Modified patch by Adam Preuss.
This builds on the existing framework for block tracing, edge profiling and optimal edge profiling.
See -help-hidden for new flags.
For documentation, see the technical report "Implementation of Path Profiling..." in llvm.org/pubs.
llvm-svn: 124515
benchmarks, and that it can be simplified to X/Y. (In general you can only
simplify (Z*Y)/Y to Z if the multiplication did not overflow; if Z has the
form "X/Y" then this is the case). This patch implements that transform and
moves some Div logic out of instcombine and into InstructionSimplify.
Unfortunately instcombine gets in the way somewhat, since it likes to change
(X/Y)*Y into X-(X rem Y), so I had to teach instcombine about this too.
Finally, thanks to the NSW/NUW flags, sometimes we know directly that "Z*Y"
does not overflow, because the flag says so, so I added that logic too. This
eliminates a bunch of divisions and subtractions in 447.dealII, and has good
effects on some other benchmarks too. It seems to have quite an effect on
tramp3d-v4 but it's hard to say if it's good or bad because inlining decisions
changed, resulting in massive changes all over.
llvm-svn: 124487
default implementation for x86, going through the stack in a similr
fashion to how the codegen implements BUILD_VECTOR. Eventually this
will get matched to VINSERTF128 if AVX is available.
llvm-svn: 124307
implementation of EXTRACT_SUBVECTOR for x86, going through the stack
in a similr fashion to how the codegen implements BUILD_VECTOR.
Eventually this will get matched to VEXTRACTF128 if AVX is available.
llvm-svn: 124292
a few loops accordingly. Should be no functional change.
This is a step for more accurate cost/benefit analysis of devirt/inlining
bonuses.
llvm-svn: 124275
This will be used to check patterns referencing a forthcoming
INSERT_SUBVECTOR SDNode and will also be used to check
EXTRACT_SUBVECTOR nodes.
llvm-svn: 124191
optimized code are:
(non-negative number)+(power-of-two) != 0 -> true
and
(x | 1) != 0 -> true
Instcombine knows about the second one of course, but only does it if X|1
has only one use. These fire thousands of times in the testsuite.
llvm-svn: 124183
with BasicAA's DecomposeGEPExpression, which recently began
using a TargetData. This fixes PR8968, though the testcase
is awkward to reduce.
Also, update several off GetUnderlyingObject's users
which happen to have a TargetData handy to pass it in.
llvm-svn: 124134
computation, the Ancestor field is always set to the Parent, so we can remove
the explicit link entirely and merge the Parent and Ancestor fields. Instead of
checking for whether an ancestor exists for a node or not, we simply check
whether the node has already been processed. This is simpler if Compress is
inlined into Eval, so I did that as well.
This is about a 3% speedup running -domtree on test-suite + SPEC2000 & SPEC2006,
but it also opens up some opportunities for further improvement.
llvm-svn: 124061
flags. They are still not enable in this revision.
Added TargetInstrInfo::isZeroCost() to fix a fundamental problem with
the scheduler's model of operand latency in the selection DAG.
Generalized unit tests to work with sched-cycles.
llvm-svn: 123969
TargetInstrInfo:
Change produceSameValue() to take MachineRegisterInfo as an optional argument.
When in SSA form, targets can use it to make more aggressive equality analysis.
Machine LICM:
1. Eliminate isLoadFromConstantMemory, use MI.isInvariantLoad instead.
2. Fix a bug which prevent CSE of instructions which are not re-materializable.
3. Use improved form of produceSameValue.
ARM:
1. Teach ARM produceSameValue to look pass some PIC labels.
2. Look for operands from different loads of different constant pool entries
which have same values.
3. Re-implement PIC GA materialization using movw + movt. Combine the pair with
a "add pc" or "ldr [pc]" to form pseudo instructions. This makes it possible
to re-materialize the instruction, allow machine LICM to hoist the set of
instructions out of the loop and make it possible to CSE them. It's a bit
hacky, but it significantly improve code quality.
4. Some minor bug fixes as well.
With the fixes, using movw + movt to materialize GAs significantly outperform the
load from constantpool method. 186.crafty and 255.vortex improved > 20%, 254.gap
and 176.gcc ~10%.
llvm-svn: 123905
checks enabled:
1) Use '<' to compare integers in a comparison function rather than '<='.
