Feed All Stories

Today

yamaguchi updated the diff for D35759: [Bash-autocompletion] Show HelpText with possible flags.

Add newline after the end of the line.

Sat, Jul 22, 2:37 AM
krytarowski added a comment to D35558: Use sys::Memory::AllocateRWX for JIT code.

I have no opinion on further design, if this patch is fine as of now - an intermediate solution for 5.0.0(svn) - great.

Sat, Jul 22, 2:27 AM
yamaguchi added inline comments to D35759: [Bash-autocompletion] Show HelpText with possible flags.
Sat, Jul 22, 2:18 AM
yamaguchi updated the diff for D35759: [Bash-autocompletion] Show HelpText with possible flags.

Fixed test and update diff according to Rui's comment.

Sat, Jul 22, 2:16 AM
fedor.sergeev updated subscribers of D35755: [Solaris] gcc toolchain handling revamp.
Sat, Jul 22, 1:29 AM
yan12125 updated the diff for D35137: [Driver] Honor -isysroot for Linux targets.

Add a test and fix repeated call to defaultSysRoot()

Sat, Jul 22, 12:39 AM
gareevroman created D35761: [Polly][WIP] Use SCEV information for the second level aliasing.
Sat, Jul 22, 12:25 AM
v.g.vassilev added a comment to D34444: Teach codegen to work in incremental processing mode..

@rjmccall, thanks for the prompt and thorough reply.

Okay. In that case, I see two problems, one major and one potentially major.

This is a very accurate diagnosis which took us 5 years to discover on an empirical basis ;)

You could've asked at any time. :)

True. I am not really sure I knew what to ask, though ;)

We're open to general "I'm trying to do this and having problems" questions on the mailing lists. You probably would've needed to know to CC me specifically, though; sadly, I can't keep up with all the lists I need to.

Good to know. Thanks! I will come back to you once I get rid of our O(100) clang patches to discuss what would be the best way of supporting incremental compilation.

That's quite brittle, because that code is only executed in a code path that only you are using, and you're not adding any tests. I would greatly prefer a change to IRGen's core assumptions, as suggested.

I am open to changing this code as well. That should probably be another review.

I agree. Are you comfortable with blocking this review until that lands? It seems like it would significantly change this.

Sat, Jul 22, 12:16 AM
davidxl added a comment to D35750: [x86] Teach the x86 backend about general fast rep+movs and rep+stos features of modern x86 CPUs, and use this feature to drastically reduce the number of places we actually emit memset and memcpy library calls..

Do you have more benchmark numbers? For reference, here is GCC does (for sandybridge and above) for mempcy when size profile data is available:

  1. when the size is <= 24, use 8 byte copy loop or straightline code.
  2. when size is is between 24 and 128, use rep movsq
  3. when size is b above that, use libcall
Sat, Jul 22, 12:04 AM

Yesterday

grosser added a comment to D35754: [ForwardOpTree] Introduce the -polly-optree pass..

One more typo.

Fri, Jul 21, 11:04 PM · Restricted Project
enlight added reviewers for D35760: Expose active and available platform lists via SBDebugger API: jingham, clayborg.
Fri, Jul 21, 10:56 PM · Restricted Project
grosser accepted D35754: [ForwardOpTree] Introduce the -polly-optree pass..
Fri, Jul 21, 10:43 PM · Restricted Project
majnemer added a comment to D34972: [CodeGen] Propagate dllexport to thunks.

OK, so we are exporting the thunks so that the linker will generate import thunks for the thunks. I think that we should have a comment to that effect near the code you added.

Fri, Jul 21, 10:31 PM
smeenai updated the summary of D34972: [CodeGen] Propagate dllexport to thunks.
Fri, Jul 21, 10:13 PM
smeenai added a comment to D34972: [CodeGen] Propagate dllexport to thunks.

Ping. I updated the description based on the email discussion with @majnemer.

