--- xvid.c 2003/03/26 11:01:03 1.45.2.2 +++ xvid.c 2006/02/25 04:41:12 1.68 @@ -3,6 +3,8 @@ * XVID MPEG-4 VIDEO CODEC * - Native API implementation - * + * 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 * the Free Software Foundation ; either version 2 of the License, or @@ -17,7 +19,7 @@ * along with this program ; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA * - * $Id: xvid.c,v 1.45.2.2 2003/03/26 11:01:03 suxen_drol Exp $ + * $Id: xvid.c,v 1.68 2006/02/25 04:41:12 suxen_drol Exp $ * ****************************************************************************/ @@ -34,22 +36,24 @@ #include "dct/fdct.h" #include "image/colorspace.h" #include "image/interpolate8x8.h" -#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_IS_IA32) +#if defined(_DEBUG) +unsigned int xvid_debug = 0; /* xvid debug mask */ +#endif -#if defined(_MSC_VER) +#if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER) # include -#else +#elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || defined(ARCH_IS_PPC) # include # include @@ -72,53 +76,62 @@ * 0 : SIGILL was *not* signalled * 1 : SIGILL was signalled */ - -int +#if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER) +static int sigill_check(void (*func)()) { -#if defined(_MSC_VER) _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_X86_64) || 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() +detect_cpu_flags(void) { /* enable native assembly optimizations by default */ unsigned int cpu_flags = XVID_CPU_ASM; -#if defined(ARCH_IS_IA32) +#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) cpu_flags |= check_cpu_features(); if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) cpu_flags &= ~XVID_CPU_SSE; @@ -128,9 +141,8 @@ #endif #if defined(ARCH_IS_PPC) -#if defined(ARCH_IS_PPC_ALTIVEC) - cpu_flags |= XVID_CPU_ALTIVEC; -#endif + if (!sigill_check(altivec_trigger)) + cpu_flags |= XVID_CPU_ALTIVEC; #endif return cpu_flags; @@ -144,7 +156,7 @@ * 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 " " " " " @@ -152,12 +164,12 @@ ****************************************************************************/ -static +static int xvid_gbl_init(xvid_gbl_init_t * init) { unsigned int cpu_flags; - if (XVID_MAJOR(init->version) != 1) /* v1.x.x */ + 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(); @@ -171,21 +183,26 @@ idct = idct_int32; /* Only needed on PPC Altivec archs */ - sadInit = 0; + sadInit = NULL; /* 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; @@ -193,8 +210,10 @@ transfer_8to16sub = transfer_8to16sub_c; transfer_8to16subro = transfer_8to16subro_c; transfer_8to16sub2 = transfer_8to16sub2_c; + transfer_8to16sub2ro = transfer_8to16sub2ro_c; transfer_16to8add = transfer_16to8add_c; transfer8x8_copy = transfer8x8_copy_c; + transfer8x4_copy = transfer8x4_copy_c; /* Interlacing functions */ MBFieldTest = MBFieldTest_c; @@ -204,6 +223,15 @@ interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_c; interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c; + interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_c; + interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_c; + interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_c; + + interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_c; + interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_c; + interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_c; + interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_c; + interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c; interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c; interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c; @@ -218,13 +246,8 @@ interpolate8x8_avg2 = interpolate8x8_avg2_c; interpolate8x8_avg4 = interpolate8x8_avg4_c; - /* reduced resoltuion */ - copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C; - add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_C; - vfilter_31 = xvid_VFilter_31_C; - hfilter_31 = xvid_HFilter_31_C; - filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_C; - filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_C; + /* postprocessing */ + image_brightness = image_brightness_c; /* Initialize internal colorspace transformation tables */ colorspace_init(); @@ -237,6 +260,7 @@ 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; @@ -246,10 +270,10 @@ 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; @@ -257,37 +281,33 @@ 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; */ + 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; + sse8_8bit = sse8_8bit_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) || (cpu_flags & XVID_CPU_SSE) || (cpu_flags & XVID_CPU_SSE2)) @@ -299,19 +319,23 @@ 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; @@ -321,6 +345,7 @@ transfer_8to16sub2 = transfer_8to16sub2_mmx; transfer_16to8add = transfer_16to8add_mmx; transfer8x8_copy = transfer8x8_copy_mmx; + transfer8x4_copy = transfer8x4_copy_mmx; /* Interlacing Functions */ MBFieldTest = MBFieldTest_mmx; @@ -330,18 +355,23 @@ interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_mmx; interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx; + interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_mmx; + interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_mmx; + interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_mmx; + + interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_mmx; + interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_mmx; + interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_mmx; + interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_mmx; + interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx; interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx; interpolate8x8_avg2 = interpolate8x8_avg2_mmx; interpolate8x8_avg4 = interpolate8x8_avg4_mmx; - /* reduced resolution */ - copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_mmx; - add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_mmx; - hfilter_31 = xvid_HFilter_31_mmx; - filter_18x18_to_8x8 = xvid_Filter_18x18_To_8x8_mmx; - filter_diff_18x18_to_8x8 = xvid_Filter_Diff_18x18_To_8x8_mmx; + /* postprocessing */ + image_brightness = image_brightness_mmx; /* image input xxx_to_yv12 related functions */ yv12_to_yv12 = yv12_to_yv12_mmx; @@ -355,18 +385,20 @@ 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; /* Motion estimation related functions */ - calc_cbp = calc_cbp_mmx; - sad16 = sad16_mmx; - sad8 = sad8_mmx; - sad16bi = sad16bi_mmx; - sad8bi = sad8bi_mmx; - dev16 = dev16_mmx; - sad16v = sad16v_mmx; + calc_cbp = calc_cbp_mmx; + sad16 = sad16_mmx; + sad8 = sad8_mmx; + sad16bi = sad16bi_mmx; + sad8bi = sad8bi_mmx; + dev16 = dev16_mmx; + sad16v = sad16v_mmx; + sse8_16bit = sse8_16bit_mmx; + sse8_8bit = sse8_8bit_mmx; } /* these 3dnow functions are faster than mmx, but slower than xmm. */ @@ -385,27 +417,34 @@ if ((cpu_flags & XVID_CPU_MMXEXT)) { - /* Inverse DCT */ + /* DCT */ + fdct = fdct_xmm_skal; idct = idct_xmm; /* Interpolation */ interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_xmm; interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_xmm; interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm; - - /* reduced resolution */ - copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_xmm; - add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm; + + interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_xmm; + interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_xmm; + interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_xmm; + + interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_xmm; + interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_xmm; + interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_xmm; + interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_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; + transfer_8to16sub2ro = transfer_8to16sub2ro_xmm; /* Colorspace transformation */ yv12_to_yv12 = yv12_to_yv12_xmm; @@ -427,65 +466,80 @@ interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dn; interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dn; interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dn; + + interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dn; + interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dn; + interpolate8x4_halfpel_hv = interpolate8x4_halfpel_hv_3dn; } if ((cpu_flags & XVID_CPU_3DNOWEXT)) { - /* Inverse DCT */ - idct = idct_3dne; - /* Buffer transfer */ transfer_8to16copy = transfer_8to16copy_3dne; transfer_16to8copy = transfer_16to8copy_3dne; transfer_8to16sub = transfer_8to16sub_3dne; transfer_8to16subro = transfer_8to16subro_3dne; - transfer_8to16sub2 = transfer_8to16sub2_3dne; transfer_16to8add = transfer_16to8add_3dne; transfer8x8_copy = transfer8x8_copy_3dne; + transfer8x4_copy = transfer8x4_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; + + interpolate8x4_halfpel_h = interpolate8x4_halfpel_h_3dne; + interpolate8x4_halfpel_v = interpolate8x4_halfpel_v_3dne; + interpolate8x4_halfpel_hv = interpolate8x4_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)) { 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; -#if defined(EXPERIMENTAL_SSE2_CODE) - /* ME; slower than xmm */ + /* SAD operators */ sad16 = sad16_sse2; dev16 = dev16_sse2; -#endif - /* Forward and Inverse DCT */ - idct = idct_sse2; - fdct = fdct_sse2; + + /* DCT operators */ + fdct = fdct_sse2_skal; + /* idct = idct_sse2_skal; */ /* Is now IEEE1180 and Walken compliant. Disabled until fully tested. */ + + /* postprocessing */ + image_brightness = image_brightness_sse2; } -#endif +#endif /* ARCH_IS_IA32 */ #if defined(ARCH_IS_IA64) if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */ @@ -500,56 +554,179 @@ 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; + 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; - DPRINTF(DPRINTF_DEBUG, "Using IA-64 assembler routines."); } -#endif +#endif #if defined(ARCH_IS_PPC) - if ((cpu_flags & XVID_CPU_ASM)) - { - calc_cbp = calc_cbp_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_halfpel_add = interpolate8x8_halfpel_add_altivec_c; + interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_altivec_c; + interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_altivec_c; + interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_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; + + dequant_mpeg_intra = dequant_mpeg_intra_altivec_c; + dequant_mpeg_inter = dequant_mpeg_inter_altivec_c; + + /* Qpel stuff */ + xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C; + xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C; + } +#endif + +#if defined(ARCH_IS_X86_64) + /* For now, only XVID_CPU_ASM is looked for, so user can still + * disable asm usage the usual way. When Intel EMT64 cpus will + * be out, maybe we'll have to check more precisely what cpu + * features there really are. */ + if (cpu_flags & XVID_CPU_ASM) { + /* SIMD state flusher */ + emms = emms_mmx; - if ((cpu_flags & XVID_CPU_ALTIVEC)) - { - calc_cbp = calc_cbp_altivec; - fdct = fdct_altivec; - idct = idct_altivec; - sadInit = sadInit_altivec; - sad16 = sad16_altivec; - sad8 = sad8_altivec; - dev16 = dev16_altivec; + /* DCT operators */ + fdct = fdct_skal_x86_64; + idct = idct_x86_64; + + /* SAD operators */ + sad16 = sad16_x86_64; + sad8 = sad8_x86_64; + sad16bi = sad16bi_x86_64; + sad8bi = sad8bi_x86_64; + dev16 = dev16_x86_64; + sad16v = sad16v_x86_64; + sse8_16bit = sse8_16bit_x86_64; + sse8_8bit = sse8_8bit_x86_64; + + /* Interpolation operators */ + interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_x86_64; + interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_x86_64; + interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_x86_64; + + interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_x86_64; + interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_x86_64; + interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_x86_64; + interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_x86_64; + + interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_x86_64; + interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_x86_64; + + interpolate8x8_avg2 = interpolate8x8_avg2_x86_64; + interpolate8x8_avg4 = interpolate8x8_avg4_x86_64; + + /* Quantization related functions */ + quant_h263_intra = quant_h263_intra_x86_64; + quant_h263_inter = quant_h263_inter_x86_64; + dequant_h263_intra = dequant_h263_intra_x86_64; + dequant_h263_inter = dequant_h263_inter_x86_64; + quant_mpeg_intra = quant_mpeg_intra_x86_64; + quant_mpeg_inter = quant_mpeg_inter_x86_64; + dequant_mpeg_intra = dequant_mpeg_intra_x86_64; + dequant_mpeg_inter = dequant_mpeg_inter_x86_64; + + /* Block related functions */ + transfer_8to16copy = transfer_8to16copy_x86_64; + transfer_16to8copy = transfer_16to8copy_x86_64; + transfer_8to16sub = transfer_8to16sub_x86_64; + transfer_8to16subro = transfer_8to16subro_x86_64; + transfer_8to16sub2 = transfer_8to16sub2_x86_64; + transfer_8to16sub2ro= transfer_8to16sub2ro_x86_64; + transfer_16to8add = transfer_16to8add_x86_64; + transfer8x8_copy = transfer8x8_copy_x86_64; + + /* Qpel stuff */ + xvid_QP_Funcs = &xvid_QP_Funcs_x86_64; + xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_x86_64; + + /* Interlacing Functions */ + MBFieldTest = MBFieldTest_x86_64; } #endif - return 0; +#if defined(_DEBUG) + xvid_debug = init->debug; +#endif + + return(0); } static int xvid_gbl_info(xvid_gbl_info_t * info) { - if (XVID_MAJOR(info->version) != 1) /* v1.x.x */ + if (XVID_VERSION_MAJOR(info->version) != 1) /* v1.x.x */ return XVID_ERR_VERSION; info->actual_version = XVID_VERSION; - info->build = "dev-api-4"; + info->build = "xvid-1.2.0-dev"; info->cpu_flags = detect_cpu_flags(); + info->num_threads = 0; -#if defined(_SMP) && defined(WIN32) - info->num_threads = pthread_num_processors_np();; -#else - info->num_threads = 0; +#if defined(WIN32) + { + DWORD dwProcessAffinityMask, dwSystemAffinityMask; + if (GetProcessAffinityMask(GetCurrentProcess(), &dwProcessAffinityMask, &dwSystemAffinityMask)) { + int i; + for(i=0; i<32; i++) { + if ((dwProcessAffinityMask & (1<num_threads++; + } + } + } #endif return 0; @@ -565,10 +742,12 @@ int height2; IMAGE img; - if (XVID_MAJOR(convert->version) != 1) /* v1.x.x */ + if (XVID_VERSION_MAJOR(convert->version) != 1) /* v1.x.x */ return XVID_ERR_VERSION; - // const int flip1 = (convert->input.colorspace & XVID_CSP_VFLIP) ^ (convert->output.colorspace & XVID_CSP_VFLIP); +#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; @@ -578,7 +757,7 @@ { case XVID_CSP_YV12 : img.y = convert->input.plane[0]; - img.v = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height; + 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, @@ -594,373 +773,6 @@ return 0; } - - -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) -{ - int i; - for (i = 0; i < size; i++) - block[i] = value; -} - -#define RANDOM(min,max) min + (rand() % (max-min)) - -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; - - return 0; -} - -int diff16(const int16_t * blockA, const int16_t * blockB, int size) -{ - int i, diff = 0; - for (i = 0; i < size; i++) - diff += ABS(blockA[i]-blockB[i]); - return diff; -} - - -#define XVID_TEST_RANDOM 0x00000001 /* random input data */ -#define XVID_TEST_VERBOSE 0x00000002 /* verbose error output */ - - -#define TEST_FORWARD 0x00000001 /* intra */ -#define TEST_FDCT (TEST_FORWARD) -#define TEST_IDCT (0) - -static int test_transform(void * funcA, void * funcB, const char * nameB, - int test, int flags) -{ - int i; - int64_t timeSTART; - int64_t timeA = 0; - int64_t timeB = 0; - DECLARE_ALIGNED_MATRIX(arrayA, 1, 64, int16_t, CACHE_LINE); - DECLARE_ALIGNED_MATRIX(arrayB, 1, 64, int16_t, CACHE_LINE); - int min, max; - int count = 0; - - int tmp; - int min_error = 0x10000*64; - int max_error = 0; - - - if ((test & TEST_FORWARD)) /* forward */ - { - min = -256; - max = 255; - }else{ /* inverse */ - min = -2048; - max = 2047; - } - - for (i = 0; i < 64*64; i++) - { - if ((flags & XVID_TEST_RANDOM)) - { - random16(arrayA, 64, min, max); - }else{ - fill16(arrayA, 64, i); - } - memcpy(arrayB, arrayA, 64*sizeof(int16_t)); - - if ((test & TEST_FORWARD)) - { - timeSTART = read_counter(); - ((fdctFunc*)funcA)(arrayA); - timeA += read_counter() - timeSTART; - - timeSTART = read_counter(); - ((fdctFunc*)funcB)(arrayB); - timeB += read_counter() - timeSTART; - } - else - { - timeSTART = read_counter(); - ((idctFunc*)funcA)(arrayA); - timeA += read_counter() - timeSTART; - - timeSTART = read_counter(); - ((idctFunc*)funcB)(arrayB); - timeB += read_counter() - timeSTART; - } - - tmp = diff16(arrayA, arrayB, 64) / 64; - if (tmp > max_error) - max_error = tmp; - if (tmp < min_error) - min_error = tmp; - - count++; - } - - /* print the "average difference" of best/worst transforms */ - printf("%s:\t%i\t(min_error:%i, max_error:%i)\n", nameB, (int)(timeB / count), min_error, max_error); - - return 0; -} - - -#define TEST_QUANT 0x00000001 /* forward quantization */ -#define TEST_INTRA 0x00000002 /* intra */ -#define TEST_QUANT_INTRA (TEST_QUANT|TEST_INTRA) -#define TEST_QUANT_INTER (TEST_QUANT) -#define TEST_DEQUANT_INTRA (TEST_INTRA) -#define TEST_DEQUANT_INTER (0) - -static int test_quant(void * funcA, void * funcB, const char * nameB, - int test, int flags) -{ - int q,i; - int64_t timeSTART; - int64_t timeA = 0; - int64_t timeB = 0; - int retA = 0, retB = 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); - int min, max; - int count = 0; - int errors = 0; - - if ((test & TEST_QUANT)) /* quant */ - { - min = -2048; - max = 2047; - }else{ /* dequant */ - min = -256; - max = 255; - } - - for (q = 1; q <= 31; q++) /* quantizer */ - { - for (i = min; i < max; i++) /* input coeff */ - { - if ((flags & XVID_TEST_RANDOM)) - { - random16(arrayX, 64, min, max); - }else{ - fill16(arrayX, 64, i); - } - - if ((test & TEST_INTRA)) /* intra */ - { - timeSTART = read_counter(); - ((quanth263_intraFunc*)funcA)(arrayA, arrayX, q, q); - timeA += read_counter() - timeSTART; - - timeSTART = read_counter(); - ((quanth263_intraFunc*)funcB)(arrayB, arrayX, q, q); - timeB += read_counter() - timeSTART; - } - else /* inter */ - { - timeSTART = read_counter(); - retA = ((quanth263_interFunc*)funcA)(arrayA, arrayX, q); - timeA += read_counter() - timeSTART; - - timeSTART = read_counter(); - retB = ((quanth263_interFunc*)funcB)(arrayB, arrayX, q); - timeB += read_counter() - timeSTART; - } - - /* compare return value from quant_inter, and compare (de)quantiz'd arrays */ - if ( ((test&TEST_QUANT) && !(test&TEST_INTRA) && retA != retB ) || - compare16(arrayA, arrayB, 64)) - { - errors++; - if ((flags & XVID_TEST_VERBOSE)) - printf("%s error: q=%i, i=%i\n", nameB, q, i); - } - - count++; - } - } - - printf("%s:\t%i", nameB, (int)(timeB / count)); - if (errors>0) - printf("\t(%i errors out of %i)", errors, count); - printf("\n"); - - return 0; -} - - - -int xvid_init_test(int flags) -{ -#if defined(ARCH_IS_IA32) - int cpu_flags; -#endif - - printf("XviD tests\n\n"); - -#if defined(ARCH_IS_IA32) - cpu_flags = detect_cpu_flags(); -#endif - - idct_int32_init(); - emms(); - - srand(time(0)); - - /* fDCT test */ - printf("--- fdct ---\n"); - test_transform(fdct_int32, fdct_int32, "c", TEST_FDCT, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_transform(fdct_int32, fdct_mmx, "mmx", TEST_FDCT, flags); - if (cpu_flags & XVID_CPU_SSE2) - test_transform(fdct_int32, fdct_sse2, "sse2", TEST_FDCT, flags); -#endif - - /* iDCT test */ - printf("\n--- idct ---\n"); - test_transform(idct_int32, idct_int32, "c", TEST_IDCT, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_transform(idct_int32, idct_mmx, "mmx", TEST_IDCT, flags); - if (cpu_flags & XVID_CPU_MMXEXT) - test_transform(idct_int32, idct_xmm, "xmm", TEST_IDCT, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_transform(idct_int32, idct_3dne, "3dne", TEST_IDCT, flags); - if (cpu_flags & XVID_CPU_SSE2) - test_transform(idct_int32, idct_sse2, "sse2", TEST_IDCT, flags); -#endif - - /* Intra quantization test */ - printf("\n--- quant intra ---\n"); - test_quant(quant_intra_c, quant_intra_c, "c", TEST_QUANT_INTRA, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(quant_intra_c, quant_intra_mmx, "mmx", TEST_QUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_quant(quant_intra_c, quant_intra_3dne, "3dne", TEST_QUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_SSE2) - test_quant(quant_intra_c, quant_intra_sse2, "sse2", TEST_QUANT_INTRA, flags); -#endif - - /* Inter quantization test */ - printf("\n--- quant inter ---\n"); - test_quant(quant_inter_c, quant_inter_c, "c", TEST_QUANT_INTER, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(quant_inter_c, quant_inter_mmx, "mmx", TEST_QUANT_INTER, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_quant(quant_inter_c, quant_inter_3dne, "3dne", TEST_QUANT_INTER, flags); - if (cpu_flags & XVID_CPU_SSE2) - test_quant(quant_inter_c, quant_inter_sse2, "sse2", TEST_QUANT_INTER, flags); -#endif - - /* Intra dequantization test */ - printf("\n--- dequant intra ---\n"); - test_quant(dequant_intra_c, dequant_intra_c, "c", TEST_DEQUANT_INTRA, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(dequant_intra_c, dequant_intra_mmx, "mmx", TEST_DEQUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_MMXEXT) - test_quant(dequant_intra_c, dequant_intra_xmm, "xmm", TEST_DEQUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_quant(dequant_intra_c, dequant_intra_3dne, "3dne", TEST_DEQUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_SSE2) - test_quant(dequant_intra_c, dequant_intra_sse2, "sse2", TEST_DEQUANT_INTRA, flags); -#endif - - /* Inter dequantization test */ - printf("\n--- dequant inter ---\n"); - test_quant(dequant_inter_c, dequant_inter_c, "c", TEST_DEQUANT_INTER, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(dequant_inter_c, dequant_inter_mmx, "mmx", TEST_DEQUANT_INTER, flags); - if (cpu_flags & XVID_CPU_MMXEXT) - test_quant(dequant_inter_c, dequant_inter_xmm, "xmm", TEST_DEQUANT_INTER, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_quant(dequant_inter_c, dequant_inter_3dne, "3dne", TEST_DEQUANT_INTER, flags); - if (cpu_flags & XVID_CPU_SSE2) - test_quant(dequant_inter_c, dequant_inter_sse2, "sse2", TEST_DEQUANT_INTER, flags); -#endif - - /* Intra quantization test */ - printf("\n--- quant4 intra ---\n"); - test_quant(quant4_intra_c, quant4_intra_c, "c", TEST_QUANT_INTRA, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(quant4_intra_c, quant4_intra_mmx, "mmx", TEST_QUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_MMXEXT) - test_quant(quant4_intra_c, quant4_intra_xmm, "xmm", TEST_QUANT_INTRA, flags); -#endif - - /* Inter quantization test */ - printf("\n--- quant4 inter ---\n"); - test_quant(quant4_inter_c, quant4_inter_c, "c", TEST_QUANT_INTER, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(quant4_inter_c, quant4_inter_mmx, "mmx", TEST_QUANT_INTER, flags); - if (cpu_flags & XVID_CPU_MMXEXT) - test_quant(quant4_inter_c, quant4_inter_xmm, "xmm", TEST_QUANT_INTER, flags); -#endif - - /* Intra dequantization test */ - printf("\n--- dequant4 intra ---\n"); - test_quant(dequant4_intra_c, dequant4_intra_c, "c", TEST_DEQUANT_INTRA, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(dequant4_intra_c, dequant4_intra_mmx, "mmx", TEST_DEQUANT_INTRA, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_quant(dequant4_intra_c, dequant4_intra_3dne, "3dne", TEST_DEQUANT_INTRA, flags); -#endif - - /* Inter dequantization test */ - printf("\n--- dequant4 inter ---\n"); - test_quant(dequant4_inter_c, dequant4_inter_c, "c", TEST_DEQUANT_INTER, flags); - -#if defined(ARCH_IS_IA32) - if (cpu_flags & XVID_CPU_MMX) - test_quant(dequant4_inter_c, dequant4_inter_mmx, "mmx", TEST_DEQUANT_INTER, flags); - if (cpu_flags & XVID_CPU_3DNOWEXT) - test_quant(dequant4_inter_c, dequant4_inter_3dne, "3dne", TEST_DEQUANT_INTER, flags); -#endif - - emms(); - - return 0; -} - - /***************************************************************************** * XviD Global Entry point * @@ -968,7 +780,7 @@ * 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. - * + * ****************************************************************************/ @@ -989,11 +801,6 @@ case XVID_GBL_CONVERT : return xvid_gbl_convert((xvid_gbl_convert_t*)param1); - case XVID_GBL_TEST : - { - ptr_t flags = (ptr_t)param1; - return xvid_init_test((int)flags); - } default : return XVID_ERR_FAIL; }