3 |
* XVID MPEG-4 VIDEO CODEC |
* XVID MPEG-4 VIDEO CODEC |
4 |
* - Native API implementation - |
* - Native API implementation - |
5 |
* |
* |
6 |
|
* Copyright(C) 2001-2004 Peter Ross <pross@xvid.org> |
7 |
|
* |
8 |
* This program is free software ; you can redistribute it and/or modify |
* This program is free software ; you can redistribute it and/or modify |
9 |
* it under the terms of the GNU General Public License as published by |
* it under the terms of the GNU General Public License as published by |
10 |
* the Free Software Foundation ; either version 2 of the License, or |
* the Free Software Foundation ; either version 2 of the License, or |
19 |
* along with this program ; if not, write to the Free Software |
* along with this program ; if not, write to the Free Software |
20 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
21 |
* |
* |
22 |
* $Id: xvid.c,v 1.47 2003-06-11 14:10:55 Isibaar Exp $ |
* $Id: xvid.c,v 1.62 2005-04-04 23:49:37 edgomez Exp $ |
23 |
* |
* |
24 |
****************************************************************************/ |
****************************************************************************/ |
25 |
|
|
36 |
#include "dct/fdct.h" |
#include "dct/fdct.h" |
37 |
#include "image/colorspace.h" |
#include "image/colorspace.h" |
38 |
#include "image/interpolate8x8.h" |
#include "image/interpolate8x8.h" |
|
#include "image/reduced.h" |
|
39 |
#include "utils/mem_transfer.h" |
#include "utils/mem_transfer.h" |
40 |
#include "utils/mbfunctions.h" |
#include "utils/mbfunctions.h" |
41 |
#include "quant/quant_h263.h" |
#include "quant/quant.h" |
|
#include "quant/quant_mpeg4.h" |
|
42 |
#include "motion/motion.h" |
#include "motion/motion.h" |
43 |
#include "motion/sad.h" |
#include "motion/sad.h" |
44 |
#include "utils/emms.h" |
#include "utils/emms.h" |
45 |
#include "utils/timer.h" |
#include "utils/timer.h" |
46 |
#include "bitstream/mbcoding.h" |
#include "bitstream/mbcoding.h" |
47 |
|
#include "image/qpel.h" |
48 |
|
#include "image/postprocessing.h" |
49 |
|
|
50 |
#if defined(ARCH_IS_IA32) |
#if defined(_DEBUG) |
51 |
|
unsigned int xvid_debug = 0; /* xvid debug mask */ |
52 |
|
#endif |
53 |
|
|
54 |
#if defined(_MSC_VER) |
#if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER) |
55 |
# include <windows.h> |
# include <windows.h> |
56 |
#else |
#elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || defined(ARCH_IS_PPC) |
57 |
# include <signal.h> |
# include <signal.h> |
58 |
# include <setjmp.h> |
# include <setjmp.h> |
59 |
|
|
76 |
* 0 : SIGILL was *not* signalled |
* 0 : SIGILL was *not* signalled |
77 |
* 1 : SIGILL was signalled |
* 1 : SIGILL was signalled |
78 |
*/ |
*/ |
79 |
|
#if (defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64)) && defined(_MSC_VER) |
80 |
int |
static int |
81 |
sigill_check(void (*func)()) |
sigill_check(void (*func)()) |
82 |
{ |
{ |
|
#if defined(_MSC_VER) |
|
83 |
_try { |
_try { |
84 |
func(); |
func(); |
85 |
} |
} _except(EXCEPTION_EXECUTE_HANDLER) { |
|
_except(EXCEPTION_EXECUTE_HANDLER) { |
|
86 |
|
|
87 |
if (_exception_code() == STATUS_ILLEGAL_INSTRUCTION) |
if (_exception_code() == STATUS_ILLEGAL_INSTRUCTION) |
88 |
return 1; |
return(1); |
89 |
} |
} |
90 |
return 0; |
return(0); |
91 |
#else |
} |
92 |
|
#elif defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) || defined(ARCH_IS_PPC) |
93 |
|
static int |
94 |
|
sigill_check(void (*func)()) |
95 |
|
{ |
96 |
void * old_handler; |
void * old_handler; |
97 |
int jmpret; |
int jmpret; |
98 |
|
|
99 |
|
/* Set our SIGILL handler */ |
100 |
old_handler = signal(SIGILL, sigill_handler); |
old_handler = signal(SIGILL, sigill_handler); |
101 |
if (old_handler == SIG_ERR) |
|
102 |
{ |
/* Check for error */ |
103 |
return -1; |
if (old_handler == SIG_ERR) { |
104 |
|
return(-1); |
105 |
} |
} |
106 |
|
|
107 |
|
/* Save stack context, so if func triggers a SIGILL, we can still roll |
108 |
|
* back to a valid CPU state */ |
109 |
jmpret = setjmp(mark); |
jmpret = setjmp(mark); |
110 |
if (jmpret == 0) |
|
111 |
{ |
/* If setjmp returned directly, then its returned value is 0, and we still |
112 |
|
* have to test the passed func. Otherwise it means the stack context has |
113 |
|
* been restored by a longjmp() call, which in our case happens only in the |
114 |
|
* signal handler */ |
115 |
|
if (jmpret == 0) { |
116 |
func(); |
func(); |
117 |
} |
} |
118 |
|
|
119 |
|
/* Restore old signal handler */ |
120 |
signal(SIGILL, old_handler); |
signal(SIGILL, old_handler); |
121 |
|
|
122 |
return jmpret; |
return(jmpret); |
|
#endif |
|
123 |
} |
} |
124 |
#endif |
#endif |
125 |
|
|
131 |
/* enable native assembly optimizations by default */ |
