the isRegSequence property.
This is a follow-up of r215394 and r215404, which respectively introduces the
isRegSequence property and uses it for ARM.
Thanks to the property introduced by the previous commits, this patch is able
to optimize the following sequence:
vmov d0, r2, r3
vmov d1, r0, r1
vmov r0, s0
vmov r1, s2
udiv r0, r1, r0
vmov r1, s1
vmov r2, s3
udiv r1, r2, r1
vmov.32 d16[0], r0
vmov.32 d16[1], r1
vmov r0, r1, d16
bx lr
into:
udiv r0, r0, r2
udiv r1, r1, r3
vmov.32 d16[0], r0
vmov.32 d16[1], r1
vmov r0, r1, d16
bx lr
This patch refactors how the copy optimizations are done in the peephole
optimizer. Prior to this patch, we had one copy-related optimization that
replaced a copy or bitcast by a generic, more suitable (in terms of register
file), copy.
With this patch, the peephole optimizer features two copy-related optimizations:
1. One for rewriting generic copies to generic copies:
PeepholeOptimizer::optimizeCoalescableCopy.
2. One for replacing non-generic copies with generic copies:
PeepholeOptimizer::optimizeUncoalescableCopy.
The goals of these two optimizations are slightly different: one rewrite the
operand of the instruction (#1), the other kills off the non-generic instruction
and replace it by a (sequence of) generic instruction(s).
Both optimizations rely on the ValueTracker introduced in r212100.
The ValueTracker has been refactored to use the information from the
TargetInstrInfo for non-generic instruction. As part of the refactoring, we
switched the tracking from the index of the definition to the actual register
(virtual or physical). This one change is to provide better consistency with
register related APIs and to ease the use of the TargetInstrInfo.
Moreover, this patch introduces a new helper class CopyRewriter used to ease the
rewriting of generic copies (i.e., #1).
Finally, this patch adds a dead code elimination pass right after the peephole
optimizer to get rid of dead code that may appear after rewriting.
This is related to <rdar://problem/12702965>.
Review: http://reviews.llvm.org/D4874
llvm-svn: 216088
legalization stage. With those two optimizations, fewer signed/zero extension
instructions can be inserted, and then we can expose more opportunities to
Machine CSE pass in back-end.
llvm-svn: 216066
Note: This was originally reverted to track down a buildbot error. This commit
exposed a latent bug that was fixed in r215753. Therefore it is reapplied
without any modifications.
I run it through SPEC2k and SPEC2k6 for AArch64 and it didn't introduce any new
regeressions.
Original commit message:
This changes the order in which FastISel tries to materialize a constant.
Originally it would try to use a simple target-independent approach, which
can lead to the generation of inefficient code.
On X86 this would result in the use of movabsq to materialize any 64bit
integer constant - even for simple and small values such as 0 and 1. Also
some very funny floating-point materialization could be observed too.
On AArch64 it would materialize the constant 0 in a register even the
architecture has an actual "zero" register.
On ARM it would generate unnecessary mov instructions or not use mvn.
This change simply changes the order and always asks the target first if it
likes to materialize the constant. This doesn't fix all the issues
mentioned above, but it enables the targets to implement such
optimizations.
Related to <rdar://problem/17420988>.
llvm-svn: 216006
When combining a pair of shuffle nodes, check if the combined shuffle mask is
trivially Undef. In case, immediately fold that pair of shuffles to Undef.
The lack of checks for undef masks was the root-cause of a poor-codegen bug
in the dag combiner.
Example:
%1 = shufflevector <4 x i32> %A, <4 x i32> %B, <4 x i32> <i32 4, i32 1, i32 1, i32 6>
%2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 0, i32 4, i32 1, i32 6>
%3 = shufflevector <4 x i32> %2, <4 x i32> undef, <4 x i32> <i32 1, i32 5, i32 3, i32 3>
Before this patch, on x86 (with -mcpu=corei7) we failed to fold the entire
sequence to Undef value and therefore we generated:
shufps $-123, %xmm1, $xmm0
pshufd $-46, %xmm0, %xmm0
With this patch, the entire shuffle sequence is folded to Undef and no
shuffles are generated in the output assembly.
