110 lines
5.6 KiB
HTML
110 lines
5.6 KiB
HTML
<!DOCTYPE html>
|
||
<html xmlns="http://www.w3.org/1999/xhtml" xmlns:svg="http://www.w3.org/2000/svg" xmlns:x86="http://www.felixcloutier.com/x86"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8"><link rel="stylesheet" type="text/css" href="style.css"></link><title>V4FMADDPS/V4FNMADDPS
|
||
— Packed Single Precision Floating-Point Fused Multiply-Add(4-Iterations)</title></head><body><header><nav><ul><li><a href='index.html'>Index</a></li><li>December 2023</li></ul></nav></header><h1>V4FMADDPS/V4FNMADDPS
|
||
— Packed Single Precision Floating-Point Fused Multiply-Add(4-Iterations)</h1>
|
||
|
||
|
||
<table>
|
||
<tr>
|
||
<th>Opcode/Instruction</th>
|
||
<th>Op/En</th>
|
||
<th>64/32 bit Mode Support</th>
|
||
<th>CPUID Feature Flag</th>
|
||
<th>Description</th></tr>
|
||
<tr>
|
||
<td>EVEX.512.F2.0F38.W0 9A /r V4FMADDPS zmm1{k1}{z}, zmm2+3, m128</td>
|
||
<td>A</td>
|
||
<td>V/V</td>
|
||
<td>AVX512_4FMAPS</td>
|
||
<td>Multiply packed single-precision floating-point values from source register block indicated by zmm2 by values from m128 and accumulate the result in zmm1.</td></tr>
|
||
<tr>
|
||
<td>EVEX.512.F2.0F38.W0 AA /r V4FNMADDPS zmm1{k1}{z}, zmm2+3, m128</td>
|
||
<td>A</td>
|
||
<td>V/V</td>
|
||
<td>AVX512_4FMAPS</td>
|
||
<td>Multiply and negate packed single-precision floating-point values from source register block indicated by zmm2 by values from m128 and accumulate the result in zmm1.</td></tr></table>
|
||
<h2 id="instruction-operand-encoding">Instruction Operand Encoding<a class="anchor" href="#instruction-operand-encoding">
|
||
¶
|
||
</a></h2>
|
||
<table>
|
||
<tr>
|
||
<th>Op/En Tuple Operand 1 Operand 2 Operand 3 Operand 4</th>
|
||
<th></th>
|
||
<th></th>
|
||
<th></th>
|
||
<th></th>
|
||
<th></th></tr>
|
||
<tr>
|
||
<td>A Tuple1_4X ModRM:reg (r, w) EVEX.vvvv (r) ModRM:r/m (r) N/A</td>
|
||
<td></td>
|
||
<td></td>
|
||
<td></td>
|
||
<td></td>
|
||
<td></td></tr></table>
|
||
<h3 id="description">Description<a class="anchor" href="#description">
|
||
¶
|
||
</a></h3>
|
||
<p>This instruction computes 4 sequential packed fused single-precision floating-point multiply-add instructions with a sequentially selected memory operand in each of the four steps.</p>
|
||
<p>In the above box, the notation of “+3” is used to denote that the instruction accesses 4 source registers based on that operand; sources are consecutive, start in a multiple of 4 boundary, and contain the encoded register operand.</p>
|
||
<p>This instruction supports memory fault suppression. The entire memory operand is loaded if any of the 16 lowest significant mask bits is set to 1 or if a “no masking” encoding is used.</p>
|
||
<p>The tuple type Tuple1_4X implies that four 32-bit elements (16 bytes) are referenced by the memory operation portion of this instruction.</p>
|
||
<p>Rounding is performed at every FMA (fused multiply and add) boundary. Exceptions are also taken sequentially. Pre- and post-computational exceptions of the first FMA take priority over the pre- and post-computational exceptions of the second FMA, etc.</p>
|
||
<h3 id="operation">Operation<a class="anchor" href="#operation">
|
||
¶
|
||
</a></h3>
|
||
<pre>src_reg_id is the 5 bit index of the vector register specified in the instruction as the src1 register.
