xvid_bench.c

mpeg4 video codec mpeg4 video codec

			   cpu->name, t, cbp, (cbp!=0x15)?"| ERROR": "");
		TEST_CBP(calc_cbp, Src2, nb_tests);
		printf("%s -   calc_cbp#2 %.3f usec       cbp=0x%02x %s\n",
			   cpu->name, t, cbp, (cbp!=0x38)?"| ERROR": "");
		TEST_CBP(calc_cbp, Src3, nb_tests);
		printf("%s -   calc_cbp#3 %.3f usec       cbp=0x%02x %s\n",
			   cpu->name, t, cbp, (cbp!=0x0f)?"| ERROR": "" );
		TEST_CBP(calc_cbp, Src4, nb_tests);
		printf("%s -   calc_cbp#4 %.3f usec       cbp=0x%02x %s\n",
			   cpu->name, t, cbp, (cbp!=0x05)?"| ERROR": "" );
		TEST_CBP(calc_cbp, Src5, nb_tests);
		printf("%s -   calc_cbp#4 %.3f usec       cbp=0x%02x %s\n",
			   cpu->name, t, cbp, (cbp!=0x3f)?"| ERROR": "" );
		printf( " --- \n" );
	}

	for(cpu = cpu_list; cpu->name!=0; ++cpu)  /* bench suggested by Carlo (carlo dot bramix at libero dot it) */
	{
		double t;
		int tst, cbp, err;

		if (!init_cpu(cpu))
			continue;

    err = 0;
    for(n=0; n<6; ++n)
    {
      for(m=0; m<64; ++m)
      {
        for(i=0; i<6*64; ++i)
          Src1[i] = (i== (m + n*64));

        TEST_CBP(calc_cbp, Src1, 1);
        if (cbp!= (((m!=0)<<(5-n))))
        {
          printf( "%s -   calc_cbp#5: ERROR at pos %d / %d!\n", cpu->name, n, m);
          err = 1;
          break;
        }
      }
    }
    if (!err)
      printf( " %s -    calc_cbp#5 : OK\n", cpu->name );

	}
}

/*********************************************************************
 * fdct/idct IEEE1180 compliance
 *********************************************************************/

typedef struct {
	long Errors[64];
	long Sqr_Errors[64];
	long Max_Errors[64];
	long Nb;
} STATS_8x8;

void init_stats(STATS_8x8 *S)
{
	int i;
	for(i=0; i<64; ++i) {
		S->Errors[i]     = 0;
		S->Sqr_Errors[i] = 0;
		S->Max_Errors[i] = 0;
	}
	S->Nb = 0;
}

void store_stats(STATS_8x8 *S, short Blk[64], short Ref[64])
{
	int i;
	for(i=0; i<64; ++i)
	{
		short Err = Blk[i] - Ref[i];
		S->Errors[i] += Err;
		S->Sqr_Errors[i] += Err * Err;
		if (Err<0) Err = -Err;
		if (S->Max_Errors[i]<Err)
			S->Max_Errors[i] = Err;
	}
	S->Nb++;
}

void print_stats(STATS_8x8 *S)
{
	int i;
	double Norm;

	assert(S->Nb>0);
	Norm = 1. / (double)S->Nb;
	printf("\n== Max absolute values of errors ==\n");
	for(i=0; i<64; i++) {
		printf("  %4ld", S->Max_Errors[i]);
		if ((i&7)==7) printf("\n");
	}

	printf("\n== Mean square errors ==\n");
	for(i=0; i<64; i++)
	{
		double Err = Norm * (double)S->Sqr_Errors[i];
		printf(" %.3f", Err);
		if ((i&7)==7) printf("\n");
	}

	printf("\n== Mean errors ==\n");
	for(i=0; i<64; i++)
	{
		double Err = Norm * (double)S->Errors[i];
		printf(" %.3f", Err);
		if ((i&7)==7) printf("\n");
	}
	printf("\n");
}

static const char *CHECK(double v, double l) {
	if (fabs(v)<=l) return "ok";
	else return "FAIL!";
}

void report_stats(STATS_8x8 *S, const double *Limits)
{
	int i;
	double Norm, PE, PMSE, OMSE, PME, OME;

	assert(S->Nb>0);
	Norm = 1. / (double)S->Nb;
	PE = 0.;
	for(i=0; i<64; i++) {
		if (PE<S->Max_Errors[i])
			PE = S->Max_Errors[i];
	}