Fri, Jul 21, 10:13 PM
enlight updated the diff for D35760: Expose active and available platform lists via SBDebugger API.

Add docstrings to SWIG bindings.

Fri, Jul 21, 10:09 PM · Restricted Project
thakis added a comment to D35757: Work around an MSVC2017 update 3 codegen bug..

From what I understand, _MSC_VER changes with each 2017 update.

Fri, Jul 21, 9:31 PM
rjmccall added a comment to D34444: Teach codegen to work in incremental processing mode..

@rjmccall, thanks for the prompt and thorough reply.

Okay. In that case, I see two problems, one major and one potentially major.

This is a very accurate diagnosis which took us 5 years to discover on an empirical basis ;)

You could've asked at any time. :)

True. I am not really sure I knew what to ask, though ;)

Fri, Jul 21, 9:23 PM
mehdi_amini added inline comments to D35633: [ThinLTO] Add FunctionAttr NoRecurse and ReadAttr propagation to ThinLTO.
Fri, Jul 21, 9:21 PM
enlight created D35760: Expose active and available platform lists via SBDebugger API.
Fri, Jul 21, 9:21 PM · Restricted Project
smeenai added inline comments to D35757: Work around an MSVC2017 update 3 codegen bug..
Fri, Jul 21, 9:18 PM
mehdi_amini added inline comments to D35702: [LTO][ThinLTO] Use the linker resolutions to mark global values as dso_local..
Fri, Jul 21, 9:10 PM
ruiu added inline comments to D35759: [Bash-autocompletion] Show HelpText with possible flags.
Fri, Jul 21, 8:52 PM
chandlerc added a comment to D35758: [LIR] Teach LIR to avoid extending the BE count prior to adding one to it when safe..

Wheee!

Fri, Jul 21, 8:32 PM
sanjoy accepted D35758: [LIR] Teach LIR to avoid extending the BE count prior to adding one to it when safe..

lgtm!

Fri, Jul 21, 8:28 PM
yamaguchi created D35759: [Bash-autocompletion] Show HelpText with possible flags.
Fri, Jul 21, 8:21 PM
chandlerc created D35758: [LIR] Teach LIR to avoid extending the BE count prior to adding one to it when safe..
Fri, Jul 21, 8:13 PM
phosek committed rL308822: Revert "Reland "[LLVM][llvm-objcopy] Added basic plumbing to get things….
Revert "Reland "[LLVM][llvm-objcopy] Added basic plumbing to get things…
Fri, Jul 21, 7:44 PM
phosek committed rL308821: Reland "[LLVM][llvm-objcopy] Added basic plumbing to get things started".
Reland "[LLVM][llvm-objcopy] Added basic plumbing to get things started"
Fri, Jul 21, 7:34 PM
zturner added a comment to D35738: Enable llvm-pdbutil to list enumerations using native PDB reader.

Is it that much extra work to implement children? It would be nice if this could just be feature complete. It seems like all you have to do is read the FieldList member of the CodeView record, load that item, and then create an enumerator that iterates each item and returns a NativeConstantSymbol or whatever we're calling it

Fri, Jul 21, 7:11 PM
thakis created D35757: Work around an MSVC2017 update 3 codegen bug..
Fri, Jul 21, 7:07 PM
srhines added a reviewer for D35756: [mips] Switch asm to __asm__ for non-GNU compiles.: petarj.
Fri, Jul 21, 7:06 PM
srhines created D35756: [mips] Switch asm to __asm__ for non-GNU compiles..
Fri, Jul 21, 7:03 PM
kcc committed rL308820: [libFuzzer bot] ensure that no process runs for more than 1 hour.
[libFuzzer bot] ensure that no process runs for more than 1 hour
Fri, Jul 21, 7:00 PM
fedor.sergeev created D35755: [Solaris] gcc toolchain handling revamp.
Fri, Jul 21, 7:00 PM
Meinersbur created D35754: [ForwardOpTree] Introduce the -polly-optree pass..
Fri, Jul 21, 7:00 PM · Restricted Project
vitalybuka committed rL308819: [sanitizer_common] Move filesystem-related code out of sanitizer_common.cc.
[sanitizer_common] Move filesystem-related code out of sanitizer_common.cc
Fri, Jul 21, 6:48 PM
vitalybuka closed D35591: [sanitizer_common] Move filesystem-related code out of sanitizer_common.cc by committing rL308819: [sanitizer_common] Move filesystem-related code out of sanitizer_common.cc.
Fri, Jul 21, 6:48 PM · Restricted Project
phosek accepted D35692: [CMake] Remove redundant logic in runtimes/CMakeList.txt.