/* enable native assembly optimizations by default */ |
132 |
unsigned int cpu_flags = XVID_CPU_ASM; |
unsigned int cpu_flags = XVID_CPU_ASM; |
133 |
|
|
134 |
#if defined(ARCH_IS_IA32) |
#if defined(ARCH_IS_IA32) || defined(ARCH_IS_X86_64) |
135 |
cpu_flags |= check_cpu_features(); |
cpu_flags |= check_cpu_features(); |
136 |
if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) |
if ((cpu_flags & XVID_CPU_SSE) && sigill_check(sse_os_trigger)) |
137 |
cpu_flags &= ~XVID_CPU_SSE; |
cpu_flags &= ~XVID_CPU_SSE; |
141 |
#endif |
#endif |
142 |
|
|
143 |
#if defined(ARCH_IS_PPC) |
#if defined(ARCH_IS_PPC) |
144 |
#if defined(ARCH_IS_PPC_ALTIVEC) |
if (!sigill_check(altivec_trigger)) |
145 |
cpu_flags |= XVID_CPU_ALTIVEC; |
cpu_flags |= XVID_CPU_ALTIVEC; |
146 |
#endif |
#endif |
|
#endif |
|
147 |
|
|
148 |
return cpu_flags; |
return cpu_flags; |
149 |
} |
} |
165 |
|
|
166 |
|
|
167 |
static |
static |
168 |
int xvid_init_init(XVID_INIT_PARAM * init_param) |
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; |
|
|
|
|
|
} else { |
|
|
|
|
|
cpu_flags = detect_cpu_flags(); |
|
|
} |
|
|
|
|
|
if ((init_param->cpu_flags & XVID_CPU_CHKONLY)) |
|
169 |
{ |
{ |
170 |
init_param->cpu_flags = cpu_flags; |
unsigned int cpu_flags; |
|
return XVID_ERR_OK; |
|
|
} |
|
171 |
|
|
172 |
init_param->cpu_flags = cpu_flags; |
if (XVID_VERSION_MAJOR(init->version) != 1) /* v1.x.x */ |
173 |
|
return XVID_ERR_VERSION; |
174 |
|
|
175 |
|
cpu_flags = (init->cpu_flags & XVID_CPU_FORCE) ? init->cpu_flags : detect_cpu_flags(); |
176 |
|
|
177 |
/* Initialize the function pointers */ |
/* Initialize the function pointers */ |
178 |
idct_int32_init(); |
idct_int32_init(); |
188 |
/* Restore FPU context : emms_c is a nop functions */ |
/* Restore FPU context : emms_c is a nop functions */ |
189 |
emms = emms_c; |
emms = emms_c; |
190 |
|
|
191 |
|
/* Qpel stuff */ |
192 |
|
xvid_QP_Funcs = &xvid_QP_Funcs_C; |
193 |
|
xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_C; |
194 |
|
xvid_Init_QP(); |
195 |
|
|
196 |
/* Quantization functions */ |
/* Quantization functions */ |
197 |
quant_intra = quant_intra_c; |
quant_h263_intra = quant_h263_intra_c; |
198 |
dequant_intra = dequant_intra_c; |
quant_h263_inter = quant_h263_inter_c; |
199 |
quant_inter = quant_inter_c; |
dequant_h263_intra = dequant_h263_intra_c; |
200 |
dequant_inter = dequant_inter_c; |
dequant_h263_inter = dequant_h263_inter_c; |
201 |
|
|
202 |
quant4_intra = quant4_intra_c; |
quant_mpeg_intra = quant_mpeg_intra_c; |
203 |
dequant4_intra = dequant4_intra_c; |
quant_mpeg_inter = quant_mpeg_inter_c; |
204 |
quant4_inter = quant4_inter_c; |
dequant_mpeg_intra = dequant_mpeg_intra_c; |
205 |
dequant4_inter = dequant4_inter_c; |
dequant_mpeg_inter = dequant_mpeg_inter_c; |
206 |
|
|
207 |
/* Block transfer related functions */ |
/* Block transfer related functions */ |
208 |
transfer_8to16copy = transfer_8to16copy_c; |
transfer_8to16copy = transfer_8to16copy_c; |
210 |
transfer_8to16sub = transfer_8to16sub_c; |
transfer_8to16sub = transfer_8to16sub_c; |
211 |
transfer_8to16subro = transfer_8to16subro_c; |
transfer_8to16subro = transfer_8to16subro_c; |
212 |
transfer_8to16sub2 = transfer_8to16sub2_c; |
transfer_8to16sub2 = transfer_8to16sub2_c; |
213 |
|
transfer_8to16sub2ro = transfer_8to16sub2ro_c; |
214 |
transfer_16to8add = transfer_16to8add_c; |
transfer_16to8add = transfer_16to8add_c; |
215 |
transfer8x8_copy = transfer8x8_copy_c; |
transfer8x8_copy = transfer8x8_copy_c; |
216 |
|
|
222 |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_c; |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_c; |
223 |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c; |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_c; |
224 |
|
|
225 |
|
interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_c; |
226 |
|
interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_c; |
227 |
|
interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_c; |
228 |
|
interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_c; |
229 |
|
|
230 |
interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c; |
interpolate16x16_lowpass_h = interpolate16x16_lowpass_h_c; |
231 |
interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c; |
interpolate16x16_lowpass_v = interpolate16x16_lowpass_v_c; |
232 |
interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c; |
interpolate16x16_lowpass_hv = interpolate16x16_lowpass_hv_c; |
241 |
interpolate8x8_avg2 = interpolate8x8_avg2_c; |
interpolate8x8_avg2 = interpolate8x8_avg2_c; |
242 |
interpolate8x8_avg4 = interpolate8x8_avg4_c; |
interpolate8x8_avg4 = interpolate8x8_avg4_c; |
243 |
|
|
244 |
/* reduced resoltuion */ |
