timesoftfloat.c

来自「基于4个mips核的noc设计」· C语言 代码 · 共 1,051 行 · 第 1/3 页

}

enum {
    numInputs_float64 = 32
};

static const struct {
    bits32 high, low;
} inputs_float64[ numInputs_float64 ] = {
    { 0x422FFFC0, 0x08000000 },
    { 0xB7E00004, 0x80000000 },
    { 0xF3FD2546, 0x120B7935 },
    { 0x3FF00000, 0x00000000 },
    { 0xCE07F766, 0xF09588D6 },
    { 0x80000000, 0x00000000 },
    { 0x3FCE0004, 0x00000000 },
    { 0x8313B60F, 0x0032BED8 },
    { 0xC1EFFFFF, 0xC0002000 },
    { 0x3FB3C75D, 0x224F2B0F },
    { 0x7FD00000, 0x004000FF },
    { 0xA12FFF80, 0x00001FFF },
    { 0x3EE00000, 0x00FE0000 },
    { 0x00100000, 0x80000004 },
    { 0x41CFFFFE, 0x00000020 },
    { 0x40303FFF, 0xFFFFFFFD },
    { 0x3FD00000, 0x3FEFFFFF },
    { 0xBFD00000, 0x10000000 },
    { 0xB7FC6B5C, 0x16CA55CF },
    { 0x413EEB94, 0x0B9D1301 },
    { 0xC7E00200, 0x001FFFFF },
    { 0x47F00021, 0xFFFFFFFE },
    { 0xBFFFFFFF, 0xF80000FF },
    { 0xC07FFFFF, 0xE00FFFFF },
    { 0x001497A6, 0x3740C5E8 },
    { 0xC4BFFFE0, 0x001FFFFF },
    { 0x96FFDFFE, 0xFFFFFFFF },
    { 0x403FC000, 0x000001FE },
    { 0xFFD00000, 0x000001F6 },
    { 0x06404000, 0x02000000 },
    { 0x479CEE1E, 0x4F789FE0 },
    { 0xC237FFFF, 0xFFFFFDFE }
};

static void time_a_float64_z_int32( int32 function( float64 ) )
{
    clock_t startClock, endClock;
    int32 count, i;
    int8 inputNum;
    float64 a;

    count = 0;
    inputNum = 0;
    startClock = clock();
    do {
        for ( i = minIterations; i; --i ) {
            a.low = inputs_float64[ inputNum ].low;
            a.high = inputs_float64[ inputNum ].high;
            function( a );
            inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
        }
        count += minIterations;
    } while ( clock() - startClock < CLOCKS_PER_SEC );
    inputNum = 0;
    startClock = clock();
    for ( i = count; i; --i ) {
        a.low = inputs_float64[ inputNum ].low;
        a.high = inputs_float64[ inputNum ].high;
        function( a );
        inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
    }
    endClock = clock();
    reportTime( count, endClock - startClock );

}

static void time_a_float64_z_float32( float32 function( float64 ) )
{
    clock_t startClock, endClock;
    int32 count, i;
    int8 inputNum;
    float64 a;

    count = 0;
    inputNum = 0;
    startClock = clock();
    do {
        for ( i = minIterations; i; --i ) {
            a.low = inputs_float64[ inputNum ].low;
            a.high = inputs_float64[ inputNum ].high;
            function( a );
            inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
        }
        count += minIterations;
    } while ( clock() - startClock < CLOCKS_PER_SEC );
    inputNum = 0;
    startClock = clock();
    for ( i = count; i; --i ) {
        a.low = inputs_float64[ inputNum ].low;
        a.high = inputs_float64[ inputNum ].high;
        function( a );
        inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
    }
    endClock = clock();
    reportTime( count, endClock - startClock );

}

static void time_az_float64( float64 function( float64 ) )
{
    clock_t startClock, endClock;
    int32 count, i;
    int8 inputNum;
    float64 a;

    count = 0;
    inputNum = 0;
    startClock = clock();
    do {
        for ( i = minIterations; i; --i ) {
            a.low = inputs_float64[ inputNum ].low;
            a.high = inputs_float64[ inputNum ].high;
            function( a );
            inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
        }
        count += minIterations;
    } while ( clock() - startClock < CLOCKS_PER_SEC );
    inputNum = 0;
    startClock = clock();
    for ( i = count; i; --i ) {
        a.low = inputs_float64[ inputNum ].low;
        a.high = inputs_float64[ inputNum ].high;
        function( a );
        inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
    }
    endClock = clock();
    reportTime( count, endClock - startClock );

