Summary:
This patch adds a generic MacroFusion pass, that is used on X86 and
AArch64, which both define target-specific shouldScheduleAdjacent
functions. This generic pass should make it easier for other targets to
implement macro fusion and I intend to add macro fusion for ARM shortly.
Reviewers: craig.topper, evandro, t.p.northover, atrick, MatzeB
Reviewed By: MatzeB
Subscribers: atrick, aemerson, mgorny, javed.absar, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D34144
llvm-svn: 305677
Add support throughout the pipeline:
- mark as legal for s32 and pointers
- map to GPRs
- lower to a sequence of instructions, which moves 0 or 1 into the
result register based on the flags set by a CMPrr
We have copied from FastISel a helper function which maps CmpInst
predicates into ARMCC codes. Ideally, we should be able to move it
somewhere that both FastISel and GlobalISel can use.
llvm-svn: 305672
Current implementation of SCEVExpander demonstrates a very naive behavior when
it deals with power calculation. For example, a SCEV for x^8 looks like
(x * x * x * x * x * x * x * x)
If we try to expand it, it generates a very straightforward sequence of muls, like:
x2 = mul x, x
x3 = mul x2, x
x4 = mul x3, x
...
x8 = mul x7, x
This is a non-efficient way of doing that. A better way is to generate a sequence of
binary power calculation. In this case the expanded calculation will look like:
x2 = mul x, x
x4 = mul x2, x2
x8 = mul x4, x4
In some cases the code size reduction for such SCEVs is dramatic. If we had a loop:
x = a;
for (int i = 0; i < 3; i++)
x = x * x;
And this loop have been fully unrolled, we have something like:
x = a;
x2 = x * x;
x4 = x2 * x2;
x8 = x4 * x4;
The SCEV for x8 is the same as in example above, and if we for some reason
want to expand it, we will generate naively 7 multiplications instead of 3.
The BinPow expansion algorithm here allows to keep code size reasonable.
This patch teaches SCEV Expander to generate a sequence of BinPow multiplications
if we have repeating arguments in SCEVMulExpressions.
Differential Revision: https://reviews.llvm.org/D34025
llvm-svn: 305663
Merge the functionality into the random access type collection.
This class was only being used in 2 places, so getting rid of it
simplifies the code.
llvm-svn: 305653
Summary:
This allows strlen to be moved out of the loop in case its argument is
not modified in the loop in LICM.
Reviewers: hfinkel, davide, sanjoy, dberlin
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34323
llvm-svn: 305641
No behavior is changed if LLVM_TARGET_TRIPLE_ENV is blank or undefined.
If LLVM_TARGET_TRIPLE_ENV is "TEST_TARGET_TRIPLE" and $TEST_TARGET_TRIPLE is not blank,
llvm::sys::getDefaultTargetTriple() returns $TEST_TARGET_TRIPLE.
Lit resets config.target_triple and config.environment[LLVM_TARGET_TRIPLE_ENV] to change the default target.
Without changing LLVM_DEFAULT_TARGET_TRIPLE nor rebuilding, lit can be run;
TEST_TARGET_TRIPLE=i686-pc-win32 bin/llvm-lit -sv path/to/test/
TEST_TARGET_TRIPLE=i686-pc-win32 ninja check-clang-tools
Differential Revision: https://reviews.llvm.org/D33662
llvm-svn: 305632
Re-apply r276044/r279124/r305516. Fixed a problem where we would refuse
to place spills as the very first instruciton of a basic block and thus
artifically increase pressure (test in
test/CodeGen/PowerPC/scavenging.mir:spill_at_begin)
This is a variant of scavengeRegister() that works for
enterBasicBlockEnd()/backward(). The benefit of the backward mode is
that it is not affected by incomplete kill flags.
This patch also changes
PrologEpilogInserter::doScavengeFrameVirtualRegs() to use the register
scavenger in backwards mode.
Differential Revision: http://reviews.llvm.org/D21885
llvm-svn: 305625
Summary:
This is my misunderstanding on isBarrier. It's not for memory barriers,
but for other control flow purposes. lwsync doesn't have it either.
