VGETEXPPS — Convert Exponents of Packed SP FP Values to SP FP Values

Instruction Operand Encoding ¶

Description ¶

Extracts the biased exponents from the normalized SP FP representation of each dword element of the source operand (the second operand) as unbiased signed integer value, or convert the denormal representation of input data to unbiased negative integer values. Each integer value of the unbiased exponent is converted to single-precision FP value and written to the corresponding dword elements of the destination operand (the first operand) as SP FP numbers.

The destination operand is a ZMM/YMM/XMM register and updated under the writemask. The source operand can be a ZMM/YMM/XMM register, a 512/256/128-bit memory location, or a 512/256/128-bit vector broadcasted from a 32-bit memory location.

EVEX.vvvv is reserved and must be 1111b, otherwise instructions will #UD.

Each GETEXP operation converts the exponent value into a FP number (permitting input value in denormal representation). Special cases of input values are listed in Table 5-15.

The formula is:

GETEXP(x) = floor(log₂(|x|))

Notation floor(x) stands for maximal integer not exceeding real number x.

Software usage of VGETEXPxx and VGETMANTxx instructions generally involve a combination of GETEXP operation and GETMANT operation (see VGETMANTPD). Thus VGETEXPxx instruction do not require software to handle SIMD FP exceptions.

Figure 5-14 illustrates the VGETEXPPS functionality on input values with normalized representation.

Operation ¶

NormalizeExpTinySPFP(SRC[31:0])
{
    // Jbit is the hidden integral bit of a FP number. In case of denormal number it has the value of ZERO.
    Src.Jbit ← 0;
    Dst.exp ← 1;
    Dst.fraction ← SRC[22:0];
    WHILE(Src.Jbit = 0)
    {
        Src.Jbit←Dst.fraction[22];
                        // Get the fraction MSB
        Dst.fraction←Dst.fraction << 1 ;
                        // One bit shift left
        Dst.exp-- ;
                // Decrement the exponent
    }
    Dst.fraction←0;
    Dst.sign←1;
    TMP[31:0] ← MXCSR.DAZ? 0 : (Dst.sign << 31) OR (Dst.exp << 23) OR (Dst.fraction) ;
    Return (TMP[31:0]);
}
ConvertExpSPFP(SRC[31:0])
{
    Src.sign←0;
                // Zero out sign bit
    Src.exp ← SRC[30:23];
    Src.fraction ← SRC[22:0];
    // Check for NaN
    IF (SRC = NaN)
    {
        IF ( SRC = SNAN ) SET IE;
        Return QNAN(SRC);
    }
    // Check for +INF
    IF (SRC = +INF) Return (SRC);
    // check if zero operand
    IF ((Src.exp = 0) AND ((Src.fraction = 0) OR (MXCSR.DAZ = 1))) Return (-INF);
    }
    ELSE // check if denormal operand (notice that MXCSR.DAZ = 0)
    {
        IF ((Src.exp = 0) AND (Src.fraction != 0))
        {
            TMP[31:0]←NormalizeExpTinySPFP(SRC[31:0]) ;
                            // Get Normalized Exponent
            Set #DE
        }
        ELSE // exponent value is correct
        {
            TMP[31:0]←(Src.sign << 31) OR (Src.exp << 23) OR (Src.fraction) ;
        }
        TMP←SAR(TMP, 23) ;
                    // Shift Arithmetic Right
        TMP←TMP – 127;
                    // Subtract Bias
        Return CvtI2D(TMP);
                    // Convert INT to Single-Precision FP number
    }
}

VGETEXPPS (EVEX encoded versions) ¶

(KL, VL) = (4, 128), (8, 256), (16, 512)
FOR j←0 TO KL-1
    i←j * 32
    IF k1[j] OR *no writemask*
        THEN
            IF (EVEX.b = 1) AND (SRC *is memory*)
                THEN
                    DEST[i+31:i] ←
            ConvertExpSPFP(SRC[31:0])
                ELSE
                    DEST[i+31:i] ←
            ConvertExpSPFP(SRC[i+31:i])
            FI;
        ELSE
            IF *merging-masking*
                        ; merging-masking
                THEN *DEST[i+31:i] remains unchanged*
                ELSE
                        ; zeroing-masking
                    DEST[i+31:i] ← 0
            FI
    FI;
ENDFOR
DEST[MAXVL-1:VL] ← 0

Intel C/C++ Compiler Intrinsic Equivalent ¶

VGETEXPPS __m512 _mm512_getexp_ps( __m512 a);

VGETEXPPS __m512 _mm512_mask_getexp_ps(__m512 s, __mmask16 k, __m512 a);

VGETEXPPS __m512 _mm512_maskz_getexp_ps( __mmask16 k, __m512 a);

VGETEXPPS __m512 _mm512_getexp_round_ps( __m512 a, int sae);

VGETEXPPS __m512 _mm512_mask_getexp_round_ps(__m512 s, __mmask16 k, __m512 a, int sae);

VGETEXPPS __m512 _mm512_maskz_getexp_round_ps( __mmask16 k, __m512 a, int sae);

VGETEXPPS __m256 _mm256_getexp_ps(__m256 a);

VGETEXPPS __m256 _mm256_mask_getexp_ps(__m256 s, __mmask8 k, __m256 a);

VGETEXPPS __m256 _mm256_maskz_getexp_ps( __mmask8 k, __m256 a);

VGETEXPPS __m128 _mm_getexp_ps(__m128 a);

VGETEXPPS __m128 _mm_mask_getexp_ps(__m128 s, __mmask8 k, __m128 a);

VGETEXPPS __m128 _mm_maskz_getexp_ps( __mmask8 k, __m128 a);

SIMD Floating-Point Exceptions ¶

Invalid, Denormal

Other Exceptions ¶

See Exceptions Type E2.