	PMSE = 0.;
	OMSE = 0.;
	for(i=0; i<64; i++)
	{
		double Err = Norm * (double)S->Sqr_Errors[i];
		OMSE += Err;
		if (PMSE < Err) PMSE = Err;
	}
	OMSE /= 64.;

	PME = 0.;
	OME = 0.;
	for(i=0; i<64; i++)
	{
		double Err = Norm * (double)S->Errors[i];
		OME += Err;
		Err = fabs(Err);
		if (PME < Err) PME = Err;
	}
	OME /= 64.;

	printf( "Peak error:   %4.4f\n", PE );
	printf( "Peak MSE:     %4.4f\n", PMSE );
	printf( "Overall MSE:  %4.4f\n", OMSE );
	printf( "Peak ME:      %4.4f\n", PME );
	printf( "Overall ME:   %4.4f\n", OME );

	if (Limits!=0)
	{
		printf( "[PE<=%.4f %s]  ", Limits[0], CHECK(PE,   Limits[0]) );
		printf( "\n" );
		printf( "[PMSE<=%.4f %s]", Limits[1], CHECK(PMSE, Limits[1]) );
		printf( "[OMSE<=%.4f %s]", Limits[2], CHECK(OMSE, Limits[2]) );
		printf( "\n" );
		printf( "[PME<=%.4f %s] ", Limits[3], CHECK(PME , Limits[3]) );
		printf( "[OME<=%.4f %s] ", Limits[4], CHECK(OME , Limits[4]) );
		printf( "\n" );
	}
}

///* ////////////////////////////////////////////////////// */
/* Pseudo-random generator specified by IEEE 1180 */

static long ieee_seed = 1;
static void ieee_reseed(long s) {
	ieee_seed = s;
}
static long ieee_rand(int Min, int Max)
{
	static double z = (double) 0x7fffffff;

	long i,j;
	double x;

	ieee_seed = (ieee_seed * 1103515245) + 12345;
	i = ieee_seed & 0x7ffffffe;
	x = ((double) i) / z;
	x *= (Max-Min+1);
	j = (long)x;
	j = j + Min;
	assert(j>=Min && j<=Max);
	return (short)j;
}

#define CLAMP(x, M)   (x) = ((x)<-(M)) ? (-(M)) : ((x)>=(M) ? ((M)-1) : (x))

static double Cos[8][8];
static void init_ref_dct()
{
	int i, j;
	for(i=0; i<8; i++)
	{
		double scale = (i == 0) ? sqrt(0.125) : 0.5;
		for (j=0; j<8; j++)
			Cos[i][j] = scale*cos( (M_PI/8.0)*i*(j + 0.5) );
	}
}

void ref_idct(short *M)
{
	int i, j, k;
	double Tmp[8][8];

	for(i=0; i<8; i++) {
		for(j=0; j<8; j++)
		{
			double Sum = 0.0;
			for (k=0; k<8; k++) Sum += Cos[k][j]*M[8*i+k];
			Tmp[i][j] = Sum;
		}
	}
	for(i=0; i<8; i++) {
		for(j=0; j<8; j++) {
			double Sum = 0.0;
			for (k=0; k<8; k++) Sum += Cos[k][i]*Tmp[k][j];
			M[8*i+j] = (short)floor(Sum + .5);
		}
	}
}

void ref_fdct(short *M)
{
	int i, j, k;
	double Tmp[8][8];

	for(i=0; i<8; i++) {
		for(j=0; j<8; j++)
		{
			double Sum = 0.0;
			for (k=0; k<8; k++) Sum += Cos[j][k]*M[8*i+k];
			Tmp[i][j] = Sum;
		}
	}
	for(i=0; i<8; i++) {
		for(j=0; j<8; j++) {
			double Sum = 0.0;
			for (k=0; k<8; k++) Sum += Cos[i][k]*Tmp[k][j];
			M[8*i+j] = (short)floor(Sum + 0.5);
		}
	}
}

void test_IEEE1180_compliance(int Min, int Max, int Sign)
{
	static const double ILimits[5] = { 1., 0.06, 0.02, 0.015, 0.0015 };
	int Loops = 10000;
	int i, m, n;
	DECLARE_ALIGNED_MATRIX(Blk0, 8, 8, short, 16); /* reference */
	DECLARE_ALIGNED_MATRIX(Blk,  8, 8, short, 16);
	DECLARE_ALIGNED_MATRIX(iBlk, 8, 8, short, 16);
	DECLARE_ALIGNED_MATRIX(Ref_FDCT, 8, 8, short, 16);
	DECLARE_ALIGNED_MATRIX(Ref_IDCT, 8, 8, short, 16);

