[Sound-open-firmware] Linux build for arm64
Hi Liam / Pierre,
Could we add a build for arm64 too in CI? With recent patches that add support for i.MX8 SOF Linux driver it would be handy to also have the build for arm64.
I can help you with the cross-compilers and build instructions.
Currently we are only building for x86_64.
https://travis-ci.org/thesofproject/linux/builds/560509972
thanks, Daniel.
On 7/18/19 2:27 PM, Daniel Baluta wrote:
Hi Liam / Pierre,
Could we add a build for arm64 too in CI? With recent patches that add support for i.MX8 SOF Linux driver it would be handy to also have the build for arm64.
I can help you with the cross-compilers and build instructions.
Currently we are only building for x86_64.
Sure, I see no issue, as long as it's just compilation it's almost free. hardware-tests are probably out-of-scope as discussed some time back.
the main point will be to agree on which kernel config you want to use. To speed things up, we use a minimal kernel based on defconfig and Kconfig fragments, so that changes are automatically applied. see https://github.com/thesofproject/kconfig. if we have a imx-defconfig fragment file then we can add support very quickly.
I use this for my own 'maintainer' tests where I compile locally with make.cross for multiple architectures, see script attached.
On Thu, Jul 18, 2019 at 10:57 PM Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com wrote:
On 7/18/19 2:27 PM, Daniel Baluta wrote:
Hi Liam / Pierre,
Could we add a build for arm64 too in CI? With recent patches that add support for i.MX8 SOF Linux driver it would be handy to also have the build for arm64.
I can help you with the cross-compilers and build instructions.
Currently we are only building for x86_64.
Sure, I see no issue, as long as it's just compilation it's almost free. hardware-tests are probably out-of-scope as discussed some time back.
the main point will be to agree on which kernel config you want to use. To speed things up, we use a minimal kernel based on defconfig and Kconfig fragments, so that changes are automatically applied. see https://github.com/thesofproject/kconfig. if we have a imx-defconfig fragment file then we can add support very quickly.
I use this for my own 'maintainer' tests where I compile locally with make.cross for multiple architectures, see script attached.
That's great! I will have a look at your scripts and provide some info tomorrow. In the arm64 world there are no separate board defconfigs.
There is one huge defconfig with everything enabled in linux/arch/arm64/configs/defconfig. So, I will trim this down to speed up compilation.
thanks, Daniel.
On 7/18/19 3:04 PM, Daniel Baluta wrote:
On Thu, Jul 18, 2019 at 10:57 PM Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com wrote:
On 7/18/19 2:27 PM, Daniel Baluta wrote:
Hi Liam / Pierre,
Could we add a build for arm64 too in CI? With recent patches that add support for i.MX8 SOF Linux driver it would be handy to also have the build for arm64.
I can help you with the cross-compilers and build instructions.
Currently we are only building for x86_64.
Sure, I see no issue, as long as it's just compilation it's almost free. hardware-tests are probably out-of-scope as discussed some time back.
the main point will be to agree on which kernel config you want to use. To speed things up, we use a minimal kernel based on defconfig and Kconfig fragments, so that changes are automatically applied. see https://github.com/thesofproject/kconfig. if we have a imx-defconfig fragment file then we can add support very quickly.
I use this for my own 'maintainer' tests where I compile locally with make.cross for multiple architectures, see script attached.
That's great! I will have a look at your scripts and provide some info tomorrow. In the arm64 world there are no separate board defconfigs.
There is one huge defconfig with everything enabled in linux/arch/arm64/configs/defconfig. So, I will trim this down to speed up compilation.
FWIW, the kernel compilation for all SOF/x86_64 targets takes about 4mn on a Skylake NUC and less than 1mn on a xeon server. This is the sort of target we'd be looking for so that CI isn't too slow (actually compilation time can be dwarfed by Github/network sluggishness).
participants (2)
-
Daniel Baluta -
Pierre-Louis Bossart