X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdct-test.c;h=80285476de43bc44d2855a4891e39b90eaaa68e7;hb=5bebe94d860cb1a0819701213e961c8269aa3284;hp=29410ca5af72a8a27235e930c5a0bb8212ce452d;hpb=0c1a9edad463bd6e22b30c19b700b099c7093fc1;p=ffmpeg diff --git a/libavcodec/dct-test.c b/libavcodec/dct-test.c index 29410ca5af7..80285476de4 100644 --- a/libavcodec/dct-test.c +++ b/libavcodec/dct-test.c @@ -1,48 +1,156 @@ -/* DCT test. (c) 2001 Fabrice Bellard. - Started from sample code by Juan J. Sierralta P. -*/ +/* + * (c) 2001 Fabrice Bellard + * 2007 Marc Hoffman + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +/** + * @file libavcodec/dct-test.c + * DCT test (c) 2001 Fabrice Bellard + * Started from sample code by Juan J. Sierralta P. + */ + #include #include #include #include #include +#include -#include "dsputil.h" +#include "libavutil/common.h" +#include "libavutil/lfg.h" -#include "i386/mmx.h" #include "simple_idct.h" +#include "aandcttab.h" +#include "faandct.h" +#include "faanidct.h" +#include "x86/idct_xvid.h" -#ifndef MAX -#define MAX(a, b) (((a) > (b)) ? (a) : (b)) -#endif +#undef printf + +void *fast_memcpy(void *a, const void *b, size_t c){return memcpy(a,b,c);}; /* reference fdct/idct */ -extern void fdct(DCTELEM *block); -extern void idct(DCTELEM *block); -extern void init_fdct(); +void ff_ref_fdct(DCTELEM *block); +void ff_ref_idct(DCTELEM *block); +void ff_ref_dct_init(void); + +void ff_mmx_idct(DCTELEM *data); +void ff_mmxext_idct(DCTELEM *data); + +void odivx_idct_c(short *block); + +// BFIN +void ff_bfin_idct(DCTELEM *block); +void ff_bfin_fdct(DCTELEM *block); + +// ALTIVEC +void fdct_altivec(DCTELEM *block); +//void idct_altivec(DCTELEM *block);?? no routine + +// ARM +void j_rev_dct_ARM(DCTELEM *data); +void simple_idct_ARM(DCTELEM *data); +void simple_idct_armv5te(DCTELEM *data); +void ff_simple_idct_armv6(DCTELEM *data); +void ff_simple_idct_neon(DCTELEM *data); + +void ff_simple_idct_axp(DCTELEM *data); + +struct algo { + const char *name; + enum { FDCT, IDCT } is_idct; + void (* func) (DCTELEM *block); + void (* ref) (DCTELEM *block); + enum formattag { NO_PERM,MMX_PERM, MMX_SIMPLE_PERM, SCALE_PERM, SSE2_PERM, PARTTRANS_PERM } format; + int mm_support; +}; -extern void j_rev_dct(DCTELEM *data); -extern void ff_mmx_idct(DCTELEM *data); -extern void ff_mmxext_idct(DCTELEM *data); +#ifndef FAAN_POSTSCALE +#define FAAN_SCALE SCALE_PERM +#else +#define FAAN_SCALE NO_PERM +#endif -extern void odivx_idct_c (short *block); +static int cpu_flags; + +struct algo algos[] = { + {"REF-DBL", 0, ff_ref_fdct, ff_ref_fdct, NO_PERM}, + {"FAAN", 0, ff_faandct, ff_ref_fdct, FAAN_SCALE}, + {"FAANI", 1, ff_faanidct, ff_ref_idct, NO_PERM}, + {"IJG-AAN-INT", 0, fdct_ifast, ff_ref_fdct, SCALE_PERM}, + {"IJG-LLM-INT", 0, ff_jpeg_fdct_islow, ff_ref_fdct, NO_PERM}, + {"REF-DBL", 1, ff_ref_idct, ff_ref_idct, NO_PERM}, + {"INT", 1, j_rev_dct, ff_ref_idct, MMX_PERM}, + {"SIMPLE-C", 1, ff_simple_idct, ff_ref_idct, NO_PERM}, + +#if HAVE_MMX + {"MMX", 0, ff_fdct_mmx, ff_ref_fdct, NO_PERM, FF_MM_MMX}, +#if HAVE_MMX2 + {"MMX2", 0, ff_fdct_mmx2, ff_ref_fdct, NO_PERM, FF_MM_MMX2}, + {"SSE2", 0, ff_fdct_sse2, ff_ref_fdct, NO_PERM, FF_MM_SSE2}, +#endif -#define AANSCALE_BITS 12 -static const unsigned short