On Thu, Nov 19, 2015 at 4:08 AM, Kyrill Tkachov <kyrylo.tkac...@arm.com> wrote:
> Hi Andrew,
>
> On 17/11/15 22:10, Andrew Pinski wrote:
>>
>> To Add support for -mcpu=thunderxt88pass1, I needed to fix up a few
>> things in the support for -mcpu=native. First was I wanted to do the same
>> cleanup that was done for some of the other .def files and move the
>> #undef into the .def files instead of the .h files.
>> Second to make it easier to understand the implemention_id and part
>> numbers
>> are really numbers, move them over to use integer to compare against
>> instead of strings which allows for easier comparisons later on.
>> Third fix the way we compare imp and part num; that is instead of finding
>> the part number and then comparing the imp, we find both numbers first
>> and then search for the ones that match.
>> Add a comment to the aarch64-cores.def file in front of each group of
>> cores
>> to signify the groups of company's cores.
>> And all of this allows for adding variant for the check of the cores
>> which allows for thunderxt88pass1 to be added.
>>
>> The reason why thunderxt88pass1 is seperate from thunderx is because
>> thunderx is changed to be an ARMv8.1 arch core while thunderxt88pass1
>> is still an ARMv8 arch core.
>>
>> I tested each of these patches seperately.
>
>
> I tried these patches out on a big.LITTLE system with 2 Cortex-A57 and 4
> Cortex-A53
> cores and the code doesn't detect it properly anymore. Can you have a look
> please?
> The Linux /proc/cpuinfo for that system looks like what's described here:
> https://lists.linaro.org/pipermail/cross-distro/2014-October/000750.html
> Look for the entry "[3] arm64, v3.17 + this patch, Juno platform "
> You can put that into a file, point the fopen call in driver-aarch64.c to
> that file
> and use that to debug the issue.
It is a simple fix, valid_bL_core_p should be checking both orders of
the core as I thought it ordering the BIG as first always. Like:
static bool
valid_bL_core_p (unsigned int *core, unsigned int bL_core)
{
return AARCH64_BIG_LITTLE (core[0], core[1]) == bL_core
|| AARCH64_BIG_LITTLE (core[1], core[0]) == bL_core;
}
Thanks,
Andrew
>
> Thanks,
> Kyrill
>
>
>> Ok for the trunk even though I missed out on stage 1?
>>
>> Thanks,
>> Andrew
>>
>> Andrew Pinski (5):
>> [AARCH64]: Move #undef into .def files.
>> [AARCH64] Change IMP and PART over to integers from strings.
>> [AARCH64] Fix part num and implement indendent.
>> {AARCH64] Add comment for the company's cores.
>> [AARCH64] Add variant support to -m*=native and add thunderxt88pass1.
>>
>> gcc/common/config/aarch64/aarch64-common.c | 5 +-
>> gcc/config/aarch64/aarch64-arches.def | 1 +
>> gcc/config/aarch64/aarch64-cores.def | 43 ++++--
>> gcc/config/aarch64/aarch64-fusion-pairs.def | 1 +
>> gcc/config/aarch64/aarch64-option-extensions.def | 2 +
>> gcc/config/aarch64/aarch64-opts.h | 4 +-
>> gcc/config/aarch64/aarch64-protos.h | 4 -
>> gcc/config/aarch64/aarch64-tune.md | 2 +-
>> gcc/config/aarch64/aarch64-tuning-flags.def | 1 +
>> gcc/config/aarch64/aarch64.c | 7 +-
>> gcc/config/aarch64/aarch64.h | 3 +-
>> gcc/config/aarch64/driver-aarch64.c | 169
>> +++++++++++++----------
>> 12 files changed, 136 insertions(+), 106 deletions(-)
>>
>
