Daniel Berlin [Tue, 5 May 2015 18:10:49 +0000 (18:10 +0000)]
Update BasicAliasAnalysis to understand that nothing aliases with undef values.
It got this in some cases (if one of them was an identified object), but not in all cases.
This caused stores to undef to block load-forwarding in some cases, etc.
Added test to Transforms/GVN to verify optimization occurs as expected.
Reid Kleckner [Tue, 5 May 2015 17:44:16 +0000 (17:44 +0000)]
Re-land "[WinEH] Add an EH registration and state insertion pass for 32-bit x86"
This reverts commit r236360.
This change exposed a bug in WinEHPrepare by opting win32 code into EH
preparation. We already knew that WinEHPrepare has bugs, and is the
status quo for x64, so I don't think that's a reason to hold off on this
change. I disabled exceptions in the sanitizer tests in r236505 and an
earlier revision.
[ShrinkWrap] Add (a simplified version) of shrink-wrapping.
This patch introduces a new pass that computes the safe point to insert the
prologue and epilogue of the function.
The interest is to find safe points that are cheaper than the entry and exits
blocks.
As an example and to avoid regressions to be introduce, this patch also
implements the required bits to enable the shrink-wrapping pass for AArch64.
** Context **
Currently we insert the prologue and epilogue of the method/function in the
entry and exits blocks. Although this is correct, we can do a better job when
those are not immediately required and insert them at less frequently executed
places.
The job of the shrink-wrapping pass is to identify such places.
** Motivating example **
Let us consider the following function that perform a call only in one branch of
a if:
define i32 @f(i32 %a, i32 %b) {
%tmp = alloca i32, align 4
%tmp2 = icmp slt i32 %a, %b
br i1 %tmp2, label %true, label %false
Therefore, we would pay the overhead of setting up/destroying the frame only if
we actually do the call.
** Proposed Solution **
This patch introduces a new machine pass that perform the shrink-wrapping
analysis (See the comments at the beginning of ShrinkWrap.cpp for more details).
It then stores the safe save and restore point into the MachineFrameInfo
attached to the MachineFunction.
This information is then used by the PrologEpilogInserter (PEI) to place the
related code at the right place. This pass runs right before the PEI.
Unlike the original paper of Chow from PLDI’88, this implementation of
shrink-wrapping does not use expensive data-flow analysis and does not need hack
to properly avoid frequently executed point. Instead, it relies on dominance and
loop properties.
The pass is off by default and each target can opt-in by setting the
EnableShrinkWrap boolean to true in their derived class of TargetPassConfig.
This setting can also be overwritten on the command line by using
-enable-shrink-wrap.
Before you try out the pass for your target, make sure you properly fix your
emitProlog/emitEpilog/adjustForXXX method to cope with basic blocks that are not
necessarily the entry block.
** Design Decisions **
1. ShrinkWrap is its own pass right now. It could frankly be merged into PEI but
for debugging and clarity I thought it was best to have its own file.
2. Right now, we only support one save point and one restore point. At some
point we can expand this to several save point and restore point, the impacted
component would then be:
- The pass itself: New algorithm needed.
- MachineFrameInfo: Hold a list or set of Save/Restore point instead of one
pointer.
- PEI: Should loop over the save point and restore point.
Anyhow, at least for this first iteration, I do not believe this is interesting
to support the complex cases. We should revisit that when we motivating
examples.
Daniel Sanders [Tue, 5 May 2015 16:29:40 +0000 (16:29 +0000)]
[bugpoint] Increase default memory limit to 400MB to fix bugpoint tests.
I tracked down the bug to an unchecked malloc in SmallVectorBase::grow_pod().
This malloc is returning NULL on my machine when running under bugpoint but not
when -enable-valgrind is given.
Daniel Sanders [Tue, 5 May 2015 10:32:24 +0000 (10:32 +0000)]
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA.
Summary:
When using the N64 ABI, element-indices use the i64 type instead of i32.
In many cases, we can use iPTR to account for this but additional patterns
and pseudo's are also required.
This fixes most (but not quite all) failures in the test-suite when using
N64 and MSA together.
the move in to R3 is moved out of the IT block so that later instructions on the same predicate can be inside this block, and we can share the IT instruction.
However, when moving the R3 copy out of the IT block, we need to clear its kill flags for anything in use at this point in time, ie, R0 here.
