> > On Feb 5, 2024, at 10:10 PM, Wathsala Wathawana Vithanage
> <wathsala.vithan...@arm.com> wrote:
> >
> > Hi Pavan,
> >
> >> 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.
> >
> > Why would one force the march specified in the part number when mcpu for
> > that part number is also available and supported by the compiler?
> >
> It would be good to explain the use case or the problem being faced.
>
The idea of this patchset is to avoid mcpu and march conflicts that can happen
with the current build flow.
#aarch64-linux-gnu-gcc -mcpu=neoverse-n2 -march=armv8.6-a shrn.c
cc1: warning: switch '-mcpu=neoverse-n2' conflicts with '-march=armv8.6-a'
In some versions of GCC mcpu=neoverse-n2 is supported but -march=armv9-a is not
so, current build flow
will choose the next supported march which is armv8.6-a and report a conflict.
> >>
> >> Example:
> >> march = armv9-a
> >> mcpu = neoverse-n2
> >>
> >> mcpu supported, march supported
> >> machine_args = ['-mcpu=neoverse-n2', '-march=armv9-a']
> >
> > -march restricts the compiler to baseline architecture of the -mcpu.
> > For instance, Neoverse-n1's baseline architecture is armv8.2-a,
> > but it has some extensions from armv8.3-a, armv8.4-a, and armv8.5-a.
> > By setting -march to armv8.2-a the compiler will strictly omit extensions
> > from 8.3, 8.4 and 8.5 resulting in a suboptimal outcome.
What if compiler only supports armv8.2-a?
Are you suggesting we don’t use march at all when mcpu is supported?
If so how do you express extensions that the SoC supports?
Neoverse-n2 has optional support for crypto and can only be enabled by
expressing it through march='armv9-a+crypto'
> >
> >>
> >> mcpu supported, march not supported
> >> machine_args = ['-mcpu=neoverse-n2']
> >
> > This will result in the best outcome.
Isn't -mcpu=neoverse-n2 -march=armv9-a+sve2+crypto also the best outcome?
> >
> >>
> >> mcpu not supported, march supported
> >> machine_args = ['-march=armv9-a']
> >
> > This too may result in a suboptimal outcome as optimization space
> > is limited to the given march (not using extensions from later
> > architectures when available).
> >
What if compiler doesn’t support mcpu=neoverse-n2 and only supports
march=armv9-a
> >>
> >> mcpu not supported, march not supported
> >> machine_args = ['-march=armv8.6-a']
> >
> > Compiler knows nothing about the target CPU or the architecture.
> > I think it's better to exit the build process with an error.
> >
Then we would need to mark all old GCC versions as not supported by a newer SoC
I don’t think that’s needed since the binaries still run but not optimally,
currently we have
a warning in place for march mismatch.
> >>
> >> Signed-off-by: Pavan Nikhilesh <pbhagavat...@marvell.com>
> >> ---
> >> v2 Changes:
> >> - Cleanup march inconsistencies. (Juraj Linkes)
> >> - Unify fallback march selection. (Juraj Linkes)
> >> - Tag along ARM WFE patch.
> >> v3 Changes:
> >> - Fix missing 'fallback_march' key check.
> >>
> >> config/arm/meson.build | 108 +++++++++++++++++++++++++-------------
> --
> >> -
> >> 1 file changed, 66 insertions(+), 42 deletions(-)
> >>
> >> diff --git a/config/arm/meson.build b/config/arm/meson.build index
> >> 36f21d22599a..ba859bd060b5 100644
> >> --- a/config/arm/meson.build
> >> +++ b/config/arm/meson.build
> >> @@ -58,18 +58,18 @@ implementer_generic = { }
> >>
> >> part_number_config_arm = {
> >> - '0xd03': {'compiler_options': ['-mcpu=cortex-a53']},
> >> - '0xd04': {'compiler_options': ['-mcpu=cortex-a35']},
> >> - '0xd05': {'compiler_options': ['-mcpu=cortex-a55']},
> >> - '0xd07': {'compiler_options': ['-mcpu=cortex-a57']},
> >> - '0xd08': {'compiler_options': ['-mcpu=cortex-a72']},
> >> - '0xd09': {'compiler_options': ['-mcpu=cortex-a73']},
> >> - '0xd0a': {'compiler_options': ['-mcpu=cortex-a75']},
> >> - '0xd0b': {'compiler_options': ['-mcpu=cortex-a76']},
> >> + '0xd03': {'mcpu': 'cortex-a53'},
> >> + '0xd04': {'mcpu': 'cortex-a35'},
> >> + '0xd05': {'mcpu': 'cortex-a55'},
> >> + '0xd07': {'mcpu': 'cortex-a57'},
> >> + '0xd08': {'mcpu': 'cortex-a72'},
> >> + '0xd09': {'mcpu': 'cortex-a73'},
> >> + '0xd0a': {'mcpu': 'cortex-a75'},
> >> + '0xd0b': {'mcpu': 'cortex-a76'},
> >> '0xd0c': {
> >> 'march': 'armv8.2-a',
> >> 'march_features': ['crypto', 'rcpc'],
> >> - 'compiler_options': ['-mcpu=neoverse-n1'],
