Skip to contentSkip to conversation

gcc: disable fortran use of libquadmath

PATCH New Matt Weber 4 participants 5 messages Sep 11, 2020 – Oct 30, 2020

lore
Share & tools
lore

Matt Weber

author
#0 lore raw

PATCH 1/1 gcc: disable fortran use of libquadmath

The GCC package has a default conf option of disabling libquadmath and the toolchain dependencies selectively enabled it if i386 / x64.

Fixes:
https://gitlab.com/bootlin/toolchains-builder/-/jobs/729359622

This patch fixes a build failure when (GCC + glibc) is being built for the IBM Power8 arch and has libgfortran enabled + libquadmath disabled. The libgfortran has a code condition for __float128 and includes the quadmath headers. The bug occurs because Power8 has emulated float128 support. The fix per GCC options is to also set --disable-libquadmath-support which disables the __float128/libquadmath support in gcc/fortran and in libgfortran [1].

Another option to fix the build failure was to enable libquadmath for IBM Power8 (ISA 2.07), however this would be soft float based as the ISA 3.0+ (Power9) first supports native float128 [2][3].

[1] https://fortran.gcc.gnu.narkive.com/8uSfoKUS/patch-build-pr-46540-add-disable-libquadmath-disable-libquadmath-support
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66382#c7
[3] https://gcc.gnu.org/onlinedocs/gcc/RS_002f6000-and-PowerPC-Options.html

Cc Thomas Petazzoni <[email protected]>
---
 package/gcc/gcc.mk | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
package/gcc/gcc.mk +1 -1
index b834269adc..c04d3b4d5c 100644
@@ -116,7 +116,7 @@ endif
 ifeq ($(BR2_USE_WCHAR)$(BR2_TOOLCHAIN_HAS_LIBQUADMATH),yy)
 HOST_GCC_COMMON_CONF_OPTS += --enable-libquadmath
 else
-HOST_GCC_COMMON_CONF_OPTS += --disable-libquadmath
+HOST_GCC_COMMON_CONF_OPTS += --disable-libquadmath --disable-libquadmath-support
 endif
 
 # Disable libsanitizer due to a build issue with gcc 7.5 and glibc 2.31.
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/1 gcc: disable fortran use of libquadmath

Hello Matt,

Le 11/09/2020 ? 14:41, Matt Weber a ?crit?:

Quoted 14 lines

The GCC package has a default conf option of disabling libquadmath and the toolchain dependencies selectively enabled it if i386 / x64.

Fixes:
https://gitlab.com/bootlin/toolchains-builder/-/jobs/729359622

This patch fixes a build failure when (GCC + glibc) is being built for the IBM Power8 arch and has libgfortran enabled + libquadmath disabled. The libgfortran has a code condition for __float128 and includes the quadmath headers. The bug occurs because Power8 has emulated float128 support. The fix per GCC options is to also set --disable-libquadmath-support which disables the __float128/libquadmath support in gcc/fortran and in libgfortran [1].

Quoted 3 lines

Another option to fix the build failure was to enable libquadmath for IBM Power8 (ISA 2.07), however this would be soft float based as the ISA 3.0+ (Power9) first supports native float128 [2][3].

I did a test with toolchain-builder this weekend using Buildroot's master branch but without your patch (I missed that it was included for toolchain builder release 2020.08-1). So I had the same issue...

I tried to fix it by enabling libquadmath even if it's a software support:
http://patchwork.ozlabs.org/project/buildroot/patch/20201004142050.2487724-1-romain.naour at gmail.com

But I can't really test the libquadmath solution since I'm only using qemu to do a boot testing. So your proposal looks safer.

Best regards,
Romain

Quoted 25 lines

[1] https://fortran.gcc.gnu.narkive.com/8uSfoKUS/patch-build-pr-46540-add-disable-libquadmath-disable-libquadmath-support
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66382#c7
[3] https://gcc.gnu.org/onlinedocs/gcc/RS_002f6000-and-PowerPC-Options.html

Cc Thomas Petazzoni <[email protected]>
---
 package/gcc/gcc.mk | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
package/gcc/gcc.mk +1 -1
index b834269adc..c04d3b4d5c 100644
@@ -116,7 +116,7 @@ endif
 ifeq ($(BR2_USE_WCHAR)$(BR2_TOOLCHAIN_HAS_LIBQUADMATH),yy)
 HOST_GCC_COMMON_CONF_OPTS += --enable-libquadmath
 else
-HOST_GCC_COMMON_CONF_OPTS += --disable-libquadmath
+HOST_GCC_COMMON_CONF_OPTS += --disable-libquadmath --disable-libquadmath-support
 endif
 
 # Disable libsanitizer due to a build issue with gcc 7.5 and glibc 2.31.
#2 lore raw

PATCH 1/1 gcc: disable fortran use of libquadmath

Romain,

Quoted 26 lines · Romain Naour

Hello Matt,

Le 11/09/2020 ? 14:41, Matt Weber a ?crit :

Quoted 14 lines

The GCC package has a default conf option of disabling libquadmath and the toolchain dependencies selectively enabled it if i386 / x64.