This fixes a simple crash with -verify-machineinstrs like below:
define void @Foo() {
entry:
%tmp = load atomic i64, i64* undef acquire, align 8
unreachable
}
I deliberately don't want to check in the test, since there is little
chance to regress on such a mistake. Such a test adds noise to the code
base.
I plan to check in first, since it fixes a crash, and the fix is obvious.
Reviewers: kbarton, echristo
Subscribers: sanjoy, nemanjai, hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D34314
llvm-svn: 305624
This ensures that symbolic relocations are generated for stack
pointer manipulations.
These relocations are of type R_WEBASSEMBLY_GLOBAL_INDEX_LEB.
This change also adds support for reading relocations of this
type in WasmObjectFile.cpp.
Since its a globally imported symbol this does mean that
the get_global/set_global instruction won't be valid until
the objects are linked that global used in no longer an
imported global.
Differential Revision: https://reviews.llvm.org/D34172
llvm-svn: 305616
Suppose we had a type index offsets array with a boundary at type index
N. Then you request the name of the type with index N+1, and that name
requires the name of index N-1 (think a parameter list, for example). We
didn't handle this, and we would print something like (<unknown UDT>,
<unknown UDT>).
The fix for this is not entirely trivial, and speaks to a larger
problem. I think we need to kill TypeDatabase, or at the very least kill
TypeDatabaseVisitor. We need a thing that doesn't do any caching
whatsoever, just given a type index it can compute the type name "the
slow way". The reason for the bug is that we don't have anything like
that. Everything goes through the type database, and if we've visited a
record, then we're "done". It doesn't know how to do the expensive thing
of re-visiting dependent records if they've not yet been visited.
What I've done here is more or less copied the code (albeit greatly
simplified) from TypeDatabaseVisitor, but wrapped it in an interface
that just returns a std::string. The logic of caching the name is now in
LazyRandomTypeCollection. Eventually I'd like to move the record
database here as well and the visited record bitfield here as well, at
which point we can actually just delete TypeDatabase. I don't see any
reason for it if a "sequential" collection is just a special case of a
random access collection with an empty partial offsets array.
Differential Revision: https://reviews.llvm.org/D34297
llvm-svn: 305612
Previously only the error codes were reported which
meant that useful information about malformed inputs
was not shown.
Differential Revision: https://reviews.llvm.org/D34008
llvm-svn: 305609
Davide Italiano reported the following issue if llvm
is compiled with gcc -Wstrict-aliasing -Werror:
.....
lib/Target/BPF/CMakeFiles/LLVMBPFCodeGen.dir/BPFISelDAGToDAG.cpp.o
../lib/Target/BPF/BPFISelDAGToDAG.cpp: In member function ‘virtual
void {anonymous}::BPFDAGToDAGISel::PreprocessISelDAG()’:
../lib/Target/BPF/BPFISelDAGToDAG.cpp:264:26: warning: dereferencing
type-punned pointer will break strict-aliasing rules
[-Wstrict-aliasing]
val = *(uint16_t *)new_val;
.....
The error is caused by my previous commit (revision 305560).
This patch fixed the issue by introducing an union to avoid
type casting.
Signed-off-by: Yonghong Song <yhs@fb.com>
llvm-svn: 305608
Summary: As far as I can tell we should be able to implement these almost the same way we do unsigned, but using signed comparisons and checks for min signed value instead of min unsigned value.
Reviewers: pete, davide, sanjoy
Reviewed By: davide
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D33815
llvm-svn: 305607
For the following motivating example
bool c();
void f();
bool start() {
bool result = c();
if (!c()) {
result = false;
goto exit;
}
f();
result = true;
exit:
return result;
}
we would previously generate a single DW_AT_const_value(1) because
only the DBG_VALUE in the second-to-last basic block survived
codegen. This patch improves the heuristic used to determine when a
DBG_VALUE is available at the beginning of its variable's enclosing
lexical scope:
- Stop giving singular constants blanket permission to take over the
entire scope. There is still a special case for constants in the
function prologue that we also miight want to retire later.
- Use the lexical scope information to determine available-at-entry
instead of proximity to the function prologue.
After this patch we generate a location list with a more accurate
narrower availability for the constant true value. As a pleasant side
effect, we also generate inline locations instead of location lists
where a loacation covers the entire range of the enclosing lexical
scope.
