ne10: set NE10_TARGET_ARCH with an override instead of anonymous python
PATCH 1/1 ne10: set NE10_TARGET_ARCH with an override instead of anonymous python
* set COMPATIBLE_MACHINE to (^$) to prevent building it for any other architectures than armv7a and aarch64
* with new arm tune files it's easy to have armv7a in OVERRIDES even when there isn't armv7a in TUNE_FEATURES:
meta/conf/machine/include/tune-cortexa9.inc:7
"${@bb.utils.contains('TUNE_FEATURES', 'cortexa9', 'armv7a:', '',d)}" in cases like this COMPATIBLE_MACHINE was satisfied thanks to the armv7a OVERRIDE, but then the anonymous python was failing with:
ne10 was skipped: Incompatible with archs other than armv7 and aarch64
--- meta-oe/recipes-support/ne10/ne10_1.2.1.bb | 19 ++++++------------- 1 file changed, 6 insertions(+), 13 deletions(-)
meta-oe/recipes-support/ne10/ne10_1.2.1.bb +6 -13
@@ -15,22 +15,15 @@ PV .= "gitr+${SRCPV}" inherit cmake -NE10_TARGET_ARCH = "" -EXTRA_OECMAKE = '-DGNULINUX_PLATFORM=ON -DNE10_BUILD_SHARED=ON -DNE10_LINUX_TARGET_ARCH="${NE10_TARGET_ARCH}"' - +# Incompatible with archs other than armv7 and aarch64 +COMPATIBLE_MACHINE = "(^$)" COMPATIBLE_MACHINE_aarch64 = "(.*)" COMPATIBLE_MACHINE_armv7a = "(.*)" +NE10_TARGET_ARCH = "" +NE10_TARGET_ARCH_aarch64 = "aarch64" +NE10_TARGET_ARCH_armv7a = "armv7" -python () { - if any(t.startswith('armv7') for t in d.getVar('TUNE_FEATURES').split()): - d.setVar('NE10_TARGET_ARCH', 'armv7') - bb.debug(2, 'Building Ne10 for armv7') - elif any(t.startswith('aarch64') for t in d.getVar('TUNE_FEATURES').split()): - d.setVar('NE10_TARGET_ARCH', 'aarch64') - bb.debug(2, 'Building Ne10 for aarch64') - else: - raise bb.parse.SkipRecipe("Incompatible with archs other than armv7 and aarch64") -} +EXTRA_OECMAKE = '-DGNULINUX_PLATFORM=ON -DNE10_BUILD_SHARED=ON -DNE10_LINUX_TARGET_ARCH="${NE10_TARGET_ARCH}"' do_install() { install -d ${D}${libdir}
Signature · 1 line
2.17.1
Quoted 50 lines · Martin Jansa
* set
COMPATIBLE_MACHINEto (^$) to prevent building it for any other architectures than armv7a and aarch64
* with new arm tune files it's easy to have armv7a in OVERRIDES even when there isn't armv7a inTUNE_FEATURES:
meta/conf/machine/include/tune-cortexa9.inc:7"${@bb.utils.contains('TUNE_FEATURES', 'cortexa9', 'armv7a:', '',d)}"in cases like this
COMPATIBLE_MACHINEwas satisfied thanks to the armv7a OVERRIDE, but then the anonymous python was failing with:
ne10 was skipped: Incompatible with archs other than armv7 and aarch64Signed-off-by Martin Jansa <[email protected]>--- meta-oe/recipes-support/ne10/ne10_1.2.1.bb | 19 ++++++------------- 1 file changed, 6 insertions(+), 13 deletions(-)meta-oe/recipes-support/ne10/ne10_1.2.1.bb +6
-14@@ -15,22 +15,15 @@ PV .= "gitr+${SRCPV}" inherit cmake -NE10_TARGET_ARCH = "" -EXTRA_OECMAKE = '-DGNULINUX_PLATFORM=ON -DNE10_BUILD_SHARED=ON -DNE10_LINUX_TARGET_ARCH="${NE10_TARGET_ARCH}"' - +# Incompatible with archs other than armv7 and aarch64 +COMPATIBLE_MACHINE = "(^$)" COMPATIBLE_MACHINE_aarch64 = "(.*)" COMPATIBLE_MACHINE_armv7a = "(.*)" +NE10_TARGET_ARCH = "" +NE10_TARGET_ARCH_aarch64 = "aarch64" +NE10_TARGET_ARCH_armv7a = "armv7" -python () { - if any(t.startswith('armv7') for t in d.getVar('TUNE_FEATURES').split()): - d.setVar('NE10_TARGET_ARCH', 'armv7') - bb.debug(2, 'Building Ne10 for armv7') - elif any(t.startswith('aarch64') for t in d.getVar('TUNE_FEATURES').split()): - d.setVar('NE10_TARGET_ARCH', 'aarch64') - bb.debug(2, 'Building Ne10 for aarch64') - else: - raise bb.parse.SkipRecipe("Incompatible with archs other than armv7 and aarch64") -} +EXTRA_OECMAKE = '-DGNULINUX_PLATFORM=ON -DNE10_BUILD_SHARED=ON -DNE10_LINUX_TARGET_ARCH="${NE10_TARGET_ARCH}"' do_install() { install -d ${D}${libdir} --