LGTM

Fri, Jul 21, 6:37 PM
qcolombet added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 6:37 PM
ctopper committed rL308818: [X86] Add nopq instruction which is a rex encoded version of nopl for gas….
[X86] Add nopq instruction which is a rex encoded version of nopl for gas…
Fri, Jul 21, 6:33 PM
ctopper committed rL308817: [X86] Add register form of NOPL and NOPW for assembler/disassembler..
[X86] Add register form of NOPL and NOPW for assembler/disassembler.
Fri, Jul 21, 6:33 PM
kcc committed rL308816: [libFuzzer bot] add one more benchmark.
[libFuzzer bot] add one more benchmark
Fri, Jul 21, 6:27 PM
MatzeB added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 6:11 PM
qcolombet added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 6:09 PM
qcolombet added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 6:06 PM
jgalenson added inline comments to D35635: Optimize {s,u}{add,sub}.with.overflow on ARM.
Fri, Jul 21, 6:04 PM
echristo added a comment to D35577: Add -flookup-tables and -fno-lookup-tables flags.

The discussion is scattered across these patches https://reviews.llvm.org/D35578 and https://reviews.llvm.org/D35579.
I will provide a brief summary here:

The idea is to control the generation of data (lookup table) generated from a function, specifically when the user is not expecting it.
For hexagon, there is tightly coupled memory and the customers usually place "text" in it.
For functions, which generate lookup tables, it is very very expensive to read the table from a far away non-TCM data section.
This option will disable the generation of lookup tables at the expense of code bloat. This is really driven by the customers of hexagon backend.

Fri, Jul 21, 6:03 PM
MatzeB added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 6:02 PM
ecbeckmann created D35753: llvm-mt: implement simple merging of manifests, not factoring namespaces..
Fri, Jul 21, 6:01 PM
ecbeckmann abandoned D35752: llvm-mt: handle initial merge, not considering namespaces..

once again wrong files in patch

Fri, Jul 21, 6:01 PM
efriedma added inline comments to D35635: Optimize {s,u}{add,sub}.with.overflow on ARM.
Fri, Jul 21, 6:00 PM
MatzeB added a reviewer for D35730: RA: Remove assert on empty live intervals: kparzysz.
Fri, Jul 21, 6:00 PM
qcolombet added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 6:00 PM
echristo added inline comments to D35750: [x86] Teach the x86 backend about general fast rep+movs and rep+stos features of modern x86 CPUs, and use this feature to drastically reduce the number of places we actually emit memset and memcpy library calls..
Fri, Jul 21, 6:00 PM
ecbeckmann updated the diff for D35752: llvm-mt: handle initial merge, not considering namespaces..

Removed build config files.

Fri, Jul 21, 6:00 PM
ecbeckmann abandoned D35751: llvm-mt: handle initial merge, not considering namespaces..

Diffed against wrong base patch

Fri, Jul 21, 5:57 PM
ecbeckmann created D35752: llvm-mt: handle initial merge, not considering namespaces..
Fri, Jul 21, 5:57 PM
MatzeB added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 5:56 PM
ecbeckmann updated the diff for D35751: llvm-mt: handle initial merge, not considering namespaces..

Diff against correct patch.

