Excellent, thanks Even - confirmed at my end and on my original use-case. Cheers, Mike
On Mon, Feb 5, 2024 at 12:04 PM Even Rouault <even.roua...@spatialys.com> wrote: > ghcr.io/osgeo/gdal:ubuntu-full-latest has been regenerated with the > rebuild of TileDB without AVX2. I've also enabled the > drivers-with-external-depencies-built-as-plugin GDAL build mode, so it is > easy to just remove a given plugin by deleting the corresponding .so in > /usr/lib/x86_64-linux-gnu/gdalplugins > > Even > Le 04/02/2024 à 22:51, Michael Sumner a écrit : > > indeed there's no avx2: > > cat /proc/cpuinfo|grep sse|head -n 1 > flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca > cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt pdpe1gb > rdtscp lm rep_good nopl cpuid extd_apicid tsc_known_freq pni pclmulqdq > ssse3 fma cx16 sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx > f16c hypervisor lahf_lm cmp_legacy svm cr8_legacy abm sse4a misalignsse > 3dnowprefetch osvw xop fma4 tbm perfctr_core ssbd ibpb vmmcall tsc_adjust > bmi1 virt_ssbd arat npt nrip_save arch_capabilities > > Cheers, Mike > > > > On Sun, Feb 4, 2024 at 10:55 PM Even Rouault <even.roua...@spatialys.com> > wrote: > >> ok, so I believe this is the AVX2 issue I was talking about, as I realize >> that enabling AVX2 is the default mode when TileDB is built from source >> (which the Docker image does), and must be explicitly disabled with >> "./bootstrap --disable-avx2" (I've just changed the build recipe to include >> that, will take effect next time the images are refreshed) >> >> To confirm, can you send or just check the output of : cat >> /proc/cpuinfo|grep sse|head -n 1 >> >> If there is no "avx2" in it, this is at 99.9% the reason of the issue. >> >> Even >> Le 04/02/2024 à 06:20, Michael Sumner a écrit : >> >> skipping TileDB does fix: >> >> ogr2ogr /tmp/newdir >> https://github.com/SymbolixAU/geojsonsf/raw/master/inst/examples/geo_melbourne.geojson >> -f >> "ESRI Shapefile" >> export GDAL_SKIP=TileDB >> ogrinfo /tmp/newdir/ >> INFO: Open of `/tmp/newdir/' >> using driver `ESRI Shapefile' successful. >> 1: geo_melbourne (Polygon) >> >> unset GDAL_SKIP >> ogrinfo /tmp/newdir/ >> Illegal instruction (core dumped) >> >> I failed to explain that I'm using gdal containers from the repo: >> >> docker run --rm -ti ghcr.io/osgeo/gdal:ubunt >> >> u-full-latest >> >> apt update >> apt install -y gdb >> >> Here's the output of under gdb as you suggested, there was a lot so I put >> it on a gist: >> https://gist.github.com/mdsumner/839ae6e05ededf640b65bfee3a20a4c0 >> >> gdb --args ogrinfo /tmp/newdir/ >> > run >> > thread apply all bt >> >> Thanks! >> >> >> >> >> >> On Sat, Feb 3, 2024 at 7:49 PM Even Rouault <even.roua...@spatialys.com> >> wrote: >> >>> - When it crashes under gdb, type "thread apply all bt" to get the stack >>> trace of all threads >>> >>> - I suspect there is a connection with >>> https://github.com/OSGeo/gdal/pull/9170 , but that pull request >>> wouldn't help here as "/tmp/newdir" could be a valid connection to TileDB >>> >>> - how did you get TileDB installed? It looks to be packaged? Which >>> distribution do you use? >>> >>> - SIGILL reminds me of issues with some TileDB builds using the AVX2 >>> instruction set by default, which could cause some crash on host CPUs that >>> don't have AVX2 (unlikely on recent hardware though) >>> >>> - Setting GDAL_SKIP=TileDB should be a workaround >>> >>> >>> Le 03/02/2024 à 07:15, Michael Sumner a écrit : >>> >>> Thanks Even, so there's something about tiledb under gdb (or maybe I am >>> mangling the context, I will try variants of the host I'm using). Run with >>> valgrind included below. >>> >>> gdb --args ogrinfo /tmp/newdir/ >>> ... >>> (gdb) run >>> Starting program: /usr/local/bin/ogrinfo /tmp/newdir/ >>> [Thread debugging using libthread_db enabled] >>> Using host libthread_db library >>> "/lib/x86_64-linux-gnu/libthread_db.so.1". >>> [New Thread 0x7fffe7757640 (LWP 988)] >>> [New Thread 0x7fffe6f56640 (LWP 989)] >>> [New Thread 0x7fffde755640 (LWP 990)] >>> [New Thread 0x7fffd5f54640 (LWP 991)] >>> [New Thread 0x7fffc5753640 (LWP 992)] >>> [New Thread 0x7fffc4f52640 (LWP 993)] >>> [New Thread 0x7fffb4751640 (LWP 