/* postprocessing */ |
245 |
copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_C; |
image_brightness = image_brightness_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; |
|
246 |
|
|
247 |
/* Initialize internal colorspace transformation tables */ |
/* Initialize internal colorspace transformation tables */ |
248 |
colorspace_init(); |
colorspace_init(); |
255 |
bgra_to_yv12 = bgra_to_yv12_c; |
bgra_to_yv12 = bgra_to_yv12_c; |
256 |
abgr_to_yv12 = abgr_to_yv12_c; |
abgr_to_yv12 = abgr_to_yv12_c; |
257 |
rgba_to_yv12 = rgba_to_yv12_c; |
rgba_to_yv12 = rgba_to_yv12_c; |
258 |
|
argb_to_yv12 = argb_to_yv12_c; |
259 |
yuyv_to_yv12 = yuyv_to_yv12_c; |
yuyv_to_yv12 = yuyv_to_yv12_c; |
260 |
uyvy_to_yv12 = uyvy_to_yv12_c; |
uyvy_to_yv12 = uyvy_to_yv12_c; |
261 |
|
|
265 |
bgrai_to_yv12 = bgrai_to_yv12_c; |
bgrai_to_yv12 = bgrai_to_yv12_c; |
266 |
abgri_to_yv12 = abgri_to_yv12_c; |
abgri_to_yv12 = abgri_to_yv12_c; |
267 |
rgbai_to_yv12 = rgbai_to_yv12_c; |
rgbai_to_yv12 = rgbai_to_yv12_c; |
268 |
|
argbi_to_yv12 = argbi_to_yv12_c; |
269 |
yuyvi_to_yv12 = yuyvi_to_yv12_c; |
yuyvi_to_yv12 = yuyvi_to_yv12_c; |
270 |
uyvyi_to_yv12 = uyvyi_to_yv12_c; |
uyvyi_to_yv12 = uyvyi_to_yv12_c; |
271 |
|
|
|
|
|
272 |
/* All colorspace transformation functions YV12->User format */ |
/* All colorspace transformation functions YV12->User format */ |
273 |
yv12_to_rgb555 = yv12_to_rgb555_c; |
yv12_to_rgb555 = yv12_to_rgb555_c; |
274 |
yv12_to_rgb565 = yv12_to_rgb565_c; |
yv12_to_rgb565 = yv12_to_rgb565_c; |
276 |
yv12_to_bgra = yv12_to_bgra_c; |
yv12_to_bgra = yv12_to_bgra_c; |
277 |
yv12_to_abgr = yv12_to_abgr_c; |
yv12_to_abgr = yv12_to_abgr_c; |
278 |
yv12_to_rgba = yv12_to_rgba_c; |
yv12_to_rgba = yv12_to_rgba_c; |
279 |
|
yv12_to_argb = yv12_to_argb_c; |
280 |
yv12_to_yuyv = yv12_to_yuyv_c; |
yv12_to_yuyv = yv12_to_yuyv_c; |
281 |
yv12_to_uyvy = yv12_to_uyvy_c; |
yv12_to_uyvy = yv12_to_uyvy_c; |
282 |
|
|
286 |
yv12_to_bgrai = yv12_to_bgrai_c; |
yv12_to_bgrai = yv12_to_bgrai_c; |
287 |
yv12_to_abgri = yv12_to_abgri_c; |
yv12_to_abgri = yv12_to_abgri_c; |
288 |
yv12_to_rgbai = yv12_to_rgbai_c; |
yv12_to_rgbai = yv12_to_rgbai_c; |
289 |
|
yv12_to_argbi = yv12_to_argbi_c; |
290 |
yv12_to_yuyvi = yv12_to_yuyvi_c; |
yv12_to_yuyvi = yv12_to_yuyvi_c; |
291 |
yv12_to_uyvyi = yv12_to_uyvyi_c; |
yv12_to_uyvyi = yv12_to_uyvyi_c; |
292 |
|
|
298 |
sad8bi = sad8bi_c; |
sad8bi = sad8bi_c; |
299 |
dev16 = dev16_c; |
dev16 = dev16_c; |
300 |
sad16v = sad16v_c; |
sad16v = sad16v_c; |
301 |
|
sse8_16bit = sse8_16bit_c; |
302 |
/* Halfpel8_Refine = Halfpel8_Refine_c; */ |
sse8_8bit = sse8_8bit_c; |
303 |
|
|
304 |
#if defined(ARCH_IS_IA32) |
#if defined(ARCH_IS_IA32) |
305 |
|
|
|
if ((cpu_flags & XVID_CPU_ASM)) |
|
|
{ |
|
|
vfilter_31 = xvid_VFilter_31_x86; |
|
|
hfilter_31 = xvid_HFilter_31_x86; |
|
|
} |
|
|
|
|
306 |
if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) || |
if ((cpu_flags & XVID_CPU_MMX) || (cpu_flags & XVID_CPU_MMXEXT) || |
307 |
(cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) || |
(cpu_flags & XVID_CPU_3DNOW) || (cpu_flags & XVID_CPU_3DNOWEXT) || |
308 |
(cpu_flags & XVID_CPU_SSE) || (cpu_flags & XVID_CPU_SSE2)) |
(cpu_flags & XVID_CPU_SSE) || (cpu_flags & XVID_CPU_SSE2)) |
314 |
if ((cpu_flags & XVID_CPU_MMX)) { |
if ((cpu_flags & XVID_CPU_MMX)) { |
315 |
|
|
316 |
/* Forward and Inverse Discrete Cosine Transformation functions */ |
/* Forward and Inverse Discrete Cosine Transformation functions */ |
317 |
fdct = fdct_mmx; |
fdct = fdct_mmx_skal; |
318 |
idct = idct_mmx; |
idct = idct_mmx; |
319 |
|
|
320 |
|
/* Qpel stuff */ |
321 |
|
xvid_QP_Funcs = &xvid_QP_Funcs_mmx; |
322 |
|
xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_mmx; |
323 |
|
|
324 |
/* Quantization related functions */ |
/* Quantization related functions */ |
325 |
quant_intra = quant_intra_mmx; |
quant_h263_intra = quant_h263_intra_mmx; |
326 |
dequant_intra = dequant_intra_mmx; |
quant_h263_inter = quant_h263_inter_mmx; |
327 |
quant_inter = quant_inter_mmx; |
dequant_h263_intra = dequant_h263_intra_mmx; |
328 |
dequant_inter = dequant_inter_mmx; |
dequant_h263_inter = dequant_h263_inter_mmx; |
329 |
|
|
330 |
quant4_intra = quant4_intra_mmx; |
quant_mpeg_intra = quant_mpeg_intra_mmx; |
331 |
dequant4_intra = dequant4_intra_mmx; |
quant_mpeg_inter = quant_mpeg_inter_mmx; |
332 |
quant4_inter = quant4_inter_mmx; |
dequant_mpeg_intra = dequant_mpeg_intra_mmx; |
333 |
dequant4_inter = dequant4_inter_mmx; |
dequant_mpeg_inter = dequant_mpeg_inter_mmx; |
334 |
|
|
335 |
/* Block related functions */ |
/* Block related functions */ |
336 |
transfer_8to16copy = transfer_8to16copy_mmx; |
transfer_8to16copy = transfer_8to16copy_mmx; |
