https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104455
Bug ID: 104455
Summary: Cannot select -march=armv7-a using GCC 11
Product: gcc
Version: 11.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: noloader at gmail dot com
Target Milestone: ---
Hi Everyone,
We are trying to fix a compile problem Debian encountered on Sid [1,2]. The
machine is armhf, but I don't have access to it. Sid is using GCC 11, but I am
not sure which version. The problem does not exist in GCC 10 and below.
I think this is the test case. Below, test.S is mostly armv4 but has some armv7
instructions. I used movw and movt as an example since they are armv7.
$ cat test.S
.globl return_magic
.type return_magic,%function
return_magic:
movw r0, #0x1234
movt r0, #0x5678
bl 0
$ g++ -g2 -O3 -Wa,--noexecstack -march=armv7-a test.S -c
When Debian compiles with GCC 11, it results in:
cc1: error: ‘-mfloat-abi=hard’: selected architecture lacks an FPU
test.S does not use floating point or neon instructions. This source file does
not need a fpu. We don't care what the platform default is because we don't use
it. But in this case, the machine is armhf so GCC should know it should use
hard floats.
When I attempt to add -mfpu=auto per [3], it results in another error:
cc1: sorry, unimplemented: -mfpu=auto not currently supported without an
explicit CPU.
cc1: error: -mfloat-abi=hard: selected processor lacks an FPU
test.S does not use floating point or neon instructions. This source file does
not need a fpu. I also tried -mfpu=none, but it results in:
g++: error: unrecognized argument in option ‘-mfpu=none’
g++: note: valid arguments to ‘-mfpu=’ are: auto crypto-neon-fp-armv8
fp-armv8 fpv4-sp-d16 fpv5-d16 fpv5-sp-d16 neon neon-fp-armv8 neon-fp16
neon-vfpv3 neon-vfpv4 vfp vfp3 vfpv2 vfpv3 vfpv3-d16 vfpv3-d16-fp16
vfpv3-fp16 vfpv3xd vfpv3xd-fp16 vfpv4 vfpv4-d16; did you mean ‘neon’?
In the past we avoided specifying a -mfpu option because the code [used to]
work with GCC 4 and above, Clang 3 and above, armel, armhf and Android with
soft floats. The compiler always knew what float abi it should use.
I would like to drop -march=armv7-a, but GCC requires me to use an ISA option
before I use instructions from the ISA. I've never liked the rule, but it is
what it is. (Microsoft's C/C++ compiler got this right).
[1] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1001995
[2] https://github.com/weidai11/cryptopp/issues/1094
[3] https://gcc.gnu.org/onlinedocs/gcc/ARM-Options.html
