Vedant Kumar [Thu, 27 Oct 2016 23:17:51 +0000 (23:17 +0000)]
[Coverage] Darwin: Move __llvm_covmap from __DATA to __LLVM_COV
Programs with very large __llvm_covmap sections may fail to link on
Darwin because because of out-of-range 32-bit RIP relative references.
It isn't possible to work around this by using the large code model
because it isn't supported on Darwin. One solution is to move the
__llvm_covmap section past the end of the __DATA segment.
=== Testing ===
In addition to check-{llvm,clang,profile}, I performed a link test on a
simple object after injecting ~4GB of padding into __llvm_covmap:
This patch should not pose any backwards-compatibility concerns. LLVM
is expected to scan all of the sections in a binary for __llvm_covmap,
so changing its segment shouldn't affect anything. I double-checked this
by loading coverage produced by an unpatched compiler with a patched
llvm-cov.
Update .debug_line section version information to match DWARF version.
In the past the compiler always emitted .debug_line version 2, though some opcodes from DWARF 3 (e.g. DW_LNS_set_prologue_end, DW_LNS_set_epilogue_begin or DW_LNS_set_isa) and from DWARF 4 could be emitted by the compiler.
This patch changes version information of .debug_line to exactly match the DWARF version. For .debug_line version 4, a new field maximum_operations_per_instruction is emitted.
Tim Shen [Thu, 27 Oct 2016 21:39:51 +0000 (21:39 +0000)]
[APFloat] Add DoubleAPFloat mode to APFloat. NFC.
Summary:
This patch adds DoubleAPFloat mode to APFloat.
Now, an APFloat with semantics PPCDoubleDouble will have DoubleAPFloat layout
(APFloat.U.Double), which contains two underlying APFloats as
PPCDoubleDoubleImpl and IEEEdouble semantics. Currently the IEEEdouble APFloat
is not used, and the first APFloat behaves exactly the same before this change.
This patch consists of three kinds of logics:
1) Construction and destruction of APFloat. Now the ctors, dtor, assign
opertors and factory functions construct different underlying layout
based on the semantics passed in.
2) s/IEEE/getIEEE()/ for normal, lifetime-unrelated computation functions.
These functions only access Floats[0] in DoubleAPFloat, which is the
same as today's semantic.
3) A "Double dispatch" function, APFloat::convert. Converting between two
different layouts requires appropriate logic.
BitcodeReader: Require clients to read the block info block at most once.
This change makes it the client's responsibility to call ReadBlockInfoBlock()
at most once. This is in preparation for a future change that will allow
there to be multiple block info blocks.
See also: http://lists.llvm.org/pipermail/llvm-dev/2016-October/106512.html
Kyle Butt [Thu, 27 Oct 2016 21:37:20 +0000 (21:37 +0000)]
CodeGen: Handle missed case of block removal during BlockPlacement.
There is a use after free bug in the existing code. Loop layout selects
a preferred exit block, and then lays out the loop. If this block is
removed during layout, it needs to be invalidated to prevent a use after
free.
Kevin Enderby [Thu, 27 Oct 2016 20:59:10 +0000 (20:59 +0000)]
Another additional error check for invalid Mach-O files for the
obsolete load commands.
Again the philosophy of the error checking in libObject for
Mach-O files, the idea behind the checking is that we never
will return a Mach-O file out of libObject that contains unknown
things the library code can’t operate on. So known obsolete
load commands will cause a hard error.
Also to make things clear I have added comments to the
values and structures in Support/Mach-O.h and
Support/MachO.def as to what is obsolete.
As noted in a TODO in the code, there may need to be a
non-default mode to allow some unknown values for well
structured Mach-O files with things like unknown load
load commands. So things like using an old lldb on a newer
Mach-O file could still provide some limited functionality.
Ehsan Amiri [Thu, 27 Oct 2016 19:10:09 +0000 (19:10 +0000)]
[PPC] Adding the removed testcase again
This testcase was originally part of r284995, but I put it in a wrong directory.