349 |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_mmx; |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_mmx; |
350 |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx; |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_mmx; |
351 |
|
|
352 |
|
interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_mmx; |
353 |
|
interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_mmx; |
354 |
|
interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_mmx; |
355 |
|
interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_mmx; |
356 |
|
|
357 |
interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx; |
interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_mmx; |
358 |
interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx; |
interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_mmx; |
359 |
|
|
360 |
interpolate8x8_avg2 = interpolate8x8_avg2_mmx; |
interpolate8x8_avg2 = interpolate8x8_avg2_mmx; |
361 |
interpolate8x8_avg4 = interpolate8x8_avg4_mmx; |
interpolate8x8_avg4 = interpolate8x8_avg4_mmx; |
362 |
|
|
363 |
/* reduced resolution */ |
/* postprocessing */ |
364 |
copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_mmx; |
image_brightness = image_brightness_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; |
|
365 |
|
|
366 |
/* image input xxx_to_yv12 related functions */ |
/* image input xxx_to_yv12 related functions */ |
367 |
yv12_to_yv12 = yv12_to_yv12_mmx; |
yv12_to_yv12 = yv12_to_yv12_mmx; |
387 |
sad8bi = sad8bi_mmx; |
sad8bi = sad8bi_mmx; |
388 |
dev16 = dev16_mmx; |
dev16 = dev16_mmx; |
389 |
sad16v = sad16v_mmx; |
sad16v = sad16v_mmx; |
390 |
|
sse8_16bit = sse8_16bit_mmx; |
391 |
|
sse8_8bit = sse8_8bit_mmx; |
392 |
} |
} |
393 |
|
|
394 |
/* these 3dnow functions are faster than mmx, but slower than xmm. */ |
/* these 3dnow functions are faster than mmx, but slower than xmm. */ |
407 |
|
|
408 |
if ((cpu_flags & XVID_CPU_MMXEXT)) { |
if ((cpu_flags & XVID_CPU_MMXEXT)) { |
409 |
|
|
410 |
/* Inverse DCT */ |
/* DCT */ |
411 |
|
fdct = fdct_xmm_skal; |
412 |
idct = idct_xmm; |
idct = idct_xmm; |
413 |
|
|
414 |
/* Interpolation */ |
/* Interpolation */ |
416 |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_xmm; |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_xmm; |
417 |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm; |
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_xmm; |
418 |
|
|
419 |
/* reduced resolution */ |
interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_xmm; |
420 |
copy_upsampled_8x8_16to8 = xvid_Copy_Upsampled_8x8_16To8_xmm; |
interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_xmm; |
421 |
add_upsampled_8x8_16to8 = xvid_Add_Upsampled_8x8_16To8_xmm; |
interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_xmm; |
422 |
|
interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_xmm; |
423 |
|
|
424 |
/* Quantization */ |
/* Quantization */ |
425 |
quant4_intra = quant4_intra_xmm; |
quant_mpeg_intra = quant_mpeg_intra_xmm; |
426 |
quant4_inter = quant4_inter_xmm; |
quant_mpeg_inter = quant_mpeg_inter_xmm; |
427 |
|
|
428 |
dequant_intra = dequant_intra_xmm; |
dequant_h263_intra = dequant_h263_intra_xmm; |
429 |
dequant_inter = dequant_inter_xmm; |
dequant_h263_inter = dequant_h263_inter_xmm; |
430 |
|
|
431 |
/* Buffer transfer */ |
/* Buffer transfer */ |
432 |
transfer_8to16sub2 = transfer_8to16sub2_xmm; |
transfer_8to16sub2 = transfer_8to16sub2_xmm; |
433 |
|
transfer_8to16sub2ro = transfer_8to16sub2ro_xmm; |
434 |
|
|
435 |
/* Colorspace transformation */ |
/* Colorspace transformation */ |
436 |
yv12_to_yv12 = yv12_to_yv12_xmm; |
yv12_to_yv12 = yv12_to_yv12_xmm; |
456 |
|
|
457 |
if ((cpu_flags & XVID_CPU_3DNOWEXT)) { |
if ((cpu_flags & XVID_CPU_3DNOWEXT)) { |
458 |
|
|
|
/* Inverse DCT */ |
|
|
idct = idct_3dne; |
|
|
|
|
459 |
/* Buffer transfer */ |
/* Buffer transfer */ |
460 |
transfer_8to16copy = transfer_8to16copy_3dne; |
transfer_8to16copy = transfer_8to16copy_3dne; |
461 |
transfer_16to8copy = transfer_16to8copy_3dne; |
transfer_16to8copy = transfer_16to8copy_3dne; |
462 |
transfer_8to16sub = transfer_8to16sub_3dne; |
transfer_8to16sub = transfer_8to16sub_3dne; |
463 |
transfer_8to16subro = transfer_8to16subro_3dne; |
transfer_8to16subro = transfer_8to16subro_3dne; |
|
transfer_8to16sub2 = transfer_8to16sub2_3dne; |
|
464 |
transfer_16to8add = transfer_16to8add_3dne; |
transfer_16to8add = transfer_16to8add_3dne; |
465 |
transfer8x8_copy = transfer8x8_copy_3dne; |
transfer8x8_copy = transfer8x8_copy_3dne; |
466 |
|
|
467 |
|
if ((cpu_flags & XVID_CPU_MMXEXT)) { |
468 |
|
/* Inverse DCT */ |
469 |
|
idct = idct_3dne; |
470 |
|
|
471 |
|
/* Buffer transfer */ |
