haraldwolff
/
CLBlast
mirror of https://github.com/CNugteren/CLBlast.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fixed an FP16 issue in the homatcopy test; added a comment about improper testing of integer returning functions for FP16

pull/222/head
Cedric Nugteren 2017-11-08 21:20:07 +01:00
parent b18cc9d3f1
commit d24138808b
2 changed files with 5 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
README.md
View File

@ -368,6 +368,10 @@ Known performance related issues:
* Performance issues on ARM Mali GPUs due to missing compiler for support for loop unrolling and array-to-register promotion.
Other known issues:
* Routines returning an integer are currently not properly tested for half-precision FP16: IHAMAX/IHAMIN/IHMAX/IHMIN
Contributing
-------------

2
test/routines/levelx/xomatcopy.hpp
View File

@ -65,7 +65,7 @@ StatusCode RunReference<half>(const Arguments<half> &args, BuffersHost<half> &bu
args2.layout = args.layout; args2.a_transpose = args.a_transpose;
args2.alpha = HalfToFloat(args.alpha);
auto status = RunReference(args2, buffers2);
FloatToHalfBuffer(buffers_host.b_mat, b_buffer2);
FloatToHalfBuffer(buffers_host.b_mat, buffers2.b_mat);
return status;
}