[PowerPC] Emit VMX loads/stores for aligned ops to avoid adding swaps on LE
Fixes PR30730.
This is a re-commit of a pulled commit. The commit was pulled because some
software projects contained uses of Altivec vectors that violated alignment
requirements. Known issues have now been fixed.
Dylan McKay [Tue, 2 May 2017 00:11:34 +0000 (00:11 +0000)]
[AVR] Fix a bug where the frame pointer is clobbered
Because it was a callee-saved register, we automatically generated code
to spill and unspill its original value so that it is restored after the
function returns.
The problem is that this code was being generated before the epilogue.
The epilogue itself uses the Y register, which could be prematurely
restored by the CSR restoration process.
This removes R29R28 from the CSR list and changes the prologue/epilogue
code to handle it explicitly.
This change caused buildbot failures, apparently because we're not
passing around types that InstSimplify is used to seeing. I'm not overly
familiar with InstSimplify, so I'm reverting this until I can figure out
what exactly is wrong.
Matthias Braun [Mon, 1 May 2017 22:32:25 +0000 (22:32 +0000)]
MachineFrameInfo: Track whether MaxCallFrameSize is computed yet; NFC
This tracks whether MaxCallFrameSize is computed yet. Ideally we would
assert and fail when the value is queried before it is computed, however
this fails various targets that need to be fixed first.
Sanjay Patel [Mon, 1 May 2017 22:25:42 +0000 (22:25 +0000)]
[InstCombine] check one-use before applying DeMorgan nor/nand folds
If we have ~(~X & Y), it only makes sense to transform it to (X | ~Y) when we do not need
the intermediate (~X & Y) value. In that case, we would need an extra instruction to
generate ~Y + 'or' (as shown in the test changes).
It's ok if we have multiple uses of ~X or Y, however. In those cases, we may not reduce the
instruction count or critical path, but we might improve throughput because we can generate
~X and ~Y in parallel. Whether that actually makes perf sense or not for a target is something
we can't answer in IR.
Greg Clayton [Mon, 1 May 2017 22:07:02 +0000 (22:07 +0000)]
Adds initial llvm-dwarfdump --verify support with unit tests.
lldb-dwarfdump gets a new "--verify" option that will verify a single file's DWARF debug info and will print out any errors that it finds. It will return an non-zero exit status if verification fails, and a zero exit status if verification succeeds. Adding the --quiet option will suppress any output the STDOUT or STDERR.
The first part of the verify does the following:
- verifies that all CU relative references (DW_FORM_ref1, DW_FORM_ref2, DW_FORM_ref4, DW_FORM_ref8, DW_FORM_ref_udata) have valid CU offsets
- verifies that all DW_FORM_ref_addr references have valid .debug_info offsets
- verifies that all DW_AT_ranges attributes have valid .debug_ranges offsets
- verifies that all DW_AT_stmt_list attributes have valid .debug_line offsets
- verifies that all DW_FORM_strp attributes have valid .debug_str offsets
Unit tests were added for each of the above cases.
Craig Topper [Mon, 1 May 2017 21:56:05 +0000 (21:56 +0000)]
[APInt] In operator!, handle single word case by comparing VAL to 0 directly and handle multiword case by comparing countLeadingZerosSlowCase() to BitWidth.
We were using operator=(0) which implicitly calls countLeadingZeros but only to compare with 64 to determine if we can compare VAL or pVal[0] to uint64_t. By handling the multiword case with countLeadingZerosSlowCase==BitWidth we can prevent a load of pVal[0] from being inserted inline at each call site. This saves a little bit of code size.
Xin Tong [Mon, 1 May 2017 17:15:37 +0000 (17:15 +0000)]
Take indirect branch into account as well when folding.
We may not be able to rewrite indirect branch target, but we also want to take it into
account when folding, i.e. if it and all its successor's predecessors go to the same
destination, we can fold, i.e. no need to thread.
Sanjoy Das [Mon, 1 May 2017 17:07:56 +0000 (17:07 +0000)]
Use WeakVH instead of WeakTrackingVH in AliasSetTracker's UnkownInsts
In cases where an instruction (a call site, say) is RAUW'ed with some
other value (this is possible via the `returned` attribute, for
instance), we want the slot in UnknownInsts to point to the original
Instruction we wanted to track, not the value it got replaced by.
Zachary Turner [Mon, 1 May 2017 16:46:39 +0000 (16:46 +0000)]
[PDB/CodeView] Rename some classes.
In preparation for introducing writing capabilities for each of
these classes, I would like to adopt a Foo / FooRef naming
convention, where Foo indicates that the class can manipulate and
serialize Foos, and FooRef indicates that it is an immutable view of
an existing Foo. In other words, Foo is a writer and FooRef is a
reader. This patch names some existing readers to conform to the
FooRef convention, while offering no functional change.
Sanjoy Das [Mon, 1 May 2017 16:28:58 +0000 (16:28 +0000)]
Emulate TrackingVH using WeakVH
Summary:
This frees up one slot in the HandleBaseKind enum, which I will use
later to add a new kind of value handle. The size of the
HandleBaseKind enum is important because we store a HandleBaseKind in
the low two bits of a (in the worst case) 4 byte aligned pointer.
Craig Topper [Mon, 1 May 2017 16:08:06 +0000 (16:08 +0000)]
[SelectionDAG] Use known ones to provide a better bound for the known zeros for CTTZ/CTLZ operations.
This is the SelectionDAG version of D32521. If know where at least one 1 is located in the input to these intrinsics we can place an upper bound on the number of bits needed to represent the count and thus increase the number of known zeros in the output.
I think we can also refine this further for CTTZ_UNDEF/CTLZ_UNDEF by assuming that the answer will never be BitWidth. I've left this out for now because it caused other test failures across multiple targets. Usually because of turning ADD into OR based on this new information.
In this patch, I introduce a new alt macro feature.
This feature adds meaning for the % when using it as a prefix to the calling macro arguments.
In the altmacro mode, the percent sign '%' before an absolute expression convert the expression first to a string.
As described in the https://sourceware.org/binutils/docs-2.27/as/Altmacro.html
"Expression results as strings
You can write `%expr' to evaluate the expression expr and use the result as a string."
expression assumptions:
1. '%' can only evaluate an absolute expression.
2. Altmacro '%' must be the first character of the evaluated expression.
3. If no '%' is located before the expression, a regular module operation is expected.
4. The result of Absolute Expressions can be only integer.
[DAGCombiner] shrink/widen a vselect to match its condition operand size (PR14657)
We discussed shrinking/widening of selects in IR in D26556, and I'll try to get back to that
patch eventually. But I'm hoping that this transform is less iffy in the DAG where we can check
legality of the select that we want to produce.
A few things to note:
1. We can't wait until after legalization and do this generically because (at least in the x86
tests from PR14657), we'll have PACKSS and bitcasts in the pattern.
2. This might benefit more of the SSE codegen if we lifted the legal-or-custom requirement, but
that requires a closer look to make sure we don't end up worse.
3. There's a 'vblendv' opportunity that we're missing that results in andn/and/or in some cases.
That should be fixed next.
4. I'm assuming that AVX1 offers the worst of all worlds wrt uneven ISA support with multiple
legal vector sizes, but if there are other targets like that, we should add more tests.
5. There's a codegen miracle in the multi-BB tests from PR14657 (the gcc auto-vectorization tests):
despite IR that is terrible for the target, this patch allows us to generate the optimal loop
code because something post-ISEL is hoisting the splat extends above the vector loops.
Sanjoy Das [Sun, 30 Apr 2017 19:41:19 +0000 (19:41 +0000)]
Rename isKnownNotFullPoison to programUndefinedIfPoison; NFC
Summary:
programUndefinedIfPoison makes more sense, given what the function
does; and I'm about to add a function with a name similar to
isKnownNotFullPoison (so do the rename to avoid confusion).
Do not legalize large add with addc/adde, introduce addcarry and do it with uaddo/addcarry
Summary: As per discution on how to get better codegen an large int legalization, it became clear that using a glue for the carry was preventing several desirable optimizations. Passing the carry down as a value allow for more flexibility.
[APInt] Remove support for wrapping from APInt::setBits.
This features isn't used anywhere in tree. It's existence seems to be preventing selfhost builds from inlining any of the setBits methods including setLowBits, setHighBits, and setBitsFrom. This is because the code makes the method recursive.
If anyone needs this feature in the future we could consider adding a setBitsWithWrap method. This way only the calls that need it would pay for it.
