phynexis-deps/armadillo-code
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

specify that hardware support for fp16 via F16C is rudimentary

Browse Source
This commit is contained in:
conrad
2025-07-23 13:07:13 +10:00
parent 583efe0eb0
commit e218004dbf
1 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs.html
+3 -2
View File
@@ -18608,10 +18608,11 @@ Support for <code>fp16</code> element types is automatically enabled if <b>both<
<li>use <code>-std=c++23</code> when compiling with gcc or clang</li>
</ul>
<br>
<li>hardware support for native FP16 arithmetic is detected
<li>hardware support for FP16 is detected
<ul>
<li>use <code>-march=native</code> when compiling with gcc or clang; this enables use of all available instructions on your CPU
<li>native FP16 arithmetic is present on x86-64 CPUs with the AVX512-FP16 instruction set extension</li>
<li>rudimentary FP16 hardware support is present on x86-64 CPUs with the F16C instruction set extensions</li>
<li>native FP16 arithmetic is present on x86-64 CPUs with the AVX512-FP16 instruction set extensions</li>
<li>native FP16 arithmetic is present on ARM CPUs with the <code>+fp16</code> feature; typically found on ARMv8.2-A+ devices</li>
</ul>
</li>