Fri, Jul 21, 5:54 PM
ecbeckmann created D35751: llvm-mt: handle initial merge, not considering namespaces..
Fri, Jul 21, 5:54 PM
arsenm added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 5:54 PM
qcolombet added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 5:54 PM
qcolombet added a comment to D35749: RA: Remove another assert on empty intervals.

Before removing all those assertions, could we talk that through first?
We started a conversion in D35730.

Fri, Jul 21, 5:54 PM
chandlerc created D35750: [x86] Teach the x86 backend about general fast rep+movs and rep+stos features of modern x86 CPUs, and use this feature to drastically reduce the number of places we actually emit memset and memcpy library calls..
Fri, Jul 21, 5:47 PM
phosek added inline comments to D34956: [ELF] - Store pointer to PT_LOAD instead of pointer to first section (FirstInPtLoad) in OutputSection.
Fri, Jul 21, 5:46 PM
jgalenson added a comment to D35641: Fix update_llc_test_checks.py ARM parsing.

Thanks! Could you commit this for me?

Fri, Jul 21, 5:45 PM
arsenm committed rL308815: AMDGPU: Remove leftover td file.
AMDGPU: Remove leftover td file
Fri, Jul 21, 5:43 PM
jgalenson added inline comments to D35635: Optimize {s,u}{add,sub}.with.overflow on ARM.
Fri, Jul 21, 5:42 PM
compnerd accepted D35265: [libunwind] Handle .ARM.exidx tables without sentinel last entry.

I wish that we could add a test for this. Unfortunately, libunwind as it stands doesn't do a very good job with tests :-(. Thanks for cleaning up the change and fixing this.

Fri, Jul 21, 5:42 PM
qcolombet added inline comments to D35730: RA: Remove assert on empty live intervals.
Fri, Jul 21, 5:38 PM
davide added a comment to D35741: Add MemorySSA alternative to AliasSetTracker in LICM..

Thanks, I'll review this soon.
A quick question. a concern people have about MemorySSA is that you pay upfront for O(1) queries, which may have some impact on compile time.

FWIW: This is and always has been false accounting in most passes as you know.
Most passes that could use MemorySSA already ask about every load and store in the function.
The only reason people think it's faster to not use MemorySSA, or to avoid the upfront cost, is because our time-passes doesn't account function time to lazy utilities/analysis properly.
If it did, AA/etc time would look *much* larger than it does now, because in reality, it *is* much larger than it seems.

IE if you avoid optimize uses, and call getClobberingAccess on every load/store, "MemorySSA" time goes down, overall time goes up.

Fri, Jul 21, 5:32 PM
arsenm closed D35749: RA: Remove another assert on empty intervals.

r308813

Fri, Jul 21, 5:26 PM
dblaikie committed rL308814: Commit missing/empty test file from r308789.
Commit missing/empty test file from r308789
Fri, Jul 21, 5:26 PM
arsenm committed rL308813: RA: Remove another assert on empty intervals.
RA: Remove another assert on empty intervals
Fri, Jul 21, 5:25 PM
rafael committed rL308812: Handle a section being more aligned than a page size..
Handle a section being more aligned than a page size.
Fri, Jul 21, 5:18 PM
MatzeB accepted D35749: RA: Remove another assert on empty intervals.

LGTM

Fri, Jul 21, 5:16 PM
phosek updated the diff for D35165: [sanitizer] Support compiler-rt builtins.
Fri, Jul 21, 5:15 PM
phosek added inline comments to D35165: [sanitizer] Support compiler-rt builtins.
Fri, Jul 21, 5:15 PM
dberlin added a comment to D35741: Add MemorySSA alternative to AliasSetTracker in LICM..

Thanks, I'll review this soon.
A quick question. a concern people have about MemorySSA is that you pay upfront for O(1) queries, which may have some impact on compile time.

