Keith Walker [Fri, 16 Sep 2016 14:07:29 +0000 (14:07 +0000)]
Place the lowered phi instruction(s) before the DEBUG_VALUE entry
When a phi node is finally lowered to a machine instruction it is
important that the lowered "load" instruction is placed before the
associated DEBUG_VALUE entry describing the value loaded.
Renamed the existing SkipPHIsAndLabels to SkipPHIsLabelsAndDebug to
more fully describe that it also skips debug entries. Then used the
"new" function SkipPHIsAndLabels when the debug information should not
be skipped when placing the lowered "load" instructions so that it is
placed before the debug entries.
Teresa Johnson [Fri, 16 Sep 2016 13:54:19 +0000 (13:54 +0000)]
[LTO] Fix handling of mixed (regular and thin) mode LTO
Summary:
In runThinLTO we start the task numbering for ThinLTO backend
tasks depending on whether there was also a regular LTO object
(CombinedModule). However, the CombinedModule is moved at
the end of runRegularLTO, so we need to save this information and
pass it into runThinLTO. Otherwise the AddOutput callback to the client
will use the same task number for both the regular LTO object
and the first ThinLTO object, which in gold-plugin caused only
one to be end up in the output filename array and therefore passed
back to gold for the final native link.
Simon Dardis [Fri, 16 Sep 2016 13:50:43 +0000 (13:50 +0000)]
[mips] Fix aui/daui/dahi/dati for MIPSR6
For compatiblity with binutils, define these instructions to take
two registers with a 16bit unsigned immediate. Both of the registers
have to be same for dahi and dati.
This is an attempt to reapply r280808: [ARM] Lower UDIV+UREM to UDIV+MLS
(and the same for SREM)
This was causing buildbot failures earlier (time outs in the LNT suite).
However, we haven't been able to reproduce this and are suspecting this
was caused by another (reverted) patch.
Eric Liu [Fri, 16 Sep 2016 11:50:57 +0000 (11:50 +0000)]
Trying to fix Mangler memory leak in TargetLoweringObjectFile.
Summary:
`TargetLoweringObjectFile` can be re-used and thus `TargetLoweringObjectFile::Initialize()`
can be called multiple times causing `Mang` pointer memory leak.
[LCG] Redesign the lazy post-order iteration mechanism for the
LazyCallGraph to support repeated, stable iterations, even in the face
of graph updates.
This is particularly important to allow the CGSCC pass manager to walk
the RefSCCs (and thus everything else) in a module more than once. Lots
of unittests and other tests were hard or impossible to write because
repeated CGSCC pass managers which didn't invalidate the LazyCallGraph
would conclude the module was empty after the first one. =[ Really,
really bad.
The interesting thing is that in many ways this simplifies the code. We
can now re-use the same code for handling reference edge insertion
updates of the RefSCC graph as we use for handling call edge insertion
updates of the SCC graph. Outside of adapting to the shared logic for
this (which isn't trivial, but is *much* simpler than the DFS it
replaces!), the new code involves putting newly created RefSCCs when
deleting a reference edge into the cached list in the correct way, and
to re-formulate the iterator to be stable and effective even in the face
of these kinds of updates.
I've updated the unittests for the LazyCallGraph to re-iterate the
postorder sequence and verify that this all works. We even check for
using alternating iterators to trigger the lazy formation of RefSCCs
after mutation has occured.
It's worth noting that there are a reasonable number of likely
simplifications we can make past this. It isn't clear that we need to
keep the "LeafRefSCCs" around any more. But I've not removed that mostly
because I want this to be a more isolated change.
James Molloy [Fri, 16 Sep 2016 10:17:04 +0000 (10:17 +0000)]
[ARM] Promote small global constants to constant pools
If a constant is unamed_addr and is only used within one function, we can save
on the code size and runtime cost of an indirection by changing the global's storage
to inside the constant pool. For example, instead of:
This can cause significant code size savings when many small strings are used in one
function (4 bytes per string).
This recommit contains fixes for a nasty bug related to fast-isel fallback - because
fast-isel doesn't know about this optimization, if it runs and emits references to
a string that we inline (because fast-isel fell back to SDAG) we will end up
with an inlined string and also an out-of-line string, and we won't emit the
out-of-line string, causing backend failures.
