| Message ID | 20250716145707.2271609-1-skandigraun@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | toolchain/clang: handle big.LITTLE Arm SoCs | expand |
Hi Gyorgy, On 7/16/25 4:57 PM, Gyorgy Sarvari via lists.openembedded.org wrote: > clang doesn't have Arm big.LITTLE specific tune options - when they are used, > the compilation fails with an error like this: > > aarch64-poky-linux-clang: error: unsupported argument 'cortex-a72.cortex-a53+crc+crypto' to option '-mcpu=' > > To avoid this, in case a big.LITTLE SoC is the target, select the tune corresponding > to the LITTLE core. > > This has been lifted from meta-clang, extended with cortexa73-cortexa35 which > was missing. > > Signed-off-by: Gyorgy Sarvari <skandigraun@gmail.com> > --- > meta/classes/toolchain/clang.bbclass | 16 ++++++++++++++++ > 1 file changed, 16 insertions(+) > > diff --git a/meta/classes/toolchain/clang.bbclass b/meta/classes/toolchain/clang.bbclass > index d7b8a3657c..b0e284be84 100644 > --- a/meta/classes/toolchain/clang.bbclass > +++ b/meta/classes/toolchain/clang.bbclass > @@ -34,4 +34,20 @@ TUNE_CCARGS += "${@bb.utils.contains("DISTRO_FEATURES", "usrmerge", " --dyld-pre > LDFLAGS:append:class-nativesdk:x86-64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-x86-64.so.2" > LDFLAGS:append:class-nativesdk:aarch64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-aarch64.so.1" > > +# clang doesn't support big.LITTLE tunes, choose always the tune for the LITTLE cores > +TUNE_CCARGS:remove = "\ > + -mcpu=cortex-a57.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a72.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a15.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a17.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a72.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a73.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a73.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a75.cortex-a55${TUNE_CCARGS_MARCH_OPTS} \ > + -mcpu=cortex-a76.cortex-a55${TUNE_CCARGS_MARCH_OPTS}" > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa53 cortexa57-cortexa53 cortexa73-cortexa53", " -mcpu=cortex-a53${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa15-cortexa7 cortexa17-cortexa7", " -mcpu=cortex-a7${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa35 cortexa73-cortexa35", " -mcpu=cortex-a35${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa75-cortexa55 cortexa76-cortexa55", " -mcpu=cortex-a55${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > + Wouldn't it be better to have this in the tune conf file directly? I'm a bit worried we'll have more big.LITTLE flavors in the future and because this is in a separate file it'll be out of date. We should be able to do that by simply using the toolchain-clang or toolchain-gcc override wherever appropriate? E.g. MCPU = "cortexa73-cortexa35" # clang doesn't have Arm big.LITTLE specific tune options, so select the tune corresponding to the LITTLE core. MCPU:toolchain-clang = "cortexa35" TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", "cortexa73-cortexa35", " -mcpu=${MCPU}", "", d)}" [... and other replacements in meta/conf/machine/include/arm/armv8a/tune-cortexa73-cortexa35.inc] (NOT TESTED) something like that? What do you think? Cheers, Quentin
On 7/16/25 17:10, Quentin Schulz wrote: > Hi Gyorgy, > > On 7/16/25 4:57 PM, Gyorgy Sarvari via lists.openembedded.org wrote: >> clang doesn't have Arm big.LITTLE specific tune options - when they are used, >> the compilation fails with an error like this: >> >> aarch64-poky-linux-clang: error: unsupported argument 'cortex-a72.cortex-a53+crc+crypto' to option '-mcpu=' >> >> To avoid this, in case a big.LITTLE SoC is the target, select the tune corresponding >> to the LITTLE core. >> >> This has been lifted from meta-clang, extended with cortexa73-cortexa35 which >> was missing. >> >> Signed-off-by: Gyorgy Sarvari <skandigraun@gmail.com> >> --- >> meta/classes/toolchain/clang.bbclass | 16 ++++++++++++++++ >> 1 file changed, 16 insertions(+) >> >> diff --git a/meta/classes/toolchain/clang.bbclass b/meta/classes/toolchain/clang.bbclass >> index d7b8a3657c..b0e284be84 100644 >> --- a/meta/classes/toolchain/clang.bbclass >> +++ b/meta/classes/toolchain/clang.bbclass >> @@ -34,4 +34,20 @@ TUNE_CCARGS += "${@bb.utils.contains("DISTRO_FEATURES", "usrmerge", " --dyld-pre >> LDFLAGS:append:class-nativesdk:x86-64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-x86-64.so.2" >> LDFLAGS:append:class-nativesdk:aarch64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-aarch64.so.1" >> >> +# clang doesn't support big.LITTLE tunes, choose always the tune for the LITTLE cores >> +TUNE_CCARGS:remove = "\ >> + -mcpu=cortex-a57.