FWIW: This is and always has been false accounting in most passes as you know.
Most passes that could use MemorySSA already ask about every load and store in the function.
The only reason people think it's faster to not use MemorySSA, or to avoid the upfront cost, is because our time-passes doesn't account function time to lazy utilities/analysis properly.
If it did, AA/etc time would look *much* larger than it does now, because in reality, it *is* much larger than it seems.

Fri, Jul 21, 5:15 PM
arsenm created D35749: RA: Remove another assert on empty intervals.
Fri, Jul 21, 5:12 PM
kcc committed rL308811: [libFuzzer] reimplement experimental_len_control=1: bump the temporary max_len….
[libFuzzer] reimplement experimental_len_control=1: bump the temporary max_len…
Fri, Jul 21, 5:11 PM
matze committed rL308810: tests/test_suite: Add --output option.
tests/test_suite: Add --output option
Fri, Jul 21, 5:11 PM
echristo accepted D35627: [ARM] Add fatal error when ARM exec mode is not available..
Fri, Jul 21, 5:11 PM
jakehehrlich updated the diff for D33964: [LLVM][llvm-objcopy] Added basic plumbing to get things started.

Somehow I messed up and uploaded the old hello-world.s test. This fixes that

Fri, Jul 21, 5:07 PM
phosek accepted D35747: [Driver] Fuchsia defaults to -fno-math-errno.

LGTM

Fri, Jul 21, 5:04 PM
phosek accepted D35748: [CodeGen][X86] Fuchsia supports sincos* libcalls and sin+cos->sincos optimization.

LGTM

Fri, Jul 21, 5:04 PM
rafael committed rL308809: Don't crash on an empty section with an ALIGN..
Don't crash on an empty section with an ALIGN.
Fri, Jul 21, 5:01 PM
arsenm closed D35730: RA: Remove assert on empty live intervals.

r308808

Fri, Jul 21, 4:57 PM
mcgrathr added a comment to D35748: [CodeGen][X86] Fuchsia supports sincos* libcalls and sin+cos->sincos optimization.

Please land it for me when approved.

Fri, Jul 21, 4:56 PM
arsenm committed rL308808: RA: Remove assert on empty live intervals.
RA: Remove assert on empty live intervals
Fri, Jul 21, 4:56 PM
mcgrathr created D35748: [CodeGen][X86] Fuchsia supports sincos* libcalls and sin+cos->sincos optimization.
Fri, Jul 21, 4:56 PM
asbirlea added a comment to D35741: Add MemorySSA alternative to AliasSetTracker in LICM..

As far as I have tested, the cost of building MemorySSA is low because it does not incur all the cost up front.
More precisely, "optimizeUses" is capped. Only when the clobbering info is requested (if this was not part of the initial optimize pass) there's a pass to obtain it.
For hoist/sink, LICM takes advantage of this.

Fri, Jul 21, 4:55 PM
chapuni committed rL308807: clangDaemon: Update libdeps in r308738..
clangDaemon: Update libdeps in r308738.
Fri, Jul 21, 4:49 PM
phosek committed rL308806: Revert "[LLVM][llvm-objcopy] Added basic plumbing to get things started".
Revert "[LLVM][llvm-objcopy] Added basic plumbing to get things started"
Fri, Jul 21, 4:43 PM
chandlerc added inline comments to D35744: Make new PM honor -fdebug-info-for-profiling.
Fri, Jul 21, 4:39 PM
mcgrathr added a comment to D35747: [Driver] Fuchsia defaults to -fno-math-errno.

Please land it for me when approved.

Fri, Jul 21, 4:37 PM
cmatthews committed rL308805: Add Form labels to run page.
Add Form labels to run page
Fri, Jul 21, 4:37 PM
cmatthews committed rL308804: Don't compare order_id either in round trip test.
Don't compare order_id either in round trip test
Fri, Jul 21, 4:37 PM
mcgrathr updated the diff for D35747: [Driver] Fuchsia defaults to -fno-math-errno.

test case

Fri, Jul 21, 4:37 PM