VFMSUB231PS

VFMSUB231PS computes op2 × op3 - op1 for each single, rounded once: the product minus the destination’s old value. It works as VFMADD231PS does, with the addend’s sign flipped, and VFNMADD231PS sets the four sign variants of the FMA family side by side. Note the order: the 231 form subtracts the accumulator from the product, so acc = x × y - acc, which is rarely what a loop wants; acc -= x × y is VFNMADD.

The error of a product#

With one rounding, x × y - p, where p is the product rounded to a single, gives exactly what that rounding lost: the classic way to split a product into a rounded part and an error. The cases recorded on Zen 5 show the effect of the single rounding. For vfmsub231ps ymm4, ymm4, ymm4, x × x - x with x = 1 + 2^-12 is 2^-12 + 2^-24, and Zen 5 returned that exactly, 39800800H. Rounding to nearest, a VMULPS would first turn x × x = 1 + 2^-11 + 2^-24 into 1 + 2^-11, and the subtraction would then give 2^-12, the 2^-24 lost.

In our builds at -O2 with -march=x86-64-v3, the loop e[i] = fmaf(x[i], y[i], -p[i]) becomes VFMSUB213PS in Clang 23 and VFMSUB132PS in GCC 14, each with the operand from memory that suits it; acc[i] = x[i] * y[i] - acc[i] compiles the same way.

In complex products#

The real part of a complex product, a.re × b.re - a.im × b.im, is one multiplication and one fused subtraction. With -ffast-math, Clang compiles a loop of _Complex float products for -march=x86-64-v3 by separating the real and imaginary parts, then VMULPS for a.im × b.im and VFMSUB231PS for the rest, with VFMADD231PS doing the imaginary part; GCC uses VFMADDSUB231PS on the interleaved values instead.

DAZ on the subtrahend#

The {evex} form recorded with MXCSR.DAZ set and FTZ clear computed 2^-126 × 0.5 - 2^-149 as 2^-127: the denormal 2^-149 counted as zero, while the denormal result itself was kept, as DAZ applies to inputs and FTZ to results.

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 // 32):
    if k1[i]:  # 1 without a write mask
        a = zmm1.f32[i]
        b = zmm2.f32[i]
        c = src3.f32[i]
        zmm1.f32[i] = fmsub(b, c, a)
    elif zeroing:
        zmm1.f32[i] = 0
    # else: zmm1.f32[i] keeps its value
# m32bcst: src3.f32[i] is the m32 for every i
# {er}: the form's rounding, not MXCSR.RC, and no flags
# 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-2

EVEX forms

AVX512-E2

Attributes

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

All forms

follows MXCSR

EVEX forms

write masking

EVEX memory forms

embedded broadcast (tuple Full); memory fault suppression

Intrinsics

The C intrinsics of Clang 23.1.2 that compile to VFMSUB231PS, 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 FMA instructions

• VFMADD213SS
• VFMADD231PD
• VFMADD231PS
• VFMADD231SD
• VFMADD231SS
• VFMADDSUB132PD
• VFMADDSUB132PS
• VFMADDSUB213PD
• VFMADDSUB213PS
• VFMADDSUB231PD
• VFMADDSUB231PS
• VFMSUB132PD
• VFMSUB132PS
• VFMSUB132SD
• VFMSUB132SS
• VFMSUB213PD
• VFMSUB213PS
• VFMSUB213SD
• VFMSUB213SS
• VFMSUB231PD
• VFMSUB231SD
• VFMSUB231SS
• VFMSUBADD132PD
• VFMSUBADD132PS
• VFMSUBADD213PD
• VFMSUBADD213PS
• VFMSUBADD231PD
• VFMSUBADD231PS
• VFNMADD132PD
• VFNMADD132PS
• VFNMADD132SD
• VFNMADD132SS
• VFNMADD213PD
• VFNMADD213PS
• VFNMADD213SD
• VFNMADD213SS
• VFNMADD231PD
• VFNMADD231PS
• VFNMADD231SD
• VFNMADD231SS

All 60 FMA instructions

More AVX10.1 instructions

• VFMADDSUB213PD
• VFMADDSUB213PH
• VFMADDSUB213PS
• VFMADDSUB231PD
• VFMADDSUB231PH
• VFMADDSUB231PS
• VFMSUB132PD
• VFMSUB132PH
• VFMSUB132PS
• VFMSUB132SD
• VFMSUB132SH
• VFMSUB132SS
• VFMSUB213PD
• VFMSUB213PH
• VFMSUB213PS
• VFMSUB213SD
• VFMSUB213SH
• VFMSUB213SS
• VFMSUB231PD
• VFMSUB231PH
• VFMSUB231SD
• VFMSUB231SH
• VFMSUB231SS
• VFMSUBADD132PD
• VFMSUBADD132PH
• VFMSUBADD132PS
• VFMSUBADD213PD
• VFMSUBADD213PH
• VFMSUBADD213PS
• VFMSUBADD231PD
• VFMSUBADD231PH
• VFMSUBADD231PS
• VFMULCPH
• VFMULCSH
• VFNMADD132PD
• VFNMADD132PH
• VFNMADD132PS
• VFNMADD132SD
• VFNMADD132SH
• VFNMADD132SS

All 686 AVX10.1 instructions