Philip Reames [Fri, 6 May 2016 18:46:45 +0000 (18:46 +0000)]
[GVN] Handle unordered atomics in cross block FRE
You'll note there are essentially no code changes here. Cross block FRE heavily reuses code from the block local FRE. All of the tricky parts were done as part of the previous patch and the refactoring that removed the original code duplication.
Philip Reames [Fri, 6 May 2016 18:17:13 +0000 (18:17 +0000)]
[GVN] Do local FRE for unordered atomic loads
This patch is the first in a small series teaching GVN to optimize unordered loads aggressively. This change just handles block local FRE because that's the simplest thing which lets me test MDA, and the AvailableValue pieces. Somewhat suprisingly, MDA appears fine and only a couple of small changes are needed in GVN.
Once this is in, I'll tackle non-local FRE and PRE. The former looks like a natural extension of this, the later will require a couple of minor changes.
Mehdi Amini [Fri, 6 May 2016 18:17:03 +0000 (18:17 +0000)]
Tweak the ThinLTO pass pipeline
Summary:
The original ThinLTO pipeline was derived from some
work I did tuning FullLTO on the test suite and SPEC. This
patch reduces the amount of work done in the "linker phase" of
the build, and extend the function simplifications passes
performed during the "compile phase". This helps the build time
by reducing the IR as much as possible during the compile phase
and limiting the work to be performed during the "link phase",
while keeping the performance "on par" with the existing pipeline.
Sanjay Patel [Fri, 6 May 2016 18:07:46 +0000 (18:07 +0000)]
[SimplifyCFG] propagate branch metadata when creating select (retry r268550 / r268751 with possible fix)
Retrying r268550/r268751 which were reverted at r268577/r268765 due a memory sanitizer failure.
I have not been able to reproduce that failure, but I've taken another guess at fixing
the problem in this version of the patch and will watch for another failure.
Original commit message:
Unlike earlier similar fixes, we need to recalculate the branch weights
in this case.
Artem Tamazov [Fri, 6 May 2016 17:48:48 +0000 (17:48 +0000)]
[AMDGPU][llvm-mc] Add support for sendmsg(...) syntax.
Added support for sendmsg(MSG[, OP[, STREAM_ID]]) syntax
in s_sendmsg and s_sendmsghalt instructions.
The syntax matches the SP3 assembler/disassembler rules.
That is why implicit inputs (like M0 and EXEC) are not printed
to disassembly output anymore.
sendmsg(...) allows only known message types and attributes,
even if literals are used instead of symbolic names.
However, raw literal (without "sendmsg") still can be used,
and that allows for any 16-bit value.
Sanjay Patel [Fri, 6 May 2016 17:07:47 +0000 (17:07 +0000)]
[SimplifyCFG] propagate branch metadata when creating select (retry r268550 with possible fix)
Retrying r268550 which was reverted at r268577 due a memory sanitizer failure.
I have not been able to reproduce that failure, but I've taken a guess at fixing
the problem in this version of the patch and will watch for another failure.
Original commit message:
Unlike earlier similar fixes, we need to recalculate the branch weights
in this case.
Geoff Berry [Fri, 6 May 2016 16:34:59 +0000 (16:34 +0000)]
[AArch64] Combine callee-save and local stack SP adjustment instructions.
Summary:
If a function needs to allocate both callee-save stack memory and local
stack memory, we currently decrement/increment the SP in two steps:
first for the callee-save area, and then for the local stack area. This
changes the code to allocate them both at once at the very beginning/end
of the function. This has two benefits:
1) there is one fewer sub/add micro-op in the prologue/epilogue
2) the stack adjustment instructions act as a scheduling barrier, so
moving them to the very beginning/end of the function increases post-RA
scheduler's ability to move instructions (that only depend on argument
registers) before any of the callee-save stores
This change can cause an increase in instructions if the original local
stack SP decrement could be folded into the first store to the stack.
This occurs when the first local stack store is to stack offset 0. In
this case we are trading off one more sub instruction for one fewer sub
micro-op (along with benefits (2) and (3) above).
Jun Bum Lim [Fri, 6 May 2016 15:08:57 +0000 (15:08 +0000)]
[AArch64] Decouple zero store promotion from narrow ld merge. NFC.
Summary: This change refactors to decouple the zero store promotion from the narrow ld merge and add a flag (enable-narrow-ld-merge=true) to control the narrow ld merge optimization.
Daniel Sanders [Fri, 6 May 2016 14:37:24 +0000 (14:37 +0000)]
[mips] Fix inconsistent .cprestore behaviour between direct object emission and assembling.
Summary:
Direct object emission has an initialization order problem where an
InitMCObjectFile is called after MipsTargetELFStreamer determines whether
PIC is enabled by default or not. There doesn't seem to be point that
initializes all cases so split the responsibility between
MipsTargetELFStreamer and MipsAsmPrinter.
Daniel Sanders [Fri, 6 May 2016 13:49:25 +0000 (13:49 +0000)]
[mips] Correct the ordering of HI/LO pairs in the relocation table.