> >> + 'mcpu': 'neoverse-n1',
> >> 'flags': [
> >> ['RTE_MACHINE', '"neoverse-n1"'],
> >> ['RTE_ARM_FEATURE_ATOMICS', true], @@ -81,7 +81,7 @@
> >> part_number_config_arm = {
> >> '0xd40': {
> >> 'march': 'armv8.4-a',
> >> 'march_features': ['sve'],
> >> - 'compiler_options': ['-mcpu=neoverse-v1'],
> >> + 'mcpu': 'neoverse-v1',
> >> 'flags': [
> >> ['RTE_MACHINE', '"neoverse-v1"'],
> >> ['RTE_ARM_FEATURE_ATOMICS', true], @@ -92,8 +92,9 @@
> >> part_number_config_arm = {
> >> 'march': 'armv8.4-a',
> >> },
> >> '0xd49': {
> >> + 'march': 'armv9-a',
> >> 'march_features': ['sve2'],
> >> - 'compiler_options': ['-mcpu=neoverse-n2'],
> >> + 'mcpu': 'neoverse-n2',
> >> 'flags': [
> >> ['RTE_MACHINE', '"neoverse-n2"'],
> >> ['RTE_ARM_FEATURE_ATOMICS', true], @@ -127,21 +128,23 @@
> >> 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'],
> >> + 'march': 'armv8-a',
> >> + 'march_features': ['crc'],
> >> + 'mcpu': 'thunderxt83',
> >> '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 +156,7 @@
> >> 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 +179,7 @@
> >> implementer_ampere = {
> >> '0x0': {
> >> 'march': 'armv8-a',
> >> 'march_features': ['crc', 'crypto'],
> >> - 'compiler_options': ['-mtune=emag'],
> >> + 'mcpu': 'emag',
> >> 'flags': [
> >> ['RTE_MACHINE', '"eMAG"'],
> >> ['RTE_MAX_LCORE', 32],
> >> @@ -186,7 +189,7 @@ 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 +209,7 @@
> >> 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 +698,21 @@
> if
> >> update_flags
> >>
> >> machine_args = [] # Clear previous machine args
> >>
> >> + candidate_mcpu = ''
> >> + if part_number_config.has_key('mcpu')
> >> + mcpu = part_number_config['mcpu']
> >> + if (cc.has_argument('-mcpu=' + mcpu))
> >> + candidate_mcpu = mcpu
> >> + 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 candidate_mcpu
> >> !=
> ''
> >> + if cc.has_argument('-march=' + part_number_config['march'])
> >> + 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'] @@
> >> -717,32 +730,43
> >> @@ if update_flags
> >> endif
> >> endforeach
> >> endif
> >> - if candidate_march == ''
> >> - error('No suitable armv8 march version found.')
> >> - endif
> >> +
> >> if candidate_march != part_number_config['march']
> >> - warning('Configuration march version is ' +
> >> - '@0@, but the compiler supports only @1@.'
> >> - .format(part_number_config['march'], candidate_march))
> >> + warning('Configuration march version is @0@, not supported.'
> >> + .format(part_number_config['march']))
> >> + if candidate_march != ''
> >> + warning('Using march version
> >> @0@.'.format(candidate_march))
> >> + endif
> >> endif
> >> - candidate_march = '-march=' + candidate_march
> >>
> >> - march_features = []
> >> - if part_number_config.has_key('march_features')
> >> - march_features += part_number_config['march_features']
> >> - endif
> >> - if soc_config.has_key('extra_march_features')
> >> - march_features += soc_config['extra_march_features']
> >> + if candidate_march == '' and candidate_mcpu == ''
> >> + error('No suitable ARM march/mcpu version found.')
> >> endif
> >> - foreach feature: march_features
> >> - if cc.has_argument('+'.join([candidate_march, feature]))
> >> - candidate_march = '+'.join([candidate_march, feature])
> >> - else
> >> - warning('The compiler does not support feature @0@'
> >> - .format(feature))
> >> +
> >> + if candidate_march != ''
> >> + candidate_march = '-march=' + candidate_march
> >> + march_features = []
> >> + if part_number_config.has_key('march_features')
> >> + march_features += part_number_config['march_features']
> >> endif
> >> - endforeach
> >> - machine_args += candidate_march
> >> + if soc_config.has_key('extra_march_features')
> >> + march_features += soc_config['extra_march_features']
> >> + endif
> >> + foreach feature: march_features
> >> + if cc.has_argument('+'.join([candidate_march, feature]))
> >> + candidate_march = '+'.join([candidate_march, feature])
> >> + else
> >> + warning('The compiler does not support feature @0@'
> >> + .format(feature))
> >> + endif
> >> + endforeach
> >> + machine_args += candidate_march
> >> + endif
> >> + endif
> >> +
> >> + if candidate_mcpu != ''
> >> + candidate_mcpu = '-mcpu=' + candidate_mcpu
> >> + machine_args += candidate_mcpu
> >> endif
> >>
> >> # apply supported compiler options
> >> --
> >> 2.43.0
> >
