VGETEXPBF16

VGETEXPBF16 returns the binary exponent of each bfloat16 value as a bfloat16, the largest integer e with 2^e no greater than |x|, 8, 16 or 32 per register. It is VGETEXPPS for bfloat16, and VGETMANTBF16 and VSCALEFBF16 are its partners for splitting a number and putting it back together. It belongs to AVX10.2; the opcode is 42h in EVEX map 6 with no prefix and W0, where VGETEXPPH has 66, and the source may be broadcast from 16 bits of memory. EVEX.vvvv must be 1111b.

Range and special inputs#

A bfloat16 has the exponent range of a single, so the results run from -126 to 127, all exact. Denormals are read as zeros, as on VADDBF16’s page: a zero or a denormal gives -infinity, where VGETEXPPH reports a denormal’s true exponent. A NaN comes back quiet, and there are no exceptions and no MXCSR.

For infinities the AVX10.2 specification (revision 7.0) disagrees with itself. Its description defines the result as floor(log2(|x|)), which is +infinity for both, as Intel’s VGETEXPPS entry gives. Its routine, reworked in revision 3.0, returns +infinity only for +infinity: -infinity falls through to the branch that subtracts the bias from the exponent field, which would give 128.0.

In our builds#

GCC 14 has no AVX10.2 option. Clang 23 at -O2 with -mavx10.2 compiles _mm512_getexp_pbh to VGETEXPBF16. logbf on a __bf16 widens it with a 16-bit shift, calls the library and narrows the result with VCVTNEPS2BF16, even with -ffast-math. 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]:
        zmm1.bf16[i] = getexp(src2.bf16[i])
    elif zeroing:
        zmm1.bf16[i] = 0
    # else: zmm1.bf16[i] keeps its value
# 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; always rounds to nearest even; 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 VGETEXPBF16, 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

• VCVTTPS2UQQS
• VCVTTSD2SIS
• VCVTTSD2USIS
• VCVTTSS2SIS
• VCVTTSS2USIS
• VDIVBF16
• VDPPHPS
• VFMADD132BF16
• VFMADD213BF16
• VFMADD231BF16
• VFMSUB132BF16
• VFMSUB213BF16
• VFMSUB231BF16
• VFNMADD132BF16
• VFNMADD213BF16
• VFNMADD231BF16
• VFNMSUB132BF16
• VFNMSUB213BF16
• VFNMSUB231BF16
• VFPCLASSBF16
• VGETMANTBF16
• VMAXBF16
• VMINBF16
• VMINMAXBF16
• VMINMAXPD
• VMINMAXPH
• VMINMAXPS
• VMINMAXSD
• VMINMAXSH
• VMINMAXSS
• VMOVD
• VMOVW
• VMPSADBW
• VMULBF16
• VPDPBSSD
• VPDPBSSDS
• VPDPBSUD
• VPDPBSUDS
• VPDPBUUD
• VPDPBUUDS

All 96 AVX10.2 instructions