A sequence of VMUL and VADD instructions always give the same or better
performance than a fused VMLA instruction on the Cortex-M4 and Cortex-M33.
Executing the VMUL and VADD back-to-back requires the same cycles, but
having separate instructions allows scheduling to avoid the hazard between
these 2 instructions.
Differential Revision: https://reviews.llvm.org/D52289
llvm-svn: 342874
There seems to be a separate command line tokenization issue that
prevents just ':\s*$' from working, since then the pattern argument
isn't quoted, and grep.exe misinterprets the backslash somehow.
llvm-svn: 342430
There is no way in the universe, that doing a full-width division in
software will be faster than doing overflowing multiplication in
software in the first place, especially given that this same full-width
multiplication needs to be done anyway.
This patch replaces the previous implementation with a direct lowering
into an overflowing multiplication algorithm based on half-width
operations.
Correctness of the algorithm was verified by exhaustively checking the
output of this algorithm for overflowing multiplication of 16 bit
integers against an obviously correct widening multiplication. Baring
any oversights introduced by porting the algorithm to DAG, confidence in
correctness of this algorithm is extremely high.
Following table shows the change in both t = runtime and s = space. The
change is expressed as a multiplier of original, so anything under 1 is
“better” and anything above 1 is worse.
+-------+-----------+-----------+-------------+-------------+
| Arch | u64*u64 t | u64*u64 s | u128*u128 t | u128*u128 s |
+-------+-----------+-----------+-------------+-------------+
| X64 | - | - | ~0.5 | ~0.64 |
| i686 | ~0.5 | ~0.6666 | ~0.05 | ~0.9 |
| armv7 | - | ~0.75 | - | ~1.4 |
+-------+-----------+-----------+-------------+-------------+
Performance numbers have been collected by running overflowing
multiplication in a loop under `perf` on two x86_64 (one Intel Haswell,
other AMD Ryzen) based machines. Size numbers have been collected by
looking at the size of function containing an overflowing multiply in
a loop.
All in all, it can be seen that both performance and size has improved
except in the case of armv7 where code size has regressed for 128-bit
multiply. u128*u128 overflowing multiply on 32-bit platforms seem to
benefit from this change a lot, taking only 5% of the time compared to
original algorithm to calculate the same thing.
The final benefit of this change is that LLVM is now capable of lowering
the overflowing unsigned multiply for integers of any bit-width as long
as the target is capable of lowering regular multiplication for the same
bit-width. Previously, 128-bit overflowing multiply was the widest
possible.
Patch by Simonas Kazlauskas!
Differential Revision: https://reviews.llvm.org/D50310
llvm-svn: 339922
The original code attempted to do this, but the std::abs() call didn't
actually do anything due to implicit type conversions. Fix the type
conversions, and perform the correct check for negative immediates.
This probably has very little practical impact, but it's worth fixing
just to avoid confusion in the future, I think.
Differential Revision: https://reviews.llvm.org/D48907
llvm-svn: 336742
Summary:
Two utils methods have essentially the same functionality. This is an attempt to merge them into one.
1. lib/Transforms/Utils/Local.cpp : MergeBasicBlockIntoOnlyPred
2. lib/Transforms/Utils/BasicBlockUtils.cpp : MergeBlockIntoPredecessor
Prior to the patch:
1. MergeBasicBlockIntoOnlyPred
Updates either DomTree or DeferredDominance
Moves all instructions from Pred to BB, deletes Pred
Asserts BB has single predecessor
If address was taken, replace the block address with constant 1 (?)
2. MergeBlockIntoPredecessor
Updates DomTree, LoopInfo and MemoryDependenceResults
Moves all instruction from BB to Pred, deletes BB
Returns if doesn't have a single predecessor
Returns if BB's address was taken
After the patch:
Method 2. MergeBlockIntoPredecessor is attempting to become the new default:
Updates DomTree or DeferredDominance, and LoopInfo and MemoryDependenceResults
Moves all instruction from BB to Pred, deletes BB
Returns if doesn't have a single predecessor
Returns if BB's address was taken
Uses of MergeBasicBlockIntoOnlyPred that need to be replaced:
1. lib/Transforms/Scalar/LoopSimplifyCFG.cpp
Updated in this patch. No challenges.