472 |
|
transfer_8to16sub2 = transfer_8to16sub2_3dne; |
473 |
|
|
474 |
|
/* Interpolation */ |
475 |
|
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_3dne; |
476 |
|
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_3dne; |
477 |
|
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_3dne; |
478 |
|
|
479 |
/* Quantization */ |
/* Quantization */ |
480 |
dequant4_intra = dequant4_intra_3dne; |
quant_h263_intra = quant_h263_intra_3dne; /* cmov only */ |
481 |
dequant4_inter = dequant4_inter_3dne; |
quant_h263_inter = quant_h263_inter_3dne; |
482 |
quant_intra = quant_intra_3dne; |
dequant_mpeg_intra = dequant_mpeg_intra_3dne; /* cmov only */ |
483 |
quant_inter = quant_inter_3dne; |
dequant_mpeg_inter = dequant_mpeg_inter_3dne; |
484 |
dequant_intra = dequant_intra_3dne; |
dequant_h263_intra = dequant_h263_intra_3dne; |
485 |
dequant_inter = dequant_inter_3dne; |
dequant_h263_inter = dequant_h263_inter_3dne; |
486 |
|
|
487 |
/* ME functions */ |
/* ME functions */ |
488 |
calc_cbp = calc_cbp_3dne; |
calc_cbp = calc_cbp_3dne; |
489 |
|
|
490 |
sad16 = sad16_3dne; |
sad16 = sad16_3dne; |
491 |
sad8 = sad8_3dne; |
sad8 = sad8_3dne; |
492 |
sad16bi = sad16bi_3dne; |
sad16bi = sad16bi_3dne; |
493 |
sad8bi = sad8bi_3dne; |
sad8bi = sad8bi_3dne; |
494 |
dev16 = dev16_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; |
|
495 |
} |
} |
496 |
|
} |
497 |
|
|
498 |
if ((cpu_flags & XVID_CPU_SSE2)) { |
if ((cpu_flags & XVID_CPU_SSE2)) { |
499 |
|
|
|
#if defined(EXPERIMENTAL_SSE2_CODE) /* many people reported crashes with SSE2 */ |
|
|
/* better deactivate it completely and fix everything */ |
|
|
/* in dev-api-4 */ |
|
500 |
calc_cbp = calc_cbp_sse2; |
calc_cbp = calc_cbp_sse2; |
501 |
|
|
502 |
/* Quantization */ |
/* Quantization */ |
503 |
quant_intra = quant_intra_sse2; |
quant_h263_intra = quant_h263_intra_sse2; |
504 |
dequant_intra = dequant_intra_sse2; |
quant_h263_inter = quant_h263_inter_sse2; |
505 |
quant_inter = quant_inter_sse2; |
dequant_h263_intra = dequant_h263_intra_sse2; |
506 |
dequant_inter = dequant_inter_sse2; |
dequant_h263_inter = dequant_h263_inter_sse2; |
507 |
|
|
508 |
/* ME; slower than xmm */ |
/* SAD operators */ |
509 |
sad16 = sad16_sse2; |
sad16 = sad16_sse2; |
510 |
dev16 = dev16_sse2; |
dev16 = dev16_sse2; |
511 |
#endif |
|
512 |
/* Forward and Inverse DCT */ |
/* DCT operators |
513 |
/* idct = idct_sse2; |
* no iDCT because it's not "Walken matching" */ |
514 |
/* fdct = fdct_sse2; Both are none to be unprecise - better deactivate for now */ |
fdct = fdct_sse2_skal; |
515 |
|
|
516 |
|
/* postprocessing */ |
517 |
|
image_brightness = image_brightness_sse2; |
518 |
} |
} |
519 |
#endif |
#endif /* ARCH_IS_IA32 */ |
520 |
|
|
521 |
#if defined(ARCH_IS_IA64) |
#if defined(ARCH_IS_IA64) |
522 |
if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */ |
if ((cpu_flags & XVID_CPU_ASM)) { /* use assembler routines? */ |
531 |
sad8 = sad8_ia64; |
sad8 = sad8_ia64; |
532 |
dev16 = dev16_ia64; |
dev16 = dev16_ia64; |
533 |
/* Halfpel8_Refine = Halfpel8_Refine_ia64; */ |
/* Halfpel8_Refine = Halfpel8_Refine_ia64; */ |
534 |
quant_intra = quant_intra_ia64; |
quant_h263_intra = quant_h263_intra_ia64; |
535 |
dequant_intra = dequant_intra_ia64; |
quant_h263_inter = quant_h263_inter_ia64; |
536 |
quant_inter = quant_inter_ia64; |
dequant_h263_intra = dequant_h263_intra_ia64; |
537 |
dequant_inter = dequant_inter_ia64; |
dequant_h263_inter = dequant_h263_inter_ia64; |
538 |
transfer_8to16copy = transfer_8to16copy_ia64; |
transfer_8to16copy = transfer_8to16copy_ia64; |
539 |
transfer_16to8copy = transfer_16to8copy_ia64; |
transfer_16to8copy = transfer_16to8copy_ia64; |
540 |
transfer_8to16sub = transfer_8to16sub_ia64; |
transfer_8to16sub = transfer_8to16sub_ia64; |
541 |
transfer_8to16sub2 = transfer_8to16sub2_ia64; |
transfer_8to16sub2 = transfer_8to16sub2_ia64; |
542 |
transfer_16to8add = transfer_16to8add_ia64; |
transfer_16to8add = transfer_16to8add_ia64; |
543 |
transfer8x8_copy = transfer8x8_copy_ia64; |
transfer8x8_copy = transfer8x8_copy_ia64; |
|
DPRINTF(DPRINTF_DEBUG, "Using IA-64 assembler routines."); |
|
544 |
} |
} |
545 |
#endif |
#endif |
546 |
|
|
547 |
#if defined(ARCH_IS_PPC) |
#if defined(ARCH_IS_PPC) |
548 |
if ((cpu_flags & XVID_CPU_ASM)) |
if ((cpu_flags & XVID_CPU_ALTIVEC)) { |
549 |
{ |
/* sad operators */ |
550 |
calc_cbp = calc_cbp_ppc; |
sad16 = sad16_altivec_c; |
551 |
} |
sad16bi = sad16bi_altivec_c; |
552 |
|
sad8 = sad8_altivec_c; |
553 |
if ((cpu_flags & XVID_CPU_ALTIVEC)) |
dev16 = dev16_altivec_c; |
554 |
{ |
|
555 |
calc_cbp = calc_cbp_altivec; |
sse8_16bit = sse8_16bit_altivec_c; |
556 |
fdct = fdct_altivec; |
|
557 |
idct = idct_altivec; |
