https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80986
Anssi Hannula <anssi.hannula at bitwise dot fi> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |anssi.hannula at bitwise dot fi
--- Comment #7 from Anssi Hannula <anssi.hannula at bitwise dot fi> ---
Created attachment 46772
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46772&action=edit
Simple testcase for the issue
The regression still exists on GCC trunk.
We also hit the issue with
https://github.com/google/skia/blob/9318a6cfbcdd160151bc4ae36425563f8cf68d3e/src/opts/SkRasterPipeline_opts.h
causing incorrect rendering.
Code is working correctly on: GCC 5.4.0, clang 6.0.1
Not working correctly on: GCC 7.3.0 and GCC trunk r274958 (and GCC 6.3.0
according to the original reporter)
Attached is a simple testcase that can be used with -std=c++11
-mfloat-abi=softfp for ARM.
Disassembly of main() shows the proper VLDR to s0 only for the latter
"non-auto" call of func() when compiled with affected GCC.