aanscales[64] = { - /* precomputed values scaled up by 14 bits */ - 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, - 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, - 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, - 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, - 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, - 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, - 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, - 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 +#if CONFIG_GPL + {"LIBMPEG2-MMX", 1, ff_mmx_idct, ff_ref_idct, MMX_PERM, FF_MM_MMX}, + {"LIBMPEG2-MMX2", 1, ff_mmxext_idct, ff_ref_idct, MMX_PERM, FF_MM_MMX2}, +#endif + {"SIMPLE-MMX", 1, ff_simple_idct_mmx, ff_ref_idct, MMX_SIMPLE_PERM, FF_MM_MMX}, + {"XVID-MMX", 1, ff_idct_xvid_mmx, ff_ref_idct, NO_PERM, FF_MM_MMX}, + {"XVID-MMX2", 1, ff_idct_xvid_mmx2, ff_ref_idct, NO_PERM, FF_MM_MMX2}, + {"XVID-SSE2", 1, ff_idct_xvid_sse2, ff_ref_idct, SSE2_PERM, FF_MM_SSE2}, +#endif + +#if HAVE_ALTIVEC + {"altivecfdct", 0, fdct_altivec, ff_ref_fdct, NO_PERM, FF_MM_ALTIVEC}, +#endif + +#if ARCH_BFIN + {"BFINfdct", 0, ff_bfin_fdct, ff_ref_fdct, NO_PERM}, + {"BFINidct", 1, ff_bfin_idct, ff_ref_idct, NO_PERM}, +#endif + +#if ARCH_ARM + {"SIMPLE-ARM", 1, simple_idct_ARM, ff_ref_idct, NO_PERM }, + {"INT-ARM", 1, j_rev_dct_ARM, ff_ref_idct, MMX_PERM }, +#if HAVE_ARMV5TE + {"SIMPLE-ARMV5TE", 1, simple_idct_armv5te, ff_ref_idct, NO_PERM }, +#endif +#if HAVE_ARMV6 + {"SIMPLE-ARMV6", 1, ff_simple_idct_armv6, ff_ref_idct, MMX_PERM }, +#endif +#if HAVE_NEON + {"SIMPLE-NEON", 1, ff_simple_idct_neon, ff_ref_idct, PARTTRANS_PERM }, +#endif +#endif /* ARCH_ARM */ + +#if ARCH_ALPHA + {"SIMPLE-ALPHA", 1, ff_simple_idct_axp, ff_ref_idct, NO_PERM }, +#endif + + { 0 } }; +#define AANSCALE_BITS 12 + uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; -int64_t gettime(void) +static int64_t gettime(void) { struct timeval tv; gettimeofday(&tv,NULL); @@ -55,34 +163,44 @@ int64_t gettime(void) static short idct_mmx_perm[64]; static short idct_simple_mmx_perm[64]={ - 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, - 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, - 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, - 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, - 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, - 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, - 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, - 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, + 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, + 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, + 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, + 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, + 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, + 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, + 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, + 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, }; -void idct_mmx_init(void) +static const uint8_t idct_sse2_row_perm[8] = {0, 4, 1, 5, 2, 6, 3, 7}; + +static void idct_mmx_init(void) { int i; /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */ for (i = 0; i < 