	STATS_8x8 FStats; /* forward dct stats */
	STATS_8x8 IStats; /* inverse dct stats */

	CPU *cpu;

	init_ref_dct();

	for(cpu = cpu_list; cpu->name!=0; ++cpu)
	{
		if (!init_cpu(cpu))
			continue;

		printf( "\n===== IEEE test for %s ==== (Min=%d Max=%d Sign=%d Loops=%d)\n",
				cpu->name, Min, Max, Sign, Loops);

		init_stats(&IStats);
		init_stats(&FStats);

		ieee_reseed(1);
		for(n=0; n<Loops; ++n)
		{
			for(i=0; i<64; ++i)
				Blk0[i] = (short)ieee_rand(Min,Max) * Sign;

			/* hmm, I'm not quite sure this is exactly */
			/* the tests described in the norm. check... */

			memcpy(Ref_FDCT, Blk0, 64*sizeof(short));
			ref_fdct(Ref_FDCT);
			for(i=0; i<64; i++) CLAMP( Ref_FDCT[i], 2048 );

			memcpy(Blk, Blk0, 64*sizeof(short));
			emms(); fdct(Blk); emms();
			for(i=0; i<64; i++) CLAMP( Blk[i], 2048 );

			store_stats(&FStats, Blk, Ref_FDCT);


			memcpy(Ref_IDCT, Ref_FDCT, 64*sizeof(short));
			ref_idct(Ref_IDCT);
			for (i=0; i<64; i++) CLAMP( Ref_IDCT[i], 256 );

			memcpy(iBlk, Ref_FDCT, 64*sizeof(short));
			emms(); idct(iBlk); emms();
			for(i=0; i<64; i++) CLAMP( iBlk[i], 256 );

			store_stats(&IStats, iBlk, Ref_IDCT);
		}


		printf( "\n  -- FDCT report --\n" );
//    print_stats(&FStats);
		report_stats(&FStats, 0); /* so far I know, IEEE1180 says nothing for fdct */

		for(i=0; i<64; i++) Blk[i] = 0;
		emms(); fdct(Blk); emms();
		for(m=i=0; i<64; i++) if (Blk[i]!=0) m++;
		printf( "FDCT(0) == 0 ?  %s\n", (m!=0) ? "NOPE!" : "yup." );

		printf( "\n  -- IDCT report --\n" );
//    print_stats(&IStats);
		report_stats(&IStats, ILimits);


		for(i=0; i<64; i++) Blk[i] = 0;
		emms(); idct(Blk); emms();
		for(m=i=0; i<64; i++) if (Blk[i]!=0) m++;
		printf( "IDCT(0) == 0 ?  %s\n", (m!=0) ? "NOPE!" : "yup." );
	}
}


void test_dct_saturation(int Min, int Max)
{
/* test behaviour on input range fringe */

	int i, n, p;
	CPU *cpu;
//  const short IDCT_MAX =  2047;  /* 12bits input */
//  const short IDCT_MIN = -2048;
//  const short IDCT_OUT =   256;  /* 9bits ouput */
	const int Partitions = 4;
	const int Loops = 10000 / Partitions;

	init_ref_dct();

	for(cpu = cpu_list; cpu->name!=0; ++cpu)
	{
		short Blk0[64], Blk[64];
		STATS_8x8 Stats;

		if (!init_cpu(cpu))
			continue;

		printf( "\n===== IEEE test for %s Min=%d Max=%d =====\n",
				cpu->name, Min, Max );

		/* FDCT tests // */

		init_stats(&Stats);