It also contains fixes for emitting .text relocations which made the sanitizer
bots unhappy.
Mehdi Amini [Fri, 16 Sep 2016 00:38:18 +0000 (00:38 +0000)]
Fix autoupgrade logic for Objective-C class properties module flag
Previous we were issuing an error when linking a module containing
the new Objective-C metadata structure for class properties with an
"old" one.
Now instead we downgrade the module flag so that the Objective-C
runtime does not expect the new metadata structure.
This is consistent with what ld64 is doing on binary files.
[codeview] Optimize the size of defranges with gaps
For small, discontiguous local variable regions, CodeView can use a
single defrange record with a gap, rather than having two defrange
records. I expect that this optimization will only have a minor impact
on debug info size.
[InstCombine] allow icmp (shr/shl) folds for vectors
These 2 helper functions were already using APInt internally, so just
change the API and caller to allow folds for splats. The scalar
regression tests look quite thorough, so I just added a couple of
tests to prove that vectors are handled too.
These folds should be grouped with the other cmp+shift folds though.
That can be an NFC follow-up.
Mehdi Amini [Thu, 15 Sep 2016 20:26:27 +0000 (20:26 +0000)]
[GlobalOpt] Dead Eliminate declarations
GlobalOpt is already dead-code-eliminating global definitions. With
this change it also takes care of declarations.
Hopefully this should make it now a strict superset of GlobalDCE.
This is important for LTO/ThinLTO as we don't want the linker to see
"undefined reference" when it processes the input files: it could
prevent proper internalization (or even load an extra file from a
static archive, changing the behavior of the program!).
[AArch64] Support for FP FMA when -ffp-contract=fast
Currently, the machine combiner can proceed matching when -ffast-math is on.
It should also match when only -ffp-contract=fast is specified as was the
case before when DAGCombiner was doing the job.
We were inadvertently adding the size of the hash value stream to
the size of the TPI stream, even though the hash value stream is
an entirely separate stream.
This pattern is matched in foldICmpBinOpEqualityWithConstant() and already works
with vectors too. I changed some comments over there to point out the current
location. The tests for this transform are currently in 'sub.ll'.
Note that the remaining folds in this block all require a sub too, so they should
get grouped with the other icmp(sub) patterns.
Simon Dardis [Thu, 15 Sep 2016 13:13:01 +0000 (13:13 +0000)]
[mips][ias] Enable IAS by default for N64 on Debian mips64el.
Unfortunately we can't enable it for all N64 because it is not yet possible to
distinguish N32 from N64.
N64 has been confirmed to produce identical (within reason) objects to GAS
during stage 2 of compiler recursion on N64-abit Fedora. Unfortunately,
Fedora's triples do not distinguish N32 from N64 so I can't enable it by
default there. I'm currently repeating this testing for Debian mips64el but
it's very unlikely to produce a different result.
James Molloy [Thu, 15 Sep 2016 12:30:27 +0000 (12:30 +0000)]
[ARM] Promote small global constants to constant pools
If a constant is unamed_addr and is only used within one function, we can save
on the code size and runtime cost of an indirection by changing the global's storage
to inside the constant pool. For example, instead of:
This can cause significant code size savings when many small strings are used in one
function (4 bytes per string).
This recommit contains fixes for a nasty bug related to fast-isel fallback - because
fast-isel doesn't know about this optimization, if it runs and emits references to
a string that we inline (because fast-isel fell back to SDAG) we will end up
with an inlined string and also an out-of-line string, and we won't emit the
out-of-line string, causing backend failures.
Tim Northover [Thu, 15 Sep 2016 10:09:59 +0000 (10:09 +0000)]
GlobalISel: remove "unsized" LLT
It was only really there as a sentinel when instructions had to have precisely
one type. Now that registers are typed, each register really has to have a type
that is sized.
[llvm-cov] Hide instantiation views for unexecuted functions
Copying in the full text of the function doesn't help at all when we
already know that it's never executed. Just say that it's unexecuted --
the relevant source text has already been printed.
Wei Mi [Thu, 15 Sep 2016 06:28:34 +0000 (06:28 +0000)]
Add some shortcuts in LazyValueInfo to reduce compile time of Correlated Value Propagation.
