PMULDQ

Multiply Packed Doubleword Integers

stableVMJITAOTinstruction

Encodings

Operand encoding

Each mode is a value of the Op/En column above. It says which field of the encoded instruction carries each operand, in the order they are written, and whether the instruction reads it, writes it or both.

Measured cost

Description#

Multiplies packed signed doubleword integers in the even-numbered (zero-based reference) elements of the first source operand with the packed signed doubleword integers in the corresponding elements of the second source operand and stores packed signed quadword results in the destination operand.

128-bit Legacy SSE version: The input signed doubleword integers are taken from the even-numbered elements of the source operands, i.e., the first (low) and third doubleword element. For 128-bit memory operands, 128 bits are fetched from memory, but only the first and third doublewords are used in the computation. The first source operand and the destination XMM operand is the same. The second source operand can be an XMM register or 128-bit memory location. Bits (MAXVL-1:128) of the corresponding destination register remain unchanged.

VEX.128 encoded version: The input signed doubleword integers are taken from the even-numbered elements of the source operands, i.e., the first (low) and third doubleword element. For 128-bit memory operands, 128 bits are fetched from memory, but only the first and third doublewords are used in the computation.The first source operand and the destination operand are XMM registers. The second source operand can be an XMM register or 128-bit memory location. Bits (MAXVL-1:128) of the corresponding destination register are zeroed.

VEX.256 encoded version: The input signed doubleword integers are taken from the even-numbered elements of the source operands, i.e., the first, 3rd, 5th, 7th doubleword element. For 256-bit memory operands, 256 bits are fetched from memory, but only the four even-numbered doublewords are used in the computation. The first source operand and the destination operand are YMM registers. The second source operand can be a YMM register or 256-bit memory location. Bits (MAXVL-1:256) of the corresponding destination ZMM register are zeroed.

EVEX encoded version: The input signed doubleword integers are taken from the even-numbered elements of the source operands. The first source operand is a ZMM/YMM/XMM registers. The second source operand can be an ZMM/YMM/XMM register, a 512/256/128-bit memory location or a 512/256/128-bit vector broadcasted from a 64-bit memory location. The destination is a ZMM/YMM/XMM register, and updated according to the writemask at 64-bit granularity.

Operation#

VPMULDQ (EVEX Encoded Versions)

(KL, VL) = (2, 128), (4, 256), (8, 512)

FOR j := 0 TO KL-1

i := j * 64

IF k1[j] OR *no writemask*

     THEN

             IF (EVEX.b = 1) AND (SRC2 *is memory*)

                  THEN DEST[i+63:i] := SignExtend64( SRC1[i+31:i]) * SignExtend64( SRC2[31:0])

                  ELSE DEST[i+63:i] := SignExtend64( SRC1[i+31:i]) * SignExtend64( SRC2[i+31:i])

             FI;

     ELSE

             IF *merging-masking*            ; merging-masking

                  THEN *DEST[i+63:i] remains unchanged*

                  ELSE *zeroing-masking*             ; zeroing-masking

                    DEST[i+63:i] := 0

             FI

FI;

ENDFOR

DEST[MAXVL-1:VL] := 0

VPMULDQ (VEX.256 Encoded Version)
DEST[63:0] := SignExtend64( SRC1[31:0]) * SignExtend64( SRC2[31:0])
DEST[127:64] := SignExtend64( SRC1[95:64]) * SignExtend64( SRC2[95:64])
DEST[191:128] := SignExtend64( SRC1[159:128]) * SignExtend64( SRC2[159:128])
DEST[255:192] := SignExtend64( SRC1[223:192]) * SignExtend64( SRC2[223:192])
DEST[MAXVL-1:256] := 0

VPMULDQ (VEX.128 Encoded Version)
DEST[63:0] := SignExtend64( SRC1[31:0]) * SignExtend64( SRC2[31:0])
DEST[127:64] := SignExtend64( SRC1[95:64]) * SignExtend64( SRC2[95:64])
DEST[MAXVL-1:128] := 0

PMULDQ (128-bit Legacy SSE Version)
DEST[63:0] := SignExtend64( DEST[31:0]) * SignExtend64( SRC[31:0])
DEST[127:64] := SignExtend64( DEST[95:64]) * SignExtend64( SRC[95:64])
DEST[MAXVL-1:128] (Unmodified)

Intel C/C++ compiler intrinsics#

VPMULDQ __m512i _mm512_mul_epi32(__m512i a, __m512i b);
VPMULDQ __m512i _mm512_mask_mul_epi32(__m512i s, __mmask8 k, __m512i a, __m512i b);
VPMULDQ __m512i _mm512_maskz_mul_epi32( __mmask8 k, __m512i a, __m512i b);
VPMULDQ __m256i _mm256_mask_mul_epi32(__m256i s, __mmask8 k, __m256i a, __m256i b);
VPMULDQ __m256i _mm256_mask_mul_epi32( __mmask8 k, __m256i a, __m256i b);
VPMULDQ __m128i _mm_mask_mul_epi32(__m128i s, __mmask8 k, __m128i a, __m128i b);
VPMULDQ __m128i _mm_mask_mul_epi32( __mmask8 k, __m128i a, __m128i b);
(V)PMULDQ __m128i _mm_mul_epi32( __m128i a, __m128i b);
VPMULDQ __m256i _mm256_mul_epi32( __m256i a, __m256i b);

SIMD Floating-Point Exceptions#

None.

Other Exceptions#

Non-EVEX-encoded instruction, see Table 2-21, "Type 4 Class Exception Conditions."

EVEX-encoded instruction, see Table 2-51, "Type E4 Class Exception Conditions."

Sources

• This instruction in the Intel manual
• Latency and throughput per microarchitecture