Quoted 2 lines
2.17.1
Have stolen the idea of this patch for another recipe. I think cortexa7 (raspi) needs COMPATIBLE_MACHINE_armv7ve and NE10_TARGET_ARCH_armv7ve entries. I know they were not handled before but... - or do I miss something
Andreas
Quoted 10 lines · Martin Jansa
* set
COMPATIBLE_MACHINEto (^$) to prevent building it for any other architectures than armv7a and aarch64
* with new arm tune files it's easy to have armv7a in OVERRIDES even when there isn't armv7a inTUNE_FEATURES:
meta/conf/machine/include/tune-cortexa9.inc:7"${@bb.utils.contains('TUNE_FEATURES', 'cortexa9', 'armv7a:', '',d)}"in cases like this
COMPATIBLE_MACHINEwas satisfied thanks to the armv7a OVERRIDE, but then the anonymous python was failing with:
ne10 was skipped: Incompatible with archs other than armv7 and aarch64
...
How does this handle Cortex A9 SoCs without NEON support?
cu
Adrian
Signature · 4 lines
"Is there not promise of rain?" Ling Tan asked suddenly out
of the darkness. There had been need of rain for many days.
"Only a promise," Lao Er said.
Pearl S. Buck - Dragon SeedQuoted 13 lines · Adrian Bunk
Quoted 10 lines · Martin Jansa
* set
COMPATIBLE_MACHINEto (^$) to prevent building it for any other architectures than armv7a and aarch64
* with new arm tune files it's easy to have armv7a in OVERRIDES even when there isn't armv7a inTUNE_FEATURES:
meta/conf/machine/include/tune-cortexa9.inc:7"${@bb.utils.contains('TUNE_FEATURES', 'cortexa9', 'armv7a:', '',d)}"in cases like this
COMPATIBLE_MACHINEwas satisfied thanks to the armv7a OVERRIDE, but then the anonymous python was failing with:
ne10 was skipped: Incompatible with archs other than armv7 and aarch64
...
Quoted 1 line
How does this handle Cortex A9 SoCs without NEON support?
this is a valid concern although rare but this combination is out there ( tegra2 )
and ne10 needs neon, so probably its better to check for neon in tune features
than anything else first.
I'm not using ne10 or libopus anywhere, it was just breaking sstate-diff-machines.sh script by adding that unjustifiable dependency based on TUNE_FEATURES.
I was going to send v2 with the armv7ve entries as requested by Andreas (I forgot that armv7ve tunes no longer include armv7a in overrides), but if someone wants to rework it to test for neon in TUNE_FEATURES feel free to do so.
It didn't check for neon before, so at least this change didn't make it worse.
Quoted 24 lines · Khem Raj
Quoted 8 lines · Adrian Bunk
Quoted 6 lines · Martin Jansa
* set
COMPATIBLE_MACHINEto (^$) to prevent building it for any other architectures than armv7a and aarch64
* with new arm tune files it's easy to have armv7a in OVERRIDES even when there isn't armv7a inTUNE_FEATURES:
meta/conf/machine/include/tune-cortexa9.inc:7"${@bb.utils.contains('TUNE_FEATURES', 'cortexa9', 'armv7a:','',d)}"
Quoted 3 lines
Quoted 3 lines
in cases like this
COMPATIBLE_MACHINEwas satisfied thanks to the armv7a OVERRIDE, but then the anonymous python was failing with:
ne10 was skipped: Incompatible with archs other than armv7 andaarch64
Quoted 3 lines
Quoted 1 line
...