Measured on compiling llc with four targets this doesn't have an
effect on compile time and reduces the size of the debug info for llc
by ~600K.
rdar://problem/30286912
llvm-svn: 305599
The verifier should not output any message in such a case.
Added test case with no .apple_name section in the file to verify new functionality.
Made existing test case more specific.
llvm-svn: 305597
In this patch, I flip the switch in DriverUtils from using the external
cvtres.exe tool to using the Windows Resource library in llvm.
I also fixed a bug where .rsrc sections were marked as discardable
memory and therefore were placed in the wrong order in the final PE.
Furthermore, I modified WindowsResource to write the coff directly to a
memory buffer instead of to file, also had it use the machine types
already declared in COFF.h instead creating my own enum.
Finally, I flipped the switch to allow all unit tests that had
previously run only on windows due to a winres dependency to run
cross-platform.
Reviewers: zturner, ruiu
Subscribers: llvm-commits, hiraditya
Differential Revision: https://reviews.llvm.org/D34265
llvm-svn: 305592
Summary:
When we fold vector constants that are operands of phi's that feed into select,
we need to set the correct insertion point for the *new* selects that get generated.
The correct insertion point is the incoming block for the phi.
Such cases can occur with patch r298845, which fixed folding of
vector constants, but the new selects could be inserted incorrectly (as the added
test case shows).
Reviewers: majnemer, spatel, sanjoy
Reviewed by: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34162
llvm-svn: 305591
The recommit fixes two bugs: The first one is to use CurrentBlock instead of
PREInstr's Parent as param of performScalarPREInsertion because the Parent
of a clone instruction may be uninitialized. The second one is stop PRE when
CurrentBlock to its predecessor is a backedge and an operand of CurInst is
defined inside of CurrentBlock. The same value defined inside of loop in last
iteration can not be regarded as available.
Right now scalarpre doesn't have phi-translate support, so it will miss some
simple pre opportunities. Like the following testcase, current scalarpre cannot
recognize the last "a * b" is fully redundent because a and b used by the last
"a * b" expr are both defined by phis.
long a[100], b[100], g1, g2, g3;
__attribute__((pure)) long goo();
void foo(long a, long b, long c, long d) {
g1 = a * b;
if (__builtin_expect(g2 > 3, 0)) {
a = c;
b = d;
g2 = a * b;
}
g3 = a * b; // fully redundant.
}
The patch adds phi-translate support in scalarpre. This is only a temporary
solution before the newpre based on newgvn is available.
Differential Revision: https://reviews.llvm.org/D32252
llvm-svn: 305578
Revert because of reports of some PPC input starting to spill when it
was predicted that it wouldn't and no spillslot was reserved.
This reverts commit r305516.
llvm-svn: 305566
If users tried to have a structure decl/init code like below
struct test_t t = { .memeber1 = 45 };
It is very likely that compiler will generate a readonly section
to hold up the init values for variable t. Later load of t members,
e.g., t.member1 will result in a read from readonly section.
BPF program cannot handle relocation. This will force users to
write:
struct test_t t = {};
t.member1 = 45;
This is just inconvenient and unintuitive.
This patch addresses this issue by implementing BPF PreprocessISelDAG.
For any load from a global constant structure or an global array of
constant struct, it attempts to
translate it into a constant directly. The traversal of the
constant struct and other constant data structures are similar
to where the assembler emits read-only sections.
Four different unit test cases are also added to cover
different scenarios.
Signed-off-by: Yonghong Song <yhs@fb.com>
llvm-svn: 305560
o This is discovered during my study of 32-bit subregister
support.
o This is no impact on current functionality since we
only support 64-bit registers.
o Searching the web, looks like the issue has been discovered
before, so fix it now.
Signed-off-by: Yonghong Song <yhs@fb.com>
llvm-svn: 305559
Summary:
Background: http://lists.llvm.org/pipermail/llvm-dev/2017-May/112779.html
This change is to alter the prototype for the atomic memcpy intrinsic. The prototype itself is being changed to more closely resemble the semantics and parameters of the llvm.memcpy intrinsic -- to ease later combination of the llvm.memcpy and atomic memcpy intrinsics. Furthermore, the name of the atomic memcpy intrinsic is being changed to make it clear that it is not a generic atomic memcpy, but specifically a memcpy is unordered atomic.
