checkasm.c

linux下编译已经通过

#define MC_TEST_AVG( name, ... ) \
    for( i = 0, ok = 1, used_asm = 0; i < 10; i++ ) \
    { \
        memcpy( buf3, buf1, 1024 ); \
        memcpy( buf4, buf1, 1024 ); \
        if( mc_a.name[i] != mc_ref.name[i] ) \
        { \
            used_asm = 1; \
            mc_c.name[i]( buf3, 32, buf2, 16, ##__VA_ARGS__ ); \
            mc_a.name[i]( buf4, 32, buf2, 16, ##__VA_ARGS__ ); \
            if( memcmp( buf3, buf4, 1024 ) )               \
            { \
                ok = 0; \
                fprintf( stderr, #name "[%d]: [FAILED]\n", i ); \
            } \
        } \
    }
    MC_TEST_AVG( avg );
    report( "mc avg :" );
    for( w = -64; w <= 128 && ok; w++ )
        MC_TEST_AVG( avg_weight, w );
    report( "mc wpredb :" );

    return ret;
}

static int check_deblock( int cpu_ref, int cpu_new )
{
    x264_deblock_function_t db_c;
    x264_deblock_function_t db_ref;
    x264_deblock_function_t db_a;
    int ret = 0, ok = 1, used_asm = 0;
    int alphas[36], betas[36];
    int8_t tcs[36][4];
    int a, c, i, j;

    x264_deblock_init( 0, &db_c );
    x264_deblock_init( cpu_ref, &db_ref );
    x264_deblock_init( cpu_new, &db_a );

    /* not exactly the real values of a,b,tc but close enough */
    a = 255; c = 250;
    for( i = 35; i >= 0; i-- )
    {
        alphas[i] = a;
        betas[i] = (i+1)/2;
        tcs[i][0] = tcs[i][2] = (c+6)/10;
        tcs[i][1] = tcs[i][3] = (c+9)/20;
        a = a*9/10;
        c = c*9/10;
    }

#define TEST_DEBLOCK( name, ... ) \
    for( i = 0; i < 36; i++ ) \
    { \
        for( j = 0; j < 1024; j++ ) \
            /* two distributions of random to excersize different failure modes */\
            buf1[j] = rand() & (i&1 ? 0xf : 0xff ); \
        memcpy( buf3, buf1, 1024 ); \
        memcpy( buf4, buf1, 1024 ); \
        if( db_a.name != db_ref.name ) \
        { \
            used_asm = 1; \
            db_c.name( &buf3[8*32], 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
            db_a.name( &buf4[8*32], 32, alphas[i], betas[i], ##__VA_ARGS__ ); \
            if( memcmp( buf3, buf4, 1024 ) )               \
            { \
                ok = 0; \
                fprintf( stderr, #name "(a=%d, b=%d): [FAILED]\n", alphas[i], betas[i] ); \
                break; \
            } \
        } \
    }

    TEST_DEBLOCK( deblock_h_luma, tcs[i] );
    TEST_DEBLOCK( deblock_v_luma, tcs[i] );
    TEST_DEBLOCK( deblock_h_chroma, tcs[i] );
    TEST_DEBLOCK( deblock_v_chroma, tcs[i] );
    TEST_DEBLOCK( deblock_h_luma_intra );
    TEST_DEBLOCK( deblock_v_luma_intra );
    TEST_DEBLOCK( deblock_h_chroma_intra );
    TEST_DEBLOCK( deblock_v_chroma_intra );

    report( "deblock :" );

    return ret;
}

static int check_quant( int cpu_ref, int cpu_new )
{
    x264_quant_function_t qf_c;
    x264_quant_function_t qf_ref;
    x264_quant_function_t qf_a;
    int16_t dct1[64]    __attribute__((__aligned__(16)));
    int16_t dct2[64]    __attribute__((__aligned__(16)));
    uint8_t cqm_buf[64] __attribute__((__aligned__(16)));
    int ret = 0, ok, used_asm;
    int oks[2] = {1,1}, used_asms[2] = {0,0};
    int i, i_cqm, qp;
    x264_t h_buf;
    x264_t *h = &h_buf;
    h->pps = h->pps_array;
    x264_param_default( &h->param );
    h->param.rc.i_qp_min = 26;