Added new test cases to test 'combine-vec-shuffle-5.ll'.
llvm-svn: 215797
We used to assume that any fixed-offset stack object was not aliased. This
meant that no IR value could point to the memory contained in such an object.
This is a reasonable default, but is not a universally-correct
target-independent fact. For example, on PowerPC (both Darwin and non-Darwin),
some byval arguments are allocated at fixed offsets by the ABI. These, however,
certainly can be pointed to by IR values. This change moves the 'isAliased'
logic out of FixedStackPseudoSourceValue and into MFI, and allows the isAliased
property to be overridden for fixed-offset objects.
This will be used by an upcoming commit to the PowerPC backend to fix PR20280.
No functionality change intended (the behavior of
FixedStackPseudoSourceValue::isAliased has been made more conservative for
callers that don't pass an MFI object, but I don't see any in-tree callers that
do that).
llvm-svn: 215794
As Jim pointed out this assert isn't really needed to test for correctness,
because the code right afterwards does the same check and falls-back to
SelectionDAG - as intended.
llvm-svn: 215735
This reverts:
r215595 "[FastISel][X86] Add large code model support for materializing floating-point constants."
r215594 "[FastISel][X86] Use XOR to materialize the "0" value."
r215593 "[FastISel][X86] Emit more efficient instructions for integer constant materialization."
r215591 "[FastISel][AArch64] Make use of the zero register when possible."
r215588 "[FastISel] Let the target decide first if it wants to materialize a constant."
r215582 "[FastISel][AArch64] Cleanup constant materialization code. NFCI."
llvm-svn: 215673
This patch allows a vector fneg of a bitcasted integer value to be optimized in the same way that we already optimize a scalar fneg. If the integer variable is a constant, we can precompute the result and not require any logic ops.
This patch is very similar to a fabs patch committed at r214892.
Differential Revision: http://reviews.llvm.org/D4852
llvm-svn: 215646
input node after manually adding it to the worklist and using CombineTo.
Once we use CombineTo the input node may have been deleted. Despite this
being *completely confusing* and somewhat broken, the only way to
"correctly" return from a DAG combine after potentially deleting the
input node is to return *that exact node*....
But really, this code should just never have used CombineTo. It won't do
what it wants (returning the node as mentioned above just causes the
combine to infloop). The correct way to combine away a casted load to
a load of the correct type is to RAUW the chain directly and then return
the loaded value to replace the actual value node.
I managed to find this with the vector shuffle fuzzer even though it
clearly has nothing at all to do with vector shuffles and rather those
happen to trigger a load of a constant pool that hits this combine *just
right*. I've included the test as it is small and a nice stress test
that the infrastructure isn't asserting.
llvm-svn: 215622
combining by replacing it with something else but not re-process the
node afterward to remove it.
In a truly remarkable stroke of bad luck, this would (in the test case
attached) end up getting some other node combined into it without ever
getting re-processed. By adding it back on to the worklist, in addition
to deleting the dead nodes more quickly we also ensure that if it
*stops* being dead for any reason it makes it back through the
legalizer. Without this, the test case will end up failing during
instruction selection due to an and node with a type we don't have an
instruction pattern for.
It took many million runs of the shuffle fuzz tester to find this.
llvm-svn: 215611
This changes the order in which FastISel tries to materialize a constant.
Originally it would try to use a simple target-independent approach, which
can lead to the generation of inefficient code.
On X86 this would result in the use of movabsq to materialize any 64bit
integer constant - even for simple and small values such as 0 and 1. Also
some very funny floating-point materialization could be observed too.