2. lib/CodeGen/CodeGenPrepare.cpp
Updated in this patch.
i. eliminateFallThrough is straightforward, but I added using a temporary array to avoid the iterator invalidation.
ii. eliminateMostlyEmptyBlock(s) methods also now use a temporary array for blocks
Some interesting aspects:
- Since Pred is not deleted (BB is), the entry block does not need updating.
- The entry block was being updated with the deleted block in eliminateMostlyEmptyBlock. Added assert to make obvious that BB=SinglePred.
- isMergingEmptyBlockProfitable assumes BB is the one to be deleted.
- eliminateMostlyEmptyBlock(BB) does not delete BB on one path, it deletes its unique predecessor instead.
- adding some test owner as subscribers for the interesting tests modified:
test/CodeGen/X86/avx-cmp.ll
test/CodeGen/AMDGPU/nested-loop-conditions.ll
test/CodeGen/AMDGPU/si-annotate-cf.ll
test/CodeGen/X86/hoist-spill.ll
test/CodeGen/X86/2006-11-17-IllegalMove.ll
3. lib/Transforms/Scalar/JumpThreading.cpp
Not covered in this patch. It is the only use case using the DeferredDominance.
I would defer to Brian Rzycki to make this replacement.
Reviewers: chandlerc, spatel, davide, brzycki, bkramer, javed.absar
Subscribers: qcolombet, sanjoy, nemanjai, nhaehnle, jlebar, tpr, kbarton, RKSimon, wmi, arsenm, llvm-commits
Differential Revision: https://reviews.llvm.org/D48202
llvm-svn: 335183
Thumb has more 16-bit encoding space dedicated to ADD than ORR, allowing both a
3-address encoding and a wider range of immediates. So, particularly when
optimizing for code size (but it doesn't make things worse elsewhere) it's
beneficial to select an OR operation to an ADD if we know overflow won't occur.
This is made even better by LLVM's penchant for putting operations in canonical
form by converting the other way.
llvm-svn: 335119
Shows some missed optimizations for the -7929856 and -2166 testcases.
-7929856 is due to a bug in ARMTargetLowering::getARMCmp, I think;
the -2166 case is a missing pattern.
llvm-svn: 335004
We need to clean up the DAG floating-point undef logic.
This process is similar to how we handled integer undef
logic in D43141.
And as we did there, I'm trying to reduce the patch by
changing tests that would probably become meaningless
once we correct FP undef folding.
llvm-svn: 332548
We need to clean up the DAG floating-point undef logic.
This process is similar to how we handled integer undef
logic in D43141.
And as we did there, I'm trying to reduce the patch by
changing tests that would probably become meaningless
once we correct FP undef folding.
llvm-svn: 332547
Summary:
This pass sinks COPY instructions into a successor block, if the COPY is not
used in the current block and the COPY is live-in to a single successor
(i.e., doesn't require the COPY to be duplicated). This avoids executing the
the copy on paths where their results aren't needed. This also exposes
additional opportunites for dead copy elimination and shrink wrapping.
These copies were either not handled by or are inserted after the MachineSink
pass. As an example of the former case, the MachineSink pass cannot sink
COPY instructions with allocatable source registers; for AArch64 these type
of copy instructions are frequently used to move function parameters (PhyReg)
into virtual registers in the entry block..
For the machine IR below, this pass will sink %w19 in the entry into its
successor (%bb.1) because %w19 is only live-in in %bb.1.
```
%bb.0:
%wzr = SUBSWri %w1, 1
%w19 = COPY %w0
Bcc 11, %bb.2
%bb.1:
Live Ins: %w19
BL @fun
%w0 = ADDWrr %w0, %w19
RET %w0
%bb.2:
%w0 = COPY %wzr
RET %w0
```
As we sink %w19 (CSR in AArch64) into %bb.1, the shrink-wrapping pass will be
able to see %bb.0 as a candidate.
With this change I observed 12% more shrink-wrapping candidate and 13% more dead copies deleted in spec2000/2006/2017 on AArch64.