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a72.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a15.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a17.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a72.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a73.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a73.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a75.cortex-a55${TUNE_CCARGS_MARCH_OPTS} \ >> + -mcpu=cortex-a76.cortex-a55${TUNE_CCARGS_MARCH_OPTS}" >> +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa53 cortexa57-cortexa53 cortexa73-cortexa53", " -mcpu=cortex-a53${TUNE_CCARGS_MARCH_OPTS}", "", d)}" >> +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa15-cortexa7 cortexa17-cortexa7", " -mcpu=cortex-a7${TUNE_CCARGS_MARCH_OPTS}", "", d)}" >> +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa35 cortexa73-cortexa35", " -mcpu=cortex-a35${TUNE_CCARGS_MARCH_OPTS}", "", d)}" >> +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa75-cortexa55 cortexa76-cortexa55", " -mcpu=cortex-a55${TUNE_CCARGS_MARCH_OPTS}", "", d)}" >> + > Wouldn't it be better to have this in the tune conf file directly? I'm a > bit worried we'll have more big.LITTLE flavors in the future and because > this is in a separate file it'll be out of date. > > We should be able to do that by simply using the toolchain-clang or > toolchain-gcc override wherever appropriate? > > E.g. > > MCPU = "cortexa73-cortexa35" > # clang doesn't have Arm big.LITTLE specific tune options, so select the > tune corresponding to the LITTLE core. > MCPU:toolchain-clang = "cortexa35" > > TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", > "cortexa73-cortexa35", " -mcpu=${MCPU}", "", d)}" > > [... and other replacements in > meta/conf/machine/include/arm/armv8a/tune-cortexa73-cortexa35.inc] > > (NOT TESTED) something like that? > > What do you think? I have no strong opinion about this... with that said, the original list was indeed slightly outdated. Will send a v2 later today with this implementation.
On Wed, Jul 16, 2025 at 8:10 AM Quentin Schulz via lists.openembedded.org <quentin.schulz=cherry.de@lists.openembedded.org> wrote: > > Hi Gyorgy, > > On 7/16/25 4:57 PM, Gyorgy Sarvari via lists.openembedded.org wrote: > > clang doesn't have Arm big.LITTLE specific tune options - when they are used, > > the compilation fails with an error like this: > > > > aarch64-poky-linux-clang: error: unsupported argument 'cortex-a72.cortex-a53+crc+crypto' to option '-mcpu=' > > > > To avoid this, in case a big.LITTLE SoC is the target, select the tune corresponding > > to the LITTLE core. > > > > This has been lifted from meta-clang, extended with cortexa73-cortexa35 which > > was missing. > > > > Signed-off-by: Gyorgy Sarvari <skandigraun@gmail.com> > > --- > > meta/classes/toolchain/clang.bbclass | 16 ++++++++++++++++ > > 1 file changed, 16 insertions(+) > > > > diff --git a/meta/classes/toolchain/clang.bbclass b/meta/classes/toolchain/clang.bbclass > > index d7b8a3657c..b0e284be84 100644 > > --- a/meta/classes/toolchain/clang.bbclass > > +++ b/meta/classes/toolchain/clang.bbclass > > @@ -34,4 +34,20 @@ TUNE_CCARGS += "${@bb.utils.contains("DISTRO_FEATURES", "usrmerge", " --dyld-pre > > LDFLAGS:append:class-nativesdk:x86-64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-x86-64.so.2" > > LDFLAGS:append:class-nativesdk:aarch64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-aarch64.so.1" > > > > +# clang doesn't support big.LITTLE tunes, choose always the tune for the LITTLE cores > > +TUNE_CCARGS:remove = "\ > > + -mcpu=cortex-a57.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a72.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a15.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a17.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a72.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a73.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a73.