So I removed it. Before adding it back I did some small enhancements. Also I
changed the assertions a little bit, to take into account the impact of some
changes performed since code review is done.
This is similar to changes done for another testcase in the original commit.
See: https://reviews.llvm.org/D23614#577749
Basically for instead of vxor we now generate xxlxor in some cases, which is
better.
This works great most of the time. However, if the body of the function is
greater than 127 bytes, the branch target limitation of cbz becomes an issue.
This occurs in the unoptimized code generation cases sometimes (like in
compiler-rt).
Since this is a matter of correctness, possibly pay a small penalty instead. We
now form this slightly differently:
cbnz .Lbody
udf #249 @ __brkdiv0
.Lbody:
...
The positive case is through the branch instead of being the next instruction.
However, because of the basic block layout, the negated branch is going to be
a short distance always (2 bytes away, after the inserted __brkdiv0).
The new t__brkdiv0 instruction is required to explicitly mark the instruction as
a terminator as the generic UDF instruction is not a terminator.
Greg Clayton [Thu, 27 Oct 2016 16:32:04 +0000 (16:32 +0000)]
Switch all DWARF variables for tags, attributes and forms over to use the llvm::dwarf enumerations instead of using raw uint16_t values. This allows easier debugging as users can see the values of the enumerations in the variables view that will show the enumeration string instead of just a number.
Dehao Chen [Thu, 27 Oct 2016 16:30:08 +0000 (16:30 +0000)]
Add Loop Sink pass to reverse the LICM based of basic block frequency.
Summary: LICM may hoist instructions to preheader speculatively. Before code generation, we need to sink down the hoisted instructions inside to loop if it's beneficial. This pass is a reverse of LICM: looking at instructions in preheader and sinks the instruction to basic blocks inside the loop body if basic block frequency is smaller than the preheader frequency.
[mips] Do not allow -opt-bisect-limit to skip the PIC call optimization pass.
r282428 added the MipsOptimizePICCall as an opt-in pass that can be
skipped when using the -opt-bisect-limit option. However, this pass is
needed because it generates code that conforms to the o32 ABI
specification by using the $t9 register for PIC calls with JALR
instructions.
This bug was exposed by the fact that skipFunction() also checks for
the "optnone" attribute. This caused functions with that attribute to
break the requirements of the o32 ABI.
Bjorn Pettersson [Thu, 27 Oct 2016 14:48:09 +0000 (14:48 +0000)]
Fix memory issue in AttrBuilder::removeAttribute uses.
Summary:
Found when running Valgrind.
This removes two unnecessary assignments when using
AttrBuilder::removeAttribute.
AttrBuilder::removeAttribute returns a reference to the object.
As the LHSes were the same as the callees, the assignments
resulted in memcpy calls where dst = src.
Simon Pilgrim [Thu, 27 Oct 2016 14:29:28 +0000 (14:29 +0000)]
[DAGCombiner] Add vector demanded elements support to computeKnownBits
Currently computeKnownBits returns the common known zero/one bits for all elements of vector data, when we may only be interested in one/some of the elements.
This patch adds a DemandedElts argument that allows us to specify the elements we actually care about. The original computeKnownBits implementation calls with a DemandedElts demanding all elements to match current behaviour. Scalar types set this to 1.
The approach was found to be easier than trying to add a per-element known bits solution, for a similar usefulness given the combines where computeKnownBits is typically used.
I've only added support for a few opcodes so far (the ones that have proven straightforward to test), all others will default to demanding all elements but can be updated in due course.
DemandedElts support could similarly be added to computeKnownBitsForTargetNode in a future commit.
Alexey Bataev [Thu, 27 Oct 2016 12:02:28 +0000 (12:02 +0000)]
[SLP] Fix for PR30626: Compiler crash inside SLP Vectorizer.