Reviewers: qcolombet, MatzeB, thegameg, mcrosier, gberry, hfinkel, john.brawn, twoh, RKSimon, sebpop, kparzysz
Reviewed By: sebpop
Subscribers: evandro, sebpop, sfertile, aemerson, mgorny, javed.absar, kristof.beyls, llvm-commits
Differential Revision: https://reviews.llvm.org/D41463
llvm-svn: 328237
Enable multiple COPY hints to eliminate more COPYs during register allocation.
Note that this is something all targets should do, see
https://reviews.llvm.org/D38128.
Review: Eli Friedman
llvm-svn: 325327
This patch combines some cases of ARMISD::CMOV for integers that arise in comparisons of the form
a != b ? x : 0
a == b ? 0 : x
and that currently (e.g. in Thumb1) are emitted as branches.
Differential Revision: https://reviews.llvm.org/D34515
llvm-svn: 325323
Discussed here:
http://lists.llvm.org/pipermail/llvm-dev/2018-January/120320.html
In preparation for adding support for named vregs we are changing the sigil for
physical registers in MIR to '$' from '%'. This will prevent name clashes of
named physical register with named vregs.
llvm-svn: 323922
In -debug output we print "pred:" whenever a MachineOperand is a
predicate operand in the instruction descriptor, and "opt:" whenever a
MachineOperand is an optional def in the instruction descriptor.
Differential Revision: https://reviews.llvm.org/D41870
llvm-svn: 322096
As part of the unification of the debug format and the MIR format, print
MBB references as '%bb.5'.
The MIR printer prints the IR name of a MBB only for block definitions.
* find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)->getNumber\(\)/" << printMBBReference(*\1)/g'
* find . \( -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#" << ([a-zA-Z0-9_]+)\.getNumber\(\)/" << printMBBReference(\1)/g'
* find . \( -name "*.txt" -o -name "*.s" -o -name "*.mir" -o -name "*.cpp" -o -name "*.h" -o -name "*.ll" \) -type f -print0 | xargs -0 sed -i '' -E 's/BB#([0-9]+)/%bb.\1/g'
* grep -nr 'BB#' and fix
Differential Revision: https://reviews.llvm.org/D40422
llvm-svn: 319665
output
As part of the unification of the debug format and the MIR format,
always use `printReg` to print all kinds of registers.
Updated the tests using '_' instead of '%noreg' until we decide which
one we want to be the default one.
Differential Revision: https://reviews.llvm.org/D40421
llvm-svn: 319445
Summary:
This fixes PR35221.
Use pseudo-instructions to let MachineCSE hoist global address computation.
Subscribers: aemerson, javed.absar, kristof.beyls, llvm-commits, hiraditya
Differential Revision: https://reviews.llvm.org/D39871
llvm-svn: 318081
As far as I can tell, this matches gcc: -mfloat-abi determines the
calling convention for all functions except those explicitly defined as
soft-float in the ARM RTABI.
This change only affects cases where the user specifies -mfloat-abi to
override the default calling convention derived from the target triple.
Fixes https://bugs.llvm.org//show_bug.cgi?id=34530.
Differential Revision: https://reviews.llvm.org/D38299
llvm-svn: 316708
Swap the compare operands if the lhs is a shift and the rhs isn't,
as in arm and T2 the shift can be performed by the compare for its
second operand.
Differential Revision: https://reviews.llvm.org/D39004
llvm-svn: 316562
The ARM backend should call setBooleanContents so that it can
use known bits to make some optimizations.
Review: D35821
Patch by Joel Galenson <jgalenson@google.com>
llvm-svn: 311446
Summary:
This patch teaches PostDominatorTree about infinite loops. It is built on top of D29705 by @dberlin which includes a very detailed motivation for this change.
What's new is that the patch also teaches the incremental updater how to deal with reverse-unreachable regions and how to properly maintain and verify tree roots. Before that, the incremental algorithm sometimes ended up preserving reverse-unreachable regions after updates that wouldn't appear in the tree if it was constructed from scratch on the same CFG.