This appeases the machine verifier which thought that R0 wasn't defined when used.
I have a test case, but its extremely register allocator specific. It would be too fragile to commit a test which depends on the register allocator here.
Lang Hames [Mon, 4 May 2015 22:03:10 +0000 (22:03 +0000)]
[Orc] Refactor the compile-on-demand layer to make module partitioning lazy,
and avoid cloning unused decls into every partition.
Module partitioning showed up as a source of significant overhead when I
profiled some trivial test cases. Avoiding the overhead of partitionging
for uncalled functions helps to mitigate this.
This change also means that it is no longer necessary to have a
LazyEmittingLayer underneath the CompileOnDemand layer, since the
CompileOnDemandLayer will not extract or emit function bodies until they are
called.
Tim Northover [Mon, 4 May 2015 20:41:51 +0000 (20:41 +0000)]
CodeGen: match up correct insertvalue indices when assessing tail calls.
When deciding whether a value comes from the aggregate or inserted value of an
insertvalue instruction, we compare the indices against those of the location
we're interested in. One of the lists needs reversing because the input data is
backwards (so that modifications take place at the end of the SmallVector), but
we were reversing both before leading to incorrect results.
Keno Fischer [Mon, 4 May 2015 20:03:01 +0000 (20:03 +0000)]
Respect object format choice on Darwin
Summary:
The object format can be set to something other than MachO, e.g.
to use ELF-on-Darwin for MCJIT. This already works on Windows, so
there's no reason it shouldn't on Darwin.
Ulrich Weigand [Mon, 4 May 2015 17:41:22 +0000 (17:41 +0000)]
[SystemZ] Reclassify f32 subregs of f64 registers
At the moment, all subregs defined by the SystemZ target can be modified
independently of the wider register. E.g. writing to a GR32 does not
change the upper 32 bits of the GR64. Writing to an FP32 does not change
the lower 32 bits of the FP64.
Hoewver, the upcoming support for the vector extension redefines FP64 as
one half of a V128. Floating-point operations leave the other half of
a V128 in an unpredictable state, so it's no longer the case that writing
to an FP32 leaves the bits of the underlying register (the V128) alone.
I'd prefer to have separate subreg_ names for this situation, so that
it's obvious at a glance whether we're talking about a subreg that leaves
the other parts of the register alone.
Ulrich Weigand [Mon, 4 May 2015 17:40:53 +0000 (17:40 +0000)]
[SystemZ] Clean up AsmParser isMem() handling
We know what MemoryKind an operand has at the time we construct it,
so we might as well just record it in an unused part of the structure.
This makes it easier to add scatter/gather addresses later.
Ulrich Weigand [Mon, 4 May 2015 17:39:40 +0000 (17:39 +0000)]
[SystemZ] Fix getTargetNodeName
It seems SystemZTargetLowering::getTargetNodeName got out of sync with
some recent changes to the SystemZISD opcode list. Add back all the
missing opcodes (and re-sort to the same order as SystemISelLowering.h).
where the kill flag was removed from the BUNDLE instruction, but not the t2ORRrr inside it. The verifier then thought that
R0 was undefined when read by the AND.
This change make the toggleKillFlags method also check for bundles and toggle flags on bundled instructions.
Setting the kill flag is special cased as we only want to set the kill flag on the last instruction in the bundle.
AVX-512: added calling convention for i1 vectors in 32-bit mode.
Fixed some bugs in extend/truncate for AVX-512 target.
Removed VBROADCASTM (masked broadcast) node, since it is not used any more.
Yaron Keren [Mon, 4 May 2015 04:48:10 +0000 (04:48 +0000)]
Replace windows_error calls with mapWindowsError.
After r210687, windows_error does nothing but call mapWindowsError.
Other Windows/*.inc files directly call mapWindowsError. This patch
updates Path.inc and Process.inc to do the same.
This is a followup from:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20150323/268067.html
Upgrade instructions:
$ mv llvm/include/llvm/Config/config.h ./config.h.BACKUP
# copy the configure line from line 7 of llvm/config.log
# (for example: `$ ./configure --no-create --no-recursion`)
$ mkdir build
$ cd build
# run the configure line, but this time with '../llvm' at the beginning:
$ ../llvm/configure --no-create --no-recursion
These warnings will soon be turned into hard errors after a week. Speak up now
if this is going to be a problem for you.
