Bug 236855
| Summary: | www/qt5-webengine: fails to build on aarch64 and armv7 after 5.12 update | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | Ports & Packages | Reporter: | Jan Beich <jbeich> | ||||||||||||||
| Component: | Individual Port(s) | Assignee: | freebsd-kde (group) <kde> | ||||||||||||||
| Status: | Closed FIXED | ||||||||||||||||
| Severity: | Affects Some People | CC: | kde, linimon, mikael, tcberner | ||||||||||||||
| Priority: | --- | Keywords: | patch, regression | ||||||||||||||
| Version: | Latest | Flags: | tcberner:
maintainer-feedback+
jbeich: merge-quarterly+ |
||||||||||||||
| Hardware: | arm64 | ||||||||||||||||
| OS: | Any | ||||||||||||||||
| URL: | http://thunderx1.nyi.freebsd.org/data/latest-per-pkg/qt5-webengine/ | ||||||||||||||||
| Bug Depends on: | 228603 | ||||||||||||||||
| Bug Blocks: | 238701 | ||||||||||||||||
| Attachments: |
|
||||||||||||||||
|
Description
Jan Beich
2019-03-28 16:38:49 UTC
13.0 aarch64:
In file included from ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/skia/src/core/SkOpts.cpp:43:
../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/skia/src/opts/SkRasterPipeline_opts.h:683:19: error: functional-style cast from 'neon::F' (aka 'V<float>') to '__fp16' is not allowed
__fp16 fp16 = __fp16(f);
^~~~~~~~
1 error generated.
13.0 armv7:
In file included from ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/protobuf/src/google/protobuf/compiler/javanano/javanano_extension.cc:33:
In file included from ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/protobuf/src/google/protobuf/compiler/javanano/javanano_extension.h:38:
../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/protobuf/src/google/protobuf/stubs/common.h:38:10: fatal error: 'algorithm' file not found
#include <algorithm>
^~~~~~~~~~~
1 error generated.
Hmmm, I don't have any sensible way to test armv7/aarch64 stuff... mfg Tobias (In reply to Tobias C. Berner from comment #2) Why not cross-compile? On -CURRENT amd64 you can build a very large set of release/architecture tuples. # NOTE: -x (native-xtools) for < 12.0 targets requires /usr/src to match requested version $ poudriere jail -cj 112aarch64 -x -a arm64.aarch64 -v 11.2-RELEASE $ poudriere jail -cj 120aarch64 -x -a arm64.aarch64 -v 12.0-RELEASE $ poudriere jail -cj head-aarch64 -x -a arm64.aarch64 -v 13.0-CURRENT $ poudriere jail -cj 120armv7 -x -a arm.armv7 -v releng/12.0 -m svn+https $ poudriere jail -cj head-armv7 -x -a arm.armv7 -v head -m svn+https More details: https://wiki.freebsd.org/QemuUserModeHowTo https://wiki.freebsd.org/Ports/BuildingPackagesThroughEmulation https://www.youtube.com/watch?v=2J9Lz3pgnbA https://www.youtube.com/watch?v=JfZIoyQhly4 Yes, I know about that. What I meant is, without wasting CPU cycles on my good machine for that :) (In reply to Tobias C. Berner from comment #4) On my i7-6700k building qt5-webengine 5.9.5 for aarch64 took ~9 minutes more than for amd64 (53 min vs. 44 min). Is that much? Or did you build jails without -x (native-xtools) which provides major speedup? If you don't want to I can test for you. OTOH, I'm not gonna waste time on amd64/i386 instead as that would be your job. ;) Patches are always welcome :D That being said, I will order an SSD and dig out my old computer ^^ mfg Tobias Created attachment 203288 [details] patch qemu-user hangs on my system, can you test the attached patch on aarch64 and revert https://github.com/qt/qtwebengine-chromium/commit/b2de8e8046dc1c558465f74b4afe11d57cbc6cf3 ? Created attachment 203289 [details] patch2 > revert https://github.com/qt/qtwebengine-chromium/commit/b2de8e8046dc1c558465f74b4afe11d57cbc6cf3 Doesn't revert cleanly. __ARM_FP16_FORMAT_IEEE shenanigans remind me of bug 231293. Comment on attachment 203289 [details]
patch2
Indeed. The old version also builds with Clang.
