Re: [Buildroot] [PATCH next 1/2] package/highway: new package

[Julien Olivain <ju.o@free.fr>]

Hi Thomas,

On 23/11/2022 15:17, Thomas Petazzoni wrote:
> Hello Julien,
> 
> On Tue, 22 Nov 2022 22:55:56 +0100
> Julien Olivain <ju.o@free.fr> wrote:
> 
>> diff --git a/package/highway/Config.in b/package/highway/Config.in
>> new file mode 100644
>> index 0000000000..f9a0bc910c
>> --- /dev/null
>> +++ b/package/highway/Config.in
>> @@ -0,0 +1,42 @@
>> +config BR2_PACKAGE_HIGHWAY_ARCH_SUPPORTS
>> +	bool
>> +	# Supports all BR architectures by default, with few exceptions
>> +	default y
>> +	# Armv7 support is only with fpv4, see:
>> +	# https://github.com/google/highway/blob/1.0.1/CMakeLists.txt#L221
>> +	depends on !BR2_arm || BR2_ARM_FPU_VFPV4
> 
> Hm, I suppose if it works on other weird architectures, it must have a
> non-optimized version, that could work on ARM platforms that do not
> have a VFPv4 ?

You are right! There is apparently an issue in the package on
detection of arm without VFPv4. I initially wrote this package recipe
for highway-1.0.1. I then updated it to 1.0.2 which seems to have
relaxed some restrictions (incl. RISC-V 32bit). I'll investigate how
those limitations can be removed and send an updated version of this
patch.

>> +ifeq ($(BR2_ARM_FPU_VFPV4),y)
>> +HIGHWAY_CONF_OPTS += -DHWY_CMAKE_ARM7=ON
>> +else
>> +HIGHWAY_CONF_OPTS += -DHWY_CMAKE_ARM7=OFF
>> +endif
> 
> So even with -DHWY_CMAKE_ARM7=OFF, it doesn't work for ARMv5, or ARMv7
> without VFPv4 ?
> 
>> diff --git a/support/testing/tests/package/test_highway.py 
>> b/support/testing/tests/package/test_highway.py
> 
> Runtime tests coming together with the new package: excellent!

Yes, this test infra is fun! If I manage to remove the arch
limitations discussed above, I'll also rewrite this test to use the
default armv5 config...

> Thomas

Thanks for the review!
Best regards,

Julien.
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot