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
rdar://problem/8893967: JM/lencod miscompile at -arch armv7 -mthumb -O3
Added ResurrectKill to remove kill flags after we decide to reused a
physical register. And (hopefully) ensure that we call it in all the
right places.
Sorry, I'm not checking in a unit test given that it's a miscompile I
can't reproduce easily with a toy example. Failures in the rewriter
depend on a series of heuristic decisions maked during one of the many
upstream phases in codegen. This case would require coercing regalloc
to generate a couple of rematerialzations in a way that causes the
scavenger to reuse the same register at just the wrong point.
The general way to test this is to implement kill flags
verification. Then we could have a simple, robust compile-only unit
test. That would be worth doing if the whole pass was not about to
disappear. At this point we focus verification work on the next
generation of regalloc.
llvm-svn: 124442
Linear scan regalloc is currently assuming that any register aliased with
a member of a regclass must also be in at least one regclass. That is not
always true. For example, for X86, RIP is in a regclass but IP is not.
If you're unlucky, this can cause a crash by invalidating the iterator.
llvm-svn: 124365
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
operand being factorized (and erased) could occur several times in Ops,
resulting in freed memory being used when the next occurrence in Ops was
analyzed.
llvm-svn: 124287
merge vector<intptr_t>::push_back() and vector<void*>::push_back() because
Enumerate() doesn't realize that "i64* null" and "i8** null" are equivalent.
llvm-svn: 124285
doesn't return immediately after then the insert position in UniqueSCEVs will
be out of date. No test because this is a memory corruption issue. Fixes PR9051!
llvm-svn: 124282
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
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
occurs because instcombine sinks loads and inserts phis. This kicks in
on such apps as 175.vpr, eon, 403.gcc, xalancbmk and a bunch of times in
spec2006 in some app that uses std::deque.
This resolves the last of rdar://7339113.
llvm-svn: 124090
common cases. This triggers a surprising number of times in SPEC2K6
because min/max idioms end up doing this. For example, code from the
STL ends up looking like this to SRoA:
%202 = load i64* %__old_size, align 8, !tbaa !3
%203 = load i64* %__old_size, align 8, !tbaa !3
%204 = load i64* %__n, align 8, !tbaa !3
%205 = icmp ult i64 %203, %204
%storemerge.i = select i1 %205, i64* %__n, i64* %__old_size
%206 = load i64* %storemerge.i, align 8, !tbaa !3
We can now promote both the __n and the __old_size allocas.
This addresses another chunk of rdar://7339113, poor codegen on
stringswitch.
llvm-svn: 124088
clang's -Wuninitialized-experimental warning.
While these don't look like real bugs, clang's
-Wuninitialized-experimental analysis is stricter
than GCC's, and these fixes have the benefit
of being general nice cleanups.
llvm-svn: 124073
that have PHI or select uses of their element pointers. This can often happen
when instcombine sinks two loads into a successor, inserting a phi or select.
With this patch, we can scalarize the alloca, but the pinned elements are not
yet promoted. This is still a win for large aggregates where only one element
is used. This fixes rdar://8904039 and part of rdar://7339113 (poor codegen
on stringswitch).
llvm-svn: 124070
handle the "Transformation preventing inst" printing,
so that -scalarrepl -debug will always print the rejected
instruction. No functionality change.
llvm-svn: 124066
1. Fixed ARM pc adjustment.
2. Fixed dynamic-no-pic codegen
3. CSE of pc-relative load of global addresses.
It's now enabled by default for Darwin.
llvm-svn: 123991
qadd and qdadd uses "rd, rm, rn", the same applies to the 'sub' variants. This
is described in ARM manuals and matches the encoding used by the gnu assembler.
llvm-svn: 123975
DAG. Disable using "-disable-sched-cycles".
For ARM, this enables a framework for modeling the cpu pipeline and
counting stalls. It also activates several heuristics to drive
scheduling based on the model. Scheduling is inherently imprecise at
this stage, and until spilling is improved it may defeat attempts to
schedule. However, this framework provides greater control over
tuning codegen.
Although the flag is not target-specific, it should have very little
affect on the default scheduler used by x86. The only two changes that
affect x86 are:
- scheduling a high-latency operation bumps the current cycle so independent
operations can have their latency covered. i.e. two independent 4
cycle operations can produce results in 4 cycles, not 8 cycles.
- Two operations with equal register pressure impact and no
latency-based stalls on their uses will be prioritized by depth before height
(height is irrelevant if no stalls occur in the schedule below this point).
llvm-svn: 123971
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
a select. A vector select is pairwise on each element so we'd need a new
condition with the right number of elements to select on. Fixes PR8994.
llvm-svn: 123963
While here, I'd like to complain about how vector is not an aggregate type
according to llvm::Type::isAggregateType(), but they're listed under aggregate
types in the LangRef and zero vectors are stored as ConstantAggregateZero.
llvm-svn: 123956
value, the "add pc" must be CSE'ed at the same time. We could follow the same
approach as T2 by adding pseudo instructions that combine the ldr + "add pc".
But the better approach is to use movw + movt (which I will enable soon), so
I'll leave this as a TODO.
llvm-svn: 123949
in cdp/cdp2 instructions. Also increase the hack with cdp/cdp2 instructions.
- Fix the encoding of cdp/cdp2 instructions for ARM (no thumb and thumb2 yet) and add testcases for t
hem.
llvm-svn: 123927
The value mapping gets confused about which original values have multiple new
definitions so they may need phi insertions.
This could probably be simplified by letting enterIntvBefore() take a live range
to be added following the instruction. As long as the range stays inside the
same basic block, value mapping shouldn't be a problem.
llvm-svn: 123926
auto-simplier the transform most missed by early-cse is (zext X) != 0 -> X != 0.
This patch adds this transform and some related logic to InstructionSimplify
and removes some of the logic from instcombine (unfortunately not all because
there are several situations in which instcombine can improve things by making
new instructions, whereas instsimplify is not allowed to do this). At -O2 this
often results in more than 15% more simplifications by early-cse, and results in
hundreds of lines of bitcode being eliminated from the testsuite. I did see some
small negative effects in the testsuite, for example a few additional instructions
in three programs. One program, 483.xalancbmk, got an additional 35 instructions,
which seems to be due to a function getting an additional instruction and then
being inlined all over the place.
llvm-svn: 123911
to add/sub by doing the normal operation and then checking for overflow
afterwards. This generally relies on the DAG handling the later invalid
operations as well.
Fixes the 64-bit part of rdar://8622122 and rdar://8774702.
llvm-svn: 123908
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
Added a check for already live regs before claiming HighRegPressure.
Fixed a few cases of checking the wrong number of successors.
Added some tracing until these heuristics are better understood.
llvm-svn: 123892
with an invalid type then split the result and perform the overflow check
normally.
Fixes the 32-bit parts of rdar://8622122 and rdar://8774702.
llvm-svn: 123864
interval after an instruction. The leaveIntvAfter() method only adds liveness
from the instruction's boundary index to the inserted copy.
Ideally, SplitKit should be smarter about this, perhaps by combining useIntv()
and leaveIntvAfter() into one method that guarantees continuity.
llvm-svn: 123858
Region splitting includes loop splitting as a subset, and it is more generic.
The splitting heuristics for variables that are live in more than one block are
now:
1. Try to create a region that covers multiple basic blocks.
2. Try to create a new live range for each block with multiple uses.
3. Spill.
Steps 2 and 3 are similar to what the standard spiller is doing.
llvm-svn: 123853
by indvars through the scev expander.
trunc(add x, y) --> add(trunc x, y). Currently SCEV largely folds the other way
which is probably wrong, but preserved to minimize churn. Instcombine doesn't
do this fold either, demonstrating a missed optz'n opportunity on code doing
add+trunc+add.
llvm-svn: 123838
of the floating point types less than 64-bits. It's somewhat of a temporary
hack but forces more accurate modeling of register pressure and results
in fewer spills.
llvm-svn: 123811
are pointing to the same object, one pointer is accessing the entire
object, and the other is access has a non-zero size. This prevents
TBAA from kicking in and saying NoAlias in such cases.
llvm-svn: 123775
Analyze the live range's behavior entering and leaving basic blocks. Compute an
interference pattern for each allocation candidate, and use SpillPlacement to
find an optimal region where that register can be live.
This code is still not enabled.
llvm-svn: 123774
These were not recommended by my auto-simplifier since they don't fire often enough.
However they do fire from time to time, for example they remove one subtraction from
the final bitcode for 483.xalancbmk.
llvm-svn: 123755
simplification in fully optimized code. It occurs sporadically in the testsuite, and
many times in 403.gcc: the final bitcode has 131 fewer subtractions after this change.
The reason that the multiplies are not eliminated is the same reason that instcombine
did not catch this: they are used by other instructions (instcombine catches this with
a more general transform which in general is only profitable if the operands have only
one use).
llvm-svn: 123754
without whatever this was trying to do. When/if someone has the time to do some empirical
evaluations, it might be worth it to figure out what this code was trying to do and see if
it's worth resurrecting/fixing.
llvm-svn: 123684
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
This shaves off 4 popcounts from the hacked 186.crafty source.
This is enabled even when a native popcount instruction is available. The
combined code is one operation longer but it should be faster nevertheless.
llvm-svn: 123621
movw r0, :lower16:(L_foo$non_lazy_ptr-(LPC0_0+4))
movt r0, :upper16:(L_foo$non_lazy_ptr-(LPC0_0+4))
LPC0_0:
add r0, pc, r0
It's not yet enabled by default as some tests are failing. I suspect bugs in
down stream tools.
llvm-svn: 123619
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
This fixes the original testcase in PR8927. It also causes a clang
binary built with a patched clang to increase in size by 0.21%.
We can probably get some of the size back by writing a pass that
detects that a global never has its pointer compared and adds
unnamed_addr to it (maybe extend global opt). It is also possible that
there are some other cases clang could add unnamed_addr to.
I will investigate extending globalopt next.
llvm-svn: 123584
into and/shift would cause nodes to move around and a dangling pointer
to happen. The code tried to avoid this with a HandleSDNode, but
got the details wrong.
llvm-svn: 123578
then don't try to decimate it into its individual pieces. This will just make a mess of the
IR and is pointless if none of the elements are individually accessed. This was generating
really terrible code for std::bitset (PR8980) because it happens to be lowered by clang
as an {[8 x i8]} structure instead of {i64}.
The testcase now is optimized to:
define i64 @test2(i64 %X) {
br label %L2
L2: ; preds = %0
ret i64 %X
}
before we generated:
define i64 @test2(i64 %X) {
%sroa.store.elt = lshr i64 %X, 56
%1 = trunc i64 %sroa.store.elt to i8
%sroa.store.elt8 = lshr i64 %X, 48
%2 = trunc i64 %sroa.store.elt8 to i8
%sroa.store.elt9 = lshr i64 %X, 40
%3 = trunc i64 %sroa.store.elt9 to i8
%sroa.store.elt10 = lshr i64 %X, 32
%4 = trunc i64 %sroa.store.elt10 to i8
%sroa.store.elt11 = lshr i64 %X, 24
%5 = trunc i64 %sroa.store.elt11 to i8
%sroa.store.elt12 = lshr i64 %X, 16
%6 = trunc i64 %sroa.store.elt12 to i8
%sroa.store.elt13 = lshr i64 %X, 8
%7 = trunc i64 %sroa.store.elt13 to i8
%8 = trunc i64 %X to i8
br label %L2
L2: ; preds = %0
%9 = zext i8 %1 to i64
%10 = shl i64 %9, 56
%11 = zext i8 %2 to i64
%12 = shl i64 %11, 48
%13 = or i64 %12, %10
%14 = zext i8 %3 to i64
%15 = shl i64 %14, 40
%16 = or i64 %15, %13
%17 = zext i8 %4 to i64
%18 = shl i64 %17, 32
%19 = or i64 %18, %16
%20 = zext i8 %5 to i64
%21 = shl i64 %20, 24
%22 = or i64 %21, %19
%23 = zext i8 %6 to i64
%24 = shl i64 %23, 16
%25 = or i64 %24, %22
%26 = zext i8 %7 to i64
%27 = shl i64 %26, 8
%28 = or i64 %27, %25
%29 = zext i8 %8 to i64
%30 = or i64 %29, %28
ret i64 %30
}
In this case, instcombine was able to eliminate the nonsense, but in PR8980 enough
PHIs are in play that instcombine backs off. It's better to not generate this stuff
in the first place.
llvm-svn: 123571
multiple uses. In some cases, all the uses are the same operation,
so instcombine can go ahead and promote the phi. In the testcase
this pushes an add out of the loop.
llvm-svn: 123568
http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel
In a silly microbenchmark on a 65 nm core2 this is 1.5x faster than the old
code in 32 bit mode and about 2x faster in 64 bit mode. It's also a lot shorter,
especially when counting 64 bit population on a 32 bit target.
I hope this is fast enough to replace Kernighan-style counting loops even when
the input is rather sparse.
llvm-svn: 123547
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
The basic issue is that isel (very reasonably!) expects conditional branches
to be folded, so CGP leaving around a bunch dead computation feeding
conditional branches isn't such a good idea. Just fold branches on constants
into unconditional branches.
llvm-svn: 123526
have objectsize folding recursively simplify away their result when it
folds. It is important to catch this here, because otherwise we won't
eliminate the cross-block values at isel and other times.
llvm-svn: 123524
disabled in this checkin. Sorry for the large diffs due to
refactoring. New functionality is all guarded by EnableSchedCycles.
Scheduling the isel DAG is inherently imprecise, but we give it a best
effort:
- Added MayReduceRegPressure to allow stalled nodes in the queue only
if there is a regpressure need.
- Added BUHasStall to allow checking for either dependence stalls due to
latency or resource stalls due to pipeline hazards.
- Added BUCompareLatency to encapsulate and standardize the heuristics
for minimizing stall cycles (vs. reducing register pressure).
- Modified the bottom-up heuristic (now in BUCompareLatency) to
prioritize nodes by their depth rather than height. As long as it
doesn't stall, height is irrelevant. Depth represents the critical
path to the DAG root.
- Added hybrid_ls_rr_sort::isReady to filter stalled nodes before
adding them to the available queue.
Related Cleanup: most of the register reduction routines do not need
to be templates.
llvm-svn: 123468
simplification present in fully optimized code (I think instcombine fails to
transform some of these when "X-Y" has more than one use). Fires here and
there all over the test-suite, for example it eliminates 8 subtractions in
the final IR for 445.gobmk, 2 subs in 447.dealII, 2 in paq8p etc.
llvm-svn: 123442
threading of shifts over selects and phis while there. This fires here and
there in the testsuite, to not much effect. For example when compiling spirit
it fires 5 times, during early-cse, resulting in 6 more cse simplifications,
and 3 more terminators being folded by jump threading, but the final bitcode
doesn't change in any interesting way: other optimizations would have caught
the opportunity anyway, only later.
llvm-svn: 123441
instead of DomTree/DomFrontier. This may be interesting for reducing compile
time. This is currently disabled, but seems to work just fine.
When this is enabled, we eliminate two runs of dominator frontier, one in the
"early per-function" optimizations and one in the "interlaced with inliner"
function passes.
llvm-svn: 123434
- 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
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
the symbolic immediate names used for these instructions, fixing their pretty-printers, and
adding proper encoding information for them.
With this, we can properly pretty-print and encode assembly like:
mrc p15, #0, r3, c13, c0, #3
Fixes <rdar://problem/8857858>.
llvm-svn: 123404
set up the source operands. The original instr has an immediate operand that
should be replaced with the frame reg operand rather than just adding the
reg operand. Previously, the instruction ended up with too many operands
causing an assert() when adding the default predicate. rdar://8825456
llvm-svn: 123387
It will still return an iterator that points to the first terminator or end(),
but there may be DBG_VALUE instructions following the first terminator.
llvm-svn: 123384
This is a minor extension of SROA to handle a special case that is
important for some ARM NEON operations. Some of the NEON intrinsics
return multiple values, which are handled as struct types containing
multiple elements of the same vector type. The corresponding return
types declared in the arm_neon.h header have equivalent arrays. We
need SROA to recognize that it can split up those arrays and structs
into separate vectors, even though they are not always accessed with
the same type. SROA already handles loads and stores of an entire
alloca by using insertvalue/extractvalue to access the individual
pieces, and that code works the same regardless of whether the type
is a struct or an array. So, all that needs to be done is to check
for compatible arrays and homogeneous structs.
llvm-svn: 123381
SROA only split up structs and arrays one level at a time, so padding can
only cause trouble if it is located in between the struct or array elements.
llvm-svn: 123380
is "X != 0 -> X" when X is a boolean. This occurs a lot because of the way
llvm-gcc converts gcc's conditional expressions. Add this, and a few other
similar transforms for completeness.
llvm-svn: 123372
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
.code 32 if the TargetMachine's isThumb() boolean does not match. The correct
fix is to switch ARM subtargets at that point and is tracked by rdar://8856789
which is bigger task.
llvm-svn: 123353
Evan Cheng