/* mem transfer */ |
558 |
sadInit = sadInit_altivec; |
transfer_8to16copy = transfer_8to16copy_altivec_c; |
559 |
sad16 = sad16_altivec; |
transfer_16to8copy = transfer_16to8copy_altivec_c; |
560 |
sad8 = sad8_altivec; |
transfer_8to16sub = transfer_8to16sub_altivec_c; |
561 |
dev16 = dev16_altivec; |
transfer_8to16subro = transfer_8to16subro_altivec_c; |
562 |
} |
transfer_8to16sub2 = transfer_8to16sub2_altivec_c; |
563 |
#endif |
transfer_16to8add = transfer_16to8add_altivec_c; |
564 |
|
transfer8x8_copy = transfer8x8_copy_altivec_c; |
|
return XVID_ERR_OK; |
|
|
} |
|
|
|
|
|
|
|
|
|
|
|
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; |
|
|
} |
|
565 |
|
|
566 |
|
/* Inverse DCT */ |
567 |
|
idct = idct_altivec_c; |
568 |
|
|
569 |
emms(); |
/* Interpolation */ |
570 |
return XVID_ERR_OK; |
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_altivec_c; |
571 |
} |
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_altivec_c; |
572 |
|
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_altivec_c; |
573 |
|
|
574 |
|
interpolate8x8_avg2 = interpolate8x8_avg2_altivec_c; |
575 |
|
interpolate8x8_avg4 = interpolate8x8_avg4_altivec_c; |
576 |
|
|
577 |
|
interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_altivec_c; |
578 |
|
interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_altivec_c; |
579 |
|
interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_altivec_c; |
580 |
|
interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_altivec_c; |
581 |
|
|
582 |
|
/* Colorspace conversion */ |
583 |
|
bgra_to_yv12 = bgra_to_yv12_altivec_c; |
584 |
|
abgr_to_yv12 = abgr_to_yv12_altivec_c; |
585 |
|
rgba_to_yv12 = rgba_to_yv12_altivec_c; |
586 |
|
argb_to_yv12 = argb_to_yv12_altivec_c; |
587 |
|
|
588 |
|
yuyv_to_yv12 = yuyv_to_yv12_altivec_c; |
589 |
|
uyvy_to_yv12 = uyvy_to_yv12_altivec_c; |
590 |
|
|
591 |
|
yv12_to_yuyv = yv12_to_yuyv_altivec_c; |
592 |
|
yv12_to_uyvy = yv12_to_uyvy_altivec_c; |
593 |
|
|
594 |
void fill8(uint8_t * block, int size, int value) |
/* Quantization */ |
595 |
{ |
quant_h263_intra = quant_h263_intra_altivec_c; |
596 |
int i; |
quant_h263_inter = quant_h263_inter_altivec_c; |
597 |
for (i = 0; i < size; i++) |
dequant_h263_intra = dequant_h263_intra_altivec_c; |
598 |
block[i] = value; |
dequant_h263_inter = dequant_h263_inter_altivec_c; |
599 |
|
|
600 |
|
dequant_mpeg_intra = dequant_mpeg_intra_altivec_c; |
601 |
|
dequant_mpeg_inter = dequant_mpeg_inter_altivec_c; |
602 |
|
|
603 |
|
/* Qpel stuff */ |
604 |
|
xvid_QP_Funcs = &xvid_QP_Funcs_Altivec_C; |
605 |
|
xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_Altivec_C; |
606 |
} |
} |
607 |
|
#endif |
608 |
|
|
609 |
void fill16(int16_t * block, int size, int value) |
#if defined(ARCH_IS_X86_64) |
610 |
{ |
/* For now, only XVID_CPU_ASM is looked for, so user can still |
611 |
int i; |
* disable asm usage the usual way. When Intel EMT64 cpus will |
612 |
for (i = 0; i < size; i++) |
* be out, maybe we'll have to check more precisely what cpu |
613 |
block[i] = value; |
* features there really are. */ |
614 |
} |
if (cpu_flags & XVID_CPU_ASM) { |
615 |
|
/* SIMD state flusher */ |
616 |
|
emms = emms_3dn; |
617 |
|
|
618 |
#define RANDOM(min,max) min + (rand() % (max-min)) |
/* DCT operators */ |
619 |
|
fdct = fdct_skal_x86_64; |
620 |
|
idct = idct_x86_64; |
621 |
|
|
622 |
|
/* SAD operators */ |
623 |
|
sad16 = sad16_x86_64; |
624 |
|
sad8 = sad8_x86_64; |
625 |
|
sad16bi = sad16bi_x86_64; |
626 |
|
sad8bi = sad8bi_x86_64; |
627 |
|
dev16 = dev16_x86_64; |
628 |
|
sad16v = sad16v_x86_64; |
629 |
|
sse8_16bit = sse8_16bit_x86_64; |
630 |
|
sse8_8bit = sse8_8bit_x86_64; |
631 |
|
|
632 |
|
/* Interpolation operators */ |
633 |
|
interpolate8x8_halfpel_h = interpolate8x8_halfpel_h_x86_64; |
634 |
|
interpolate8x8_halfpel_v = interpolate8x8_halfpel_v_x86_64; |
635 |
|
interpolate8x8_halfpel_hv = interpolate8x8_halfpel_hv_x86_64; |
636 |
|
|
637 |
|
interpolate8x8_halfpel_add = interpolate8x8_halfpel_add_x86_64; |
638 |
|
interpolate8x8_halfpel_h_add = interpolate8x8_halfpel_h_add_x86_64; |
639 |
|
interpolate8x8_halfpel_v_add = interpolate8x8_halfpel_v_add_x86_64; |
640 |
|
interpolate8x8_halfpel_hv_add = interpolate8x8_halfpel_hv_add_x86_64; |
641 |
|
|
642 |
void random8(uint8_t * block, int size, int min, int max) |
interpolate8x8_6tap_lowpass_h = interpolate8x8_6tap_lowpass_h_x86_64; |
643 |
{ |
interpolate8x8_6tap_lowpass_v = interpolate8x8_6tap_lowpass_v_x86_64; |
|
int i; |
|
|
for (i = 0; i < size; i++) |
|
|
block[i] = RANDOM(min,max); |
|
|
} |
|
644 |
|
|
645 |
void random16(int16_t * block, int size, int min, int max) |
interpolate8x8_avg2 = interpolate8x8_avg2_x86_64; |
