126 lines
5.3 KiB
HTML
126 lines
5.3 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>VCVTNEPS2BF16
|
|||
|
|
— Convert Packed Single Data to Packed BF16 Data</title></head><body><header><nav><ul><li><a href='index.html'>Index</a></li><li>December 2023</li></ul></nav></header><h1>VCVTNEPS2BF16
|
|||
|
|
— Convert Packed Single Data to Packed BF16 Data</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.128.F3.0F38.W0 72 /r VCVTNEPS2BF16 xmm1{k1}{z}, xmm2/m128/m32bcst</td>
|
|||
|
|
<td>A</td>
|
|||
|
|
<td>V/V</td>
|
|||
|
|
<td>AVX512VL AVX512_BF16</td>
|
|||
|
|
<td>Convert packed single data from xmm2/m128 to packed BF16 data in xmm1 with writemask k1.</td></tr>
|
|||
|
|
<tr>
|
|||
|
|
<td>EVEX.256.F3.0F38.W0 72 /r VCVTNEPS2BF16 xmm1{k1}{z}, ymm2/m256/m32bcst</td>
|
|||
|
|
<td>A</td>
|
|||
|
|
<td>V/V</td>
|
|||
|
|
<td>AVX512VL AVX512_BF16</td>
|
|||
|
|
<td>Convert packed single data from ymm2/m256 to packed BF16 data in xmm1 with writemask k1.</td></tr>
|
|||
|
|
<tr>
|
|||
|
|
<td>EVEX.512.F3.0F38.W0 72 /r VCVTNEPS2BF16 ymm1{k1}{z}, zmm2/m512/m32bcst</td>
|
|||
|
|
<td>A</td>
|
|||
|
|
<td>V/V</td>
|
|||
|
|
<td>AVX512F AVX512_BF16</td>
|
|||
|
|
<td>Convert packed single data from zmm2/m512 to packed BF16 data in ymm1 with writemask k1.</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</th>
|
|||
|
|
<th>Tuple</th>
|
|||
|
|
<th>Operand 1</th>
|
|||
|
|
<th>Operand 2</th>
|
|||
|
|
<th>Operand 3</th>
|
|||
|
|
<th>Operand 4</th></tr>
|
|||
|
|
<tr>
|
|||
|
|
<td>A</td>
|
|||
|
|
<td>Full</td>
|
|||
|
|
<td>ModRM:reg (w)</td>
|
|||
|
|
<td>ModRM:r/m (r)</td>
|
|||
|
|
<td>N/A</td>
|
|||
|
|
<td>N/A</td></tr></table>
|
|||
|
|
<h3 id="description">Description<a class="anchor" href="#description">
|
|||
|
|
¶
|
|||
|
|
</a></h3>
|
|||
|
|
<p>Converts one SIMD register of packed single data into a single register of packed BF16 data.</p>
|
|||
|
|
<p>This instruction uses “Round to nearest (even)” rounding mode. Output denormals are always flushed to zero and input denormals are always treated as zero. MXCSR is not consulted nor updated.</p>
|
|||
|
|
<p>As the instruction operand encoding table shows, the EVEX.vvvv field is not used for encoding an operand. EVEX.vvvv is reserved and must be 0b1111 otherwise instructions will #UD.</p>
|
|||
|
|
<h3 id="operation">Operation<a class="anchor" href="#operation">
|
|||
|
|
¶
|
|||
|
|
</a></h3>
|
|||
|
|
<pre>Define convert_fp32_to_bfloat16(x):
|
|||
|
|
IF x is zero or denormal:
|
|||
|
|
dest[15] := x[31] // sign preserving zero (denormal go to zero)
|
|||
|
|
dest[14:0] := 0
|
|||
|
|
ELSE IF x is infinity:
|
|||
|
|
dest[15:0] := x[31:16]
|
|||
|
|
ELSE IF x is NAN:
|
|||
|
|
dest[15:0] := x[31:16] // truncate and set MSB of the mantissa to force QNAN
|
|||
|
|
dest[6] := 1
|
|||
|
|
ELSE // normal number
|
|||
|
|
LSB := x[16]
|
|||
|
|
rounding_bias := 0x00007FFF + LSB
|
|||
|
|
temp[31:0] := x[31:0] + rounding_bias // integer add
|
|||
|
|
dest[15:0] := temp[31:16]
|
|||
|
|
RETURN dest
|
|||
|
|
</pre>
|
|||
|
|
<h4 id="vcvtneps2bf16-dest--src">VCVTNEPS2BF16 dest, src<a class="anchor" href="#vcvtneps2bf16-dest--src">
|
|||
|
|
¶
|
|||
|
|
</a></h4>
|
|||
|
|
<pre>VL = (128, 256, 512)
|
|||
|
|
KL = VL/16
|
|||
|
|
origdest := dest
|
|||
|
|
FOR i := 0 to KL/2-1:
|
|||
|
|
IF k1[ i ] or *no writemask*:
|
|||
|
|
IF src is memory and evex.b == 1:
|
|||
|
|
t := src.fp32[0]
|
|||
|
|
ELSE:
|
|||
|
|
t := src.fp32[ i ]
|
|||
|
|
dest.word[i] := convert_fp32_to_bfloat16(t)
|
|||
|
|
ELSE IF *zeroing*:
|
|||
|
|
dest.word[ i ] := 0
|
|||
|
|
ELSE: // Merge masking, dest element unchanged
|
|||
|
|
dest.word[ i ] := origdest.word[ i ]
|
|||
|
|
DEST[MAXVL-1:VL/2] := 0
|
|||
|
|
</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>VCVTNEPS2BF16 __m128bh _mm_cvtneps_pbh (__m128);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m128bh _mm_mask_cvtneps_pbh (__m128bh, __mmask8, __m128);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m128bh _mm_maskz_cvtneps_pbh (__mmask8, __m128);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m128bh _mm256_cvtneps_pbh (__m256);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m128bh _mm256_mask_cvtneps_pbh (__m128bh, __mmask8, __m256);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m128bh _mm256_maskz_cvtneps_pbh (__mmask8, __m256);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m256bh _mm512_cvtneps_pbh (__m512);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m256bh _mm512_mask_cvtneps_pbh (__m256bh, __mmask16, __m512);
|
|||
|
|
</pre>
|
|||
|
|
<pre>VCVTNEPS2BF16 __m256bh _mm512_maskz_cvtneps_pbh (__mmask16, __m512);
|
|||
|
|
</pre>
|
|||
|
|
<h3 class="exceptions" id="simd-floating-point-exceptions">SIMD Floating-Point Exceptions<a class="anchor" href="#simd-floating-point-exceptions">
|
|||
|
|
¶
|
|||
|
|
</a></h3>
|
|||
|
|
<p>None.</p>
|
|||
|
|
<h3 class="exceptions" id="other-exceptions">Other Exceptions<a class="anchor" href="#other-exceptions">
|
|||
|
|
¶
|
|||
|
|
</a></h3>
|
|||
|
|
<p>See <span class="not-imported">Table 2-49</span>, “Type E4 Class Exception Conditions.”</p><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>
|