[PATCH] D16869: Fix printing of f16 machine operands

Hal Finkel via llvm-commits llvm-commits at lists.llvm.org
Wed Feb 3 16:28:59 PST 2016

hfinkel accepted this revision.
hfinkel added a reviewer: hfinkel.
hfinkel added a comment.
This revision is now accepted and ready to land.

In http://reviews.llvm.org/D16869#343632, @rampitec wrote:

> In http://reviews.llvm.org/D16869#343619, @hfinkel wrote:
> > Please include a test case (you'll need to put '; REQUIRES: asserts' in it if you're checking debug output. Maybe, however, you can get the same problem if you use MIR, in which case that would be better.
> The testcase is simple:
> ; RUN: llc -march=hsail -print-after=dead-mi-elimination < %s 2>&1 | FileCheck %s
>  ; CHECK: half 4.000000e+00
>  define half @half_const(half %in) {
> entry:
>   %add = fadd half %in, 0xH4400
>   ret half %add
> }
> Though I found no backend in the trunk where a half immediate survives instruction selection and present as machine operand, thus I do not see what to put into -march for testing.

Ah, okay. In that case, with `unused` capatilized to match project coding conventions, this LGTM. Just mention in the commit message that we can't (yet) hit this with any in-tree target.


More information about the llvm-commits mailing list