Zachary Turner [Fri, 1 May 2015 20:24:26 +0000 (20:24 +0000)]
[llvm-pdbdump] Support dynamic load address and external symbols.
This patch adds the --load-address command line option to
llvm-pdbdump, which dumps all addresses assuming the module has
loaded at the specified address.
Additionally, this patch adds an option to llvm-pdbdump to support
dumping of public symbols (i.e. symbols with external linkage).
Keno Fischer [Fri, 1 May 2015 20:21:45 +0000 (20:21 +0000)]
Reapply [RuntimeDyldELF] Fold Placeholder into Addend
This reapplies r235060 and 235070, which were reverted because of test failures
in LLDB. The failure was caused because at moment RuntimeDyld is processing
relocations for all sections, irrespective of whether we actually load them
into memory or not, but RuntimeDyld was not actually remembering where in memory
the unrelocated section is. This commit includes a fix for that issue by
remembering that pointer, though the longer term fix should be to stop processing
unneeded sections.
Original Summary:
This allows us to get rid of the original unrelocated object file after
we're done processing relocations (but before applying them).
MachO and COFF already do not require this (currently we have temporary hacks
to prevent ownership from being released, but those are brittle and should be
removed soon).
The placeholder mechanism allowed the relocation resolver to look at original
object file to obtain more information that are required to apply the
relocations. This is usually necessary in two cases:
- For relocations targetting sub-word memory locations, there may be pieces
of the instruction at the target address which we should not override.
- Some relocations on some platforms allow an extra addend to be encoded in
their immediate fields.
The problem is that in the second case the information cannot be recovered
after the relocations have been applied once because they will have been
overridden. In the first case we also need to be careful to not use any bits
that aren't fixed and may have been overriden by applying a first relocation.
In the past both have been fixed by just looking at original object file. This
patch attempts to recover the information from the first by looking at the
relocated object file, while the extra addend in the second case is read
upon relocation processing and addend to the regular addend.
I have tested this on X86. Other platforms represent my best understanding
of how those relocations should work, but I may have missed something because
I do not have access to those platforms.
We will keep the ugly workarounds in place for a couple of days, so this commit
can be reverted if it breaks the bots.
Reid Kleckner [Fri, 1 May 2015 20:04:54 +0000 (20:04 +0000)]
[WinEH] Add an EH registration and state insertion pass for 32-bit x86
This pass is responsible for constructing the EH registration object
that gets linked into fs:00, which is all it does in this change. In the
future, it will also insert stores to update the EH state number.
I considered keeping this functionality in WinEHPrepare, but it's pretty
separable and X86 specific. It has conceptually very little to do with
the task of WinEHPrepare, which is currently outlining. WinEHPrepare is
also in theory useful on ARM, but this logic is pretty x86 specific.
Pete Cooper [Fri, 1 May 2015 18:57:32 +0000 (18:57 +0000)]
[ARM] Transfer the internal flag in thumb2 size reduction.
Converting from t2LDRs to tLDRr caused the shift argument to drop the internal flag. This would then throw machine verifier errors.
Unfortunately i'm having trouble reducing a test case. I'm going to keep trying, but so far its a scary combination of machine sinking, an 'and i1', loads feeding loads, and a bunch of code which shouldn't change IT block formation, but does. Its not useful to commit a test in that state as we have no way of knowing if it even hits this code reliably in future.
Benjamin Kramer [Fri, 1 May 2015 15:16:11 +0000 (15:16 +0000)]
Remove std::move on return when it could prevent copy elision.
Found by -Wpessimizing-move, no functional change. The APFloat and
PassManager change doesn't affect codegen as returning a by-value
argument will always result in a move.
Toma Tabacu [Fri, 1 May 2015 12:19:27 +0000 (12:19 +0000)]
[mips] [IAS] Fix error messages for using LI with 64-bit immediates.
Summary:
LI should never accept immediates larger than 32 bits.
The additional Is32BitImm boolean also paves the way for unifying the functionality that LA and LI have in common.
Toma Tabacu [Fri, 1 May 2015 10:26:47 +0000 (10:26 +0000)]
[mips] [IAS] Slightly improve shift instruction generation in expandLoadImm.
Summary:
Generate one DSLL32 of 0 instead of two consecutive DSLL of 16.
