- User Since
- Jun 17 2015, 7:07 AM (314 w, 1 d)
The breaking buildbots:
Mon, Jun 21
Some bikeshedding: Calls to AvoidConcat could be avoided by always inserting a space between tokens at the cost of making the output larger. In the current form, the flag could also be named -fminimize-whitespace.
If the sizes of the output changes a mismatch with the old reference output is expected (the removed border contains undefined values anyway). In that case, just replace the reference-output.out with D103092-output.txt.
Sat, Jun 19
- Undo "typechange" changes
Thu, Jun 17
Could you please address the build failure mentioned in https://reviews.llvm.org/rG57e85622bbdb2eb18cc03df2ea457019c58f6912#inline-6002:
This patch seem to have broken the Polly buildbot:
Wed, Jun 16
Could you remove the backticks from the title before committing? It is not interpreted as Markdown and its shortness is comparatively important (Many guidelines recommend at most 50 characters)
Mon, Jun 14
LGTM with one nitpick.
Fri, Jun 11
Could you delete unused driver_parse and driver_run?
Does this expect on loop unrolling to happened? Did you consider adding #pragma clang loop unroll(full) to make sure of it?
Thu, Jun 10
Obsolete since D94973 was committed.
- Merge branch 'arcpatch-D99459' into pragma-omp-unroll
This was the last patch to make the windows buildbot green: http://meinersbur.de:8011/#/builders/146/builds/593
- Remove allow_test_fail
LGTM, thank you
Wed, Jun 9
The patch in the current form does both at the same time:
Tue, Jun 8
- Integreate isConstantExpression into VerifyPositiveIntegerConstantInClause
OK, disregard that, I was looking at an old language reference (the first found by Google).
Could you also update the language reference? It currently reads:
The metadata on memory reads also implies that if conversion (i.e. speculative execution within a loop iteration) is safe.
Mon, Jun 7
Sun, Jun 6
Could you mention in the commit message/summary that the goal/motivation is to make the isl-noexceptions.h converge towards the official bindings?
Fri, Jun 4
Thu, Jun 3
- Address review
Wed, Jun 2
Instead of reverting, I fixed the test in rG64e5a3bbdde2.
A lot has changed since I put up this patch and I will need some time rebasing.
Tue, Jun 1
Besides the switch from os.path to pathlib, is the change to first build the path, then to resolve it (instead of first using abspath to resolve it, then concat the path) intentional?
Ideally, how a test is executed is entirely specified in one file (the CMakeLists.txt), instead split between two files. In the standard case, a lit.local.cfg should not be required.
Sun, May 30
I suggest to put each test into its own directly, i.e. the hello test into Fortran/UnitTests/hello. This would allow selecting each test separately using TEST_SUITE_SUBDIRS (instead of having to select all Fortran UnitTests/Benchmarks at the same time) and each test can itself decide whether it is using llvm_singlesource() or llvm_multisource().
Fri, May 28
Is there interest in this? Windows buildbot is still failing (http://meinersbur.de:8011/#/builders/146/builds/169).
- Remove clean=False which is the default anyway.
Thu, May 27
Thank you @hubert.reinterpretcast for the analysis.
The test is sufficient for me. Thank you for the fix.
A concern that I have is that the bash script will only on platforms that have bash. In addition to Windows, some distribution use different default shells and may not have it installed (MacOS uses zsh and ships only an old version of bash).
However, as long as this is transitional, and to be removed again once the flang toolchain is fully functional, I would be ok with it.
Wed, May 26
May 25 2021
If I recall correctly, I tried this already. Unfortunately I encountered a design problem.
image is meant to be a pointer to a (HEIGHT+2)*(WIDTH+2) image, where each "pixel" is a represented as a grayscale float (4 bytes) value. sizeof(float*) * (2 + HEIGHT) * (2 + WIDTH) allocates a pointer-size for each pixel (8 bytes on a 64-bit system), i.e. typically twice as much. However, the data is not supposed to be a 2-dimensional array of pointers.
This only changes code in ScalarEvolution, so is this a SCEV fix or was Polly using ScalarEvolution in a way that was not originally intended?
LGTM, thank you.
May 24 2021
- Address review
- Merge branch 'arcpatch-D102180' into HEAD
- Use SmallVector<T,0>