    for( i_cqm = 0; i_cqm < 4; i_cqm++ )
    {
        if( i_cqm == 0 )
            for( i = 0; i < 6; i++ )
                h->pps->scaling_list[i] = x264_cqm_flat16;
        else if( i_cqm == 1 )
            for( i = 0; i < 6; i++ )
                h->pps->scaling_list[i] = x264_cqm_jvt[i];
        else
        {
            if( i_cqm == 2 )
                for( i = 0; i < 64; i++ )
                    cqm_buf[i] = 10 + rand() % 246;
            else
                for( i = 0; i < 64; i++ )
                    cqm_buf[i] = 1;
            for( i = 0; i < 6; i++ )
                h->pps->scaling_list[i] = cqm_buf;
        }

        x264_cqm_init( h );
        x264_quant_init( h, 0, &qf_c );
        x264_quant_init( h, cpu_ref, &qf_ref );
        x264_quant_init( h, cpu_new, &qf_a );

#define INIT_QUANT8() \
        { \
            static const int scale1d[8] = {32,31,24,31,32,31,24,31}; \
            int x, y; \
            for( y = 0; y < 8; y++ ) \
                for( x = 0; x < 8; x++ ) \
                { \
                    unsigned int scale = (255*scale1d[y]*scale1d[x])/16; \
                    dct1[y*8+x] = dct2[y*8+x] = (rand()%(2*scale+1))-scale; \
                } \
        }

#define INIT_QUANT4() \
        { \
            static const int scale1d[4] = {4,6,4,6}; \
            int x, y; \
            for( y = 0; y < 4; y++ ) \
                for( x = 0; x < 4; x++ ) \
                { \
                    unsigned int scale = 255*scale1d[y]*scale1d[x]; \
                    dct1[y*4+x] = dct2[y*4+x] = (rand()%(2*scale+1))-scale; \
                } \
        }

#define TEST_QUANT_DC( name, cqm ) \
        if( qf_a.name != qf_ref.name ) \
        { \
            used_asms[0] = 1; \
            for( qp = 51; qp > 0; qp-- ) \
            { \
                for( i = 0; i < 16; i++ ) \
                    dct1[i] = dct2[i] = (rand() & 0x1fff) - 0xfff; \
                qf_c.name( (void*)dct1, h->quant4_mf[CQM_4IY][qp][0], h->quant4_bias[CQM_4IY][qp][0] ); \
                qf_a.name( (void*)dct2, h->quant4_mf[CQM_4IY][qp][0], h->quant4_bias[CQM_4IY][qp][0] ); \
                if( memcmp( dct1, dct2, 16*2 ) )       \
                { \
                    oks[0] = 0; \
                    fprintf( stderr, #name "(cqm=%d): [FAILED]\n", i_cqm ); \
                    break; \
                } \
            } \
        }

#define TEST_QUANT( qname, block, w ) \
        if( qf_a.qname != qf_ref.qname ) \
        { \
            used_asms[0] = 1; \
            for( qp = 51; qp > 0; qp-- ) \
            { \
                INIT_QUANT##w() \
                qf_c.qname( (void*)dct1, h->quant##w##_mf[block][qp], h->quant##w##_bias[block][qp] ); \
                qf_a.qname( (void*)dct2, h->quant##w##_mf[block][qp], h->quant##w##_bias[block][qp] ); \
                if( memcmp( dct1, dct2, w*w*2 ) ) \
                { \
                    oks[0] = 0; \
                    fprintf( stderr, #qname "(qp=%d, cqm=%d, block="#block"): [FAILED]\n", qp, i_cqm ); \
                    break; \
                } \
            } \
        }

