Current SCEV expansion will expand SCEV as a sequence of operations
and doesn't utilize the value already existed. This will introduce
redundent computation which may not be cleaned up throughly by
following optimizations.
This patch introduces an ExprValueMap which is a map from SCEV to the
set of equal values with the same SCEV. When a SCEV is expanded, the
set of values is checked and reused whenever possible before generating
a sequence of operations.
The original commit triggered regressions in Polly tests. The regressions
exposed two problems which have been fixed in current version.
1. Polly will generate a new function based on the old one. To generate an
instruction for the new function, it builds SCEV for the old instruction,
applies some tranformation on the SCEV generated, then expands the transformed
SCEV and insert the expanded value into new function. Because SCEV expansion
may reuse value cached in ExprValueMap, the value in old function may be
inserted into new function, which is wrong.
In SCEVExpander::expand, there is a logic to check the cached value to
be used should dominate the insertion point. However, for the above
case, the check always passes. That is because the insertion point is
in a new function, which is unreachable from the old function. However
for unreachable node, DominatorTreeBase::dominates thinks it will be
dominated by any other node.
The fix is to simply add a check that the cached value to be used in
expansion should be in the same function as the insertion point instruction.
2. When the SCEV is of scConstant type, expanding it directly is cheaper than
reusing a normal value cached. Although in the cached value set in ExprValueMap,
there is a Constant type value, but it is not easy to find it out -- the cached
Value set is not sorted according to the potential cost. Existing reuse logic
in SCEVExpander::expand simply chooses the first legal element from the cached
value set.
The fix is that when the SCEV is of scConstant type, don't try the reuse
logic. simply expand it.
Differential Revision: http://reviews.llvm.org/D12090
llvm-svn: 259736
Current SCEV expansion will expand SCEV as a sequence of operations
and doesn't utilize the value already existed. This will introduce
redundent computation which may not be cleaned up throughly by
following optimizations.
This patch introduces an ExprValueMap which is a map from SCEV to the
set of equal values with the same SCEV. When a SCEV is expanded, the
set of values is checked and reused whenever possible before generating
a sequence of operations.
Differential Revision: http://reviews.llvm.org/D12090
llvm-svn: 259662
This regresses a test in LoopVectorize, so I'll need to go away and think about how to solve this in a way that isn't broken.
From the writeup in PR26071:
What's happening is that ComputeKnownZeroes is telling us that all bits except the LSB are zero. We're then deciding that only the LSB needs to be demanded from the icmp's inputs.
This is where we're wrong - we're assuming that after simplification the bits that were known zero will continue to be known zero. But they're not - during trivialization the upper bits get changed (because an XOR isn't shrunk), so the icmp fails.
The fault is in demandedbits - its contract does clearly state that a non-demanded bit may either be zero or one.
llvm-svn: 259649
differentiate between indirect references to functions an direct calls.
This doesn't do a whole lot yet other than change the print out produced
by the analysis, but it lays the groundwork for a very major change I'm
working on next: teaching the call graph to actually be a call graph,
modeling *both* the indirect reference graph and the call graph
simultaneously. More details on that in the next patch though.
The rest of this is essentially a bunch of over-engineering that won't
be interesting until the next patch. But this also isolates essentially
all of the churn necessary to introduce the edge abstraction from the
very important behavior change necessary in order to separately model
the two graphs. So it should make review of the subsequent patch a bit
easier at the cost of making this patch seem poorly motivated. ;]
Differential Revision: http://reviews.llvm.org/D16038
llvm-svn: 259463
The computation of ICmp demanded bits is independent of the individual operand being evaluated. We simply return a mask consisting of the minimum leading zeroes of both operands.
We were incorrectly passing "I" to ComputeKnownBits - this should be "UserI->getOperand(0)". In cases where we were evaluating the 1th operand, we were taking the minimum leading zeroes of it and itself.
This should fix PR26266.
llvm-svn: 258690
Some patterns of select+compare allow us to know exactly the value of the uppermost bits in the select result. For example:
%b = icmp ugt i32 %a, 5
%c = select i1 %b, i32 2, i32 %a
Here we know that %c is bounded by 5, and therefore KnownZero = ~APInt(5).getActiveBits() = ~7.
There are several such patterns, and this patch attempts to understand a reasonable subset of them - namely when the base values are the same (as above), and when they are related by a simple (add nsw), for example (add nsw %a, 4) and %a.
llvm-svn: 257769
Summary:
Since globals may escape as function arguments (even when they have been
found to be non-escaping, because of optimizations such as memcpyoptimizer
that replaces stores with memcpy), all arguments to a function are checked
during query to make sure they are identifiable. At that time, also ensure
we return a conservative result only if the arguments don't alias to our global.