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a75.cortex-a55${TUNE_CCARGS_MARCH_OPTS} \ > > + -mcpu=cortex-a76.cortex-a55${TUNE_CCARGS_MARCH_OPTS}" > > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa53 cortexa57-cortexa53 cortexa73-cortexa53", " -mcpu=cortex-a53${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa15-cortexa7 cortexa17-cortexa7", " -mcpu=cortex-a7${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa35 cortexa73-cortexa35", " -mcpu=cortex-a35${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > > +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa75-cortexa55 cortexa76-cortexa55", " -mcpu=cortex-a55${TUNE_CCARGS_MARCH_OPTS}", "", d)}" > > + > > Wouldn't it be better to have this in the tune conf file directly? I'm a > bit worried we'll have more big.LITTLE flavors in the future and because > this is in a separate file it'll be out of date. > > We should be able to do that by simply using the toolchain-clang or > toolchain-gcc override wherever appropriate? > > E.g. > > MCPU = "cortexa73-cortexa35" > # clang doesn't have Arm big.LITTLE specific tune options, so select the > tune corresponding to the LITTLE core. > MCPU:toolchain-clang = "cortexa35" > > TUNE_CCARGS .= "${@bb.utils.contains("TUNE_FEATURES", > "cortexa73-cortexa35", " -mcpu=${MCPU}", "", d)}" > > [... and other replacements in > meta/conf/machine/include/arm/armv8a/tune-cortexa73-cortexa35.inc] > > (NOT TESTED) something like that? > > What do you think? Agreed. Having it in tune files is definitely easier to maintain > > Cheers, > Quentin > > -=-=-=-=-=-=-=-=-=-=-=- > Links: You receive all messages sent to this group. > View/Reply Online (#220469): https://lists.openembedded.org/g/openembedded-core/message/220469 > Mute This Topic: https://lists.openembedded.org/mt/114186273/1997914 > Group Owner: openembedded-core+owner@lists.openembedded.org > Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [raj.khem@gmail.com] > -=-=-=-=-=-=-=-=-=-=-=- >
diff --git a/meta/classes/toolchain/clang.bbclass b/meta/classes/toolchain/clang.bbclass index d7b8a3657c..b0e284be84 100644 --- a/meta/classes/toolchain/clang.bbclass +++ b/meta/classes/toolchain/clang.bbclass @@ -34,4 +34,20 @@ TUNE_CCARGS += "${@bb.utils.contains("DISTRO_FEATURES", "usrmerge", " --dyld-pre LDFLAGS:append:class-nativesdk:x86-64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-x86-64.so.2" LDFLAGS:append:class-nativesdk:aarch64 = " -Wl,-dynamic-linker,${base_libdir}/ld-linux-aarch64.so.1" +# clang doesn't support big.LITTLE tunes, choose always the tune for the LITTLE cores +TUNE_CCARGS:remove = "\ + -mcpu=cortex-a57.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a72.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a15.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a17.cortex-a7${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a72.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a73.cortex-a35${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a73.cortex-a53${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a75.cortex-a55${TUNE_CCARGS_MARCH_OPTS} \ + -mcpu=cortex-a76.cortex-a55${TUNE_CCARGS_MARCH_OPTS}" +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa53 cortexa57-cortexa53 cortexa73-cortexa53", " -mcpu=cortex-a53${TUNE_CCARGS_MARCH_OPTS}", "", d)}" +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa15-cortexa7 cortexa17-cortexa7", " -mcpu=cortex-a7${TUNE_CCARGS_MARCH_OPTS}", "", d)}" +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa72-cortexa35 cortexa73-cortexa35", " -mcpu=cortex-a35${TUNE_CCARGS_MARCH_OPTS}", "", d)}" +TUNE_CCARGS:append = "${@bb.utils.contains_any("TUNE_FEATURES", "cortexa75-cortexa55 cortexa76-cortexa55", " -mcpu=cortex-a55${TUNE_CCARGS_MARCH_OPTS}", "", d)}" + TCOVERRIDE = "toolchain-clang"
clang doesn't have Arm big.LITTLE specific tune options - when they are used, the compilation fails with an error like this: aarch64-poky-linux-clang: error: unsupported argument 'cortex-a72.cortex-a53+crc+crypto' to option '-mcpu=' To avoid this, in case a big.LITTLE SoC is the target, select the tune corresponding to the LITTLE core. This has been lifted from meta-clang, extended with cortexa73-cortexa35 which was missing. Signed-off-by: Gyorgy Sarvari <skandigraun@gmail.com> --- meta/classes/toolchain/clang.bbclass | 16 ++++++++++++++++ 1 file changed, 16 insertions(+)