64; i++) { - idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); -// idct_simple_mmx_perm[i] = simple_block_permute_op(i); + idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); +// idct_simple_mmx_perm[i] = simple_block_permute_op(i); } } -static DCTELEM block[64] __attribute__ ((aligned (8))); -static DCTELEM block1[64] __attribute__ ((aligned (8))); -static DCTELEM block_org[64] __attribute__ ((aligned (8))); +DECLARE_ALIGNED(16, static DCTELEM, block)[64]; +DECLARE_ALIGNED(8, static DCTELEM, block1)[64]; +DECLARE_ALIGNED(8, static DCTELEM, block_org)[64]; + +static inline void mmx_emms(void) +{ +#if HAVE_MMX + if (cpu_flags & FF_MM_MMX) + __asm__ volatile ("emms\n\t"); +#endif +} -void dct_error(const char *name, int is_idct, +static void dct_error(const char *name, int is_idct, void (*fdct_func)(DCTELEM *block), - void (*fdct_ref)(DCTELEM *block), int test) + void (*fdct_ref)(DCTELEM *block), int form, int test) { int it, i, scale; int err_inf, v; @@ -90,8 +208,9 @@ void dct_error(const char *name, int is_idct, int64_t sysErr[64], sysErrMax=0; int maxout=0; int blockSumErrMax=0, blockSumErr; + AVLFG prng; - srandom(0); + av_lfg_init(&prng, 1); err_inf = 0; err2 = 0; @@ -100,23 +219,23 @@ void dct_error(const char *name, int is_idct, for(i=0;i<64;i++) block1[i] = 0; switch(test){ - case 0: + case 0: for(i=0;i<64;i++) - block1[i] = (random() % 512) -256; + block1[i] = (av_lfg_get(&prng) % 512) -256; if (is_idct){ - fdct(block1); + ff_ref_fdct(block1); for(i=0;i<64;i++) block1[i]>>=3; } break; case 1:{ - int num= (random()%10)+1; + int num = av_lfg_get(&prng) % 10 + 1; for(i=0;i>3)&3)] = block1[i]; + } else { for(i=0; i<64; i++) block[i]= block1[i]; } @@ -150,16 +274,16 @@ void dct_error(const char *name, int is_idct, for(i=0;i<64;i++) sum+=block[i]; - if((sum&1)==0) block[63]^=1; + if((sum&1)==0) block[63]^=1; } #endif fdct_func(block); - emms(); /* for ff_mmx_idct */ + mmx_emms(); - if (fdct_func == fdct_ifast) { + if (form == SCALE_PERM) { for(i=0; i<64; i++) { - scale = 8*(1 << (AANSCALE_BITS + 11)) / aanscales[i]; + scale = 8*(1 << (AANSCALE_BITS + 11)) / ff_aanscales[i]; block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS; } } @@ -172,9 +296,9 @@ void dct_error(const char *name, int is_idct, if (v > err_inf) err_inf = v; err2 += v * v; - sysErr[i] += block[i] - block1[i]; - blockSumErr += v; - if( abs(block[i])>maxout) maxout=abs(block[i]); + sysErr[i] += block[i] - block1[i]; + blockSumErr += v; + if( abs(block[i])>maxout) maxout=abs(block[i]); } if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr; #if 0 // print different matrix pairs @@ -191,16 +315,16 @@ void dct_error(const char *name, int is_idct, } #endif } - for(i=0; i<64; i++) sysErrMax= MAX(sysErrMax, ABS(sysErr[i])); - + for(i=0; i<64; i++) sysErrMax= FFMAX(sysErrMax, FFABS(sysErr[i])); + #if 1 // dump systematic errors for(i=0; i<64; i++){ - if(i%8==0) printf("\n"); - printf("%5d ", (int)sysErr[i]); + if(i%8==0) printf("\n"); + printf("%7d ", (int)sysErr[i]); } printf("\n"); #endif - + printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n", is_idct ? "IDCT" : "DCT", name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax); @@ -209,11 +333,11 @@ void dct_error(const char *name, int is_idct, for(i=0;i<64;i++) block1[i] = 0; switch(test){ - case 0: + case 0: for(i=0;i<64;i++) - block1[i] = (random() % 512) -256; + block1[i] = av_lfg_get(&prng) % 512 -256; if (is_idct){ - fdct(block1); + ff_ref_fdct(block1); for(i=0;i<64;i++) block1[i]>>=3; @@ -221,18 +345,17 @@ void dct_error(const char *name, int is_idct, break; case 1:{ case 2: - block1[0] = (random() % 512) -256; - block1[1] = (random() % 512) -256; - block1[2] = (random() % 512) -256; - block1[3] = (random() % 512) -256; + block1[0] = av_lfg_get(&prng) % 512 -256; + block1[1] = av_lfg_get(&prng) % 512 -256; + block1[2] = av_lfg_get(&prng) % 512 -256; + block1[3] = av_lfg_get(&prng) % 512 -256; }break; } - if (fdct_func == ff_mmx_idct || - fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) { + if (form == MMX_PERM) { for(i=0;i<64;i++) block[idct_mmx_perm[i]] = block1[i]; - } else if(fdct_func == ff_simple_idct_mmx ) { + } else if(form == MMX_SIMPLE_PERM) { for(i=0;i<64;i++) block[idct_simple_mmx_perm[i]] = block1[i]; } else { @@ -247,13 +370,13 @@ void dct_error(const char *name, int is_idct, for(i=0; i<64; i++) block[i]= block1[i]; // memcpy(block, block1, sizeof(DCTELEM) * 64); -// dont memcpy especially not fastmemcpy because it does movntq !!! +// do not memcpy especially not fastmemcpy because it does movntq !!! fdct_func(block); } it1 += NB_ITS_SPEED; ti1 = gettime() - ti; } while (ti1 < 1000000); - emms(); + mmx_emms(); printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", @@ -261,10 +384,10 @@ void dct_error(const char *name, int is_idct, #endif } -static uint8_t img_dest[64] __attribute__ ((aligned (8))); -static uint8_t img_dest1[64] __attribute__ ((aligned (8))); +DECLARE_ALIGNED(8, static uint8_t, img_dest)[64]; +DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64]; -void idct248_ref(uint8_t *dest, int linesize, int16_t *block) +static void idct248_ref(uint8_t *dest, int linesize, int16_t *block) { static int init; static double c8[8][8]; @@ -284,7 +407,7 @@ void idct248_ref(uint8_t *dest, int linesize, int16_t *block) sum += c8[i][j] * c8[i][j]; } } - + for(i=0;i<4;i++) { sum = 0; for(j=0;j<4;j++) { @@ -344,31 +467,33 @@ void idct248_ref(uint8_t *dest, int linesize, int16_t *block) } } -void idct248_error(const char *name, +static void idct248_error(const char *name, void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block)) { int it, i, it1, ti, ti1, err_max, v; - srandom(0); - + AVLFG prng; + + av_lfg_init(&prng, 1); + /* just one test to see if code is correct (precision is less important here) */ err_max = 0; for(it=0;it]\n" "test-number 0 -> test with random matrixes\n" @@ -428,7 +553,6 @@ void help(void) " 2 -> do 3. test from mpeg4 std\n" "-i test IDCT implementations\n" "-4 test IDCT248 implementations\n"); - exit(1); } int main(int argc, char **argv) @@ -436,8 +560,9 @@ int main(int argc, char **argv) int test_idct = 0, test_248_dct = 0; int c,i; int test=1; + cpu_flags = mm_support(); - init_fdct(); + ff_ref_dct_init(); idct_mmx_init(); for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i; @@ -445,7 +570,7 @@ int main(int argc, char **argv) cropTbl[i] = 0; cropTbl[i + MAX_NEG_CROP + 256] = 255; } - + for(;;) { c = getopt(argc, argv, "ih4"); if (c == -1) @@ -460,38 +585,20 @@ int main(int argc, char **argv) default : case 'h': help(); - break; + return 0; } } - + if(optind