}

static void time_ab_float64_z_flag( flag function( float64, float64 ) )
{
    clock_t startClock, endClock;
    int32 count, i;
    int8 inputNumA, inputNumB;
    float64 a, b;

    count = 0;
    inputNumA = 0;
    inputNumB = 0;
    startClock = clock();
    do {
        for ( i = minIterations; i; --i ) {
            a.low = inputs_float64[ inputNumA ].low;
            a.high = inputs_float64[ inputNumA ].high;
            b.low = inputs_float64[ inputNumB ].low;
            b.high = inputs_float64[ inputNumB ].high;
            function( a, b );
            inputNumA = ( inputNumA + 1 ) & ( numInputs_float64 - 1 );
            if ( inputNumA == 0 ) ++inputNumB;
            inputNumB = ( inputNumB + 1 ) & ( numInputs_float64 - 1 );
        }
        count += minIterations;
    } while ( clock() - startClock < CLOCKS_PER_SEC );
    inputNumA = 0;
    inputNumB = 0;
    startClock = clock();
    for ( i = count; i; --i ) {
        a.low = inputs_float64[ inputNumA ].low;
        a.high = inputs_float64[ inputNumA ].high;
        b.low = inputs_float64[ inputNumB ].low;
        b.high = inputs_float64[ inputNumB ].high;
        function( a, b );
        inputNumA = ( inputNumA + 1 ) & ( numInputs_float64 - 1 );
        if ( inputNumA == 0 ) ++inputNumB;
        inputNumB = ( inputNumB + 1 ) & ( numInputs_float64 - 1 );
    }
    endClock = clock();
    reportTime( count, endClock - startClock );

}

static void time_abz_float64( float64 function( float64, float64 ) )
{
    clock_t startClock, endClock;
    int32 count, i;
    int8 inputNumA, inputNumB;
    float64 a, b;

    count = 0;
    inputNumA = 0;
    inputNumB = 0;
    startClock = clock();
    do {
        for ( i = minIterations; i; --i ) {
            a.low = inputs_float64[ inputNumA ].low;
            a.high = inputs_float64[ inputNumA ].high;
            b.low = inputs_float64[ inputNumB ].low;
            b.high = inputs_float64[ inputNumB ].high;
            function( a, b );
            inputNumA = ( inputNumA + 1 ) & ( numInputs_float64 - 1 );
            if ( inputNumA == 0 ) ++inputNumB;
            inputNumB = ( inputNumB + 1 ) & ( numInputs_float64 - 1 );
        }
        count += minIterations;
    } while ( clock() - startClock < CLOCKS_PER_SEC );
    inputNumA = 0;
    inputNumB = 0;
    startClock = clock();
    for ( i = count; i; --i ) {
        a.low = inputs_float64[ inputNumA ].low;
        a.high = inputs_float64[ inputNumA ].high;
        b.low = inputs_float64[ inputNumB ].low;
        b.high = inputs_float64[ inputNumB ].high;
        function( a, b );
        inputNumA = ( inputNumA + 1 ) & ( numInputs_float64 - 1 );
        if ( inputNumA == 0 ) ++inputNumB;
        inputNumB = ( inputNumB + 1 ) & ( numInputs_float64 - 1 );
    }
    endClock = clock();
    reportTime( count, endClock - startClock );

}

static const struct {
    bits32 high, low;
} inputs_float64_pos[ numInputs_float64 ] = {
    { 0x422FFFC0, 0x08000000 },
    { 0x37E00004, 0x80000000 },
    { 0x73FD2546, 0x120B7935 },
    { 0x3FF00000, 0x00000000 },
    { 0x4E07F766, 0xF09588D6 },
    { 0x00000000, 0x00000000 },
    { 0x3FCE0004, 0x00000000 },
    { 0x0313B60F, 0x0032BED8 },
    { 0x41EFFFFF, 0xC0002000 },
    { 0x3FB3C75D, 0x224F2B0F },
    { 0x7FD00000, 0x004000FF },
    { 0x212FFF80, 0x00001FFF },
    { 0x3EE00000, 0x00FE0000 },
    { 0x00100000, 0x80000004 },
    { 0x41CFFFFE, 0x00000020 },
    { 0x40303FFF, 0xFFFFFFFD },
    { 0x3FD00000, 0x3FEFFFFF },
    { 0x3FD00000, 0x10000000 },
    { 0x37FC6B5C, 0x16CA55CF },
    { 0x413EEB94, 0x0B9D1301 },
    { 0x47E00200, 0x001FFFFF },
    { 0x47F00021, 0xFFFFFFFE },
    { 0x3FFFFFFF, 0xF80000FF },
    { 0x407FFFFF, 0xE00FFFFF },
    { 0x001497A6, 0x3740C5E8 },
    { 0x44BFFFE0, 0x001FFFFF },
    { 0x16FFDFFE, 0xFFFFFFFF },
    { 0x403FC000, 0x000001FE },
    { 0x7FD00000, 0x000001F6 },
    { 0x06404000, 0x02000000 },
    { 0x479CEE1E, 0x4F789FE0 },
    { 0x4237FFFF, 0xFFFFFDFE }
};