This patch makes the following assumptions:
- A sequence of updates should produce the same tree as a recalculating it.
- Any sequence of the same updates should lead to the same tree.
- Siblings and roots are unordered.
The last two properties are essential to efficiently perform batch updates in the future.
When it comes to the first one, we can decide later that the consistency between freshly built tree and an updated one doesn't matter match, as there are many correct ways to pick roots in infinite loops, and to relax this assumption. That should enable us to recalculate postdominators less frequently.
This patch is pretty conservative when it comes to incremental updates on reverse-unreachable regions and ends up recalculating the whole tree in many cases. It should be possible to improve the performance in many cases, if we decide that it's important enough.
That being said, my experiments showed that reverse-unreachable are very rare in the IR emitted by clang when bootstrapping clang. Here are the statistics I collected by analyzing IR between passes and after each removePredecessor call:
```
# functions: 52283
# samples: 337609
# reverse unreachable BBs: 216022
# BBs: 247840796
Percent reverse-unreachable: 0.08716159869015269 %
Max(PercRevUnreachable) in a function: 87.58620689655172 %
# > 25 % samples: 471 ( 0.1395104988314885 % samples )
... in 145 ( 0.27733680163724345 % functions )
```
Most of the reverse-unreachable regions come from invalid IR where it wouldn't be possible to construct a PostDomTree anyway.
I would like to commit this patch in the next week in order to be able to complete the work that depends on it before the end of my internship, so please don't wait long to voice your concerns :).
Reviewers: dberlin, sanjoy, grosser, brzycki, davide, chandlerc, hfinkel
Reviewed By: dberlin
Subscribers: nhaehnle, javed.absar, kparzysz, uabelho, jlebar, hiraditya, llvm-commits, dberlin, david2050
Differential Revision: https://reviews.llvm.org/D35851
llvm-svn: 310940
`llc -march` is problematic because it only switches the target
architecture, but leaves the operating system unchanged. This
occasionally leads to indeterministic tests because the OS from
LLVM_DEFAULT_TARGET_TRIPLE is used.
However we can simply always use `llc -mtriple` instead. This changes
all the tests to do this to avoid people using -march when they copy and
paste parts of tests.
See also the discussion in https://reviews.llvm.org/D35287
llvm-svn: 309755
When we have a diamond ifcvt the fallthough block will have a branch at the end
of it that disappears when predicated, so discount it from the predication cost.
Differential Revision: https://reviews.llvm.org/D34952
llvm-svn: 307788
The current heuristic in isProfitableToIfCvt assumes we have a branch predictor,
and so gives the wrong answer in some cases when we don't. This patch adds a
subtarget feature to indicate that a subtarget has no branch predictor, and
changes the heuristic in isProfitableToiIfCvt when it's present. This gives a
slight overall improvement in a set of embedded benchmarks on Cortex-M4 and
Cortex-M33.
Differential Revision: https://reviews.llvm.org/D34398
llvm-svn: 306547
The benchmarking summarized in
http://lists.llvm.org/pipermail/llvm-dev/2017-May/113525.html showed
this is beneficial for a wide range of cores.
As is to be expected, quite a few small adaptations are needed to the
regressions tests, as the difference in scheduling results in:
- Quite a few small instruction schedule differences.
- A few changes in register allocation decisions caused by different
instruction schedules.
- A few changes in IfConversion decisions, due to a difference in
instruction schedule and/or the estimated cost of a branch mispredict.
llvm-svn: 306514
This has been deprecated since ARMARM v7-AR, release C.b, published back
in 2012.
This also removes test/CodeGen/Thumb2/ifcvt-neon.ll that originally was
introduced to check that conditionalization of Neon instructions did
happen when generating Thumb2. However, the test had evolved and was no
longer testing that. Rather than trying to adapt that test, this commit
introduces test/CodeGen/Thumb2/ifcvt-neon-deprecated.mir, since we can
now use the MIR framework to write nicer/more maintainable tests.
llvm-svn: 305998
Summary:
In SelectionDAG, when a store is immediately chained to another store
to the same address, elide the first store as it has no observable
effects. This is causes small improvements dealing with intrinsics
lowered to stores.