2) Use the uniqued set DefBlocks rather than Info.DefiningBlocks to initialize
the priority queue.
The speedup of scalarrepl on test-suite + SPEC2000 + SPEC2006 is a bit less, at
just under 16% rather than 17%.
llvm-svn: 123662
eliminating a potentially quadratic data structure, this also gives a 17%
speedup when running -scalarrepl on test-suite + SPEC2000 + SPEC2006. My initial
experiment gave a greater speedup around 25%, but I moved the dominator tree
level computation from dominator tree construction to PromoteMemToReg.
Since this approach to computing IDFs has a much lower overhead than the old
code using precomputed DFs, it is worth looking at using this new code for the
second scalarrepl pass as well.
llvm-svn: 123609
half a million non-local queries, each of which would otherwise have triggered a
linear scan over a basic block.
Also fix a fixme for memory intrinsics which dereference pointers. With this,
we prove that a pointer is non-null because it was dereferenced by an intrinsic
112 times in llvm-test.
llvm-svn: 123533
these would try hard to match constants by inverting the bits
and recursively matching. There are two problems with this:
1) some patterns would match when we didn't want them to (theoretical)
2) this is insanely expensive to do, and most often pointless.
This was apparently useful in just 2 instcombine cases, which I
added code to handle explicitly. This change speeds up 'opt'
time on 176.gcc by 1% and produces bitwise identical code.
llvm-svn: 123518
early in the cleanup code and one late interlaced with the inliner. The second one is
important because inlining and other scalar optzns can unpin allocas, allowing them to
be split up and promoted. While important for performance, this is also relatively
rare, and we would previously force a (non-lazy) computation of DomFrontiers, which
happened even if nothing became unpinned.
With this patch, the first pass of scalarrepl still promotes the vast bulk of allocas
in programs, but hte second pass has changed to use SSAUpdater, which is more "sparse"
and lazy. This speeds up opt -O3 time on kimwitu++ (a c++ app) by about 1%. The
numbers are interesting: the first pass promotes ~17500 allocas. The second pass
promotes about 1600. For non-C++ codes, the compile time win should be greater,
because the second pass of scalarrepl does less.
llvm-svn: 123437
- Fixed :upper16: fix up routine. It should be shifting down the top 16 bits first.
- Added support for Thumb2 :lower16: and :upper16: fix up.
- Added :upper16: and :lower16: relocation support to mach-o object writer.
llvm-svn: 123424
most important simplifications, as well as resolving phase ordering issues where instcombine
would inhibit important CSE'ing opportunities, for instance on BitBench/drop3.
llvm-svn: 123418
While there, I noticed that the transform "undef >>a X -> undef" was wrong.
For example if X is 2 then the top two bits must be equal, so the result can
not be anything. I fixed this in the constant folder as well. Also, I made
the transform for "X << undef" stronger: it now folds to undef always, even
though X might be zero. This is in accordance with the LangRef, but I must
admit that it is fairly aggressive. Also, I added "i32 X << 32 -> undef"
following the LangRef and the constant folder, likewise fairly aggressive.
llvm-svn: 123417
Add methods for accessing the (single) entry / exit edge of a region. If no such
edge exists, null is returned. Both accessors return the start block of the
corresponding edge. The edge can finally be formed by utilizing
Region::getEntry() or Region::getExit();
Contributed by: Andreas Simbuerger <simbuerg@fim.uni-passau.de>
llvm-svn: 123410
in the right direction. It eliminated some hacks and will unblock codegen
work. But it's far from being done. It doesn't reject illegal expressions,
e.g. (FOO - :lower16:BAR). It also doesn't work in Thumb2 mode at all.
llvm-svn: 123369
"this" pointer for any subclass of User, you could static_cast it to
User* and then reinterpret_cast that to Use* to get the end of the
operand list. This isn't a safe assumption in general, because the
static_cast might adjust the "this" pointer. Fixed by having these
OperandTraits classes take an extra template parameter, which is the
subclass of User. This is groundwork for PR889.
llvm-svn: 123235
phi nodes. It is called from MergeBlockIntoPredecessor which is
called from GVN, which claims to preserve these.