994)] >>> [New Thread 0x7fffabf50640 (LWP 995)] >>> [New Thread 0x7fffab74f640 (LWP 996)] >>> [New Thread 0x7fffa2f4e640 (LWP 997)] >>> [New Thread 0x7fff9a74d640 (LWP 998)] >>> [New Thread 0x7fff91f4c640 (LWP 999)] >>> [New Thread 0x7fff8974b640 (LWP 1000)] >>> [New Thread 0x7fff78f4a640 (LWP 1001)] >>> [New Thread 0x7fff78749640 (LWP 1002)] >>> [New Thread 0x7fff6f5ff640 (LWP 1003)] >>> >>> Thread 1 "ogrinfo" received signal SIGILL, Illegal instruction. >>> 0x00007ffff3773c9e in tiledb::common::ThreadPool::ThreadPool(unsigned >>> long) () from /lib/x86_64-linux-gnu/libtiledb.so.2.16 >>> >>> >>> >>> >>> valgrind -s ogrinfo /tmp/newdir >>> ==704== Memcheck, a memory error detector >>> ==704== Copyright (C) 2002-2017, and GNU GPL'd, by Julian Seward et al. >>> ==704== Using Valgrind-3.18.1 and LibVEX; rerun with -h for copyright >>> info >>> ==704== Command: ogrinfo /tmp/newdir >>> ==704== >>> INFO: Open of `/tmp/newdir' >>> using driver `ESRI Shapefile' successful. >>> 1: geo_melbourne (Polygon) >>> ==704== >>> ==704== HEAP SUMMARY: >>> ==704== in use at exit: 25,486 bytes in 216 blocks >>> ==704== total heap usage: 15,761 allocs, 15,545 frees, 2,390,169 bytes >>> allocated >>> ==704== >>> ==704== LEAK SUMMARY: >>> ==704== definitely lost: 0 bytes in 0 blocks >>> ==704== indirectly lost: 0 bytes in 0 blocks >>> ==704== possibly lost: 544 bytes in 1 blocks >>> ==704== still reachable: 24,942 bytes in 215 blocks >>> ==704== suppressed: 0 bytes in 0 blocks >>> ==704== Rerun with --leak-check=full to see details of leaked memory >>> ==704== >>> ==704== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0) >>> >>> >>> ogrinfo /tmp/newdir >>> Illegal instruction (core dumped) >>> >>> Cheers, Mike >>> >>> >>> >>> >>> On Sat, Feb 3, 2024 at 12:46 PM Even Rouault <even.roua...@spatialys.com> >>> wrote: >>> >>>> Michael, >>>> >>>> I'm wondering if there not might be something wrong with your build or >>>> runtime environment. Or there's something subtle, because that works fine >>>> for me with my dev build or in the >>>> ghcr.io/osgeo/gdal:alpine-normal-3.8.3 Docker image >>>> >>>> Try running "valgrind ogrinfo /tmp/newdir/" or "gdb --args ogrinfo >>>> /tmp/newdir/" (type "run") to get more useful information >>>> >>>> Even >>>> Le 03/02/2024 à 02:35, Michael Sumner via gdal-dev a écrit : >>>> >>>> I'm getting Illegal instruction / core dumped on ogrinfo of a >>>> directory: >>>> >>>> ogr2ogr /tmp/newdir >>>> https://github.com/SymbolixAU/geojsonsf/raw/master/inst/examples/geo_melbourne.geojson >>>> -f "ESRI Shapefile" >>>> >>>> ogrinfo /tmp/newdir/ >>>> Illegal instruction (core dumped) >>>> >>>> I've worked back through some docker images and it wasn't a problem in >>>> 3.6.0, but I'm getting it since 3.7.0 - or I'm doing something wrong >>>> entirely. >>>> >>>> Cheers, Mike >>>> >>>> >>>> -- >>>> Michael Sumner >>>> Software and Database Engineer >>>> Australian Antarctic Division >>>> Hobart, Australia >>>> e-mail: mdsum...@gmail.com >>>> >>>> _______________________________________________ >>>> gdal-dev mailing >>>> listgdal-dev@lists.osgeo.orghttps://lists.osgeo.org/mailman/listinfo/gdal-dev >>>> >>>> -- http://www.spatialys.com >>>> My software is free, but my time generally not. >>>> >>>> >>> >>> -- >>> Michael Sumner >>> Software and Database Engineer >>> Australian Antarctic Division >>> Hobart, Australia >>> e-mail: mdsum...@gmail.com >>> >>> -- http://www.spatialys.com >>> My software is free, but my time generally not. >>> >>> >> >> -- >> Michael Sumner >> Software and Database Engineer >> Australian Antarctic Division >> Hobart, Australia >> e-mail: mdsum...@gmail.com >> >> -- http://www.spatialys.com >> My software is free, but my time generally not. >> >> > > -- > Michael Sumner > Software and Database Engineer > Australian Antarctic Division > Hobart, Australia > e-mail: mdsum...@gmail.com > > -- http://www.spatialys.com > My software is free, but my time generally not. > > -- Michael Sumner Software and Database Engineer Australian Antarctic Division Hobart, Australia e-mail: mdsum...@gmail.com
_______________________________________________ gdal-dev mailing list gdal-dev@lists.osgeo.org https://lists.osgeo.org/mailman/listinfo/gdal-dev
