL, "sse2", -2048, 2047); - - printf("\n--- quant inter ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_inter(quant_inter_c, quant_inter_mmx, "c", "mmx", -2048, 2047); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_h263_inter(quant_inter_c, quant_inter_3dne, NULL, "3dne", -2048, 2047); - if (cpu_flags & XVID_CPU_SSE2) - test_h263_inter(quant_inter_c, quant_inter_sse2, NULL, "sse2", -2048, 2047); - - printf("\n--- dequan intra ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_intra(dequant_intra_c, dequant_intra_mmx, "c", "mmx", -256, 255); - if (cpu_flags & XVID_CPU_MMXEXT) - test_h263_intra(dequant_intra_c, dequant_intra_xmm, NULL, "xmm", -256, 255); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_h263_intra(dequant_intra_c, dequant_intra_3dne, NULL, "3dne", -256, 255); - if (cpu_flags & XVID_CPU_SSE2) - test_h263_intra(dequant_intra_c, dequant_intra_sse2, NULL, "sse2", -256, 255); - - printf("\n--- dequant inter ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_inter((quanth263_interFunc*)dequant_inter_c, - (quanth263_interFunc*)dequant_inter_mmx, "c", "mmx", -256, 255); - - if (cpu_flags & XVID_CPU_MMXEXT) - test_h263_inter((quanth263_interFunc*)dequant_inter_c, - (quanth263_interFunc*)dequant_inter_xmm, NULL, "xmm", -256, 255); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_h263_inter((quanth263_interFunc*)dequant_inter_c, - (quanth263_interFunc*)dequant_inter_3dne, NULL, "3dne", -256, 255); - if (cpu_flags & XVID_CPU_SSE2) - test_h263_inter((quanth263_interFunc*)dequant_inter_c, - (quanth263_interFunc*)dequant_inter_sse2, NULL, "sse2", -256, 255); - - printf("\n--- quant4_intra ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_intra((quanth263_intraFunc*)quant4_intra_c, - (quanth263_intraFunc*)quant4_intra_mmx, "c", "mmx", -2048, 2047); - if (cpu_flags & XVID_CPU_MMXEXT) - test_h263_intra((quanth263_intraFunc*)quant4_intra_c, - (quanth263_intraFunc*)quant4_intra_xmm, NULL, "xmm", -2048, 2047); - - printf("\n--- quant4_inter ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_inter((quanth263_interFunc*)quant4_inter_c, - (quanth263_interFunc*)quant4_inter_mmx, "c", "mmx", -2048, 2047); - if (cpu_flags & XVID_CPU_MMXEXT) - test_h263_inter((quanth263_interFunc*)quant4_inter_c, - (quanth263_interFunc*)quant4_inter_xmm, NULL, "xmm", -2048, 2047); - - - printf("\n--- dequant4_intra ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_intra((quanth263_intraFunc*)dequant4_intra_c, - (quanth263_intraFunc*)dequant4_intra_mmx, "c", "mmx", -256, 255); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_h263_intra((quanth263_intraFunc*)dequant4_intra_c, - (quanth263_intraFunc*)dequant4_intra_3dne, NULL, "sse2", -256, 255); - - printf("\n--- dequant4_inter ---\n"); - if (cpu_flags & XVID_CPU_MMX) - test_h263_inter((quanth263_interFunc*)dequant4_inter_c, - (quanth263_interFunc*)dequant4_inter_mmx, "c", "mmx", -256, 255); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_h263_inter((quanth263_interFunc*)dequant4_inter_c, - (quanth263_interFunc*)dequant4_inter_3dne, NULL, "sse2", -256, 255); - - emms_mmx(); - -#endif - - return XVID_ERR_OK; -} +/***************************************************************************** + * XviD Global Entry point + * + * Well this function initialize all internal function pointers according + * to the CPU features forced by the library client or autodetected (depending + * on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all + * image colorspace transformation tables. + * + ****************************************************************************/ int -xvid_init(void *handle, +xvid_global(void *handle, int opt, void *param1, void *param2) { switch(opt) { - case XVID_INIT_INIT : - return xvid_init_init((XVID_INIT_PARAM*)param1); + case XVID_GBL_INIT : + return xvid_gbl_init((xvid_gbl_init_t*)param1); - case XVID_INIT_CONVERT : - return xvid_init_convert((XVID_INIT_CONVERTINFO*)param1); + case XVID_GBL_INFO : + return xvid_gbl_info((xvid_gbl_info_t*)param1); - case XVID_INIT_TEST : - return xvid_init_test(); + case XVID_GBL_CONVERT : + return xvid_gbl_convert((xvid_gbl_convert_t*)param1); default : return XVID_ERR_FAIL; @@ -837,15 +717,15 @@ void *param2) { switch (opt) { - case XVID_DEC_DECODE: - return decoder_decode((DECODER *) handle, (XVID_DEC_FRAME *) param1, (XVID_DEC_STATS*) param2); - case XVID_DEC_CREATE: - return decoder_create((XVID_DEC_PARAM *) param1); + return decoder_create((xvid_dec_create_t *) param1); case XVID_DEC_DESTROY: return decoder_destroy((DECODER *) handle); + case XVID_DEC_DECODE: + return decoder_decode((DECODER *) handle, (xvid_dec_frame_t *) param1, (xvid_dec_stats_t*) param2); + default: return XVID_ERR_FAIL; } @@ -871,18 +751,15 @@ switch (opt) { case XVID_ENC_ENCODE: - if (((Encoder *) handle)->mbParam.max_bframes >= 0) - return encoder_encode_bframes((Encoder *) handle, (XVID_ENC_FRAME *) param1, - (XVID_ENC_STATS *) param2); - else - return encoder_encode((Encoder *) handle, (XVID_ENC_FRAME *) param1, - (XVID_ENC_STATS *) param2); + return enc_encode((Encoder *) handle, + (xvid_enc_frame_t *) param1, + (xvid_enc_stats_t *) param2); case XVID_ENC_CREATE: - return encoder_create((XVID_ENC_PARAM *) param1); + return enc_create((xvid_enc_create_t *) param1); case XVID_ENC_DESTROY: - return encoder_destroy((Encoder *) handle); + return enc_destroy((Encoder *) handle); default: return XVID_ERR_FAIL;