646 |
{ |
interpolate8x8_avg4 = interpolate8x8_avg4_x86_64; |
|
int i; |
|
|
for (i = 0; i < size; i++) |
|
|
block[i] = RANDOM(min,max); |
|
|
} |
|
647 |
|
|
648 |
int compare16(const int16_t * blockA, const int16_t * blockB, int size) |
/* Quantization related functions */ |
649 |
{ |
quant_h263_intra = quant_h263_intra_x86_64; |
650 |
int i; |
quant_h263_inter = quant_h263_inter_x86_64; |
651 |
for (i = 0; i < size; i++) |
dequant_h263_intra = dequant_h263_intra_x86_64; |
652 |
if (blockA[i] != blockB[i]) |
dequant_h263_inter = dequant_h263_inter_x86_64; |
653 |
return 1; |
quant_mpeg_intra = quant_mpeg_intra_x86_64; |
654 |
|
quant_mpeg_inter = quant_mpeg_inter_x86_64; |
655 |
|
dequant_mpeg_intra = dequant_mpeg_intra_x86_64; |
656 |
|
dequant_mpeg_inter = dequant_mpeg_inter_x86_64; |
657 |
|
|
658 |
return 0; |
/* Block related functions */ |
659 |
} |
transfer_8to16copy = transfer_8to16copy_x86_64; |
660 |
|
transfer_16to8copy = transfer_16to8copy_x86_64; |
661 |
|
transfer_8to16sub = transfer_8to16sub_x86_64; |
662 |
|
transfer_8to16subro = transfer_8to16subro_x86_64; |
663 |
|
transfer_8to16sub2 = transfer_8to16sub2_x86_64; |
664 |
|
transfer_8to16sub2ro= transfer_8to16sub2ro_x86_64; |
665 |
|
transfer_16to8add = transfer_16to8add_x86_64; |
666 |
|
transfer8x8_copy = transfer8x8_copy_x86_64; |
667 |
|
|
668 |
|
/* Qpel stuff */ |
669 |
|
xvid_QP_Funcs = &xvid_QP_Funcs_x86_64; |
670 |
|
xvid_QP_Add_Funcs = &xvid_QP_Add_Funcs_x86_64; |
671 |
|
|
672 |
int diff16(const int16_t * blockA, const int16_t * blockB, int size) |
/* Interlacing Functions */ |
673 |
{ |
MBFieldTest = MBFieldTest_x86_64; |
|
int i, diff = 0; |
|
|
for (i = 0; i < size; i++) |
|
|
diff += ABS(blockA[i]-blockB[i]); |
|
|
return diff; |
|
674 |
} |
} |
675 |
|
#endif |
676 |
|
|
677 |
|
#if defined(_DEBUG) |
678 |
|
xvid_debug = init->debug; |
679 |
|
#endif |
680 |
|
|
681 |
#define XVID_TEST_RANDOM 0x00000001 /* random input data */ |
return(0); |
|
#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; |
|
682 |
} |
} |
683 |
|
|
|
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)); |
|
684 |
|
|
685 |
if ((test & TEST_FORWARD)) |
static int |
686 |
{ |
xvid_gbl_info(xvid_gbl_info_t * info) |
|
timeSTART = read_counter(); |
|
|
((fdctFunc*)funcA)(arrayA); |
|
|
timeA += read_counter() - timeSTART; |
|
|
|
|
|
timeSTART = read_counter(); |
|
|
((fdctFunc*)funcB)(arrayB); |
|
|
timeB += read_counter() - timeSTART; |
|
|
} |
|
|
else |
|
687 |
{ |
{ |
688 |
timeSTART = read_counter(); |
if (XVID_VERSION_MAJOR(info->version) != 1) /* v1.x.x */ |
689 |
((idctFunc*)funcA)(arrayA); |
return XVID_ERR_VERSION; |
|
timeA += read_counter() - timeSTART; |
|
690 |
|
|
691 |
timeSTART = read_counter(); |
info->actual_version = XVID_VERSION; |
692 |
((idctFunc*)funcB)(arrayB); |
info->build = "xvid-1.1.0-beta2"; |
693 |
timeB += read_counter() - timeSTART; |
info->cpu_flags = detect_cpu_flags(); |
|
} |
|
|
|
|
|
tmp = diff16(arrayA, arrayB, 64) / 64; |
|
|
if (tmp > max_error) |
|
|
max_error = tmp; |
|
|
if (tmp < min_error) |
|
|
min_error = tmp; |
|
|
|
|
|
count++; |
|
|
} |
|
694 |
|
|
695 |
/* print the "average difference" of best/worst transforms */ |
#if defined(_SMP) && defined(WIN32) |
696 |
printf("%s:\t%i\t(min_error:%i, max_error:%i)\n", nameB, (int)(timeB / count), min_error, max_error); |
info->num_threads = pthread_num_processors_np();; |
697 |
|
#else |
698 |
|
info->num_threads = 0; |
699 |
|
#endif |
700 |
|
|
701 |
return 0; |
return 0; |
702 |
} |
} |
703 |
|
|
704 |
|
|
705 |
#define TEST_QUANT 0x00000001 /* forward quantization */ |
static int |
706 |
#define TEST_INTRA 0x00000002 /* intra */ |
xvid_gbl_convert(xvid_gbl_convert_t* convert) |
|
#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 */ |
|
707 |
{ |
{ |
708 |
min = -2048; |
int width; |
709 |
max = 2047; |
int height; |
710 |
}else{ /* dequant */ |
int width2; |
711 |
min = -256; |
int height2; |
712 |
max = 255; |
IMAGE img; |
|
} |
|
713 |
|
|
714 |
for (q = 1; q <= 31; q++) /* quantizer */ |
if (XVID_VERSION_MAJOR(convert->version) != 1) /* v1.x.x */ |
715 |
{ |
return XVID_ERR_VERSION; |
|
for (i = min; i < max; i++) /* input coeff */ |
|
|
{ |
|
|
if ((flags & XVID_TEST_RANDOM)) |
|
|
{ |
|
|
random16(arrayX, 64, min, max); |
|
|
}else{ |
|
|
fill16(arrayX, 64, i); |
|
|
} |
|
716 |
|
|
717 |
if ((test & TEST_INTRA)) /* intra */ |
#if 0 |
718 |
{ |
const int flip1 = (convert->input.colorspace & XVID_CSP_VFLIP) ^ (convert->output.colorspace & XVID_CSP_VFLIP); |
719 |
timeSTART = read_counter(); |
#endif |
720 |
((quanth263_intraFunc*)funcA)(arrayA, arrayX, q, q); |
width = convert->width; |
721 |
timeA += read_counter() - timeSTART; |