Comment on attachment 203288 [details] patch (In reply to mikael.urankar from comment #7) > can you test the attached patch on aarch64 and revert ...? 11.2/12.0/13.0 aarch64 fails in the same place: ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/zlib/arm_features.c:16:10: fatal error: 'asm/hwcap.h' file not found #include <asm/hwcap.h> ^~~~~~~~~~~~~ 1 error generated. The fp16 thing on ARM reminds me of something I had to fix at work a while ago (I had to make the GCC case work though). My fix was https://github.com/OSSystems/meta-browser/blob/master/recipes-browser/chromium/files/aarch64-skia-build-fix.patch, and I assume the webengine guys needed a GCC fix as well. It's weird that their change breaks our clang build because it should also break their clang builds (unless they're not testing clang+arm64). In any case, it'd be good to file a bug in Qt's bug tracker about this. The arm_features.c error could be solved by adjusting patch-src_3rdparty_chromium_third__party_zlib_arm__features.c and only including those headers when we're not building on BSD (the patch came from www/chromium, but I'm not sure if the chromium port enables OS_LINUX in builds like we do for webengine). Created attachment 203292 [details]
patch
and with this patch for arm_features.c? Not tested on my side, it takes ages to build on my jetson.
Created attachment 203301 [details]
patch
this one compiles
Comment on attachment 203301 [details] patch Builds fine now. https://people.freebsd.org/~jbeich/logs/112aarch64/qt5-webengine-5.12.1_1.log https://people.freebsd.org/~jbeich/logs/120aarch64/qt5-webengine-5.12.1_1.log https://people.freebsd.org/~jbeich/logs/head-aarch64/qt5-webengine-5.12.1_1.log AFAIU we still don't have one single patch with all fixes, right? If that's indeed the case can someone take care of doing that? Created attachment 203348 [details] aarch64 fixes The version tested in comment 14. I've updated build logs since ports r497572. Comment on attachment 203348 [details] aarch64 fixes >+--- src/3rdparty/chromium/third_party/skia/src/opts/SkRasterPipeline_opts.h.orig 2019-01-16 10:59:47 UTC >++++ src/3rdparty/chromium/third_party/skia/src/opts/SkRasterPipeline_opts.h >+@@ -658,9 +658,7 @@ SI F approx_powf(F x, F y) { >+ >+ SI F from_half(U16 h) { >+ #if defined(__ARM_FP16_FORMAT_IEEE) >+- __fp16 fp16; >+- memcpy(&fp16, &h, sizeof(U16)); >+- return float(fp16); >++ return vcvt_f32_f16(h); >+ >+ #elif defined(JUMPER_IS_HSW) || defined(JUMPER_IS_AVX512) >+ return _mm256_cvtph_ps(h); >+@@ -680,10 +678,7 @@ SI F from_half(U16 h) { >+ >+ SI U16 to_half(F f) { >+ #if defined(__ARM_FP16_FORMAT_IEEE) >+- __fp16 fp16 = __fp16(f); >+- U16 u16; >+- memcpy(&u16, &fp16, sizeof(U16)); >+- return u16; >++ return vcvt_f16_f32(f); armv7 didn't like this change: In file included from ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/skia/src/core/SkOpts.cpp:43: ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/skia/src/opts/SkRasterPipeline_opts.h:661:12: error: use of undeclared identifier 'vcvt_f32_f16' return vcvt_f32_f16(h); ^ ../../../../qtwebengine-everywhere-src-5.12.1/src/3rdparty/chromium/third_party/skia/src/opts/SkRasterPipeline_opts.h:681:12: error: use of undeclared identifier 'vcvt_f16_f32' return vcvt_f16_f32(f); ^ 2 errors generated. Created attachment 203356 [details] aarch64 + armv7 fixes 12.0 armv7 builds fine but 13.0 armv7 fails in what maybe a qemu-user-static issue. https://people.freebsd.org/~jbeich/logs/120armv7/qt5-webengine-5.12.1_1.log https://people.freebsd.org/~jbeich/logs/head-armv7/qt5-webengine-5.12.1_1.log Comment on attachment 203356 [details] aarch64 + armv7 fixes armv6 also builds fine, an improvement compared to qt5-webengine-5.9.* series. ;) https://people.freebsd.org/~jbeich/logs/112armv6/qt5-webengine-5.12.1_1.log https://people.freebsd.org/~jbeich/logs/120armv6/qt5-webengine-5.12.1_1.log https://people.freebsd.org/~jbeich/logs/head-armv6/qt5-webengine-5.12.1_1.log Comment on attachment 203356 [details]