Reviewers: hfinkel, jmolloy
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D16140
llvm-svn: 257750
The early return seems to be missed. This causes a radical and wrong loop
optimization on powerpc. It isn't reproducible on x86_64, because
"UseInterleaved" is false.
Patch by Tim Shen.
llvm-svn: 257134
See PR25822 for a more full summary, but we were conflating the concepts of "capture" and "escape". We were proving nocapture and using that proof to infer noescape, which is not true. Escaped-ness is a function-local property - as soon as a value is used in a call argument it escapes. Capturedness is a related but distinct property. It implies a *temporally limited* escape. Consider:
static int a;
int b;
int g(int * nocapture arg);
int f() {
a = 2; // Even though a escapes to g, it is not captured so can be treated as non-escaping here.
g(&a); // But here it must be treated as escaping.
g(&b); // Now that g(&a) has returned we know it was not captured so we can treat it as non-escaping again.
}
The original commit did not sufficiently understand this nuance and so caused PR25822 and PR26046.
r248576 included both a performance improvement (which has been backed out) and a related conformance fix (which has been kept along with its testcase).
llvm-svn: 257058
Summary:
This reverts commit 5a9e526f29cf8510ab5c3d566fbdcf47ac24e1e9.
As per discussion in D15665
This also add a test case so that regression introduced by that diff are not reintroduced.
Reviewers: vaivaswatha, jmolloy, hfinkel, reames
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D15919
llvm-svn: 256932
Most of the properties of memset_pattern16 can be now covered by the generic attributes and inferred by InferFunctionAttrs. The only exceptions are:
- We don't yet have a writeonly attribute for the first argument.
- We don't have an attribute for modeling the access size facts encoded in MemoryLocation.cpp.
Differential Revision: http://reviews.llvm.org/D15879
llvm-svn: 256911
The cost is calculated for all X86 targets. When gather/scatter instruction
is not supported we calculate the cost of scalar sequence.
Differential revision: http://reviews.llvm.org/D15677
llvm-svn: 256519
Summary: This patch changes gc.statepoint intrinsic's return type to token type instead of i32 type. Using token types could prevent LLVM to merge different gc.statepoint nodes into PHI nodes and cause further problems with gc relocations. The patch also changes the way on how gc.relocate and gc.result look for their corresponding gc.statepoint on unwind path. The current implementation uses the selector value extracted from a { i8*, i32 } landingpad as a hook to find the gc.statepoint, while the patch directly uses a token type landingpad (http://reviews.llvm.org/D15405) to find the gc.statepoint.
Reviewers: sanjoy, JosephTremoulet, pgavlin, igor-laevsky, mjacob
Subscribers: reames, mjacob, sanjoy, llvm-commits
Differential Revision: http://reviews.llvm.org/D15662
llvm-svn: 256443
This patch removes all weight-related interfaces from BPI and replace
them by probability versions. With this patch, we won't use edge weight
anymore in either IR or MC passes. Edge probabilitiy is a better
representation in terms of CFG update and validation.
Differential revision: http://reviews.llvm.org/D15519
llvm-svn: 256263
This is recommit of r256028 with minor fixes in unittests:
CodeGen/Mips/eh.ll
CodeGen/Mips/insn-zero-size-bb.ll
Original commit message:
When identifying blocks post-dominated by an unreachable-terminated block
in BranchProbabilityInfo, consider only the edge to the normal destination
block if the terminator is InvokeInst and let calcInvokeHeuristics() decide
edge weights for the InvokeInst.
llvm-svn: 256202
This patch transforms truncation between vectors of integers into
X86ISD::PACKUS/PACKSS operations during DAG combine. We don't do it in
lowering phase because after type legalization, the original truncation
will be turned into a BUILD_VECTOR with each element that is extracted
from a vector and then truncated, and from them it is difficult to do
this optimization. This greatly improves the performance of truncations
on some specific types.
Cost table is updated accordingly.
Differential revision: http://reviews.llvm.org/D14588
llvm-svn: 256194
Summary:
The analysis of shader inputs was completely wrong. We were passing the
wrong index to AttributeSet::hasAttribute() and the logic for which
inputs where in SGPRs was wrong too.