The patch is to partially fix PR10584. Correlated Value Propagation queries LVI
to check non-null for pointer params of each callsite. If we know the def of
param is an alloca instruction, we know it is non-null and can return early from
LVI. Similarly, CVP queries LVI to check whether pointer for each mem access is
constant. If the def of the pointer is an alloca instruction, we know it is not
a constant pointer. These shortcuts can reduce the cost of CVP significantly.
Jonas Hahnfeld [Thu, 15 Sep 2016 06:14:13 +0000 (06:14 +0000)]
[CMake] Fixing lit for runtimes directory
Copy variable LLVM_BUILD_MAIN_SRC_DIR from LLVMConfig.cmake to
LLVM_MAIN_SRC_DIR as it is named for in-tree builds. This ensures that
add_lit_target() can reliably find llvm-lit which is not necessarily
in the PATH.
[libFuzzer] implement print_pcs with trace-pc-guard. Change the trace-pc-guard heuristic for 8-bit counters to look more like in AFL (not that it's provable better, but the existin test preferes this heuristic)
Wei Mi [Thu, 15 Sep 2016 04:06:44 +0000 (04:06 +0000)]
Add a C++ unittest to test the fix for PR30213.
The test exercises the branch in scev expansion when the value in ValueOffsetPair
is a ptr and the offset is not divisible by the elem type size of value.
The `CVType` had two redundant fields which were confusing and
error-prone to fill out. By treating member records as a distinct
type from leaf records, we are able to simplify this quite a bit.
Summary:
This fixes a dumpbin warning on objects produced by the MC assembler
when starting from text. All .debug$S sections are supposed to be marked
IMAGE_SCN_MEM_DISCARDABLE. The main, non-COMDAT .debug$S section had
this set, but any comdat ones were not being marked discardable because
there was no .section flag for it.
This change does two things:
- If the section name starts with .debug, implicitly mark the section as
discardable. This means we do the same thing as gas on .s files with
DWARF from GCC, which is important.
- Adds the 'D' flag to the .section directive on COFF to explicitly say
a section is discardable. We only emit this flag if the section name
does not start with .debug. This allows the user to explicitly tweak
their section flags without worrying about magic section names.
The only thing you can't do in this scheme is to create a
non-discardable section with a name starting with ".debug", but
hopefully users don't need that.
Mehdi Amini [Wed, 14 Sep 2016 22:29:59 +0000 (22:29 +0000)]
Fix auto-upgrade of TBAA tags in Bitcode Reader
If TBAA is on an intrinsic and it gets upgraded, it'll delete the call
instruction that we collected in a vector. Even if we were to use
WeakVH, it'll drop the TBAA and we'll hit the assert on the upgrade
path.
r263673 gave a shot to make sure the TBAA upgrade happens before
intrinsics upgrade, but failed to account for all cases.
Instead of collecting instructions in a vector, this patch makes it
just upgrade the TBAA on the fly, because metadata are always
already loaded at this point.
Mehdi Amini [Wed, 14 Sep 2016 21:05:04 +0000 (21:05 +0000)]
[LTO] Fix commons handling
Previously the prevailing information was not honored, and commons
symbols could override a strong definition. This patch fixes it and
propose the following semantic for commons: the client should mark
as prevailing the commons that it expects the LTO implementation to
merge (i.e. take the maximum size and alignment).
It implies that commons are allowed to have multiple prevailing
definitions.
Sanjoy Das [Wed, 14 Sep 2016 20:22:03 +0000 (20:22 +0000)]
[Stackmap] Added callsite counts to emitted function information.
Summary:
It was previously not possible for tools to use solely the stackmap
information emitted to reconstruct the return addresses of callsites in
the map, which is necessary to use the information to walk a stack. This
patch adds per-function callsite counts when emitting the stackmap
section in order to resolve the problem. Note that this slightly alters
the stackmap format, so external tools parsing these maps will need to
be updated.
**Problem Details:**
Records only store their offset from the beginning of the function they
belong to. While these records and the functions are output in program
order, it is not possible to determine where the end of one function's
records are without the callsite count when processing the records to
compute return addresses.
projects / llvm / log