I'm skeptical that this is the actual problem behind PR8954, but
this is a stab in the right direction.
llvm-svn: 123222
Fix the TargetRegisterInfo::NoRegister places where someone preferred
typing 'TargetRegisterInfo::NoRegister' instead of typing '0'.
Note that TableGen is already emitting xx::NoRegister in xxGenRegisterNames.inc.
llvm-svn: 123140
The numbering plan is now:
0 NoRegister.
[1;2^30) Physical registers.
[2^30;2^31) Stack slots.
[2^31;2^32) Virtual registers. (With -1u and -2u used by DenseMapInfo.)
Each segment is filled from the left, so any mistaken interpretation should
quickly cause crashes.
FirstVirtualRegister has been removed. TargetRegisterInfo provides predicates
conversion functions that should be used instead of interpreting register
numbers manually.
It is now legal to pass NoRegister to isPhysicalRegister() and
isVirtualRegister(). The result is false in both cases.
It is quite rare to represent stack slots in this way, so isPhysicalRegister()
and isVirtualRegister() require that isStackSlot() be checked first if it can
possibly return true. This allows a very fast implementation of the common
predicates.
llvm-svn: 123137
void f(int* begin, int* end) { std::fill(begin, end, 0); }
which turns into a != exit expression where one pointer is
strided and (thanks to step #1) known to not overflow, and
the other is loop invariant.
The observation here is that, though the IV is strided by
4 in this case, that the IV *has* to become equal to the
end value. It cannot "miss" the end value by stepping over
it, because if it did, the strided IV expression would
eventually wrap around.
Handle this by turning A != B into "A-B != 0" where the A-B
part is known to be NUW.
llvm-svn: 123131
when no virtual registers have been allocated.
It was only used to resize IndexedMaps, so provide an IndexedMap::resize()
method such that
Map.grow(MRI.getLastVirtReg());
can be replaced with the simpler
Map.resize(MRI.getNumVirtRegs());
This works correctly when no virtuals are allocated, and it bypasses the to/from
index conversions.
llvm-svn: 123130
physical register numbers.
This makes the hack used in LiveInterval official, and lets LiveInterval be
oblivious of stack slots.
The isPhysicalRegister() and isVirtualRegister() predicates don't know about
this, so when a variable may contain a stack slot, isStackSlot() should always
be tested first.
llvm-svn: 123128
of using a Location class with the same information.
When making a copy of a MachineOperand that was already stored in a
MachineInstr, it is necessary to clear the parent pointer on the copy. Otherwise
the register use-def lists become inconsistent.
Add MachineOperand::clearParent() to do that. An alternative would be a custom
MachineOperand copy constructor that cleared ParentMI. I didn't want to do that
because of the performance impact.
llvm-svn: 123109
Print virtual registers numbered from 0 instead of the arbitrary
FirstVirtualRegister. The first virtual register is printed as %vreg0.
TRI::NoRegister is printed as %noreg.
llvm-svn: 123107
Provide MRI::getNumVirtRegs() and TRI::index2VirtReg() functions to allow
iteration over virtual registers without depending on the representation of
virtual register numbers.
llvm-svn: 123098
Add a unnamed_addr bit to global variables and functions. This will be used
to indicate that the address is not significant and therefore the constant
or function can be merged with others.
If an optimization pass can show that an address is not used, it can set this.
Examples of things that can have this set by the FE are globals created to
hold string literals and C++ constructors.
Adding unnamed_addr to a non-const global should have no effect unless
an optimization can transform that global into a constant.
Aliases are not allowed to have unnamed_addr since I couldn't figure
out any use for it.
llvm-svn: 123063
1. Take a flags argument instead of a bool. This makes
it more clear to the reader what it is used for.
2. Add a flag that says that "remapping a value not in the
map is ok".
3. Reimplement MapValue to share a bunch of code and be a lot
more efficient. For lookup failures, don't drop null values
into the map.
4. Using the new flag a bunch of code can vaporize in LinkModules
and LoopUnswitch, kill it.
No functionality change.
llvm-svn: 123058
Instead encode llvm IR level property "HasSideEffects" in an operand (shared
with IsAlignStack). Added MachineInstrs::hasUnmodeledSideEffects() to check
the operand when the instruction is an INLINEASM.
This allows memory instructions to be moved around INLINEASM instructions.
llvm-svn: 123044
Also fix an off-by-one in SelectionDAGBuilder that was preventing shuffle
vectors from being translated to EXTRACT_SUBVECTOR.
