X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdct-test.c;h=f337576836edfccc1c0888d5cd3198a4554d7de8;hb=3a651f599a18b023602370b67a77eb0efa309b20;hp=69195e497d6b71d73f158ee208c64bc7c4702641;hpb=875f31259b8d14d091002a980eb722ef06f4d2f8;p=ffmpeg diff --git a/libavcodec/dct-test.c b/libavcodec/dct-test.c index 69195e497d6..f337576836e 100644 --- a/libavcodec/dct-test.c +++ b/libavcodec/dct-test.c @@ -2,365 +2,257 @@ * (c) 2001 Fabrice Bellard * 2007 Marc Hoffman * - * This file is part of FFmpeg. + * This file is part of Libav. * - * FFmpeg is free software; you can redistribute it and/or + * Libav 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, + * Libav 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 + * License along with Libav; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ /** - * @file dct-test.c - * DCT test. (c) 2001 Fabrice Bellard. + * @file + * DCT test (c) 2001 Fabrice Bellard * Started from sample code by Juan J. Sierralta P. */ +#include "config.h" #include #include #include -#include +#if HAVE_UNISTD_H #include +#endif #include +#include "libavutil/cpu.h" #include "libavutil/common.h" +#include "libavutil/lfg.h" +#include "libavutil/time.h" +#include "dct.h" +#include "idctdsp.h" #include "simple_idct.h" +#include "xvididct.h" #include "aandcttab.h" #include "faandct.h" #include "faanidct.h" -#include "i386/idct_xvid.h" - -#undef printf -#undef random - -void *fast_memcpy(void *a, const void *b, size_t c){return memcpy(a,b,c);}; - -/* reference fdct/idct */ -void fdct(DCTELEM *block); -void idct(DCTELEM *block); -void init_fdct(); - -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 - +#include "dctref.h" 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; + const char *name; + void (*func)(int16_t *block); + enum idct_permutation_type perm_type; + int cpu_flag; + int nonspec; }; -#ifndef FAAN_POSTSCALE -#define FAAN_SCALE SCALE_PERM -#else -#define FAAN_SCALE NO_PERM -#endif - -static int cpu_flags; - -struct algo algos[] = { - {"REF-DBL", 0, fdct, fdct, NO_PERM}, - {"FAAN", 0, ff_faandct, fdct, FAAN_SCALE}, - {"FAANI", 1, ff_faanidct, idct, NO_PERM}, - {"IJG-AAN-INT", 0, fdct_ifast, fdct, SCALE_PERM}, - {"IJG-LLM-INT", 0, ff_jpeg_fdct_islow, fdct, NO_PERM}, - {"REF-DBL", 1, idct, idct, NO_PERM}, - {"INT", 1, j_rev_dct, idct, MMX_PERM}, - {"SIMPLE-C", 1, ff_simple_idct, idct, NO_PERM}, - -#ifdef HAVE_MMX - {"MMX", 0, ff_fdct_mmx, fdct, NO_PERM, FF_MM_MMX}, -#ifdef HAVE_MMX2 - {"MMX2", 0, ff_fdct_mmx2, fdct, NO_PERM, FF_MM_MMXEXT}, - {"SSE2", 0, ff_fdct_sse2, fdct, NO_PERM, FF_MM_SSE2}, -#endif - -#ifdef CONFIG_GPL - {"LIBMPEG2-MMX", 1, ff_mmx_idct, idct, MMX_PERM, FF_MM_MMX}, - {"LIBMPEG2-MMXEXT", 1, ff_mmxext_idct, idct, MMX_PERM, FF_MM_MMXEXT}, -#endif - {"SIMPLE-MMX", 1, ff_simple_idct_mmx, idct, MMX_SIMPLE_PERM, FF_MM_MMX}, - {"XVID-MMX", 1, ff_idct_xvid_mmx, idct, NO_PERM, FF_MM_MMX}, - {"XVID-MMX2", 1, ff_idct_xvid_mmx2, idct, NO_PERM, FF_MM_MMXEXT}, - {"XVID-SSE2", 1, ff_idct_xvid_sse2, idct, SSE2_PERM, FF_MM_SSE2}, -#endif +static const struct algo fdct_tab[] = { + { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE }, + { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE }, + { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE }, +#if CONFIG_FAANDCT + { "FAAN", ff_faandct, FF_IDCT_PERM_NONE }, +#endif /* CONFIG_FAANDCT */ +}; -#ifdef HAVE_ALTIVEC - {"altivecfdct", 0, fdct_altivec, fdct, NO_PERM, FF_MM_ALTIVEC}, -#endif +static const struct algo idct_tab[] = { + { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE }, + { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 }, + { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE }, +#if CONFIG_FAANIDCT + { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE }, +#endif /* CONFIG_FAANIDCT */ +#if CONFIG_MPEG4_DECODER + { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 }, +#endif /* CONFIG_MPEG4_DECODER */ +}; -#ifdef ARCH_BFIN - {"BFINfdct", 0, ff_bfin_fdct, fdct, NO_PERM}, - {"BFINidct", 1, ff_bfin_idct, idct, NO_PERM}, +#if ARCH_ARM +#include "arm/dct-test.c" +#elif ARCH_PPC +#include "ppc/dct-test.c" +#elif ARCH_X86 +#include "x86/dct-test.c" +#else +static const struct algo fdct_tab_arch[] = { 0 }; +static const struct algo idct_tab_arch[] = { 0 }; #endif - { 0 } -}; - #define AANSCALE_BITS 12 -uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; - -int64_t gettime(void) -{ - struct timeval tv; - gettimeofday(&tv,NULL); - return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec; -} - #define NB_ITS 20000 #define NB_ITS_SPEED 50000 -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, -}; - -static const uint8_t idct_sse2_row_perm[8] = {0, 4, 1, 5, 2, 6, 3, 7}; +DECLARE_ALIGNED(16, static int16_t, block)[64]; +DECLARE_ALIGNED(8, static int16_t, block1)[64]; -void idct_mmx_init(void) +static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng) { - int i; + int i, j; - /* 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); + memset(block, 0, 64 * sizeof(*block)); + + switch (test) { + case 0: + for (i = 0; i < 64; i++) + block[i] = (av_lfg_get(prng) % 512) - 256; + if (is_idct) { + ff_ref_fdct(block); + for (i = 0; i < 64; i++) + block[i] >>= 3; + } + break; + case 1: + j = av_lfg_get(prng) % 10 + 1; + for (i = 0; i < j; i++) + block[av_lfg_get(prng) % 64] = av_lfg_get(prng) % 512 - 256; + break; + case 2: + block[ 0] = av_lfg_get(prng) % 4096 - 2048; + block[63] = (block[0] & 1) ^ 1; + break; } } -static DCTELEM block[64] __attribute__ ((aligned (16))); -static DCTELEM block1[64] __attribute__ ((aligned (8))); -static DCTELEM block_org[64] __attribute__ ((aligned (8))); - -static inline void mmx_emms(void) +static void permute(int16_t dst[64], const int16_t src[64], + enum idct_permutation_type perm_type) { -#ifdef HAVE_MMX - if (cpu_flags & FF_MM_MMX) - __asm__ volatile ("emms\n\t"); + int i; + +#if ARCH_X86 + if (permute_x86(dst, src, perm_type)) + return; #endif + + switch (perm_type) { + case FF_IDCT_PERM_LIBMPEG2: + for (i = 0; i < 64; i++) + dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i]; + break; + case FF_IDCT_PERM_PARTTRANS: + for (i = 0; i < 64; i++) + dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i]; + break; + default: + for (i = 0; i < 64; i++) + dst[i] = src[i]; + break; + } } -void dct_error(const char *name, int is_idct, - void (*fdct_func)(DCTELEM *block), - void (*fdct_ref)(DCTELEM *block), int form, int test) +static int dct_error(const struct algo *dct, int test, int is_idct, int speed) { + void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct; int it, i, scale; int err_inf, v; - int64_t err2, ti, ti1, it1; - int64_t sysErr[64], sysErrMax=0; - int maxout=0; - int blockSumErrMax=0, blockSumErr; + int64_t err2, ti, ti1, it1, err_sum = 0; + int64_t sysErr[64], sysErrMax = 0; + int maxout = 0; + int blockSumErrMax = 0, blockSumErr; + AVLFG prng; + double omse, ome; + int spec_err; - srandom(0); + av_lfg_init(&prng, 1); err_inf = 0; err2 = 0; - for(i=0; i<64; i++) sysErr[i]=0; - for(it=0;it>=3; + for (i = 0; i < 64; i++) + sysErr[i] = 0; + for (it = 0; it < NB_ITS; it++) { + init_block(block1, test, is_idct, &prng); + permute(block, block1, dct->perm_type); + + dct->func(block); + emms_c(); + + if (!strcmp(dct->name, "IJG-AAN-INT")) { + for (i = 0; i < 64; i++) { + scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i]; + block[i] = (block[i] * scale) >> AANSCALE_BITS; } - break; - case 1:{ - int num= (random()%10)+1; - for(i=0;i>3)&3)] = block1[i]; - } else { - for(i=0; i<64; i++) - block[i]= block1[i]; - } -#if 0 // simulate mismatch control for tested IDCT but not the ref -{ int sum=0; - for(i=0;i<64;i++) - sum+=block[i]; - - if((sum&1)==0) block[63]^=1; -} -#endif - - fdct_func(block); - mmx_emms(); - - if (form == SCALE_PERM) { - for(i=0; i<64; i++) { - scale = 8*(1 << (AANSCALE_BITS + 11)) / ff_aanscales[i]; - block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS; - } - } - - fdct_ref(block1); - - blockSumErr=0; - for(i=0;i<64;i++) { - v = abs(block[i] - block1[i]); + blockSumErr = 0; + for (i = 0; i < 64; i++) { + int err = block[i] - block1[i]; + err_sum += err; + v = abs(err); 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]); - } - if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr; -#if 0 // print different matrix pairs - if(blockSumErr){ - printf("\n"); - for(i=0; i<64; i++){ - if((i&7)==0) printf("\n"); - printf("%4d ", block_org[i]); - } - for(i=0; i<64; i++){ - if((i&7)==0) printf("\n"); - printf("%4d ", block[i] - block1[i]); - } + if (abs(block[i]) > maxout) + maxout = abs(block[i]); } -#endif + if (blockSumErrMax < blockSumErr) + blockSumErrMax = blockSumErr; } - for(i=0; i<64; i++) sysErrMax= FFMAX(sysErrMax, FFABS(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]); + for (i = 0; i < 64; 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); -#if 1 //Speed test - /* speed test */ - for(i=0;i<64;i++) - block1[i] = 0; - switch(test){ - case 0: - for(i=0;i<64;i++) - block1[i] = (random() % 512) -256; - if (is_idct){ - fdct(block1); + omse = (double) err2 / NB_ITS / 64; + ome = (double) err_sum / NB_ITS / 64; - for(i=0;i<64;i++) - block1[i]>>=3; - } - 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; - }break; - } + spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015); - if (form == MMX_PERM) { - for(i=0;i<64;i++) - block[idct_mmx_perm[i]] = block1[i]; - } else if(form == MMX_SIMPLE_PERM) { - for(i=0;i<64;i++) - block[idct_simple_mmx_perm[i]] = block1[i]; - } else { - for(i=0; i<64; i++) - block[i]= block1[i]; - } + printf("%s %s: ppe=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n", + is_idct ? "IDCT" : "DCT", dct->name, err_inf, + omse, ome, (double) sysErrMax / NB_ITS, + maxout, blockSumErrMax); + + if (spec_err && !dct->nonspec) + return 1; - ti = gettime(); + if (!speed) + return 0; + + /* speed test */ + init_block(block, test, is_idct, &prng); + permute(block1, block, dct->perm_type); + + ti = av_gettime_relative(); it1 = 0; do { - for(it=0;itfunc(block); } it1 += NB_ITS_SPEED; - ti1 = gettime() - ti; + ti1 = av_gettime_relative() - ti; } while (ti1 < 1000000); - mmx_emms(); + emms_c(); - printf("%s %s: %0.1f kdct/s\n", - is_idct ? "IDCT" : "DCT", - name, (double)it1 * 1000.0 / (double)ti1); -#endif + printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name, + (double) it1 * 1000.0 / (double) ti1); + + return 0; } -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]; @@ -372,19 +264,19 @@ void idct248_ref(uint8_t *dest, int linesize, int16_t *block) if (!init) { init = 1; - for(i=0;i<8;i++) { + for (i = 0; i < 8; i++) { sum = 0; - for(j=0;j<8;j++) { - s = (i==0) ? sqrt(1.0/8.0) : sqrt(1.0/4.0); + for (j = 0; j < 8; j++) { + s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0); c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0); sum += c8[i][j] * c8[i][j]; } } - for(i=0;i<4;i++) { + for (i = 0; i < 4; i++) { sum = 0; - for(j=0;j<4;j++) { - s = (i==0) ? sqrt(1.0/4.0) : sqrt(1.0/2.0); + for (j = 0; j < 4; j++) { + s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0); c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0); sum += c4[i][j] * c4[i][j]; } @@ -393,184 +285,187 @@ void idct248_ref(uint8_t *dest, int linesize, int16_t *block) /* butterfly */ s = 0.5 * sqrt(2.0); - for(i=0;i<4;i++) { - for(j=0;j<8;j++) { - block1[8*(2*i)+j] = (block[8*(2*i)+j] + block[8*(2*i+1)+j]) * s; - block1[8*(2*i+1)+j] = (block[8*(2*i)+j] - block[8*(2*i+1)+j]) * s; + for (i = 0; i < 4; i++) { + for (j = 0; j < 8; j++) { + block1[8 * (2 * i) + j] = + (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s; + block1[8 * (2 * i + 1) + j] = + (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s; } } /* idct8 on lines */ - for(i=0;i<8;i++) { - for(j=0;j<8;j++) { + for (i = 0; i < 8; i++) { + for (j = 0; j < 8; j++) { sum = 0; - for(k=0;k<8;k++) - sum += c8[k][j] * block1[8*i+k]; - block2[8*i+j] = sum; + for (k = 0; k < 8; k++) + sum += c8[k][j] * block1[8 * i + k]; + block2[8 * i + j] = sum; } } /* idct4 */ - for(i=0;i<8;i++) { - for(j=0;j<4;j++) { + for (i = 0; i < 8; i++) { + for (j = 0; j < 4; j++) { /* top */ sum = 0; - for(k=0;k<4;k++) - sum += c4[k][j] * block2[8*(2*k)+i]; - block3[8*(2*j)+i] = sum; + for (k = 0; k < 4; k++) + sum += c4[k][j] * block2[8 * (2 * k) + i]; + block3[8 * (2 * j) + i] = sum; /* bottom */ sum = 0; - for(k=0;k<4;k++) - sum += c4[k][j] * block2[8*(2*k+1)+i]; - block3[8*(2*j+1)+i] = sum; + for (k = 0; k < 4; k++) + sum += c4[k][j] * block2[8 * (2 * k + 1) + i]; + block3[8 * (2 * j + 1) + i] = sum; } } /* clamp and store the result */ - for(i=0;i<8;i++) { - for(j=0;j<8;j++) { - v = block3[8*i+j]; - if (v < 0) - v = 0; - else if (v > 255) - v = 255; - dest[i * linesize + j] = (int)rint(v); + for (i = 0; i < 8; i++) { + for (j = 0; j < 8; j++) { + v = block3[8 * i + j]; + if (v < 0) v = 0; + else if (v > 255) v = 255; + dest[i * linesize + j] = (int) rint(v); } } } -void idct248_error(const char *name, - void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block)) +static void idct248_error(const char *name, + void (*idct248_put)(uint8_t *dest, int line_size, + int16_t *block), + int speed) { int it, i, it1, ti, ti1, err_max, v; + AVLFG prng; - srandom(0); + 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 err_max) err_max = v; } -#if 0 - printf("ref=\n"); - for(i=0;i<8;i++) { - int j; - for(j=0;j<8;j++) { - printf(" %3d", img_dest1[i*8+j]); - } - printf("\n"); - } - - printf("out=\n"); - for(i=0;i<8;i++) { - int j; - for(j=0;j<8;j++) { - printf(" %3d", img_dest[i*8+j]); - } - printf("\n"); - } -#endif } - printf("%s %s: err_inf=%d\n", - 1 ? "IDCT248" : "DCT248", - name, err_max); + printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max); + + if (!speed) + return; - ti = gettime(); + ti = av_gettime_relative(); it1 = 0; do { - for(it=0;it]\n" "test-number 0 -> test with random matrixes\n" " 1 -> test with random sparse matrixes\n" " 2 -> do 3. test from mpeg4 std\n" "-i test IDCT implementations\n" - "-4 test IDCT248 implementations\n"); + "-4 test IDCT248 implementations\n" + "-t speed test\n"); } +#if !HAVE_GETOPT +#include "compat/getopt.c" +#endif + 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(); - idct_mmx_init(); + int c, i; + int test = 1; + int speed = 0; + int err = 0; - for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i; - for(i=0;i