Summary:
There seems to have been a misunderstanding as to the meaning of 'offset' in
the rules laid down by our ABI. The previous code believed that 'offset' meant
the offset within the section that the relocation is applied to. However, it
should have meant the offset from the symbol used in the relocation expression.
This patch adds two fields to ELFRelocationEntry and uses them to correct the
order of relocations for MIPS. These fields contain:
* The original symbol before shouldRelocateWithSymbol() is considered. This
ensures that R_MIPS_GOT16 is able to correctly distinguish between local and
external symbols, allowing us to tell whether %got() requires a matching
%lo() or not (local symbols require one, external symbols don't). It also
prevents confusing cases where the fuzzy matching rules cause things like
%hi(foo)/%lo(foo+3) and %hi(bar)/%lo(bar+1) to swap their %lo()'s.
* The original offset before shouldRelocateWithSymbol() is considered. The
existing Addend field is always zero when the object uses in place addends
(because it's already moved it to the encoding) but MIPS needs to use the
original offset to ensure that the linker correctly calculates the carry-in
bit for %hi() and %got().
IAS ensures that unmatchable %hi()/%got() relocations are placed at the end of
the table to ensure that the linker rejects the table (we're unable to report
such errors directly). The alternatives to this risk accidental matching
against inappropriate relocations which may silently compute incorrect values
due to an incorrect carry bit between the %lo() and %hi()/%got().
Daniel Sanders [Fri, 6 May 2016 13:23:51 +0000 (13:23 +0000)]
[mips][mips16] Use isUnconditionalBranch() in AnalyzeBranch() and constant island pass.
Summary:
This stops it misidentifying unconditional branches as conditional branches
which fixes a -verify-machineinstrs error about exiting a function via fall through.
Sam Kolton [Fri, 6 May 2016 11:31:17 +0000 (11:31 +0000)]
[TableGen] AsmMatcher: support for default values for optional operands
Summary:
This change allows to specify "DefaultMethod" for optional operand (IsOptional = 1) in AsmOperandClass that return default value for operand. This is used in convertToMCInst to set default values in MCInst.
Previously if you wanted to set default value for operand you had to create custom converter method. With this change it is possible to use standard converters even when optional operands presented.
Ryan Govostes [Fri, 6 May 2016 10:25:22 +0000 (10:25 +0000)]
[asan] add option to set shadow mapping offset
Allowing overriding the default ASAN shadow mapping offset with the
-asan-shadow-offset option, and allow zero to be specified for both offset and
scale.
Adam Nemet [Thu, 5 May 2016 23:41:28 +0000 (23:41 +0000)]
[LAA] Fix confusing debug message
This message used to be correct, when all we cared about was whether the
dependence was safe (i.e. NoDep) or unsafe. With the current more
precise characterization, this is a forward dep.
Justin Bogner [Thu, 5 May 2016 23:19:08 +0000 (23:19 +0000)]
SDAG: Rename Select->SelectImpl and repurpose Select as returning void
This is a step towards removing the rampant undefined behaviour in
SelectionDAG, which is a part of llvm.org/PR26808.
We rename SelectionDAGISel::Select to SelectImpl and update targets to
match, and then change Select to return void and consolidate the
sketchy behaviour we're trying to get away from there.
Next, we'll update backends to implement `void Select(...)` instead of
SelectImpl and eventually drop the base Select implementation.
Justin Bogner [Thu, 5 May 2016 22:37:45 +0000 (22:37 +0000)]
SDAG: Remove OPC_MarkGlueResults and associated logic. NFC
This opcode never happens in practice, and yet the logic we have in
place to handle it would be undefined behaviour if we ever executed
it. Remove it rather than trying to refactor code that's never
reached.
Mehdi Amini [Thu, 5 May 2016 20:05:33 +0000 (20:05 +0000)]
Degrade assertions to a warning in LTOCodeGenerator for preserved linkonce
The assertions were assuming that the linker will not ask to preserve
a global that is internal or available_externally, as it does not
really make sense. In practice this break the bootstrap of clang,
I degrade to a warning for now.
Chris Bieneman [Thu, 5 May 2016 19:57:03 +0000 (19:57 +0000)]
Remove LLVM_ENABLE_TIMESTAMPS
Summary:
As per the discussion on LLVM-dev this patch proposes removing LLVM_ENABLE_TIMESTAMPS.
The only complicated bit of this patch is the Windows support. On windows we used to log an error if /INCREMENTAL was passed to the linker when timestamps were disabled.
With this change since timestamps in code are always disabled we will always compile on windows with /Brepro unless /INCREMENTAL is specified, and we will log a warning when /INCREMENTAL is specified to notify the user that the build will be non-deterministic.
Tim Northover [Thu, 5 May 2016 18:38:53 +0000 (18:38 +0000)]
ARM: don't attempt to merge litpools referencing different PC-anchors.