Reviewers: reames, sanjoy, efriedma
Reviewed By: reames
Subscribers: mzolotukhin, anna, llvm-commits, skatkov
Differential Revision: https://reviews.llvm.org/D33240
llvm-svn: 305558
This reverts commit r305455. This commit was reported as breaking one of
the sanitizer buildbots. Reverting until lab.llvm.org comes back online.
llvm-svn: 305557
The variant generation for commutative/associative patterns would simply
delete the first output from the list assuming that it was identical to
the original pattern. This does not have to be the case, and a legitimate
variant could actually be removed that way.
llvm-svn: 305556
The second part of r305300: when placing the mux at the later location,
make sure that it won't use any register that was killed between the
two original instructions. Remove any such kills and transfer them to
the mux.
llvm-svn: 305553
- Topologocal is abbreviated as "topo" in comments, but "top" is used in only one comment. Modify it for consistency.
- Capitalize "succ" and "pred" for consistency in one figure.
- Other trivial fixes.
llvm-svn: 305552
Summary: This is the demorganed version of the case we already handle for the OR of iszero.
Reviewers: spatel
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34244
llvm-svn: 305548
This patch should fix sanitizer-x86_64-linux-fast bot.
The problem was that the contents of this stream are aligned to 4 byte,
and the paddings were created just by incrementing `Offset`, so paddings
had undefined values. When the entire stream is written to an output,
it triggered msan.
llvm-svn: 305541
Summary: We were using the system compiler to run the test suite.
Reviewers: hansw
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34246
llvm-svn: 305525
This resubmits commit c0c249e9f2ef83e1d1e5f166b50673d92f3579d7.
It was broken due to some weird template issues, which have
since been fixed.
llvm-svn: 305517
Re-apply r276044/r279124. Trying to reproduce or disprove the ppc64
problems reported in the stage2 build last time, which I cannot
reproduce right now.
This is a variant of scavengeRegister() that works for
enterBasicBlockEnd()/backward(). The benefit of the backward mode is
that it is not affected by incomplete kill flags.
This patch also changes
PrologEpilogInserter::doScavengeFrameVirtualRegs() to use the register
scavenger in backwards mode.
Differential Revision: http://reviews.llvm.org/D21885
llvm-svn: 305516
This reverts commit 83ea17ebf2106859a51fbc2a86031b44d33696ad.
This is failing due to some strange template problems, so reverting
until it can be straightened out.
llvm-svn: 305505
Summary:
Split the PGOMemOPSizeOpt pass out from IndirectCallPromotion.cpp into
its own file.
Reviewers: davidxl
Subscribers: mgorny, llvm-commits
Differential Revision: https://reviews.llvm.org/D34248
llvm-svn: 305501
After some internal discussions, we agreed that the raw output style had
outlived its usefulness. It was originally created before we had even
thought of dumping to YAML, and it was intended to give us some insight
into the internals of a PDB file. Now we have YAML mode which does
almost exactly this but is more powerful in that it can round-trip back
to a PDB, which the raw mode could not do. So the raw mode had become
purely a maintenance burden.
One option was to just delete it. However, its original goal was to be
as readable as possible while staying close to the "metal" - i.e.
presenting the output in a way that maps directly to the underlying file
format. We don't actually need that last requirement anymore since it's
covered by the yaml mode, so we could repurpose "raw" mode to actually
just be as readable as possible.
This patch implements about 80% of the functionality previously in raw
mode, but in a completely different style that is more akin to what
cvdump outputs. Records are very compressed, often times appearing on
just one line. One nice thing about this is that it makes full record
matching easier, because you can grep for indices, names, and leaf types
on a single line often.
See the tests for some examples of what the new output looks like.
Note that this patch actually regresses the functionality of raw mode in
a few areas, but only because the patch was already unreasonably large
and going 100% would have been even worse. Specifically, this patch is
missing:
The ability to dump module debug subsections (checksums, lines, etc)
The ability to dump section headers
Aside from that everything is here. While goign through the tests fixing
them all up, I found many duplicate tests. They've been deleted. In
subsequent patches I will go through and re-add the missing
functionality.