On AArch64 it would materialize the constant 0 in a register even the
architecture has an actual "zero" register.
On ARM it would generate unnecessary mov instructions or not use mvn.
This change simply changes the order and always asks the target first if it
likes to materialize the constant. This doesn't fix all the issues
mentioned above, but it enables the targets to implement such
optimizations.
Related to <rdar://problem/17420988>.
llvm-svn: 215588
This is a cleaner solution to the problem described in r215431.
When instructions are combined a dangling DBG_VALUE is removed.
This resolves bug 20598.
llvm-svn: 215587
New function to erase a machine instruction and mark DBG_VALUE
for removal. A DBG_VALUE is marked for removal when it references
an operand defined in the instruction.
Use the new function to cleanup code in dead machine instruction
removal pass.
llvm-svn: 215580
critical edge has been split. The MachineDominatorTree will when lazy update the
underlying dominance properties when require.
** Context **
This is a follow-up of r215410.
Each time a critical edge is split this invalidates the dominator tree
information. Thus, subsequent queries of that interface will be slow until the
underlying information is actually recomputed (costly).
** Problem **
Prior to this patch, splitting a critical edge needed to query the dominator
tree to update the dominator information.
Therefore, splitting a bunch of critical edges will likely produce poor
performance as each query to the dominator tree will use the slow query path.
This happens a lot in passes like MachineSink and PHIElimination.
** Proposed Solution **
Splitting a critical edge is a local modification of the CFG. Moreover, as soon
as a critical edge is split, it is not critical anymore and thus cannot be a
candidate for critical edge splitting anymore. In other words, the predecessor
and successor of a basic block inserted on a critical edge cannot be inserted by
critical edge splitting.
Using these observations, we can pile up the splitting of critical edge and
apply then at once before updating the DT information.
The core of this patch moves the update of the MachineDominatorTree information
from MachineBasicBlock::SplitCriticalEdge to a lazy MachineDominatorTree.
** Performance **
Thanks to this patch, the motivating example compiles in 4- minutes instead of
6+ minutes. No test case added as the motivating example as nothing special but
being huge!
The binaries are strictly identical for all the llvm test-suite + SPECs with and
without this patch for both Os and O3.
Regarding compile time, I observed only noise, although on average I saw a
small improvement.
<rdar://problem/17894619>
llvm-svn: 215576
Add header guards to files that were missing guards. Remove #endif comments
as they don't seem common in LLVM (we can easily add them back if we decide
they're useful)
Changes made by clang-tidy with minor tweaks.
llvm-svn: 215558
This patch improves the existing algorithm in DAGCombiner that
attempts to fold shuffles according to rule:
shuffle(shuffle(x, y, M1), undef, M2) -> shuffle(y, undef, M3)
Before this change, there were cases where the DAGCombiner conservatively
avoided folding shuffles even if the resulting mask would have been legal.
That is because the algorithm wrongly assumed that commuting
an illegal shuffle mask would always produce an illegal mask.
With this change, we now correctly compute the commuted shuffle mask before
calling method 'isShuffleMaskLegal' on it.
On X86, this improves for example the codegen for the following function:
define <4 x i32> @test(<4 x i32> %A, <4 x i32> %B) {
%1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 2, i32 6, i32 7>
%2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
ret <4 x i32> %2
}
Before this change the X86 backend (-mcpu=corei7) generated
the following assembly code for function @test:
shufps $-23, %xmm0, %xmm1 # xmm1 = xmm1[1,2],xmm0[2,3]
movhlps %xmm1, %xmm1 # xmm1 = xmm1[1,1]
movaps %xmm1, %xmm0
Now we produce:
movhlps %xmm0, %xmm0 # xmm0 = xmm0[1,1]
Added extra test cases in combine-vec-shuffle-2.ll to verify that we correctly
fold according to the above-mentioned rule.
llvm-svn: 215555
This implements PPCTargetLowering::getTgtMemIntrinsic for Altivec load/store
intrinsics. As with the construction of the MachineMemOperands for the
intrinsic calls used for unaligned load/store lowering, the only slight
complication is that we need to represent a larger memory range than the
loaded/stored value-type size (because the address is rounded down to an
aligned address, and we need to conservatively represent the entire possible
range of the actual access). This required adding an extra size field to
TargetLowering::IntrinsicInfo, and this was done in a way that required no
modifications to other targets (the size defaults to the store size of the
provided memory data type).