Patch by Tim Northover.
The test changes are needed to keep those spill-q tests from testing aligned
spills and restores. If the only aligned stack objects are spill slots, we
no longer realign the stack frame. Prior to this patch, an EXTRACT_SUBVECTOR
was legalized by loading from the stack, which created an aligned frame index.
Now, however, there is nothing except the spill slot in the stack frame, so
I added an aligned alloca.
llvm-svn: 122995
We were never generating any of these nodes with variable indices, and there
was one legalizer function asserting on a non-constant index. If we ever have
a need to support variable indices, we can add this back again.
llvm-svn: 122993
This pass precomputes CFG block frequency information that can be used by the
register allocator to find optimal spill code placement.
Given an interference pattern, placeSpills() will compute which basic blocks
should have the current variable enter or exit in a register, and which blocks
prefer the stack.
The algorithm is ready to consume block frequencies from profiling data, but for
now it gets by with the static estimates used for spill weights.
This is a work in progress and still not hooked up to RegAllocGreedy.
llvm-svn: 122938
My i386 llvm-gcc nightly tester found a regression for
SingleSource/Benchmarks/McGill/chomp that a bisect blamed on 122743.
That seems strange but apparently the combination of earlycse and instcombine
did something bad. Chris says he intended to remove the instcombine pass, so
let's go ahead and try that. We'll see if there are any performance losses.
llvm-svn: 122907
It forms memset and memcpy's, and will someday form popcount and
other stuff. All of this is bad when compiling the implementation
of memset, memcpy, popcount, etc.
llvm-svn: 122854
The analysis will be needed by both the greedy register allocator and the
X86FloatingPoint pass. It only needs to be computed once when the CFG doesn't
change.
This pass is very fast, usually showing up as 0.0% wall time.
llvm-svn: 122832
a pointer value has potentially become escaping. Implementations can choose to either fall back to
conservative responses for that value, or may recompute their analysis to accomodate the change.
llvm-svn: 122777
improvement in the generated code, and speeds up 'opt -std-compile-opts'
compile time on 176.gcc from 24.84s to 23.2s (about 7%).
This also resolves a specific code quality issue in rdar://7352081 which
was generating poor code for:
int t(int a, int b) {
if (a & b & 1)
return a & b;
return 3;
}
llvm-svn: 122740
update a callGraph when performing the common operation of splicing the body to
a new function and updating all callers (such as via RAUW).
No users yet, though this is intended for DeadArgumentElimination as part of
PR8887.
llvm-svn: 122728
of instcombine that is currently in the middle of the loop pass pipeline. This
commit only checks in the pass; it will hopefully be enabled by default later.
llvm-svn: 122719
compile, and everyone's tests have shown it to be slower in practice, even for
quite large graphs.
I also hope to do an optimization that is only correct with the simpler data
structure, which would break this even further.
llvm-svn: 122684
naively implemented, the Lengauer-Tarjan algorithm requires a separate bucket
for each vertex. However, this is unnecessary, because each vertex is only
placed into a single bucket (that of its semidominator), and each vertex's
bucket is processed before it is added to any bucket itself.
Instead of using a bucket per vertex, we use a single array Buckets that has two
purposes. Before the vertex V with DFS number i is processed, Buckets[i] stores
the index of the first element in V's bucket. After V's bucket is processed,
Buckets[i] stores the index of the next element in the bucket to which V now
belongs, if any.
Reading from the buckets can also be optimized. Instead of processing the bucket
of V's parent at the end of processing V, we process the bucket of V itself at
the beginning of processing V. This means that the case of the root vertex can
be simplified somewhat. It also means that we don't need to look up the DFS
number of the semidominator of every node in the bucket we are processing,
since we know it is the current index being processed.
This is a 6.5% speedup running -domtree on test-suite + SPEC2000/2006, with
larger speedups of around 12% on the larger benchmarks like GCC.
llvm-svn: 122680
limitations, this kicks in dozens of times in the 4 specfp2000 benchmarks,
and hundreds of times in the int part. It also kicks in hundreds of times
in multisource.
This kicks in right before loop deletion, which has the pleasant effect of
deleting loops that *just* do a memset.
llvm-svn: 122664
Evan Cheng