Summary:
Apply canonicalization rules:
1. Input vectors with no elements selected from can be replaced with undef.
2. If only one input vector is constant it shall be the second one.
This allows constant-folding to cover more ad-hoc simplifications that
were in place and avoid duplication for RHS and LHS checks.
There are more rules we may want to add in the future when we see a
justification. e.g. mask elements that select undef elements can be
replaced with undef.
Daniel Sanders [Sat, 29 Apr 2017 17:30:09 +0000 (17:30 +0000)]
[globalisel][tablegen] Compute available feature bits correctly.
Summary:
Predicate<> now has a field to indicate how often it must be recomputed.
Currently, there are two frequencies, per-module (RecomputePerFunction==0)
and per-function (RecomputePerFunction==1). Per-function predicates are
currently recomputed more frequently than necessary since the only predicate
in this category is cheap to test. Per-module predicates are now computed in
getSubtargetImpl() while per-function predicates are computed in selectImpl().
Tablegen now manages the PredicateBitset internally. It should only be
necessary to add the required includes.
Also fixed a problem revealed by the test case where
constrainSelectedInstRegOperands() would attempt to tie operands that
BuildMI had already tied.
[KnownBits] Add methods for determining if the known bits represent a negative/nonnegative number and add methods for changing the negative/nonnegative state
Summary: This patch adds isNegative, isNonNegative for querying whether the sign bit is known. It also adds makeNegative and makeNonNegative for controlling the sign bit.
[ConstantRange] Improve the efficiency of one of the ConstantRange constructors.
We were default constructing the Lower/Upper APInts. Then creating min or max value, then doing a move assignment to Lower and copy assignment to upper. The copy assignment operator in particular has an out of line function call that has to examine whether or not a previous allocation exists that can be reused which of course it can't in this case.
The new code creates the min/max value first, move constructs Lower from it then copy constructs Upper from Lower.
This also seems to have convinced a self host build that this constructor can be inlined more readily into other methods in ConstantRange.
[llvm-pdbdump] Abstract some of the YAML/Raw printing code.
There is a lot of duplicate code for printing line info between
YAML and the raw output printer. This introduces a base class
that can be shared between the two, and makes some minor
cleanups in the process.
[ObjCARC] Do not move a release between a call and a
retainAutoreleasedReturnValue that retains the returned value.
This commit fixes a bug in ARC optimizer where it moves a release
between a call and a retainAutoreleasedReturnValue, causing the returned
object to be released before the retainAutoreleasedReturnValue can
retain it.
This commit accomplishes that by doing a lookahead and checking whether
the call prevents the release from moving upwards. In the long term, we
should treat the region between the retainAutoreleasedReturnValue and
the call as a critical section and disallow moving anything there
(possibly using operand bundles).
Davide Italiano [Sat, 29 Apr 2017 00:18:26 +0000 (00:18 +0000)]
[LoopUnswitch] Make DEBUG output more readable (part 2).
I fixed my miscompile in r301722 and I hope I don't have to take
a look at this code again now that Chandler has a new LoopUnswitch
pass, but maybe this could be of use for somebody else in the
meanwhile.
Fuzzer: Mark test/cxxstring.test UNSUPPORTED: windows
This has been mysteriously failing since r301593, which cleaned up the
types of things like size_t and SIZE_MAX for freestanding targets. Reid
and Kostya suggested marking it as UNSUPPORTED on windows, given that no
one has been able to reproduce locally.
Hans Wennborg [Fri, 28 Apr 2017 23:01:32 +0000 (23:01 +0000)]
Revert r301697 "[IR] Make add/remove Attributes use AttrBuilder instead of AttributeList"
This broke the Clang build. (Clang-side patch missing?)
Original commit message:
> [IR] Make add/remove Attributes use AttrBuilder instead of
> AttributeList
>
> This change cleans up call sites and avoids creating temporary
> AttributeList objects.
>
> NFC
Adrian Prantl [Fri, 28 Apr 2017 22:25:46 +0000 (22:25 +0000)]
Remove line and file from DINamespace.
Fixes the issue highlighted in
http://lists.llvm.org/pipermail/cfe-dev/2014-June/037500.html.
The DW_AT_decl_file and DW_AT_decl_line attributes on namespaces can
prevent LLVM from uniquing types that are in the same namespace. They
also don't carry any meaningful information.
projects / llvm / log