After successfull horizontal reduction vectorization attempt for PHI node
vectorizer tries to update root binary op by combining vectorized tree
and the ReductionPHI node. But during vectorization this ReductionPHI
can be vectorized itself and replaced by the `undef` value, while the
instruction itself is marked for deletion. This 'marked for deletion'
PHI node then can be used in new binary operation, causing "Use still
stuck around after Def is destroyed" crash upon PHI node deletion.
Also the test is fixed to make it perform actual testing.
George Rimar [Thu, 27 Oct 2016 11:50:04 +0000 (11:50 +0000)]
[Object/ELF] - Fixed behavior when SectionHeaderTable->sh_size is too large.
Elf.h already has code checking that section table does not go past end of file.
Problem is that this check may not work on values greater than UINT64_MAX / Header->e_shentsize
because of calculation overflow.
Sam Parker [Thu, 27 Oct 2016 09:47:10 +0000 (09:47 +0000)]
[ARM] Predicate UMAAL selection on hasDSP.
UMAAL is a DSP instruction and it is not available on thumbv7m
(Cortex-M3) and thumbv6m (Cortex-M0+1) targets. Also fix wrong
CHECK prefix in longMAC.ll test.
Nicolai Haehnle [Thu, 27 Oct 2016 08:15:07 +0000 (08:15 +0000)]
AMDGPU: Fix SILoadStoreOptimizer when writes cannot be merged due register dependencies
Summary:
When finding a match for a merge and collecting the instructions that must
be moved, keep in mind that the instruction we merge might actually use one
of the defs that are being moved.
The fact that the ds_read in the test case is not eliminated suggests that
there might be another problem related to alias analysis, but that's a
separate problem: this pass should still work correctly even when earlier
optimization passes missed something or were disabled.
Justin Bogner [Wed, 26 Oct 2016 22:37:52 +0000 (22:37 +0000)]
llvm-objdump: Make some error messages more consistent
Most of the version of report_error were quoting the filename and
printing a colon between the file name and the error message, but this
one wasn't doing either of those. Fix the output to be more
consistent.
Vedant Kumar [Wed, 26 Oct 2016 22:07:39 +0000 (22:07 +0000)]
[utils] Add a '--unified-report' option to the code coverage prep script
In --unified-report mode, a single coverage report is prepared for all
specified binaries and written to *report_dir*. This mode is compatible
with all existing script options, including the --restrict mode which is
used to limit coverage reporting to certain files or directories.
This should not break any existing users of the script.
Vedant Kumar [Wed, 26 Oct 2016 22:07:35 +0000 (22:07 +0000)]
[utils] Add an '--only-merge' option to the code coverage prep script
In --only-merge mode, the script terminates after the profile merging
step. This makes the script less stateful: it's more natural to split
the merge out into a separate step instead of relying on the first
invocation of the script to do it.
This should not break any existing users of the script.
Evandro Menezes [Wed, 26 Oct 2016 22:06:20 +0000 (22:06 +0000)]
[AArch64] Create feature set for Samsung Exynos-M2
Since Exynos-M2 improved the FP square root unit a bit over the one in
Exynos-M1, it does not benefit from using the Newton series for such
operations.
Ehsan Amiri [Wed, 26 Oct 2016 20:16:59 +0000 (20:16 +0000)]
[PPC] Remove testcase from incorrect directory
During my last commit this testcase was put in an incorrect directory. Removing
it. Will put it in the right directory when I can verify everything is correct.
Tim Northover [Wed, 26 Oct 2016 20:01:00 +0000 (20:01 +0000)]
ARM: don't rely on push/pop reglists being in order when folding SP adjust.
It would be a very nice invariant to rely on, but unfortunately it doesn't
necessarily hold (and the causes of mis-sorted reglists appear to be quite
varied) so to be robust the frame lowering code can't assume that the first
register in the list is also the first one that actually gets pushed.