Given something like:
ldr r0, .LCPI0_0 (== pc-rel var)
add r0, pc
ldr r1, .LCPI0_1 (== pc-rel var)
add r1, pc
we cannot combine the 2 ldr instructions and litpools because they get added to
a different pc to form the correct address. I think the original logic came
from a time when we fused the LDRpci/PICADD instructions into one
pseudo-instruction so the PC was always immediately at-hand. That's no longer
the case.
Should fix general-dynamic TLS access on Linux, and quite possibly other -fPIC
code that relies on litpools (e.g. v6m and -Oz compilations) though trivial
tweaks of the .ll test didn't provoke anything.
Vitaly Buka [Thu, 5 May 2016 18:31:00 +0000 (18:31 +0000)]
Revert "[ThinLTO] Emit individual index files for distributed backends"
MemorySanitizer: use-of-uninitialized-value in lib/Bitcode/Writer/BitcodeWriter.cpp:364:70
http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast/builds/12544/steps/check-llvm%20msan/logs/stdio
Kevin Enderby [Thu, 5 May 2016 17:43:35 +0000 (17:43 +0000)]
Clean up the specific error message for a malformed Mach-O files with bad segment
load commands.
The existing test case in test/Object/macho-invalid.test for
macho-invalid-too-small-segment-load-command has a cmdsize of 55, while
being too small also it is not a multiple of 4. So when that check is added
this test case will produce a different error. So I constructed a new test case
that will trigger the intended error.
I also changed the error message to be consistent with the other malformed Mach-O
file error messages which prints the load command index. I also removed both
object_error::macho_load_segment_too_small and
object_error::macho_load_segment_too_many_sections from Object/Error.h
as they are not needed and can just use object_error::parse_failed and let the
error message string distinguish the specific error.
Mehdi Amini [Thu, 5 May 2016 16:33:51 +0000 (16:33 +0000)]
llvm-lto: add a -thinlto-module-id that enables to force the Module identifier.
ThinLTO is using the Module Identifier to find the corresponding entry
in the index. However when reproducing part of the flow from temporary
files generated from the linker, you'd like to process a file and
force llvm-lto to use another module identifier than the current
filename. The alternative would be to tweak the index, which would be
more involved.
[Hexagon] Handle operand type differences for A2_tfrpi
The instruction A2_tfrpi has a 64-bit operand, while the corresponding
intrinsic takes a 32-bit value. The actual value has only 8 significant
bits, so the difference is only in the type used to represent it.
In order to map the intrinsic to the instruction, the operand needs to
be extended to the correct type.
Silviu Baranga [Thu, 5 May 2016 15:20:39 +0000 (15:20 +0000)]
[LV] Identify more induction PHIs by coercing expressions to AddRecExprs
Summary:
Some PHIs can have expressions that are not AddRecExprs due to the presence
of sext/zext instructions. In order to prevent the Loop Vectorizer from
bailing out when encountering these PHIs, we now coerce the SCEV
expressions to AddRecExprs using SCEV predicates (when possible).
We only do this when the alternative would be to not vectorize.
James Y Knight [Thu, 5 May 2016 14:35:40 +0000 (14:35 +0000)]
Remove bit-rotten CppBackend.
This backend was supposed to generate C++ code which will re-construct
the LLVM IR passed as input. This seems to me to have very marginal
usefulness in the first place.
However, the code has never been updated to use IRBuilder, which makes
its current value negative -- people who look at the output may be
steered to use the *wrong* C++ APIs to construct IR.
Furthermore, it's generated code that doesn't compile since at least
2013.
Teresa Johnson [Thu, 5 May 2016 13:44:56 +0000 (13:44 +0000)]
[ThinLTO] Emit individual index files for distributed backends
Summary:
When launching ThinLTO backends in a distributed build (currently
supported in gold via the thinlto-index-only plugin option), emit
an individual index file for each backend process as described here:
http://lists.llvm.org/pipermail/llvm-dev/2016-April/098272.html
The individual index file encodes the summary and module information
required for implementing the importing/exporting decisions made
for a given module in the thin link step.
This is in place of the current mechanism that uses the combined index
to make importing decisions in each back end independently. It is an
enabler for doing global summary based optimizations in the thin link
step (which will be recorded in the individual index files), and reduces
the size of the index that must be sent to each backend process, and
the amount of work to scan it in the backends.
Rather than create entirely new ModuleSummaryIndex structures (and all
the included unique_ptrs) for each backend index file, a map is created
to record all of the GUID and summary pointers needed for a particular
index file. The IndexBitcodeWriter walks this map instead of the full
index (hiding the details of managing the appropriate summary iteration
in a new iterator subclass). This is more efficient than walking the
entire combined index and filtering out just the needed summaries during
each backend bitcode index write.
[X86] Extend some Linux special cases to cover kFreeBSD.
Both Linux and kFreeBSD use glibc, so follow similiar code paths.
Add isTargetGlibc to check for this, and use it instead of isTargetLinux
in a few places.
Craig Topper [Thu, 5 May 2016 06:19:27 +0000 (06:19 +0000)]
[TableGen] Make sure to recursively factor any ScopeMatchers created while forming a SwitchType node. Remove a couple hundred bytes from the X86 matcher table.
projects / llvm / log