Reviewers: arsenm
Subscribers: arsenm, llvm-commits
Differential Revision: http://reviews.llvm.org/D15608
llvm-svn: 256082
This patch removes all getEdgeWeight() interfaces from CodeGen directory. As
getEdgeProbability() is a little more expensive than getEdgeWeight(), I will
compose a patch soon in which BPI only stores probabilities instead of edge
weights so that getEdgeProbability() will have O(1) time.
Differential revision: http://reviews.llvm.org/D15489
llvm-svn: 256039
When identifying blocks post-dominated by an unreachable-terminated block
in BranchProbabilityInfo, consider only the edge to the normal destination
block if the terminator is InvokeInst and let calcInvokeHeuristics() decide
edge weights for the InvokeInst.
llvm-svn: 256028
Summary:
1. Modify AnalyzeCallGraph() to retain function info for external functions
if the function has [InaccessibleMemOr]ArgMemOnly flags.
2. When analyzing the use of a global is function parameter at a call site,
mark the callee also as modifying the global appropriately.
3. Add additional test cases.
Depends on D15499
Reviewers: hfinkel, jmolloy
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D15605
llvm-svn: 255994
Previously in the conversion cost table there are no entries for integer-integer
conversions on SSE2. This will result in imprecise costs for certain vectorized
operations. This patch adds those entries for SSE2 and SSE4.1. The cost numbers
are counted from the result of running llc on the new test case in this patch.
Differential revision: http://reviews.llvm.org/D15132
llvm-svn: 255315
Currently in LLVM's cost model, a vectorized arithmetic instruction will have
high cost if its type is split into multiple registers. However, this
punishment is too heavy and unnecessary. The overhead of the split should not
be on arithmetic instructions but instructions that implement the split. Note
that during vectorization we have calculated the register pressure, and we
only choose proper interleaving factor (and also vectorization factor) so
that we don't use more registers than the maximum number.
Here is a very simple example: if a vadd has the cost 1, and if we double VF
so that we need two registers to perform it, then its cost will become 4 with
the current implementation, which will prevent us to use larger VF.
Differential revision: http://reviews.llvm.org/D15159
llvm-svn: 254671
I checked and updated the cost of AVX-512 conversion operations. Added cost of conversion operations in DQ mode.
Conversion of illegal types that requires vector split is not calculated right now (like for other X86 targets).
Differential Revision: http://reviews.llvm.org/D15074
llvm-svn: 254494
The cost for scalarized operations is computed as N * (scalar operation
cost + 1 extractelement + 1 insertelement). This partially fixes
inflating the cost of scalarized operations since every operation is
scalarized and free. I don't think we want any cost asociated with
scalarization, but for now insertelement is still counted. I'm not sure
if we should pretend that insertelement is also free, or add a way
to compute a custom scalarization cost.
llvm-svn: 254438
Cost calculation for vector GEP failed with due to invalid cast to GEP index operand.
The bug is fixed, added a test.
http://reviews.llvm.org/D14976
llvm-svn: 254408
Note, this was reviewed (and more details are in) http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20151109/312083.html
These intrinsics currently have an explicit alignment argument which is
required to be a constant integer. It represents the alignment of the
source and dest, and so must be the minimum of those.
This change allows source and dest to each have their own alignments
by using the alignment attribute on their arguments. The alignment
argument itself is removed.
There are a few places in the code for which the code needs to be
checked by an expert as to whether using only src/dest alignment is
safe. For those places, they currently take the minimum of src/dest
alignments which matches the current behaviour.
For example, code which used to read:
call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 500, i32 8, i1 false)
will now read:
call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 8 %dest, i8* align 8 %src, i32 500, i1 false)
For out of tree owners, I was able to strip alignment from calls using sed by replacing:
(call.*llvm\.memset.*)i32\ [0-9]*\,\ i1 false\)
with:
$1i1 false)
and similarly for memmove and memcpy.
I then added back in alignment to test cases which needed it.
A similar commit will be made to clang which actually has many differences in alignment as now
IRBuilder can generate different source/dest alignments on calls.
In IRBuilder itself, a new argument was added. Instead of calling:
CreateMemCpy(Dst, Src, getInt64(Size), DstAlign, /* isVolatile */ false)
you now call
CreateMemCpy(Dst, Src, getInt64(Size), DstAlign, SrcAlign, /* isVolatile */ false)
There is a temporary class (IntegerAlignment) which takes the source alignment and rejects
implicit conversion from bool. This is to prevent isVolatile here from passing its default
parameter to the source alignment.