|
||
define NFMA_PS(kl, vl, dest, k1, msrc, regs_loaded, src_base, posneg):
|
||
tmpdest := dest
|
||
// reg[] is an array representing the SIMD register file.
|
||
FOR j := 0 to regs_loaded-1:
|
||
FOR i := 0 to kl-1:
|
||
IF k1[i] or *no writemask*:
|
||
IF posneg = 0:
|
||
tmpdest.single[i] := RoundFPControl_MXCSR(tmpdest.single[i] - reg[src_base + j ].single[i] * msrc.single[j])
|
||
ELSE:
|
||
tmpdest.single[i] := RoundFPControl_MXCSR(tmpdest.single[i] + reg[src_base + j ].single[i] * msrc.single[j])
|
||
ELSE IF *zeroing*:
|
||
tmpdest.single[i] := 0
|
||
dest := tmpdst
|
||
dest[MAX_VL-1:VL] := 0
|
||
V4FMADDPS and V4FNMADDPS dest{k1}, src1, msrc (AVX512)
|
||
KL, VL = (16,512)
|
||
regs_loaded := 4
|
||
src_base := src_reg_id & ~3 // for src1 operand
|
||
posneg := 0 if negative form, 1 otherwise
|
||
NFMA_PS(kl, vl, dest, k1, msrc, regs_loaded, src_base, posneg)
|
||
</pre>
|
||
<h3 id="intel-c-c++-compiler-intrinsic-equivalent">Intel C/C++ Compiler Intrinsic Equivalent<a class="anchor" href="#intel-c-c++-compiler-intrinsic-equivalent">
|
||
¶
|
||
</a></h3>
|
||
<pre>V4FMADDPS __m512 _mm512_4fmadd_ps( __m512, __m512x4, __m128 *);
|
||
</pre>
|
||
<pre>V4FMADDPS __m512 _mm512_mask_4fmadd_ps(__m512, __mmask16, __m512x4, __m128 *);
|
||
</pre>
|
||
<pre>V4FMADDPS __m512 _mm512_maskz_4fmadd_ps(__mmask16, __m512, __m512x4, __m128 *);
|
||
</pre>
|
||
<pre>V4FNMADDPS __m512 _mm512_4fnmadd_ps(__m512, __m512x4, __m128 *);
|
||
</pre>
|
||
<pre>V4FNMADDPS __m512 _mm512_mask_4fnmadd_ps(__m512, __mmask16, __m512x4, __m128 *);
|
||
</pre>
|
||
<pre>V4FNMADDPS __m512 _mm512_maskz_4fnmadd_ps(__mmask16, __m512, __m512x4, __m128 *);
|
||
</pre>
|
||
<h3 class="exceptions" id="simd-floating-point-exceptions">SIMD Floating-Point Exceptions<a class="anchor" href="#simd-floating-point-exceptions">
|
||
¶
|
||
</a></h3>
|
||
<p>Overflow, Underflow, Invalid, Precision, Denormal.</p>
|
||
<h3 class="exceptions" id="other-exceptions">Other Exceptions<a class="anchor" href="#other-exceptions">
|
||
¶
|
||
</a></h3>
|
||
<p>See Type E2; additionally:</p>
|
||
<table>
|
||
<tr>
|
||
<td>#UD</td>
|
||
<td>If the EVEX broadcast bit is set to 1.</td></tr>
|
||
<tr>
|
||
<td>#UD</td>
|
||
<td>If the MODRM.mod = 0b11.</td></tr></table><footer><p>
|
||
This UNOFFICIAL, mechanically-separated, non-verified reference is provided for convenience, but it may be
|
||
inc<span style="opacity: 0.2">omp</span>lete or b<sub>r</sub>oke<sub>n</sub> in various obvious or non-obvious
|
||
ways. Refer to <a href="https://software.intel.com/en-us/download/intel-64-and-ia-32-architectures-sdm-combined-volumes-1-2a-2b-2c-2d-3a-3b-3c-3d-and-4">Intel® 64 and IA-32 Architectures Software Developer’s Manual</a> for anything serious.
|
||
</p></footer></body></html>
|