static void time_az_float64_pos( float64 function( float64 ) )
{
    clock_t startClock, endClock;
    int32 count, i;
    int8 inputNum;
    float64 a;

    count = 0;
    inputNum = 0;
    startClock = clock();
    do {
        for ( i = minIterations; i; --i ) {
            a.low = inputs_float64_pos[ inputNum ].low;
            a.high = inputs_float64_pos[ inputNum ].high;
            function( a );
            inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
        }
        count += minIterations;
    } while ( clock() - startClock < CLOCKS_PER_SEC );
    inputNum = 0;
    startClock = clock();
    for ( i = count; i; --i ) {
        a.low = inputs_float64_pos[ inputNum ].low;
        a.high = inputs_float64_pos[ inputNum ].high;
        function( a );
        inputNum = ( inputNum + 1 ) & ( numInputs_float64 - 1 );
    }
    endClock = clock();
    reportTime( count, endClock - startClock );

}

enum {
    INT32_TO_FLOAT32 = 1,
    INT32_TO_FLOAT64,
    FLOAT32_TO_INT32,
    FLOAT32_TO_INT32_ROUND_TO_ZERO,
    FLOAT32_TO_FLOAT64,
    FLOAT32_ROUND_TO_INT,
    FLOAT32_ADD,
    FLOAT32_SUB,
    FLOAT32_MUL,
    FLOAT32_DIV,
    FLOAT32_REM,
    FLOAT32_SQRT,
    FLOAT32_EQ,
    FLOAT32_LE,
    FLOAT32_LT,
    FLOAT32_EQ_SIGNALING,
    FLOAT32_LE_QUIET,
    FLOAT32_LT_QUIET,
    FLOAT64_TO_INT32,
    FLOAT64_TO_INT32_ROUND_TO_ZERO,
    FLOAT64_TO_FLOAT32,
    FLOAT64_ROUND_TO_INT,
    FLOAT64_ADD,
    FLOAT64_SUB,
    FLOAT64_MUL,
    FLOAT64_DIV,
    FLOAT64_REM,
    FLOAT64_SQRT,
    FLOAT64_EQ,
    FLOAT64_LE,
    FLOAT64_LT,
    FLOAT64_EQ_SIGNALING,
    FLOAT64_LE_QUIET,
    FLOAT64_LT_QUIET,
    NUM_FUNCTIONS
};

static struct {
    char *name;
    int8 numInputs;
    flag roundingMode, tininessMode;
} functions[ NUM_FUNCTIONS ] = {
    { 0, 0, 0, 0 },
    { "int32_to_float32",                1, TRUE,  FALSE },
    { "int32_to_float64",                1, FALSE, FALSE },
    { "float32_to_int32",                1, TRUE,  FALSE },
    { "float32_to_int32_round_to_zero",  1, FALSE, FALSE },
    { "float32_to_float64",              1, FALSE, FALSE },
    { "float32_round_to_int",            1, TRUE,  FALSE },
    { "float32_add",                     2, TRUE,  FALSE },
    { "float32_sub",                     2, TRUE,  FALSE },
    { "float32_mul",                     2, TRUE,  TRUE, },
    { "float32_div",                     2, TRUE,  FALSE },
    { "float32_rem",                     2, FALSE, FALSE },
    { "float32_sqrt",                    1, TRUE,  FALSE },
    { "float32_eq",                      2, FALSE, FALSE },
    { "float32_le",                      2, FALSE, FALSE },
    { "float32_lt",                      2, FALSE, FALSE },