VPBROADCASTW

VPBROADCASTW copies one word to every word of a vector: 8, 16 or 32 copies of the low word of an XMM register or of a word in memory, or, in the EVEX forms of AVX512BW, of the low 16 bits of a 32-bit general-purpose register. VPBROADCASTD describes the family and how the sources differ by encoding, and VPBROADCASTB what the byte form is used for. On Zen 5, vpbroadcastw ymm0{k4}{z}, ecx wrote 46C2h, the low word of ECX, into the ten elements whose bits are set in the low 16 bits of K4 (6DE5h) and zeroed the other six, and every recorded XMM and YMM form cleared the register above the vector it wrote.

In compiled code#

In our builds at -O2, GCC 14 and Clang 23 compile _mm256_set1_epi16(x) and a loop adding a run-time int16_t to every element of an array to VMOVD and VPBROADCASTW with -march=x86-64-v3, and to VPBROADCASTW straight from the register with -march=x86-64-v4. A word already in memory, _mm256_set1_epi16(*p), is the memory form in both, which reads just those two bytes at any alignment.

A constant is another matter. For _mm256_add_epi16(a, _mm256_set1_epi16(0x1234)), GCC builds 12341234h in a general-purpose register and broadcasts it as a doubleword with VPBROADCASTD, since two copies of the word make one doubleword; Clang adds a 32-byte constant from memory with VPADDW, without any broadcast.

Before AVX2 there is no word broadcast. For the SSE2 baseline, _mm_set1_epi16(x) is a MOVD and two shuffles: GCC interleaves the register with itself by PUNPCKLWD and copies the doubleword with PSHUFD, Clang fills the low four words with PSHUFLW and copies them with PSHUFD. With SSSE3, GCC uses one PSHUFB with a constant selector.

AMD’s latency spreadsheet for Zen 5 gives the 128-bit VEX form a latency of one cycle on all four vector pipes and the 256-bit VEX form one cycle on two of them, and lists the forms with a general-purpose source as two macro-ops. VPBROADCASTQ does the same for quadwords, and VPBROADCASTMW2D copies a mask register into every doubleword instead.

In the optimization guide

What the optimization guide says about VPBROADCASTW, a line per claim. Each links to the claim in the guide, with its sources and where they differ.

Skylake to Cooper Lake

• AVX-512 and frequency on the servers: A broadcast of a 32-bit or 64-bit element from memory runs on the load ports, sparing port 5; a 16-bit one still goes through port 5, so it gains nothing from the memory form.

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 // 16):
    if k1[i]:  # 1 without a write mask
        zmm1.w[i] = src2.w[0]
    elif zeroing:
        zmm1.w[i] = 0
    # else: zmm1.w[i] keeps its value
# zmm1 is zeroed from bit VL up

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-6

VPBROADCASTW xmm {k}{z}, xmm/m16 VPBROADCASTW ymm {k}{z}, xmm/m16 VPBROADCASTW zmm {k}{z}, xmm/m16

AVX512-E6

VPBROADCASTW xmm {k}{z}, r32 VPBROADCASTW ymm {k}{z}, r32 VPBROADCASTW zmm {k}{z}, r32

AVX512-E7NM

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 VPBROADCASTW, 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 AVX2 instructions

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

All 138 AVX2 instructions

More AVX10.1 instructions

• VPADDSW
• VPADDUSB
• VPADDUSW
• VPADDW
• VPALIGNR
• VPANDD
• VPANDND
• VPANDNQ
• VPANDQ
• VPAVGB
• VPAVGW
• VPBLENDMB
• VPBLENDMD
• VPBLENDMQ
• VPBLENDMW
• VPBROADCASTB
• VPBROADCASTD
• VPBROADCASTMB2Q
• VPBROADCASTMW2D
• VPBROADCASTQ
• VPCLMULQDQ
• VPCMPB
• VPCMPD
• VPCMPEQB
• VPCMPEQD
• VPCMPEQQ
• VPCMPEQW
• VPCMPGTB
• VPCMPGTD
• VPCMPGTQ
• VPCMPGTW
• VPCMPQ
• VPCMPUB
• VPCMPUD
• VPCMPUQ
• VPCMPUW
• VPCMPW
• VPCOMPRESSB
• VPCOMPRESSD
• VPCOMPRESSQ

All 686 AVX10.1 instructions