Note, changes in future can now be made to codegen. I didn't change anything here, but this
change should enable better memcpy code sequences.
Reviewed by Hal Finkel.
llvm-svn: 253511
The underlying issues surrounding codegen for 32-bit vselects have been resolved. The pessimistic costs for 64-bit vselects remain due to the bad
scalarization that is still happening there.
I tested this on A57 in T32, A32 and A64 modes. I saw no regressions, and some improvements.
From my benchmarks, I saw these improvements in A57 (T32)
spec.cpu2000.ref.177_mesa 5.95%
lnt.SingleSource/Benchmarks/Shootout/strcat 12.93%
lnt.MultiSource/Benchmarks/MiBench/telecomm-CRC32/telecomm-CRC32 11.89%
I also measured A57 A32, A53 T32 and A9 T32 and found no performance regressions. I see much bigger wins in third-party benchmarks with this change
Differential Revision: http://reviews.llvm.org/D14743
llvm-svn: 253349
A function can be marked as norecurse if:
* The SCC to which it belongs has cardinality 1; and either
a) It does not call any non-norecurse function. This includes self-recursion; or
b) It only has one callsite and the function that callsite is within is marked norecurse.
a) is best propagated bottom-up and b) is best propagated top-down.
We build up the norecurse attributes bottom-up using the existing SCC pass, and mark functions with no obvious recursion (but not provably norecurse) to sweep later, top-down.
llvm-svn: 252862
This is fix for PR24059.
When we are hoisting instruction above some condition it may turn out
that metadata on this instruction was control dependant on the condition.
This metadata becomes invalid and we need to drop it.
This patch should cover most obvious places of speculative execution (which
I have found by greping isSafeToSpeculativelyExecute). I think there are more
cases but at least this change covers the severe ones.
Differential Revision: http://reviews.llvm.org/D14398
llvm-svn: 252604
Summary:
GetUnderlyingObjects() can return "null" among its list of objects,
we don't want to deduce that two pointers can point to the same
memory in this case, so filter it out.
Reviewers: anemet
Subscribers: dexonsmith, llvm-commits
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 252149
Summary:
We now collect all types of dependences including lexically forward
deps not just "interesting" ones.
Reviewers: hfinkel
Subscribers: rengolin, llvm-commits
Differential Revision: http://reviews.llvm.org/D13256
llvm-svn: 251985
Summary:
When the dependence distance in zero then we have a loop-independent
dependence from the earlier to the later access.
No current client of LAA uses forward dependences so other than
potentially hitting the MaxDependences threshold earlier, this change
shouldn't affect anything right now.
This and the previous patch were tested together for compile-time
regression. None found in LNT/SPEC.
Reviewers: hfinkel
Subscribers: rengolin, llvm-commits
Differential Revision: http://reviews.llvm.org/D13255
llvm-svn: 251973
Summary:
Before this change, we didn't use to collect forward dependences since
none of the current clients (LV, LDist) required them.
The motivation to also collect forward dependences is a new pass
LoopLoadElimination (LLE) which discovers store-to-load forwarding
opportunities across the loop's backedge. The pass uses both lexically
forward or backward loop-carried dependences to detect these
opportunities.
The new pass also analyzes loop-independent (forward) dependences since
they can conflict with the loop-carried dependences in terms of how the
data flows through memory.
The newly added test only covers loop-carried forward dependences
because loop-independent ones are currently categorized as NoDep. The
next patch will fix this.
The two patches were tested together for compile-time regression. None
found in LNT/SPEC.
Note that with this change LAA provides all dependences rather than just
"interesting" ones. A subsequent NFC patch will remove the now trivial
isInterestingDependence and rename the APIs.
Reviewers: hfinkel
Subscribers: jmolloy, rengolin, llvm-commits
Differential Revision: http://reviews.llvm.org/D13254
llvm-svn: 251972
Have `getConstantEvolutionLoopExitValue` work correctly with multiple
entry loops.
As far as I can tell, `getConstantEvolutionLoopExitValue` never did the
right thing for multiple entry loops; and before r249712 it would
silently return an incorrect answer. r249712 changed SCEV to fail an
assert on a multiple entry loop, and this change fixes the underlying
issue.
llvm-svn: 251770
Prevent `createNodeFromSelectLikePHI` from creating SCEV expressions
that break LCSSA.
A better fix for the same issue is to teach SCEVExpander to not break
LCSSA by inserting PHI nodes at appropriate places. That's planned for
the future.
Fixes PR25360.
llvm-svn: 251756