Should fix an issue where we were turning something like:
Nemanja Ivanovic [Wed, 26 Oct 2016 19:51:35 +0000 (19:51 +0000)]
Do not assume that FP vector operands are never legalized by expanding
This patch ensures that if a floating point vector operand is legalized by
expanding, it is legalized through the stack rather than by calling
DAGTypeLegalizer::IntegerToVector which will cause a failure since the operand
is a non-integer type.
Sanjoy Das [Wed, 26 Oct 2016 19:18:43 +0000 (19:18 +0000)]
Simplify `x >=u x >> y` and `x >=u x udiv y`
Summary:
Extends InstSimplify to handle both `x >=u x >> y` and `x >=u x udiv y`.
This is a folloup of rL258422 and
https://github.com/rust-lang/rust/pull/30917 where llvm failed to
optimize away the bounds checking in a binary search.
Dan Gohman [Wed, 26 Oct 2016 17:44:09 +0000 (17:44 +0000)]
[WebAssembly] Update the README.txt.
Update the README.txt with newer information, add a link to the Emscripten
page explaining the current easiest way to use the LLVM wasm backend, and
mention that other ways of using the LLVM wasm backend are in development.
Robert Lougher [Wed, 26 Oct 2016 17:01:47 +0000 (17:01 +0000)]
Reapply: "Remove debug location from common tail when tail-merging"
This reapplies revision 285093. Original commit message:
The branch folding pass tail merges blocks into a common-tail. However, the
tail retains the debug information from one of the original inputs to the
merge (chosen randomly). This is a problem for sampled-based PGO, as hits
on the common-tail will be attributed to whichever block was chosen,
irrespective of which path was actually taken to the common-tail.
This patch fixes the issue by nulling the debug location for the common-tail.
Dehao Chen [Wed, 26 Oct 2016 15:48:45 +0000 (15:48 +0000)]
Introduce updateDiscriminator interface to DILocation to make it cleaner assigning discriminators.
Summary: This patch introduces updateDiscriminator to DILocation so that it can be directly called by AddDiscriminator. It also makes it easier to update the discriminator later.
Matt Arsenault [Wed, 26 Oct 2016 14:53:50 +0000 (14:53 +0000)]
Fix nondeterministic output in local stack slot alloc pass
This finds all of the references to a frame index in a function, and
sorts by the offset. If multiple instructions use the same offset,
nothing was breaking the tie for sorting.
This avoids the test failures the reverted r282999 introduced.
Victor Leschuk [Wed, 26 Oct 2016 11:59:03 +0000 (11:59 +0000)]
DebugInfo: support for DWARFv5 DW_AT_alignment attribute
* Assume that clang passes non-zero alignment value to DIBuilder
only in case when it was forced by C++11 'alignas', C11 '_Alignas'
or compiler attribute '__attribute__((aligned (N)))'.
* Emit DW_AT_alignment if alignment is specified for type/object.
Andrea Di Biagio [Wed, 26 Oct 2016 10:28:32 +0000 (10:28 +0000)]
[IndVarSimplify][DebugLoc] When widening the exit loop condition, correctly reuse the debug location of the original comparison.
When the loop exit condition is canonicalized as a != compaison, reuse the
debug location of the original (non canonical) comparison.
Before this patch, the debug location of the new icmp was obtained from the
loop latch terminator. This patch fixes the issue by correctly setting the
IRBuilder's "current debug location" to the location of the original compare.
Victor Leschuk [Wed, 26 Oct 2016 08:55:27 +0000 (08:55 +0000)]
DebugInfo: support for DWARFv5 DW_AT_alignment attribute
* Assume that clang passes non-zero alignment value to DIBuilder
only in case when it was forced by C++11 'alignas', C11 '_Alignas'
or compiler attribute '__attribute__((aligned (N)))'.
* Emit DW_AT_alignment if alignment is specified for type/object.
Summary: Clang's intrinsic header currently tries to negate the third operand of a vfmadd mask3 in order to create vfmsub, but this fails isel. This patch adds scalar vfmsub and vfnmsub mask3 that we can use instead to avoid the negate. This is consistent with the packed instructions.
projects / llvm / log