Differential Revision: https://reviews.llvm.org/D34191
llvm-svn: 305495
Add condition for MachineLICM to safely hoist instructions that utilize
non constant registers that are reserved.
On PPC, global variable access is done through the table of contents (TOC)
which is always in register X2. The ABI reserves this register in any
functions that have calls or access global variables.
A call through a function pointer involves saving, changing and restoring
this register around the call and thus MachineLICM does not consider it to
be invariant. We can however guarantee the register is preserved across the
call and thus is invariant.
Differential Revision: https://reviews.llvm.org/D33562
llvm-svn: 305490
Currently we expect A to be on the same side in both Ands but nothing guarantees that.
While there also switch to using matchers for some of the code.
Differential Revision: https://reviews.llvm.org/D34230
llvm-svn: 305487
The code assumed that we process instructions in basic block order. FastISel
processes instructions in reverse basic block order. We need to pre-assign
virtual registers before selecting otherwise we get def-use relationships wrong.
This only affects code with swifterror registers.
rdar://32659327
llvm-svn: 305484
If a regular LTO module has a summary index, then instead of linking
it into the combined regular LTO module right away, add it to the
combined summary index and associate it with a special module that
represents the combined regular LTO module.
Any such modules are linked during LTO::run(), at which time we use
the results of summary-based dead stripping to control whether to
link prevailing symbols.
Differential Revision: https://reviews.llvm.org/D33922
llvm-svn: 305482
This is a fix for the test case in PR32314.
Basic Alias Analysis can ask if two nodes are known non-equal after looking through a phi node to find a GEP. isAddOfNonZero saw an add of a constant from the same phi and said that its output couldn't be equal. But Basic Alias Analysis was really asking about the value from the previous loop iteration.
This patch at least makes that case not happen anymore, I'm not sure if there were still other ways this can fail. As was discussed in the bug, it looks like fixing BasicAA would be difficult so this patch seemed like a possible workaround
Differential Revision: https://reviews.llvm.org/D33136
llvm-svn: 305481
This patch fixes a potential verification error (64-bit register operands for cmpw) with -verify-machineinstrs.
Differential Revision: https://reviews.llvm.org/D34208
llvm-svn: 305479
Summary:
Points to existing documentation for branch_weights and
function_entry_count, and adds an example for VP value profile metadata.
Reviewers: davidxl, reames
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D34218
llvm-svn: 305475
In preparation for doing storemerge post-legalization, reorder
visitSTORE passes to move pre/post-index combining after store
merge. Reordered passes other than store merge are unaffected.
llvm-svn: 305473
As all store merges checks are based on the memory operation
performed, allow use of truncated stores and extended loads as valid
input candidates for merging.
llvm-svn: 305468
AVX512 compare instructions return v*i1 types.
In cases where the number of elements in the returned value are less than 8, clang adds zeroes to get a mask of v8i1 type.
Later on it's replaced with CONCAT_VECTORS, which then is lowered to many DAG nodes including insert/extract element and shift right/left nodes.
The fact that AVX512 compare instructions put the result in a k register and zeroes all its upper bits allows us to remove the extra nodes simply by copying the result to the required register class.
When lowering, identify these cases and transform them into an INSERT_SUBVECTOR node (marked legal), then catch this pattern in instructions selection phase and transform it into one avx512 cmp instruction.
Differential Revision: https://reviews.llvm.org/D33188
llvm-svn: 305465
This is a fix for PR33292 that shows a case of extremely long compilation
of a single .c file with clang, with most time spent within SCEV.
We have a mechanism of limiting recursion depth for getAddExpr to avoid
long analysis in SCEV. However, there are calls from getAddExpr to getMulExpr
and back that do not propagate the info about depth. As result of this, a chain
getAddExpr -> ... .> getAddExpr -> getMulExpr -> getAddExpr -> ... -> getAddExpr
can be extremely long, with every segment of getAddExpr's being up to max depth long.
This leads either to long compilation or crash by stack overflow. We face this situation while
analyzing big SCEVs in the test of PR33292.
This patch applies the same limit on max expression depth for getAddExpr and getMulExpr.
Differential Revision: https://reviews.llvm.org/D33984
llvm-svn: 305463
Add support for modulo for targets that have hardware division and for
those that don't. When hardware division is not available, we have to
choose the correct libcall to use. This is generally straightforward,
except for AEABI.