In order to do this I had to change createLShiftOri's template argument from a bool to an unsigned.
This also gave me the opportunity to rewrite the mips64-expansions.s test, as it was testing the same cases multiple times and skipping over other cases.
It was also somewhat unreadable, as the CHECK lines were grouped in a huge block of text at the beginning of the file.
This patch fixes issues with vector constant folding not correctly handling scalar input operands if they require implicit truncation - this was tested with llvm-stress as recommended by Patrik H Hagglund.
The patch ensures that integer input scalars from a build vector are correctly truncated before folding, and that constant integer scalar results are promoted to a legal type before inclusion in the new folded build vector.
I have added another crash test case and also a test for UINT_TO_FP / SINT_TO_FP using an non-truncated scalar input, which was failing before this patch.
Tom Stellard [Fri, 1 May 2015 03:44:08 +0000 (03:44 +0000)]
R600/SI: Fix verifier errors from the SIAnnotateControlFlow pass
This pass was generating 'Instruction does not dominate all uses!'
errors for programs which had loops with a condition variable that
depended on the result of a phi instruction from outside of the loop.
The pass was inserting new phi nodes outside of the loop which used values
defined inside the loop.
Pete Cooper [Thu, 30 Apr 2015 23:57:47 +0000 (23:57 +0000)]
[ARM] optimizeSelect should clear kill flags.
If we move an instruction from one block down to a MOVC and predicate it,
then the original instruction could be moved in to a loop. In this case,
its invalid for any kill flags to remain on there.
InstrProf: Instrumenter support for setting profile output from command line
This change is the second of 3 patches to add support for specifying
the profile output from the command line via -fprofile-instr-generate=<path>,
where the specified output path/file will be overridden by the
LLVM_PROFILE_FILE environment variable.
This patch adds the necessary support to the llvm instrumenter, specifically
a new member of GCOVOptions for clang to save the specified filename, and
support for calling the new compiler-rt interface from __llvm_profile_init.
Pete Cooper [Thu, 30 Apr 2015 23:14:14 +0000 (23:14 +0000)]
Commute the internal flag on MachineOperands.
When commuting a thumb instruction in the size reduction pass, thumb
instructions are represented as a bundle and so some operands may be marked
as internal. The internal flag has to move with the operand when commuting.
This test is sensitive to register allocation so can't specifically check that
this error was happening, but so long as it continues to pass with -verify then
hopefully its still ok.
Pete Cooper [Thu, 30 Apr 2015 22:15:59 +0000 (22:15 +0000)]
Don't always apply kill flag in thumb2 ABS pseudo expansion.
The expansion for t2ABS was always setting the kill flag on the rsb instruction.
It should instead only be set on rsb if it was set on the original ABS instruction.
[X86] Use 4 byte preferred aggregate alignment on Win32
This helps reduce the frequency of stack realignment prologues in 32-bit
X86 Windows code. Before this change and the corresponding clang change,
we would take the max of the type preferred alignment and the explicit
alignment on the alloca.
If you don't override aggregate alignment in datalayout, you get a
default of 8. This dates back to 2007 / r34356, and changing it seems
prohibitively difficult at this point.
InstCombineSimplifyDemanded: Remove nsw/nuw flags when optimizing demanded bits
When optimizing demanded bits of the operands of an Add we have to
remove the nsw/nuw flags as we have no guarantee anymore that we don't
wrap. This is legal here because the top bit is not demanded. In fact
this operaion was already performed but missed in the case of an Add
with a constant on the right side. To fix this this patch refactors the
code to unify the code paths in SimplifyDemandedUseBits() handling of
Add/Sub:
- The transformation of Add->Or is removed from the simplify demand
code because the equivalent transformation exists in
InstCombiner::visitAdd()
- KnownOnes/KnownZero are not adjusted for Add x, C anymore as
computeKnownBits() already performs these computations.
- The simplification of the operands is unified. In this new version
constant on the right side of a Sub are shrunk now as I could not find
a reason why not to do so.
- The special case for clearing nsw/nuw in ShrinkDemandedConstant() is
not necessary anymore as the caller does that already.
InstCombine: Move Sub->Xor rule from SimplifyDemanded to InstCombine
The rule that turns a sub to xor if the LHS is 2^n-1 and the remaining bits
are known zero, does not use the demanded bits at all: Move it to the
normal InstCombine code path.