VPCMPEQB

VPCMPEQB compares bytes for equality: the VEX form writes FFh or 00h per byte into a vector, as SSE2’s PCMPEQB does, and the EVEX form a bit per byte into a mask register. VPCMPEQD describes the two encodings, and the PCMPEQB page the byte search with VPMOVMSKB and the all-ones idiom. The byte width adds:

• AVX512BW and 64 bits. The EVEX form needs AVX512BW (or AVX10.1) and fills up to 64 mask bits, clearing the rest: on Zen 5, {evex} vpcmpeqb k1, ymm5, ymm5 set K1 to FFFFFFFFh, 32 ones, whatever it held before. A write mask clears the bits it leaves out: vpcmpeqb k4{k1}, xmm0, xmm31 found bytes 2, 5, 10, 11 and 15 equal, but of these K1 selected only bit 5, so K4 became 20h.
• No broadcast. Like the other byte forms, the compare cannot repeat a byte from memory, so the wanted byte is spread first with VPBROADCASTB.

Counting bytes#

A byte counter would overflow after 255 matches, so the subtraction of -1 that counts doublewords does not carry over. For n += a[i] == k over 1,024 bytes, GCC 14 and Clang 23 at -O2 with -march=x86-64-v3 compare with VPCMPEQB, keep 1 per match with VPAND, before or after widening into doubleword sums: GCC with VPMOVZXBW and VPMOVZXWD from 32 compared bytes, Clang with VPMOVZXBD from eight. With -march=x86-64-v4, GCC compares into K1 with VPCMPB, predicate 0, and turns the mask into bytes of 1 with a zero-masked VMOVDQU8.

Neither compiler vectorizes a search loop that returns at the first match, if (a[i] == k) return i;: both keep it scalar, Clang unrolled eight times. The vector search comes from intrinsics or a C library.

With a write mask, _mm512_mask_cmpeq_epi8_mask(m, a, b) becomes a masked VPCMPB in GCC; Clang compares without the mask and ANDs the result in a general-purpose register.

AMD’s latency spreadsheet for Zen 5 gives the vector form one cycle, four per cycle, and the mask form 3, 4 and 6 cycles for xmm, ymm and zmm sources, two per cycle. Zen 5’s 10 recorded cases cover every form, the EVEX ones masked and unmasked.

Forms

Latency and throughput

Data flow

Where the elements of the sources go, element 0 on the right; results that depend on a control value are drawn for an example. How to read.

Operation

What the forms do, in pseudocode that our tests run against results recorded on real processors. How to read.

for i in range(VL // 8):
    a = ymm2.b[i]
    b = src3.b[i]
    ymm1.b[i] = -1 if a == b else 0
# ymm1 is zeroed from bit VL up

for i in range(VL // 8):
    k1[i] = k2[i] and zmm2.b[i] == src3.b[i]
# the other bits of k1 are cleared

Operands

Flags

None: XED records no flag that the forms test or change.

Exceptions

The exception class is XED's name for the set of exception conditions that a form follows. How to read.

VEX forms

avx-type-4

EVEX forms

AVX512-E4

Attributes

What else XED records about the forms. How to read.

EVEX forms

write masking

EVEX memory forms

memory fault suppression

Intrinsics

The C intrinsics of Clang 23.1.2 that compile to VPCMPEQB, each with the form it compiles to and the Clang options it needs. The forms come from compiling every intrinsic and decoding the result with Intel XED; all intrinsics A–Z explains the method.

CPUID and processors

Each form belongs to an XED ISA set. A form can be used when the processor reports every CPUID bit of one of its ISA set's alternatives.

More AVX instructions

• VPACKSSDW
• VPACKSSWB
• VPACKUSDW
• VPACKUSWB
• VPADDB
• VPADDD
• VPADDQ
• VPADDSB
• VPADDSW
• VPADDUSB
• VPADDUSW
• VPADDW
• VPALIGNR
• VPAND
• VPANDN
• VPAVGB
• VPAVGW
• VPBLENDVB
• VPBLENDW
• VPCLMULQDQ
• VPCMPEQD
• VPCMPEQQ
• VPCMPEQW
• VPCMPESTRI
• VPCMPESTRM
• VPCMPGTB
• VPCMPGTD
• VPCMPGTQ
• VPCMPGTW
• VPCMPISTRI
• VPCMPISTRM
• VPERM2F128
• VPERMILPD
• VPERMILPS
• VPEXTRB
• VPEXTRD
• VPEXTRQ
• VPEXTRW
• VPHADDD
• VPHADDSW

All 253 AVX instructions

More AVX2 instructions

• VPADDB
• VPADDD
• VPADDQ
• VPADDSB
• VPADDSW
• VPADDUSB
• VPADDUSW
• VPADDW
• VPALIGNR
• VPAND
• VPANDN
• VPAVGB
• VPAVGW
• VPBLENDD
• VPBLENDVB
• VPBLENDW
• VPBROADCASTB
• VPBROADCASTD
• VPBROADCASTQ
• VPBROADCASTW
• VPCMPEQD
• VPCMPEQQ
• VPCMPEQW
• VPCMPGTB
• VPCMPGTD
• VPCMPGTQ
• VPCMPGTW
• VPERM2I128
• VPERMD
• VPERMPD
• VPERMPS
• VPERMQ
• VPGATHERDD
• VPGATHERDQ
• VPGATHERQD
• VPGATHERQQ
• VPHADDD
• VPHADDSW
• VPHADDW
• VPHSUBD

All 138 AVX2 instructions

More AVX10.1 instructions

• VPALIGNR
• VPANDD
• VPANDND
• VPANDNQ
• VPANDQ
• VPAVGB
• VPAVGW
• VPBLENDMB
• VPBLENDMD
• VPBLENDMQ
• VPBLENDMW
• VPBROADCASTB
• VPBROADCASTD
• VPBROADCASTMB2Q
• VPBROADCASTMW2D
• VPBROADCASTQ
• VPBROADCASTW
• VPCLMULQDQ
• VPCMPB
• VPCMPD
• VPCMPEQD
• VPCMPEQQ
• VPCMPEQW
• VPCMPGTB
• VPCMPGTD
• VPCMPGTQ
• VPCMPGTW
• VPCMPQ
• VPCMPUB
• VPCMPUD
• VPCMPUQ
• VPCMPUW
• VPCMPW
• VPCOMPRESSB
• VPCOMPRESSD
• VPCOMPRESSQ
• VPCOMPRESSW
• VPCONFLICTD
• VPCONFLICTQ
• VPDPBUSD

All 686 AVX10.1 instructions