From mboxrd@z Thu Jan 1 00:00:00 1970 From: Thomas Petazzoni Date: Sun, 7 Aug 2016 11:29:35 +0200 Subject: [Buildroot] [PATCH] openblas: fix build issues on ARM In-Reply-To: <9111A0DF-B21B-4A2E-9CA7-FEAE55D1E91A@gmail.com> References: <1470520434-15516-1-git-send-email-thomas.petazzoni@free-electrons.com> <9111A0DF-B21B-4A2E-9CA7-FEAE55D1E91A@gmail.com> Message-ID: <20160807112935.3a9ca090@free-electrons.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: buildroot@busybox.net Hello, On Sun, 7 Aug 2016 01:13:40 -0700, Khem Raj wrote: > > + - Assuming that one uses -mfloat-abi=hard on ARMv7 is completely > > + wrong. -mfloat-abi=softfp is a perfectly valid choice on ARMv7, and > > + OpenBLAS should not make such assumptions. > > perhaps they do not support softfp, this should be confirmed. Hum, right, they might have some assembly code that assumes floating point arguments are passed in FP registers. > > +-ifeq ($(CORE), ARMV5) > > +-CCOMMON_OPT += -marm -march=armv5 > > +-FCOMMON_OPT += -marm -march=armv5 > > -marm option is also removed. I wonder if this code will now build when we have thumb/thumb2 ISA > selected as default for an arm machine. So maybe I should keep the -marm flag here, and in fact just remove -march=armv5. Also, for the ARMv7-A, I could keep the flag, and make the ARMV7 OpenBLAS platform only selectable when ARMv7 + EABIhf + VFPv3/v4 is available. Thanks for the feedback! Thomas -- Thomas Petazzoni, CTO, Free Electrons Embedded Linux, Kernel and Android engineering http://free-electrons.com