[PATCH] aarch64: Fix f{max,min}{f} build for GCC 4.9 and 5

[PATCH] aarch64: Fix f{max,min}{f} build for GCC 4.9 and 5

[Adhemerval Zanella]

GCC 4.9 and 5 do not generate a correct f{max,min}nm instruction for
__builtin_{fmax,fmin}{f} without -ffinite-math-only.  It is clear a
compiler issue since the instruction can handle NaN and Inf correctly
and GCC6+ does not show this issue.

We backport a fix to GCC 5, raise the minimum required GCC version
for aarch64 (since GCC 4.9 branch is now closed) and/or add configure
check to check for this issue.  However I think -ffinit-math-only
should be safe for these specific implementation and it is a much
simpler solution.

Checked on aarch64-linux-gnu with GCC 5.3.1.

	* sysdeps/aarch64/fpu/Makefile (CFLAGS-s_fmax.c, CFLAGS-s_fmaxf.c,
	CFLAGS-s_fmin.c, CFLAGS-s_fminf.c): New rule: add -ffinite-math-only.

[1] https://gcc.gnu.org/ml/gcc/2016-08/msg00010.html

Signed-off-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
---
 ChangeLog                    | 5 +++++
 sysdeps/aarch64/fpu/Makefile | 6 ++++++
 2 files changed, 11 insertions(+)

diff --git a/sysdeps/aarch64/fpu/Makefile b/sysdeps/aarch64/fpu/Makefile
index bf38f2c..4a182bd 100644
--- a/sysdeps/aarch64/fpu/Makefile
+++ b/sysdeps/aarch64/fpu/Makefile
@@ -5,4 +5,10 @@ CFLAGS-s_lroundf.c += -fno-math-errno
 CFLAGS-s_lround.c += -fno-math-errno
 CFLAGS-s_llroundf.c += -fno-math-errno
 CFLAGS-s_llround.c += -fno-math-errno
+# GCC 4.9 and 5 requires the flag to correct emits a f{max,min}nm
+# for a __builtin_{fmax,fmin}{f}.
+CFLAGS-s_fmax.c += -ffinite-math-only
+CFLAGS-s_fmaxf.c += -ffinite-math-only
+CFLAGS-s_fmin.c += -ffinite-math-only
+CFLAGS-s_fminf.c += -ffinite-math-only
 endif
-- 
2.7.4

Re: [PATCH] aarch64: Fix f{max,min}{f} build for GCC 4.9 and 5

[Szabolcs Nagy]

On 17/11/17 10:32, Adhemerval Zanella wrote:
> GCC 4.9 and 5 do not generate a correct f{max,min}nm instruction for
> __builtin_{fmax,fmin}{f} without -ffinite-math-only.  It is clear a
> compiler issue since the instruction can handle NaN and Inf correctly
> and GCC6+ does not show this issue.
> 
> We backport a fix to GCC 5, raise the minimum required GCC version
> for aarch64 (since GCC 4.9 branch is now closed) and/or add configure
> check to check for this issue.  However I think -ffinit-math-only
> should be safe for these specific implementation and it is a much
> simpler solution.
> 
> Checked on aarch64-linux-gnu with GCC 5.3.1.
> 
> 	* sysdeps/aarch64/fpu/Makefile (CFLAGS-s_fmax.c, CFLAGS-s_fmaxf.c,
> 	CFLAGS-s_fmin.c, CFLAGS-s_fminf.c): New rule: add -ffinite-math-only.
> 

OK.