How does this handle Cortex A9 SoCs without NEON support?
this is a valid concern although rare but this combination is out there ( tegra2 )
and ne10 needs neon, so probably its better to check for neon in tune features
than anything else first.
Quoted 5 lines · Martin Jansa
I'm not using ne10 or libopus anywhere, it was just breaking
sstate-diff-machines.shscript by adding that unjustifiable dependency based onTUNE_FEATURES.
Quoted 7 lines
I was going to send v2 with the armv7ve entries as requested by Andreas (I forgot that armv7ve tunes no longer include armv7a in overrides), but if someone wants to rework it to test for neon in
TUNE_FEATURESfeel free to do so.It didn't check for neon before, so at least this change didn't make it worse.
Yeah - you fix something you are not interested in and people (either not using it) open up wishlists..
Thanks (and sorry :)
Andreas
Quoted 12 lines · Andreas Müller
Quoted 10 lines · Martin Jansa
I'm not using ne10 or libopus anywhere, it was just breaking
sstate-diff-machines.shscript by adding that unjustifiable dependency based onTUNE_FEATURES.I was going to send v2 with the armv7ve entries as requested by Andreas (I forgot that armv7ve tunes no longer include armv7a in overrides), but if someone wants to rework it to test for neon in
TUNE_FEATURESfeel free to do so.
Quoted 4 lines
Quoted 2 lines
It didn't check for neon before, so at least this change didn't make it worse.
Yeah - you fix something you are not interested in and people (either not using it) open up wishlists..
This got merged and backported to warrior without the _armv7ve over-ride, so Cortex-A15 etc targets are going to see some regression.
PATCH 1/1 ne10, libopus: add armv7ve override as well
* fix the override in packagegroup-meta-oe.bb
* alllow building ne10 for armv7ve (e.g. Cortex-A15)
--- meta-oe/recipes-core/packagegroups/packagegroup-meta-oe.bb | 3 ++- meta-oe/recipes-multimedia/libopus/libopus_1.1.3.bb | 3 ++- meta-oe/recipes-support/ne10/ne10_1.2.1.bb | 4 +++- 3 files changed, 7 insertions(+), 3 deletions(-)
meta-oe/recipes-core/packagegroups/packagegroup-meta-oe.bb +2 -1
@@ -220,7 +220,8 @@ RDEPENDS_packagegroup-meta-oe-shells ="\ NE10 = "" NE10_aarch64 = "ne10" -NE10_arm7 = "ne10" +NE10_armv7a = "ne10" +NE10_armv7ve = "ne10" RDEPENDS_packagegroup-meta-oe-support ="\ anthy asio atop augeas avro-c bdwgc frame grail \
meta-oe/recipes-multimedia/libopus/libopus_1.1.3.bb +2 -1
@@ -28,9 +28,10 @@ EXTRA_OECONF = " \ --enable-custom-modes \ " -# ne10 is available only for armv7a and aarch64 +# ne10 is available only for armv7a, armv7ve and aarch64 DEPENDS_append_aarch64 = " ne10" DEPENDS_append_armv7a = " ne10" +DEPENDS_append_armv7ve = " ne10" python () { if d.getVar('TARGET_FPU') in [ 'soft' ]:
meta-oe/recipes-support/ne10/ne10_1.2.1.bb +3 -1
@@ -15,13 +15,15 @@ PV .= "gitr+${SRCPV}" inherit cmake -# Incompatible with archs other than armv7 and aarch64 +# Incompatible with archs other than armv7, armv7ve and aarch64 COMPATIBLE_MACHINE = "(^$)" COMPATIBLE_MACHINE_aarch64 = "(.*)" COMPATIBLE_MACHINE_armv7a = "(.*)" +COMPATIBLE_MACHINE_armv7ve = "(.*)" NE10_TARGET_ARCH = "" NE10_TARGET_ARCH_aarch64 = "aarch64" NE10_TARGET_ARCH_armv7a = "armv7" +NE10_TARGET_ARCH_armv7ve = "armv7" EXTRA_OECMAKE = '-DGNULINUX_PLATFORM=ON -DNE10_BUILD_SHARED=ON -DNE10_LINUX_TARGET_ARCH="${NE10_TARGET_ARCH}"'
Signature · 1 line
2.17.1