        TEST_QUANT( quant_8x8, CQM_8IY, 8 );
        TEST_QUANT( quant_8x8, CQM_8PY, 8 );
        TEST_QUANT( quant_4x4, CQM_4IY, 4 );
        TEST_QUANT( quant_4x4, CQM_4PY, 4 );
        TEST_QUANT_DC( quant_4x4_dc, **h->quant4_mf[CQM_4IY] );
        TEST_QUANT_DC( quant_2x2_dc, **h->quant4_mf[CQM_4IC] );

#define TEST_DEQUANT( qname, dqname, block, w ) \
        if( qf_a.dqname != qf_ref.dqname ) \
        { \
            used_asms[1] = 1; \
            for( qp = 51; qp > 0; qp-- ) \
            { \
                INIT_QUANT##w() \
                qf_c.qname( (void*)dct1, h->quant##w##_mf[block][qp], h->quant##w##_bias[block][qp] ); \
                memcpy( dct2, dct1, w*w*2 ); \
                qf_c.dqname( (void*)dct1, h->dequant##w##_mf[block], qp ); \
                qf_a.dqname( (void*)dct2, h->dequant##w##_mf[block], qp ); \
                if( memcmp( dct1, dct2, w*w*2 ) ) \
                { \
                    oks[1] = 0; \
                    fprintf( stderr, #dqname "(qp=%d, cqm=%d, block="#block"): [FAILED]\n", qp, i_cqm ); \
                    break; \
                } \
            } \
        }

        TEST_DEQUANT( quant_8x8, dequant_8x8, CQM_8IY, 8 );
        TEST_DEQUANT( quant_8x8, dequant_8x8, CQM_8PY, 8 );
        TEST_DEQUANT( quant_4x4, dequant_4x4, CQM_4IY, 4 );
        TEST_DEQUANT( quant_4x4, dequant_4x4, CQM_4PY, 4 );
    }

    ok = oks[0]; used_asm = used_asms[0];
    report( "quant :" );

    ok = oks[1]; used_asm = used_asms[1];
    report( "dequant :" );

    return ret;
}

static int check_intra( int cpu_ref, int cpu_new )
{
    int ret = 0, ok = 1, used_asm = 0;
    int i;
    DECLARE_ALIGNED( uint8_t, edge[33], 8 );
    struct
    {
        x264_predict_t      predict_16x16[4+3];
        x264_predict_t      predict_8x8c[4+3];
        x264_predict8x8_t   predict_8x8[9+3];
        x264_predict_t      predict_4x4[9+3];
    } ip_c, ip_ref, ip_a;

    x264_predict_16x16_init( 0, ip_c.predict_16x16 );
    x264_predict_8x8c_init( 0, ip_c.predict_8x8c );
    x264_predict_8x8_init( 0, ip_c.predict_8x8 );
    x264_predict_4x4_init( 0, ip_c.predict_4x4 );

    x264_predict_16x16_init( cpu_ref, ip_ref.predict_16x16 );
    x264_predict_8x8c_init( cpu_ref, ip_ref.predict_8x8c );
    x264_predict_8x8_init( cpu_ref, ip_ref.predict_8x8 );
    x264_predict_4x4_init( cpu_ref, ip_ref.predict_4x4 );

    x264_predict_16x16_init( cpu_new, ip_a.predict_16x16 );
    x264_predict_8x8c_init( cpu_new, ip_a.predict_8x8c );
    x264_predict_8x8_init( cpu_new, ip_a.predict_8x8 );
    x264_predict_4x4_init( cpu_new, ip_a.predict_4x4 );

    x264_predict_8x8_filter( buf1+48, edge, ALL_NEIGHBORS, ALL_NEIGHBORS );

