From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from foss.arm.com (foss.arm.com [217.140.110.172]) by sourceware.org (Postfix) with ESMTP id 575B33858D39 for ; Mon, 20 Nov 2023 15:58:44 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 575B33858D39 Authentication-Results: sourceware.org; dmarc=none (p=none dis=none) header.from=foss.arm.com Authentication-Results: sourceware.org; spf=none smtp.mailfrom=foss.arm.com ARC-Filter: OpenARC Filter v1.0.0 sourceware.org 575B33858D39 Authentication-Results: server2.sourceware.org; arc=none smtp.remote-ip=217.140.110.172 ARC-Seal: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1700495926; cv=none; b=d3bat4y1puEZsJNpBMCxUo8vbcc7A9JwbxxjhTmidIgqVKBE4kBDT/eGxUUP+9mFGfsXDVFg6AipuMZbiI7MFjtIEdELzmunyJ7YM08vvZO3nm9s/NhsJVKJJxPwYRM0V6vRRfW2KVGC5ErRffpdLwtpKNwMOM8mgnHNxc6N274= ARC-Message-Signature: i=1; a=rsa-sha256; d=sourceware.org; s=key; t=1700495926; c=relaxed/simple; bh=CuxlAD3X971Imz0YcD/d5CKQqvBl2fVFvfFsIyKr1I8=; h=Message-ID:Date:MIME-Version:Subject:To:From; b=KuB63zETCWYiIs0j0Wp6MlLH4o9ptMqgwTbewawhsnoNRdVJviiMCNLTffTeKefpJdaI6ZD0mxy1VJlgeJhlLdlxigDWb06jV297x1oaSI4uAAz2q3acQD46QYdREterWk6JAaBFi4RW8bXZwUkZYpySZKZk+YuGdS3qfv1VOhQ= ARC-Authentication-Results: i=1; server2.sourceware.org Received: from usa-sjc-imap-foss1.foss.arm.com (unknown [10.121.207.14]) by usa-sjc-mx-foss1.foss.arm.com (Postfix) with ESMTP id 3BA06FEC; Mon, 20 Nov 2023 07:59:30 -0800 (PST) Received: from [10.57.38.109] (unknown [10.57.38.109]) by usa-sjc-imap-foss1.foss.arm.com (Postfix) with ESMTPSA id 3617C3F6C4; Mon, 20 Nov 2023 07:58:43 -0800 (PST) Message-ID: <0badcfd7-5c21-43e7-8143-ab402bb37c7d@foss.arm.com> Date: Mon, 20 Nov 2023 15:58:41 +0000 MIME-Version: 1.0 User-Agent: Mozilla Thunderbird Subject: Re: [committed 03/22] arm: testsuite: avoid hard-float ABI incompatibility with -march Content-Language: en-GB To: Christophe Lyon Cc: Richard Earnshaw , "gcc-patches@gcc.gnu.org" References: <20231113142658.69039-1-rearnsha@arm.com> <20231113142658.69039-4-rearnsha@arm.com> <724d822d-7bf1-4284-8688-cc834fd90817@foss.arm.com> <5f0f6e18-7c6a-4873-b9d9-722a431d4a3b@foss.arm.com> From: Richard Earnshaw In-Reply-To: Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit X-Spam-Status: No, score=-3489.1 required=5.0 tests=BAYES_00,KAM_DMARC_STATUS,KAM_LAZY_DOMAIN_SECURITY,KAM_SHORT,SPF_HELO_NONE,SPF_NONE,TXREP,T_SCC_BODY_TEXT_LINE autolearn=no autolearn_force=no version=3.4.6 X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on server2.sourceware.org List-Id: On 20/11/2023 14:50, Christophe Lyon wrote: > On Mon, 20 Nov 2023 at 15:39, Richard Earnshaw > wrote: >> >> >> >> On 20/11/2023 14:24, Christophe Lyon wrote: >>> On Mon, 20 Nov 2023 at 14:58, Richard Earnshaw >>> wrote: >>>> >>>> >>>> >>>> On 20/11/2023 13:36, Christophe Lyon wrote: >>>>> On Mon, 20 Nov 2023 at 13:44, Richard Earnshaw >>>>> wrote: >>>>>> >>>>>> >>>>>> >>>>>> On 20/11/2023 10:23, Christophe Lyon wrote: >>>>>>> Hi Richard, >>>>>>> >>>>>>> On Mon, 13 Nov 2023 at 15:28, Richard Earnshaw wrote: >>>>>>>> >>>>>>>> >>>>>>>> A number of tests in the gcc testsuite, especially for arm-specific >>>>>>>> targets, add various flags to control the architecture. These run >>>>>>>> into problems when the compiler is configured with -mfpu=auto if the >>>>>>>> new architecture lacks an architectural feature that implies we have >>>>>>>> floating-point instructions. >>>>>>>> >>>>>>>> The testsuite makes this worse as it falls foul of this requirement in >>>>>>>> the base architecture strings provided by target-supports.exp. >>>>>>>> >>>>>>>> To fix this we add "+fp", or something equivalent to this, to all the >>>>>>>> base architecture specifications. The feature will be ignored if the >>>>>>>> float ABI is set to soft. >>>>>>>> >>>>>>>> gcc/testsuite: >>>>>>>> >>>>>>>> * lib/target-supports.exp (check_effective_target_arm_arch_FUNC_ok): >>>>>>>> Add base FPU specifications to all architectures that can support >>>>>>>> one. >>>>>>>> --- >>>>>>>> gcc/testsuite/lib/target-supports.exp | 50 +++++++++++++-------------- >>>>>>>> 1 file changed, 25 insertions(+), 25 deletions(-) >>>>>>>> >>>>>>> >>>>>>> Our CI has detected some regressions with this patch, in particular >>>>>>> when testing for cortex-m55: >>>>>>> >>>>>>> with >>>>>>> -mthumb/-march=armv8.1-m.main+mve.fp+fp.dp/-mtune=cortex-m55/-mfloat-abi=hard/-mfpu=auto >>>>>>> and GCC configured with --disable-multilib --with-mode=thumb >>>>>>> --with-arch=armv8.1-m.main+mve.fp+fp.dp --with-float=hard >>>>>>> >>>>>>> you can see our logs here: >>>>>>> https://ci.linaro.org/job/tcwg_gnu_embed_check_gcc--master-thumb_m55_hard_eabi-build/209/artifact/artifacts/00-sumfiles/ >>>>>>> >>>>>>> Thanks, >>>>>>> >>>>>>> Christophe >>>>>> >>>>>> What exactly am I supposed to be looking at here? I see no description >>>>>> of what those logs represent. If they are supposed to be before and >>>>>> after, then why does the after only run a tiny fraction of the testsuite >>>>>> (Running gcc.git~master/gcc/testsuite/gcc.target/arm/arm.exp ... >>>>>> Running gcc.git~master/gcc/testsuite/gcc.target/arm/cmse/cmse.exp ... >>>>>> Running gcc.git~master/gcc/testsuite/gcc.target/arm/lto/lto.exp ...) >>>>>> >>>>>> The logs give no clue as to why the reminder of the testsuite wasn't run. >>>>>> >>>>>> Please don't make me guess. >>>>>> >>>>> >>>>> Here is a summary with the list of regressions: >>>>> https://ci.linaro.org/job/tcwg_gnu_embed_check_gcc--master-thumb_m55_hard_eabi-build/209/artifact/artifacts/notify/regressions.sum/*view*/ >>>> >>>> OK, that's much more useful. But how was I supposed to know that link >>>> existed? >>>> >>> The full notification email contains a lot of information, with >>> several pointers to our Jenkins artifacts. >>> The notification email is not yet automatically sent to contributors >>> because we are still polishing, and I thought I'd save you some time >>> by just sending the useful links. >>> >>> Looks like it's time to send those automatically too. >>> >>>>> >>>>> I thought you'd be able to find your way in the logs above, the .0 >>>>> files contain the logs of the initial full testsuite run, and .1 ones >>>>> contain the logs of the second run of the testsuite, restricted to the >>>>> .exp files where we detected regressions. So looking at gcc.log.1.xz >>>>> will give you details of the regressions shown in the link above. >>>> >>>> There's nothing in the page you sent me to that gives any clue as to how >>>> to read the logs there. So my assumption was that the .0 was a before >>>> run and .1 an after. Please, if you're going to direct people to the >>>> log files, provide some way for them to understand what the log files show. >>>> >>>> Now, to the specific issues: >>>> >>>> Running gcc:gcc.target/arm/arm.exp ... >>>> FAIL: gcc.target/arm/attr_thumb-static2.c (test for excess errors) >>>> UNRESOLVED: gcc.target/arm/attr_thumb-static2.c scan-assembler-times blx 2 >>>> >>>> This was fixed with "arm: testsuite: avoid problems with -mfpu=auto in >>>> attr_thumb-static2.c", which is a later patch in the series (patch 6). >>>> >>>> I don't think it's useful to try to regression test each individual >>>> patch, it wasn't practical to try to get every patch into order in the >>>> series (it would have made for a lot of churn on some files, especially >>>> target-supports.exp), so only a fully before and a fully after run is >>>> useful. If there are issues once the whole series has been applied, >>>> then that is much more interesting. >>>> >>> >>> I looked at this in more detail. >>> That specific bisection build was triggered because we detected >>> regressions, after the full series was committed. >>> What happens is that at the first bad commit (this one) there were >>> more regressions than after the full series was committed. >>> >>> So, the extract of >>> https://ci.linaro.org/job/tcwg_gnu_embed_check_gcc--master-thumb_m55_hard_eabi-build/209/artifact/artifacts/notify/regressions.sum/*view*/ >>> which remains valid on current trunk is: >>> Running gcc:gcc.target/arm/cmse/cmse.exp ... >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O0 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O1 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O2 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O3 -g scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -Os scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O0 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O1 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O2 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O3 -g scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/hard/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -Os scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/soft/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -mfloat-abi=soft -O0 scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/soft/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -mfloat-abi=soft -O1 scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/soft/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -mfloat-abi=soft -O2 scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/soft/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -mfloat-abi=soft -O3 -g >>> scan-assembler msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/soft/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -mfloat-abi=soft -Os scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O0 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O1 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O2 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O3 -g scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp-sp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -Os scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O0 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O1 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O2 scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -O3 -g scan-assembler >>> msr\tAPSR_nzcvqg, lr >>> FAIL: gcc.target/arm/cmse/mainline/8m/softfp/cmse-5.c >>> -march=armv8-m.main+fp -mthumb -Os scan-assembler msr\tAPSR_nzcvqg, >>> lr >>> >>>> R. >> >> The compiled output contains (at least for the case I tried -mthumb >> -march=armv8.1-m.main+mve.fp+fp.dp -mtune=cortex-m55 -mfloat-abi=hard >> -mfpu=auto -fdiagnostics-plain-output -march=armv8-m.base -mthumb >> -mfloat-abi=soft -O1 -mcmse -ffat-lto-objects -fno-ident -S -o cmse-2.s): >> >> msr APSR_nzcvq, r1 >> >> So this will never match the expected pattern, which is looking for 'lr' >> not 'r1'. Are you sure these tests were running before? >> > > No, your patch enabled them. > > I already noticed/reported a long time ago that cmse.exp was not > executed by all the configurations we currently run and I was > surprised to see errors in my manual runs with more specific > configurations. > > Good to see it enabled now. [TLDR] I was initially confused about why this test had suddenly started to run, but it turns out it's because cmse.exp has the gating operation, not the test itself. The problem in this case is because the specific subtest is checking for the DSP extension. But we have a command line of -mthumb -march=armv8.1-m.main+mve.fp+fp.dp -mtune=cortex-m55 \ -mfloat-abi=hard -mfpu=auto -fdiagnostics-plain-output \ -march=armv8-m.main+fp -mthumb -mfloat-abi=soft -O1 -mcmse \ -mfloat-abi=soft When compiling the test, but a command line of -mthumb -march=armv8.1-m.main+mve.fp+fp.dp -mtune=cortex-m55 \ -mfloat-abi=hard -mfpu=auto when checking for the DSP extension. The former is built up from -mthumb -march=armv8.1-m.main+mve.fp+fp.dp -mtune=cortex-m55 \ -mfloat-abi=hard -mfpu=auto for the testsuite run -fdiagnostics-plain-output from the general framework -march=armv8-m.main+fp -mthumb from the cmse.exp script [add_options_for...] -mfloat-abi=soft from the framework - variant check and finally -O1 -mcmse -mfloat-abi=soft from the dg-options in the test. So we're going to get a mismatch in features here which is not trivial to work around. The problem in this particular case stems from the fact that the DSP extension is mandatory if MVE is present, but optional otherwise and the option combinations imply it via +mve.fp. We then override the architecture with a non-MVE variant and end up not having DSP. Ultimately, therefore, we compile the test without DSP, but then check for DSP with a set of architecture flags that has DSP. Given the way this is all put together, I think you'd see the same problem if you ran the test with -mthumb -march=armv8.1-m.main+fp+dsp -mfloat-abi=softfp I don't think there's a trivial way to work around the inconsistent options problem without weakening the test overall (by just testing for "APSR_nzcvqg?"). So perhaps the best solution is to do that, but then add some additional tests that are outside this base CMSE behaviour that test the APSR register more explicitly. I'll give this some thought. R.