height = convert->height; |
722 |
|
width2 = convert->width/2; |
723 |
timeSTART = read_counter(); |
height2 = convert->height/2; |
|
((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; |
|
|
} |
|
724 |
|
|
725 |
/* compare return value from quant_inter, and compare (de)quantiz'd arrays */ |
switch (convert->input.csp & ~XVID_CSP_VFLIP) |
|
if ( ((test&TEST_QUANT) && !(test&TEST_INTRA) && retA != retB ) || |
|
|
compare16(arrayA, arrayB, 64)) |
|
726 |
{ |
{ |
727 |
errors++; |
case XVID_CSP_YV12 : |
728 |
if ((flags & XVID_TEST_VERBOSE)) |
img.y = convert->input.plane[0]; |
729 |
printf("%s error: q=%i, i=%i\n", nameB, q, i); |
img.v = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height; |
730 |
} |
img.u = (uint8_t*)convert->input.plane[0] + convert->input.stride[0]*height + (convert->input.stride[0]/2)*height2; |
731 |
|
image_output(&img, width, height, width, |
732 |
|
(uint8_t**)convert->output.plane, convert->output.stride, |
733 |
|
convert->output.csp, convert->interlacing); |
734 |
|
break; |
735 |
|
|
736 |
count++; |
default : |
737 |
} |
return XVID_ERR_FORMAT; |
738 |
} |
} |
739 |
|
|
|
printf("%s:\t%i", nameB, (int)(timeB / count)); |
|
|
if (errors>0) |
|
|
printf("\t(%i errors out of %i)", errors, count); |
|
|
printf("\n"); |
|
740 |
|
|
741 |
|
emms(); |
742 |
return 0; |
return 0; |
743 |
} |
} |
744 |
|
|
745 |
|
/***************************************************************************** |
746 |
|
* XviD Global Entry point |
747 |
int xvid_init_test(int flags) |
* |
748 |
{ |
* Well this function initialize all internal function pointers according |
749 |
#if defined(ARCH_IS_IA32) |
* to the CPU features forced by the library client or autodetected (depending |
750 |
int cpu_flags; |
* on the XVID_CPU_FORCE flag). It also initializes vlc coding tables and all |
751 |
#endif |
* image colorspace transformation tables. |
752 |
|
* |
753 |
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 XVID_ERR_OK; |
|
|
} |
|
754 |
|
|
755 |
|
|
756 |
int |
int |
757 |
xvid_init(void *handle, |
xvid_global(void *handle, |
758 |
int opt, |
int opt, |
759 |
void *param1, |
void *param1, |
760 |
void *param2) |
void *param2) |
761 |
{ |
{ |
762 |
switch(opt) |
switch(opt) |
763 |
{ |
{ |
764 |
case XVID_INIT_INIT : |
case XVID_GBL_INIT : |
765 |
return xvid_init_init((XVID_INIT_PARAM*)param1); |
return xvid_gbl_init((xvid_gbl_init_t*)param1); |
766 |
|
|
767 |
case XVID_INIT_CONVERT : |
case XVID_GBL_INFO : |
768 |
return xvid_init_convert((XVID_INIT_CONVERTINFO*)param1); |
return xvid_gbl_info((xvid_gbl_info_t*)param1); |
769 |
|
|
770 |
|
case XVID_GBL_CONVERT : |
771 |
|
return xvid_gbl_convert((xvid_gbl_convert_t*)param1); |
772 |
|
|
|
case XVID_INIT_TEST : |
|
|
{ |
|
|
ptr_t flags = (ptr_t)param1; |
|
|
return xvid_init_test((int)flags); |
|
|
} |
|
773 |
default : |
default : |
774 |
return XVID_ERR_FAIL; |
return XVID_ERR_FAIL; |
775 |
} |
} |
792 |
void *param2) |
void *param2) |
793 |
{ |
{ |
794 |
switch (opt) { |
switch (opt) { |
|
case XVID_DEC_DECODE: |
|
|
return decoder_decode((DECODER *) handle, (XVID_DEC_FRAME *) param1, (XVID_DEC_STATS*) param2); |
|
|
|
|
795 |
case XVID_DEC_CREATE: |
case XVID_DEC_CREATE: |
796 |
return decoder_create((XVID_DEC_PARAM *) param1); |
return decoder_create((xvid_dec_create_t *) param1); |
797 |
|
|
798 |
case XVID_DEC_DESTROY: |
case XVID_DEC_DESTROY: |
799 |
return decoder_destroy((DECODER *) handle); |
return decoder_destroy((DECODER *) handle); |
800 |
|
|
801 |
|
case XVID_DEC_DECODE: |
802 |
|
return decoder_decode((DECODER *) handle, (xvid_dec_frame_t *) param1, (xvid_dec_stats_t*) param2); |
803 |
|
|
804 |
default: |
default: |
805 |
return XVID_ERR_FAIL; |
return XVID_ERR_FAIL; |
806 |
} |
} |
826 |
switch (opt) { |
switch (opt) { |
827 |
case XVID_ENC_ENCODE: |
case XVID_ENC_ENCODE: |
828 |
|
|
829 |
if (((Encoder *) handle)->mbParam.max_bframes >= 0) |
return enc_encode((Encoder *) handle, |
830 |
return encoder_encode_bframes((Encoder *) handle, |
(xvid_enc_frame_t *) param1, |
831 |
(XVID_ENC_FRAME *) param1, |
(xvid_enc_stats_t *) param2); |
|
(XVID_ENC_STATS *) param2); |
|
|
else |
|
|
return encoder_encode((Encoder *) handle, |
|
|
(XVID_ENC_FRAME *) param1, |
|
|
(XVID_ENC_STATS *) param2); |
|
832 |
|
|
833 |
case XVID_ENC_CREATE: |
case XVID_ENC_CREATE: |
834 |
return encoder_create((XVID_ENC_PARAM *) param1); |
return enc_create((xvid_enc_create_t *) param1); |
835 |
|
|
836 |
case XVID_ENC_DESTROY: |
case XVID_ENC_DESTROY: |
837 |
return encoder_destroy((Encoder *) handle); |
return enc_destroy((Encoder *) handle); |
838 |
|
|
839 |
default: |
default: |
840 |
return XVID_ERR_FAIL; |
return XVID_ERR_FAIL; |