#define INTRA_TEST( name, dir, ... ) \
    if( ip_a.name[dir] != ip_ref.name[dir] )\
    { \
        used_asm = 1; \
        memcpy( buf3, buf1, 32*20 );\
        memcpy( buf4, buf1, 32*20 );\
        ip_c.name[dir]( buf3+48, ##__VA_ARGS__ );\
        ip_a.name[dir]( buf4+48, ##__VA_ARGS__ );\
        if( memcmp( buf3, buf4, 32*20 ) )\
        {\
            fprintf( stderr, #name "[%d] :  [FAILED]\n", dir );\
            ok = 0;\
            int j,k;\
            for(k=-1; k<16; k++)\
                printf("%2x ", edge[16+k]);\
            printf("\n");\
            for(j=0; j<8; j++){\
                printf("%2x ", edge[j]);\
                for(k=0; k<8; k++)\
                    printf("%2x ", buf4[48+k+j*32]);\
                printf("\n");\
            }\
            printf("\n");\
            for(j=0; j<8; j++){\
                printf("   ");\
                for(k=0; k<8; k++)\
                    printf("%2x ", buf3[48+k+j*32]);\
                printf("\n");\
            }\
        }\
    }

    for( i = 0; i < 12; i++ )
        INTRA_TEST( predict_4x4, i );
    for( i = 0; i < 7; i++ )
        INTRA_TEST( predict_8x8c, i );
    for( i = 0; i < 7; i++ )
        INTRA_TEST( predict_16x16, i );
    for( i = 0; i < 12; i++ )
        INTRA_TEST( predict_8x8, i, edge );

    report( "intra pred :" );
    return ret;
}

int check_all( int cpu_ref, int cpu_new )
{
    return check_pixel( cpu_ref, cpu_new )
         + check_dct( cpu_ref, cpu_new )
         + check_mc( cpu_ref, cpu_new )
         + check_intra( cpu_ref, cpu_new )
         + check_deblock( cpu_ref, cpu_new )
         + check_quant( cpu_ref, cpu_new );
}

int main(int argc, char *argv[])
{
    int ret = 0;
    int cpu0 = 0, cpu1 = 0;
    int i;

    buf1 = x264_malloc( 1024 ); /* 32 x 32 */
    buf2 = x264_malloc( 1024 );
    buf3 = x264_malloc( 1024 );
    buf4 = x264_malloc( 1024 );
    buf5 = x264_malloc( 1024 );

    i = ( argc > 1 ) ? atoi(argv[1]) : x264_mdate();
    fprintf( stderr, "x264: using random seed %u\n", i );
    srand( i );

    for( i = 0; i < 1024; i++ )
    {
        buf1[i] = rand() & 0xFF;
        buf2[i] = rand() & 0xFF;
        buf3[i] = buf4[i] = 0;
    }

#ifdef HAVE_MMX
    fprintf( stderr, "x264: MMXEXT against C\n" );
    cpu1 = X264_CPU_MMX | X264_CPU_MMXEXT;
    ret = check_all( 0, cpu1 );

    if( x264_cpu_detect() & X264_CPU_SSE2 )
    {
        fprintf( stderr, "\nx264: SSE2 against C\n" );
        cpu0 = cpu1;
        cpu1 |= X264_CPU_SSE | X264_CPU_SSE2;
        ret |= check_all( cpu0, cpu1 );

        if( x264_cpu_detect() & X264_CPU_SSSE3 )
        {
            fprintf( stderr, "\nx264: SSSE3 against C\n" );
            cpu0 = cpu1;
            cpu1 |= X264_CPU_SSE3 | X264_CPU_SSSE3;
            ret |= check_all( cpu0, cpu1 );
        }
    }
#elif ARCH_PPC
    if( x264_cpu_detect() & X264_CPU_ALTIVEC )
    {
        fprintf( stderr, "x264: ALTIVEC against C\n" );
        ret = check_all( 0, X264_CPU_ALTIVEC );
    }
#endif

    if( ret == 0 )
    {
        fprintf( stderr, "x264: All tests passed Yeah :)\n" );
        return 0;
    }
    fprintf( stderr, "x264: at least one test has failed. Go and fix that Right Now!\n" );
    return -1;
}