Fixes:
https://gitlab.com/bootlin/toolchains-builder/-/jobs/729359622

This patch fixes a build failure when (GCC + glibc) is being built for the IBM Power8 arch and has libgfortran enabled + libquadmath disabled. The libgfortran has a code condition for __float128 and includes the quadmath headers. The bug occurs because Power8 has emulated float128 support. The fix per GCC options is to also set --disable-libquadmath-support which disables the __float128/libquadmath support in gcc/fortran and in libgfortran [1].

Quoted 3 lines

Another option to fix the build failure was to enable libquadmath for IBM Power8 (ISA 2.07), however this would be soft float based as the ISA 3.0+ (Power9) first supports native float128 [2][3].

I did a test with toolchain-builder this weekend using Buildroot's master branch but without your patch (I missed that it was included for toolchain builder release 2020.08-1). So I had the same issue...

Quoted 5 lines

I tried to fix it by enabling libquadmath even if it's a software support:
http://patchwork.ozlabs.org/project/buildroot/patch/20201004142050.2487724-1-romain.naour at gmail.com

But I can't really test the libquadmath solution since I'm only using qemu to do a boot testing. So your proposal looks safer.

I ran into a similar test situation. I agree there are more ppc archs that could have libquadmath enabled. However after reading about some of the performance regressions when hardware doesn't support float128 but the quadmath library is used, I picked a conservative path of disabling.

Best Regards,
Matt

#3 lore raw

PATCH 1/1 gcc: disable fortran use of libquadmath

In reply to Matt Weber
Quoted 22 lines · On Fri, 11 Sep 2020 07:41:31 -0500 Matt Weber…

The GCC package has a default conf option of disabling libquadmath and the toolchain dependencies selectively enabled it if i386 / x64.

Fixes:
https://gitlab.com/bootlin/toolchains-builder/-/jobs/729359622

This patch fixes a build failure when (GCC + glibc) is being built for the IBM Power8 arch and has libgfortran enabled + libquadmath disabled. The libgfortran has a code condition for __float128 and includes the quadmath headers. The bug occurs because Power8 has emulated float128 support. The fix per GCC options is to also set --disable-libquadmath-support which disables the __float128/libquadmath support in gcc/fortran and in libgfortran [1].

Another option to fix the build failure was to enable libquadmath for IBM Power8 (ISA 2.07), however this would be soft float based as the ISA 3.0+ (Power9) first supports native float128 [2][3].

[1] https://fortran.gcc.gnu.narkive.com/8uSfoKUS/patch-build-pr-46540-add-disable-libquadmath-disable-libquadmath-support
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66382#c7
[3] https://gcc.gnu.org/onlinedocs/gcc/RS_002f6000-and-PowerPC-Options.html

Quoted 5 lines
Cc Thomas Petazzoni <[email protected]>
---
 package/gcc/gcc.mk | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Applied to master, thanks.

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#4 lore raw

PATCH 1/1 gcc: disable fortran use of libquadmath

In reply to Matt Weber
Quoted 20 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"Matt" == Matt Weber <[email protected]> writes:

The GCC package has a default conf option of disabling libquadmath and the toolchain dependencies selectively enabled it if i386 / x64.

Fixes:
https://gitlab.com/bootlin/toolchains-builder/-/jobs/729359622

This patch fixes a build failure when (GCC + glibc) is being built for the IBM Power8 arch and has libgfortran enabled + libquadmath disabled. The libgfortran has a code condition for __float128 and includes the quadmath headers. The bug occurs because Power8 has emulated float128 support. The fix per GCC options is to also set --disable-libquadmath-support which disables the __float128/libquadmath support in gcc/fortran and in libgfortran [1].

Another option to fix the build failure was to enable libquadmath for IBM Power8 (ISA 2.07), however this would be soft float based as the ISA 3.0+ (Power9) first supports native float128 [2][3].

Quoted 6 lines

[1] https://fortran.gcc.gnu.narkive.com/8uSfoKUS/patch-build-pr-46540-add-disable-libquadmath-disable-libquadmath-support
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66382#c7
[3] https://gcc.gnu.org/onlinedocs/gcc/RS_002f6000-and-PowerPC-Options.html

Cc Thomas Petazzoni <[email protected]>

Committed to 2020.02.x and 2020.08.x, thanks.

Signature · 1 line
Bye, Peter Korsgaard