This fixes test/CodeGen/PowerPC/unal-altivec-wint.ll (so it can be un-XFAILed).
llvm-svn: 215512
refactoring in 215384. This way it can unique multiple entries describing
the same piece even if they don't have the exact same location.
(The same piece may get merged in and be added from OpenRanges).
There ought to be a more elegant solution for this, though.
llvm-svn: 215418
as long as possible.
** Context **
Each time the dominance information is modified, the dominator tree analysis
switches in a slow query mode. After a few queries without any modification on
the dominator tree, it performs an expensive update of its internal structure to
provide fast queries again.
** Problem **
Prior to this patch, the MachineSink pass was splitting the critical edges on
demand while relying heavy on the dominator tree information. In some cases,
this leads to pathological behavior where:
- We end up in the slow query mode right after splitting an edge.
- We update the dominance information.
- We break the dominance information again, thus ending up in the slow query
mode and so on.
** Proposed Solution **
To mitigate this effect, this patch postpones all the splitting of the edges at
the end of each iteration of the main loop.
The benefits are:
- The dominance information is valid for the life time of an iteration.
- This simplifies the code as we do not have to special treat instructions that
are sunk on critical edges. Indeed, the related block will be available
through the next iteration.
The downside is that when edges splitting is required, this incurs an additional
iteration of the main loop compared to the previous scheme.
** Performance **
Thanks to this patch, the motivating example compiles in 6+ minutes instead of
10+ minutes. No test case added as the motivating example as nothing special but
being huge!
I have measured only noise for both the compile time and the runtime on the llvm
test-suite + SPECs with Os and O3.
Note: The current implementation of MachineBasicBlock::SplitCriticalEdge also
uses the dominance information and therefore, hits this problem. A subsequent
patch will address that.
<rdar://problem/17894619>
llvm-svn: 215410
This patch adds a new property: isRegSequence and the related target hooks:
TargetIntrInfo::getRegSequenceInputs and
TargetInstrInfo::getRegSequenceLikeInputs to specify that a target specific
instruction is a (kind of) REG_SEQUENCE.
<rdar://problem/12702965>
llvm-svn: 215394
buildLocationLists easier to read.
The previous implementation conflated the merging of individual pieces
and the merging of entire DebugLocEntries.
By splitting this functionality into two separate functions the intention
of the code should be clearer.
llvm-svn: 215383
be propagated to all its users, and this propagation could increase the
probability of finding common subexpressions. If the COPY has only one user,
the COPY itself can be removed.
llvm-svn: 215344
That broke the build:
/data/buildslave/clang-amd64-freebsd/src-llvm/lib/CodeGen/PeepholeOptimizer.cpp:729:46: error: non-const lvalue reference to type 'SmallPtrSet<[...], 8>' cannot bind to a value of unrelated type 'SmallPtrSet<[...], 16>'
Changed |= optimizeExtInstr(MI, MBB, LocalMIs);
^~~~~~~~
/data/buildslave/clang-amd64-freebsd/src-llvm/lib/CodeGen/PeepholeOptimizer.cpp:265:49: note: passing argument to parameter 'LocalMIs' here
SmallPtrSet<MachineInstr*, 8> &LocalMIs) {
^
llvm-svn: 215341
Follow up to r214266. Add missing case in ScalarizeVectorResult() for
cttz_zero_undef.
Differential Revision: http://reviews.llvm.org/D4813
llvm-svn: 215330