		/* test each computation channels separately */
		for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? Max : 0;
		ref_fdct(Blk0);
		emms(); fdct(Blk); emms();
		store_stats(&Stats, Blk, Blk0);

		for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? Min : 0;
		ref_fdct(Blk0);
		emms(); fdct(Blk); emms();
		store_stats(&Stats, Blk, Blk0);

		/* randomly saturated inputs */
		for(p=0; p<Partitions; ++p)
		{
			for(n=0; n<Loops; ++n)
			{
				for(i=0; i<64; ++i)
					Blk0[i] = Blk[i] = (ieee_rand(0,Partitions)>=p)? Max : Min;
				ref_fdct(Blk0);
				emms(); fdct(Blk); emms();
				store_stats(&Stats, Blk, Blk0);
			}
		}
		printf( "\n  -- FDCT saturation report --\n" );
		report_stats(&Stats, 0);


		/* IDCT tests // */
#if 0
		/* no finished yet */

		init_stats(&Stats);

/* test each computation channel separately */
		for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? IDCT_MAX : 0;
		ref_idct(Blk0);
		emms(); idct(Blk); emms();
		for(i=0; i<64; i++) { CLAMP(Blk0[i], IDCT_OUT); CLAMP(Blk[i], IDCT_OUT); }
		store_stats(&Stats, Blk, Blk0);

		for(i=0; i<64; i++) Blk[i] = Blk0[i] = ((i/8)==(i%8)) ? IDCT_MIN : 0;
		ref_idct(Blk0);
		emms(); idct(Blk); emms();
		for(i=0; i<64; i++) { CLAMP(Blk0[i], IDCT_OUT); CLAMP(Blk[i], IDCT_OUT); }
		store_stats(&Stats, Blk, Blk0);

		/* randomly saturated inputs */
		for(p=0; p<Partitions; ++p)
		{
			for(n=0; n<Loops; ++n)
			{
				for(i=0; i<64; ++i)
					Blk0[i] = Blk[i] = (ieee_rand(0,Partitions)>=p)? IDCT_MAX : IDCT_MIN;
				ref_idct(Blk0);
				emms(); idct(Blk); emms();
				for(i=0; i<64; i++) { CLAMP(Blk0[i],IDCT_OUT); CLAMP(Blk[i],IDCT_OUT); }
				store_stats(&Stats, Blk, Blk0);
			}
		}

		printf( "\n  -- IDCT saturation report --\n" );
		print_stats(&Stats);
		report_stats(&Stats, 0);
#endif
	}
}

/*********************************************************************
 * measure raw decoding speed
 *********************************************************************/

void test_dec(const char *name, int width, int height, int ref_chksum)
{
	FILE *f = 0;
	void *dechandle = 0;
	int xerr;
	xvid_gbl_init_t xinit;
	xvid_dec_create_t xparam;
	xvid_dec_frame_t xframe;
	double t = 0.;
	int nb = 0;
	uint8_t *buf = 0;
	uint8_t *yuv_out = 0;
	int buf_size, pos;
	uint32_t chksum = 0;
	int bps = (width+31) & ~31;

	memset(&xinit, 0, sizeof(xinit));
	xinit.cpu_flags = cpu_mask;
	xinit.version = XVID_VERSION;
	xvid_global(NULL, 0, &xinit, NULL);

	memset(&xparam, 0, sizeof(xparam));
	xparam.width  = width;
	xparam.height = height;
	xparam.version = XVID_VERSION;
	xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL);
	if (xerr==XVID_ERR_FAIL) {
		printf("ERROR: can't init decoder (err=%d)\n", xerr);
		return;
	}
	dechandle = xparam.handle;


	f = fopen(name, "rb");
	if (f==0) {
		printf( "ERROR: can't open file '%s'\n", name);
		return;
	}
	fseek(f, 0, SEEK_END);
	buf_size = ftell(f);
	fseek(f, 0, SEEK_SET);
	if (buf_size<=0) {
		printf("ERROR: error while stating file\n");
		goto End;
	}

	buf = malloc(buf_size);
	yuv_out = calloc(1, bps*height*3/2 + 15);
	if (buf==0 || yuv_out==0) {
		printf( "ERROR: malloc failed!\n" );
		goto End;
	}

	if (fread(buf, buf_size, 1, f)!=1) {
		printf( "ERROR: file-read failed\n" );
		goto End;
	}