Vesta
English
Source code

VPDPBUSDS

Multiply and Add Unsigned and Signed Bytes With Saturation

stableVMJITAOTinstruction

Encodings

OpcodeInstructionOp/En64-bitCompat/LegacyDescription
VEX.128.66.0F38.W0 51 /rVPDPBUSDS xmm1, xmm2, xmm3/m128AValidValidMultiply groups of 4 pairs signed bytes in xmm3/m128 with corresponding unsigned bytes of xmm2, summing those products and adding them to doubleword result, with signed saturation in xmm1.
VEX.256.66.0F38.W0 51 /rVPDPBUSDS ymm1, ymm2, ymm3/m256AValidValidMultiply groups of 4 pairs signed bytes in ymm3/m256 with corresponding unsigned bytes of ymm2, summing those products and adding them to doubleword result, with signed saturation in ymm1.
EVEX.128.66.0F38.W0 51 /rVPDPBUSDS xmm1{k1}{z}, xmm2, xmm3/m128/m32bcstBValidValidMultiply groups of 4 pairs signed bytes in AND AVX512VL) xmm3/m128/m32bcst with corresponding OR AVX10.1 unsigned bytes of xmm2, summing those products and adding them to doubleword result, with signed saturation in xmm1, under writemask k1.
EVEX.256.66.0F38.W0 51 /rVPDPBUSDS ymm1{k1}{z}, ymm2, ymm3/m256/m32bcstBValidValidMultiply groups of 4 pairs signed bytes in AND AVX512VL) ymm3/m256/m32bcst with corresponding OR AVX10.1 unsigned bytes of ymm2, summing those products and adding them to doubleword result, with signed saturation in ymm1, under writemask k1.
EVEX.512.66.0F38.W0 51 /rVPDPBUSDS zmm1{k1}{z}, zmm2, zmm3/m512/m32bcstBValidValidMultiply groups of 4 pairs signed bytes in OR AVX10.1 zmm3/m512/m32bcst with corresponding unsigned bytes of zmm2, summing those products and adding them to doubleword result, with signed saturation in zmm1, under writemask k1.

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.

A

  1. modrm.reg lectura y escrituraModRM byte, reg field (bits 5-3)
  2. vex.vvvv lecturaVEX prefix, vvvv field (inverted)
  3. modrm.rm lecturaModRM byte, r/m field (bits 2-0); with the SIB byte and the displacement when the mod field asks for them

B

  1. modrm.reg lectura y escrituraModRM byte, reg field (bits 5-3)
  2. evex.vvvv lecturaEVEX prefix, vvvv field (inverted)
  3. modrm.rm lecturaModRM byte, r/m field (bits 2-0); with the SIB byte and the displacement when the mod field asks for them

Tupla: Full

Measured cost

Loading measurements from arch-data...

Description

Multiplies the individual unsigned bytes of the first source operand by the corresponding signed bytes of the second source operand, producing intermediate signed word results. The word results are then summed and accumulated in the destination dword element size operand. If the intermediate sum overflows a 32b signed number the result is saturated to either 0x7FFF_FFFF for positive numbers of 0x8000_0000 for negative numbers.

This instruction supports memory fault suppression.

Operation

VPDPBUSDS dest, src1, src2 (VEX encoded versions)
VL=(128, 256)
KL=VL/32

ORIGDEST := DEST
FOR i := 0 TO KL-1:

    // Extending to 16b
    // src1extend := ZERO_EXTEND
    // src2extend := SIGN_EXTEND

    p1word := src1extend(SRC1.byte[4*i+0]) * src2extend(SRC2.byte[4*i+0])
    p2word := src1extend(SRC1.byte[4*i+1]) * src2extend(SRC2.byte[4*i+1])
    p3word := src1extend(SRC1.byte[4*i+2]) * src2extend(SRC2.byte[4*i+2])
    p4word := src1extend(SRC1.byte[4*i+3]) * src2extend(SRC2.byte[4*i+3])
    DEST.dword[i] := SIGNED_DWORD_SATURATE(ORIGDEST.dword[i] + p1word + p2word + p3word + p4word)

DEST[MAX_VL-1:VL] := 0

VPDPBUSDS dest, src1, src2 (EVEX encoded versions)
(KL,VL)=(4,128), (8,256), (16,512)
ORIGDEST := DEST
FOR i := 0 TO KL-1:

    IF k1[i] or *no writemask*:
          // Byte elements of SRC1 are zero-extended to 16b and
          // byte elements of SRC2 are sign extended to 16b before multiplication.
          IF SRC2 is memory and EVEX.b == 1:
                t := SRC2.dword[0]
          ELSE:
                t := SRC2.dword[i]
          p1word := ZERO_EXTEND(SRC1.byte[4*i]) * SIGN_EXTEND(t.byte[0])
          p2word := ZERO_EXTEND(SRC1.byte[4*i+1]) * SIGN_EXTEND(t.byte[1])
          p3word := ZERO_EXTEND(SRC1.byte[4*i+2]) * SIGN_EXTEND(t.byte[2])
          p4word := ZERO_EXTEND(SRC1.byte[4*i+3]) *SIGN_EXTEND(t.byte[3])
          DEST.dword[i] := SIGNED_DWORD_SATURATE(ORIGDEST.dword[i] + p1word + p2word + p3word + p4word)

    ELSE IF *zeroing*:
          DEST.dword[i] := 0

    ELSE: // Merge masking, dest element unchanged
          DEST.dword[i] := ORIGDEST.dword[i]

DEST[MAX_VL-1:VL] := 0

Intel C/C++ compiler intrinsics

VPDPBUSDS __m128i _mm_dpbusds_avx_epi32(__m128i, __m128i, __m128i);
VPDPBUSDS __m128i _mm_dpbusds_epi32(__m128i, __m128i, __m128i);
VPDPBUSDS __m128i _mm_mask_dpbusds_epi32(__m128i, __mmask8, __m128i, __m128i);
VPDPBUSDS __m128i _mm_maskz_dpbusds_epi32(__mmask8, __m128i, __m128i, __m128i);
VPDPBUSDS __m256i _mm256_dpbusds_avx_epi32(__m256i, __m256i, __m256i);
VPDPBUSDS __m256i _mm256_dpbusds_epi32(__m256i, __m256i, __m256i);
VPDPBUSDS __m256i _mm256_mask_dpbusds_epi32(__m256i, __mmask8, __m256i, __m256i);
VPDPBUSDS __m256i _mm256_maskz_dpbusds_epi32(__mmask8, __m256i, __m256i, __m256i);
VPDPBUSDS __m512i _mm512_dpbusds_epi32(__m512i, __m512i, __m512i);
VPDPBUSDS __m512i _mm512_mask_dpbusds_epi32(__m512i, __mmask16, __m512i, __m512i);
VPDPBUSDS __m512i _mm512_maskz_dpbusds_epi32(__mmask16, __m512i, __m512i, __m512i);

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