[PATCH 1/2] config/arm: avoid mcpu and march conflicts

[Juraj Linkeš]

On Sun, Jan 21, 2024 at 10:37 AM <pbhagavatula at marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula at marvell.com>
>
> The compiler options march and mtune are a subset
> of mcpu and will lead to conflicts if improper march
> is chosen for a given mcpu.
> To avoid conflicts, force part number march when
> mcpu is available and is supported by the compiler.
>
> Example:
>         march = armv9-a
>         mcpu = neoverse-n2
>
>         mcpu supported, march supported
>         machine_args = ['-mcpu=neoverse-n2', '-march=armv9-a']
>
>         mcpu supported, march not supported
>         machine_args = ['-mcpu=neoverse-n2']
>
>         mcpu not supported, march supported
>         machine_args = ['-march=armv9-a']
>
>         mcpu not supported, march not supported
>         machine_args = ['-march=armv8.6-a']
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula at marvell.com>
> ---
>  config/arm/meson.build | 109 +++++++++++++++++++++++++----------------
>  1 file changed, 67 insertions(+), 42 deletions(-)
>
> diff --git a/config/arm/meson.build b/config/arm/meson.build
> index 36f21d2259..8c8cfccca0 100644
> --- a/config/arm/meson.build
> +++ b/config/arm/meson.build
<snip>
> @@ -127,21 +128,22 @@ implementer_cavium = {
>      ],
>      'part_number_config': {
>          '0xa1': {
> -            'compiler_options': ['-mcpu=thunderxt88'],
> +            'mcpu': 'thunderxt88',
>              'flags': flags_part_number_thunderx
>          },
>          '0xa2': {
> -            'compiler_options': ['-mcpu=thunderxt81'],
> +            'mcpu': 'thunderxt81',
>              'flags': flags_part_number_thunderx
>          },
>          '0xa3': {
> -            'compiler_options': ['-march=armv8-a+crc', '-mcpu=thunderxt83'],
> +            'mcpu': 'thunderxt83',
> +            'compiler_options': ['-march=armv8-a+crc'],

Let's unify this with the rest and specify 'march': 'armv8-a+crc'
instead of having it under compiler_options.

>              'flags': flags_part_number_thunderx
>          },
>          '0xaf': {
>              'march': 'armv8.1-a',
>              'march_features': ['crc', 'crypto'],
> -            'compiler_options': ['-mcpu=thunderx2t99'],
> +            'mcpu': 'thunderx2t99',
>              'flags': [
>                  ['RTE_MACHINE', '"thunderx2"'],
>                  ['RTE_ARM_FEATURE_ATOMICS', true],
> @@ -153,7 +155,7 @@ implementer_cavium = {
>          '0xb2': {
>              'march': 'armv8.2-a',
>              'march_features': ['crc', 'crypto', 'lse'],
> -            'compiler_options': ['-mcpu=octeontx2'],
> +            'mcpu': 'octeontx2',
>              'flags': [
>                  ['RTE_MACHINE', '"cn9k"'],
>                  ['RTE_ARM_FEATURE_ATOMICS', true],
> @@ -176,7 +178,7 @@ implementer_ampere = {
>          '0x0': {
>              'march': 'armv8-a',
>              'march_features': ['crc', 'crypto'],
> -            'compiler_options':  ['-mtune=emag'],
> +            'mcpu': 'emag',

We're changing mtune to mcpu, is this equivalent?

>              'flags': [
>                  ['RTE_MACHINE', '"eMAG"'],
>                  ['RTE_MAX_LCORE', 32],
> @@ -186,7 +188,7 @@ implementer_ampere = {
>          '0xac3': {
>              'march': 'armv8.6-a',
>              'march_features': ['crc', 'crypto'],
> -            'compiler_options':  ['-mcpu=ampere1'],
> +            'mcpu': 'ampere1',
>              'flags': [
>                  ['RTE_MACHINE', '"AmpereOne"'],
>                  ['RTE_MAX_LCORE', 320],
> @@ -206,7 +208,7 @@ implementer_hisilicon = {
>          '0xd01': {
>              'march': 'armv8.2-a',
>              'march_features': ['crypto'],
> -            'compiler_options': ['-mtune=tsv110'],
> +            'mcpu': 'tsv110',
>              'flags': [
>                  ['RTE_MACHINE', '"Kunpeng 920"'],
>                  ['RTE_ARM_FEATURE_ATOMICS', true],
> @@ -695,11 +697,23 @@ if update_flags
>
>      machine_args = [] # Clear previous machine args
>
> +    candidate_mcpu = ''
> +    support_mcpu = false
> +    if part_number_config.has_key('mcpu')
> +        mcpu = part_number_config['mcpu']
> +        if (cc.has_argument('-mcpu=' + mcpu))
> +            candidate_mcpu = mcpu
> +            support_mcpu = true
> +        endif
> +    endif
> +
>      # probe supported archs and their features
>      candidate_march = ''
>      if part_number_config.has_key('march')
> -        if part_number_config.get('force_march', false)
> -            candidate_march = part_number_config['march']
> +        if part_number_config.get('force_march', false) or support_mcpu

Instead of using the extra "support_mcpu" variable, we could do the
same check as with candidate march (if candidate_mcpu != '', which we
actually do below in the last lines of the patch).

If I understand the logic correctly, we don't want to do the march
fallback if mcpu is specified - either the march works with the given
mcpu or we do without it (because we don't actually need it with
mcpu). Is that correct?

> +            if cc.has_argument('-march=' +  part_number_config['march'])

Now that we've added mcpu into the mix, is this still the right
condition? Can the below happen?

This check finds that machine_args = ['-march=armv9-a'] is supported.

But taken together with mcpu (machine_args = ['-mcpu=neoverse-n2',
'-march=armv9-a']), it is not supported? In this case we'll end up
with invalid configuration.

> +                candidate_march = part_number_config['march']
> +            endif
>          else
>              supported_marchs = ['armv8.6-a', 'armv8.5-a', 'armv8.4-a', 'armv8.3-a',
>                                  'armv8.2-a', 'armv8.1-a', 'armv8-a']