VGETMANTBF16

VGETMANTBF16 returns the significand of each bfloat16 value, scaled into an interval that bits 1:0 of the immediate choose, with bits 3:2 deciding the sign or turning negative inputs into the default NaN. It is VGETMANTPS for bfloat16, and that page has the table of both fields; VGETEXPBF16 returns the exponent that goes with it. It belongs to AVX10.2: the opcode is 26h in map 0F3A with the F2 prefix and W0, where VGETMANTPH has no prefix and VGETMANTPS 66, and the source may be broadcast from 16 bits of memory.

Seven bits of significand#

A bfloat16 has 7 fraction bits, so VGETMANTBF16 returns one of 128 significands in the chosen interval, and the instruction never rounds: it replaces the exponent and keeps the fraction. Like the other bfloat16 instructions, it raises no exceptions and leaves MXCSR alone, as VADDBF16’s page explains. Denormals are read as zeros: the specification’s routine (revision 7.0) clears a denormal’s fraction, so with the interval from 1 to 2 a positive denormal gives 1.0, as +0 does, where VGETMANTPH, which ignores DAZ, returns its real significand. A logarithm built from VGETEXP and VGETMANT, as VGETEXPPH’s page shows for halves, therefore sees every denormal as a zero.

In our builds#

GCC 14 has no AVX10.2 option. Clang 23 at -O2 with -mavx10.2 compiles _mm512_getmant_pbh(x, _MM_MANT_NORM_1_2, _MM_MANT_SIGN_src) to vgetmantbf16 zmm0, zmm0, 0; its header builds the immediate as the sign field shifted left by 2 joined with the interval. No C function maps to it. No AVX10.2 processor appears in this site’s recorded cases.

Forms

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]:
        a = src2.bf16[i]
        zmm1.bf16[i] = getmant(a, imm8)
    elif zeroing:
        zmm1.bf16[i] = 0
    # else: zmm1.bf16[i] keeps its value
# imm8[1:0]: [1, 2), [1/2, 2), [1/2, 1), [3/4, 3/2); bit 2: no sign; bit 3: negative gives NaN
# m16bcst: src2.bf16[i] is the m16 for every i
# BF16: to nearest even, denormals flushed, no flags; MXCSR unused
# 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.

All forms

AVX512-E4

Attributes

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

All forms

write masking; denormal inputs read as zero; denormal results flushed to zero

Memory forms

embedded broadcast (tuple Full); memory fault suppression

Intrinsics

The C intrinsics of Clang 23.1.2 that compile to VGETMANTBF16, 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 AVX10.2 instructions

• VCVTTSD2SIS
• VCVTTSD2USIS
• VCVTTSS2SIS
• VCVTTSS2USIS
• VDIVBF16
• VDPPHPS
• VFMADD132BF16
• VFMADD213BF16
• VFMADD231BF16
• VFMSUB132BF16
• VFMSUB213BF16
• VFMSUB231BF16
• VFNMADD132BF16
• VFNMADD213BF16
• VFNMADD231BF16
• VFNMSUB132BF16
• VFNMSUB213BF16
• VFNMSUB231BF16
• VFPCLASSBF16
• VGETEXPBF16
• VMAXBF16
• VMINBF16
• VMINMAXBF16
• VMINMAXPD
• VMINMAXPH
• VMINMAXPS
• VMINMAXSD
• VMINMAXSH
• VMINMAXSS
• VMOVD
• VMOVW
• VMPSADBW
• VMULBF16
• VPDPBSSD
• VPDPBSSDS
• VPDPBSUD
• VPDPBSUDS
• VPDPBUUD
• VPDPBUUDS
• VPDPWSUD

All 96 AVX10.2 instructions