The AEABI variant is trickier than the other libcalls because it
returns { quotient, remainder }, instead of just one value like the
other libcalls that we've seen so far. Therefore, we need to use custom
lowering for it. However, we don't want to have too much special code,
so we refactor the target-independent code in the legalizer by adding a
helper for replacing an instruction with a libcall. This helper is used
by the legalizer itself when dealing with simple calls, and also by the
custom ARM legalization for the more complicated AEABI divmod calls.
llvm-svn: 305459
Lowering mixed struct args, params and returns used G_INSERT, which is a
bit more convoluted to support through the entire pipeline. Since they
don't occur that often in practice, it's probably wiser to leave them
out until later.
Meanwhile, we can lower homogeneous structs using G_MERGE_VALUES, which
has good support in the legalizer. These occur e.g. as the return of
__aeabi_idivmod, so it's nice to be able to support them.
llvm-svn: 305458
Summary:
Scheduling AESE/AESMC and AESD/AESIMC instruction pairs back-to-back
gives a double digit speedup on benchmarks using those instructions on
Cortex-A processors. In GCC, this optimization is part of the generic
processor model as well.
This change should not have a major performance impact on processors
that do not optimize AES instruction pairs, although I only had access
to Cortex-A processors for benchmarking.
Reviewers: rengolin, kristof.beyls, javed.absar, evandro, silviu.baranga, MatzeB, mcrosier, joelkevinjones, joel_k_jones, bmakam, t.p.northover
Reviewed By: evandro
Subscribers: sbaranga, aemerson, llvm-commits
Differential Revision: https://reviews.llvm.org/D33836
llvm-svn: 305457
Author: milena.vujosevic.janicic
Reviewers: sdardis
The patch extends size reduction pass for MicroMIPS.
The following instructions are examined and transformed, if possible:
ADDIU instruction is transformed into 16-bit instruction ADDIUSP
ADDIU instruction is transformed into 16-bit instruction ADDIUR1SP
Differential Revision: https://reviews.llvm.org/D33887
llvm-svn: 305455
Previously if you used fmt_align(7, Center) you would get the
output ' 7 '. It may be desirable for the user to specify
the fill character though, for example producing '---7---'. This
patch adds that.
llvm-svn: 305449
The current name (addModulePath) and return value
(ModulePathStringTableTy::iterator) is a little confusing. This
API adds a module, not just a path. And the iterator is basically
just an implementation detail of the summary index. Address
both of those issues by renaming to addModule and introducing a
ModuleSummaryIndex::ModuleInfo type that the function returns.
Differential Revision: https://reviews.llvm.org/D34124
llvm-svn: 305422
Instead use target_link_libraries directly. Thanks to
Juergen Ributzka for the suggestion, which fixes an issue
when llvm is configured with no targets.
llvm-svn: 305421
Summary:
At present, `-profile-guided-section-prefix` is a `cl::Optional` option, which means it demands to be passed exactly zero or one times. Our build system makes it pretty tricky to guarantee this. We often accidentally pass the flag more than once (but always with the same "false" value) which results in an error, after which compilation fails:
```
clang (LLVM option parsing): for the -profile-guided-section-prefix option: may only occur zero or one times!
```
While we work on improving our build system, it also seems reasonable just to allow `-profile-guided-section-prefix` to be passed more than once, by to `cl::ZeroOrMore`. Quoting [[ http://llvm.org/docs/CommandLine.html#controlling-the-number-of-occurrences-required-and-allowed | the documentation ]]:
> The cl::ZeroOrMore modifier ... indicates that your program will allow the option to be specified zero or more times.
> ...
> If an option is specified multiple times for an option of the cl::opt class, only the last value will be retained.
Reviewers: danielcdh
Reviewed By: danielcdh
Subscribers: twoh, david2050, llvm-commits
Differential Revision: https://reviews.llvm.org/D34219
llvm-svn: 305413
This way we end up not looking at PHI args already removed.
MemSSA now goes through the updater so we can prune
it to avoid having redundant MemoryPHI arguments, but that
doesn't quite work for the general case.
Discussed with Daniel Berlin, fixes PR33406.
llvm-svn: 305409
Florian Hahn