Issues with ?lamch routines: use O0 to compile them and static variables #497
Description
It seems that BLIS compiles slamch and dlamch using optimization flags enabled.
My configuration is
./configure --enable-blas --enable-cblas --enable-pba-pools --enable-sba-pools --enable-threading=openmp --prefix=$HOME/amd/src/local zen2
and the compilation of these routines is
[zampins@cn514-15-r blis]$ grep lamch.c make2.log
/sw/csi/gcc/10.2.0/bin/gcc -O3 -Wall -Wno-unused-function -Wfatal-errors -fPIC -std=c99 -D_POSIX_C_SOURCE=200112L -fopenmp -Iinclude/zen2 -I./frame/3/ -I./frame/ind/ukernels/ -I./frame/3/ -I./frame/1m/ -I./frame/1f/ -I./frame/1/ -I./frame/include -DBLIS_VERSION_STRING=\"3.0.0\" -DBLIS_IS_BUILDING_LIBRARY -fvisibility=hidden -c frame/base/noopt/bli_dlamch.c -o obj/zen2/frame/base/noopt/bli_dlamch.o
/sw/csi/gcc/10.2.0/bin/gcc -O3 -Wall -Wno-unused-function -Wfatal-errors -fPIC -std=c99 -D_POSIX_C_SOURCE=200112L -fopenmp -Iinclude/zen2 -I./frame/3/ -I./frame/ind/ukernels/ -I./frame/3/ -I./frame/1m/ -I./frame/1f/ -I./frame/1/ -I./frame/include -DBLIS_VERSION_STRING=\"3.0.0\" -DBLIS_IS_BUILDING_LIBRARY -fvisibility=hidden -c frame/base/noopt/bli_slamch.c -o obj/zen2/frame/base/noopt/bli_slamch.o
Using gcc 10.2 with -march=znver2 on an AMD EPYC 7702 let my code (compiled with -fopenmp) emit this warning
(// WARNING. The value EMIN may be incorrect:-, EMIN = ,i8,/ If, after inspection, the value EMIN looks, acceptable please comment out ,/ the IF block as marked within the code of routine, DLAMC2,,/ otherwise supply EMIN explicitly.,/)(// WARNING. The value EMIN may be incorrect:-, EMIN = ,i8,/ If, after inspection, the value EMIN looks, acceptable please comment out ,/ the IF block as marked within the code of routine, DLAMC2,,/ otherwise supply EMIN explicitly.,/)
I'm not familiar with BLIS makefile rule defintions to be able to provide a fix for this. However, I see the ?lamch routines are in frame/base/noopt/ so there's probably hope that these routines will be compiled with -O0. Same happens with the amd fork of blis as well as with the libflame code. Should I submit an issue for libflame too or this one is enough?