aarch64 + armv7 fixes
ship it
A commit references this bug: Author: jbeich Date: Thu Apr 4 23:29:37 UTC 2019 New revision: 497877 URL: https://svnweb.freebsd.org/changeset/ports/497877 Log: www/qt5-webengine: unbreak on aarch64 and armv7 after r496989 PR: 236855 Reported by: pkg-fallout Submitted by: Mika?l Urankar Approved by: kde (tcberner) Changes: head/www/qt5-webengine/files/patch-src_3rdparty_chromium_build_config_compiler_BUILD.gn head/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_crc32c_src_src_crc32c__arm64__linux__check.h head/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_skia_src_core_SkCpu.cpp head/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_skia_src_opts_SkRasterPipeline__opts.h head/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_zlib_arm__features.c head/www/qt5-webengine/files/patch-src_3rdparty_chromium_v8_src_arm_cpu-arm.cc head/www/qt5-webengine/files/patch-src_3rdparty_chromium_v8_src_base_platform_platform-freebsd.cc head/www/qt5-webengine/files/patch-src_3rdparty_chromium_v8_src_libsampler_sampler.cc Thanks for the work here. Do you plan on MFHing this change? A commit references this bug: Author: jbeich Date: Fri Apr 5 19:24:01 UTC 2019 New revision: 497980 URL: https://svnweb.freebsd.org/changeset/ports/497980 Log: MFH: r497877 www/qt5-webengine: unbreak on aarch64 and armv7 after r496989 PR: 236855 Reported by: pkg-fallout Submitted by: Mika?l Urankar Approved by: kde (tcberner) Approved by: ports-secteam blanket Changes: _U branches/2019Q2/ branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_build_config_compiler_BUILD.gn branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_crc32c_src_src_crc32c__arm64__linux__check.h branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_skia_src_core_SkCpu.cpp branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_skia_src_opts_SkRasterPipeline__opts.h branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_third__party_zlib_arm__features.c branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_v8_src_arm_cpu-arm.cc branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_v8_src_base_platform_platform-freebsd.cc branches/2019Q2/www/qt5-webengine/files/patch-src_3rdparty_chromium_v8_src_libsampler_sampler.cc (In reply to Raphael Kubo da Costa from comment #22) I planned to and then forgot. Thanks for the reminder. After ports r497990 still builds fine: https://people.freebsd.org/~jbeich/logs/112aarch64/qt5-webengine-5.12.2.log https://people.freebsd.org/~jbeich/logs/120aarch64/qt5-webengine-5.12.2.log https://people.freebsd.org/~jbeich/logs/head-aarch64/qt5-webengine-5.12.2.log https://people.freebsd.org/~jbeich/logs/112armv6/qt5-webengine-5.12.2.log https://people.freebsd.org/~jbeich/logs/120armv6/qt5-webengine-5.12.2.log https://people.freebsd.org/~jbeich/logs/head-armv6/qt5-webengine-5.12.2.log (false negative) https://people.freebsd.org/~jbeich/logs/120armv7/qt5-webengine-5.12.2.log https://people.freebsd.org/~jbeich/logs/head-armv7/qt5-webengine-5.12.2.log (false negative) |