Test notes:
* Many testcases overwrite store addresses multiple times and needed
minor changes, mainly making stores volatile to prevent the
optimization from optimizing the test away.
* Many X86 test cases optimized out instructions associated with
associated with va_start.
* Note that test_splat in CodeGen/AArch64/misched-stp.ll no longer has
dependencies to check and can probably be removed and potentially
replaced with another test.
Reviewers: rnk, john.brawn
Subscribers: aemerson, rengolin, qcolombet, jyknight, nemanjai, nhaehnle, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D33206
llvm-svn: 303198
Shrink-wrapping uses post-dominators to find a restore point that
post-dominates all the uses of CSR / stack.
The way dominator trees are modeled in LLVM today is that unreachable
blocks are not present in a generic dominator tree, so, an unreachable node is
dominated by anything: include/llvm/Support/GenericDomTree.h:467.
Since for post-dominators, a no-return block is considered
"unreachable", calling findNearestCommonDominator on an unreachable node
A and a non-unreachable node B, will return B, which can be false. If we
find such node, we bail out since there is no good restore point
available.
rdar://problem/30186931
llvm-svn: 303130
During the optimisation of jump tables in the constant island pass,
an extra ADD could be left over, now dead but not removed.
Differential Revision: https://reviews.llvm.org/D31389
llvm-svn: 299634
I previously removed the T2XtPk feature from the ARM backend, but it
looks like I missed some of the tests that were using the feature.
Differential Revision: https://reviews.llvm.org/D30778
llvm-svn: 297386
Removed the HasT2ExtractPack feature and replaced its references
with HasDSP. This then allows the Thumb2 extend instructions to be
selected for ARMv8M +dsp. These instruction descriptions have also
been refactored and more target tests have been added for their isel.
Differential Revision: https://reviews.llvm.org/D29623
llvm-svn: 295452
When generating a floating point comparison we currently unconditionally
generate VCMPE. This has the sideeffect of setting the cumulative Invalid
bit in FPSCR if any of the operands are QNaN.
It is expected that use of a relational predicate on a QNaN value should
raise Invalid. Quoting from the C standard:
The relational and equality operators support the usual mathematical
relationships between numeric values. For any ordered pair of numeric
values exactly one of relationships the less, greater, equal and is true.
Relational operators may raise the floating-point exception when argument
values are NaNs.
The standard doesn't explicitly state the expectation for equality operators,
but the implication and obvious expectation is that equality operators
should not raise Invalid on a QNaN input, as those predicates are wholly
defined on unordered inputs (to return not equal).
Therefore, add a new operand to ARMISD::FPCMP and FPCMPZ indicating if
QNaN should raise Invalid, and pipe that through to TableGen.
llvm-svn: 294945
When choosing the best successor for a block, ordinarily we would have preferred
a block that preserves the CFG unless there is a strong probability the other
direction. For small blocks that can be duplicated we now skip that requirement
as well, subject to some simple frequency calculations.
Differential Revision: https://reviews.llvm.org/D28583
llvm-svn: 293716
The Requires class overrides the target requirements of an instruction,
rather than adding to them, so all ARM instructions need to include the
IsARM predicate when they have overwitten requirements.
This caused the swp and swpb instructions to be allowed in thumb mode
assembly, and the ARM encoding of CDP to be selected in codegen (which
is different for conditional instructions).
Differential Revision: https://reviews.llvm.org/D29283
llvm-svn: 293634
GCC changes the CC between the user-code and the builtins based on the
value of `-target` rather than `-mfloat-abi`. When a HF target is used,
the VFP variant of the AAPCS CC is used. Otherwise, the AAPCS variant
is used. In all cases, the AEABI functions use the AAPCS CC. Adjust
the calling convention based on the target.
Resolves PR30543!
llvm-svn: 291909
This reverts commit ada6595a526d71df04988eb0a4b4fe84df398ded.
This needs a simple probability check because there are some cases where it is
not profitable.
llvm-svn: 291695
When choosing the best successor for a block, ordinarily we would have preferred
a block that preserves the CFG unless there is a strong probability the other
direction. For small blocks that can be duplicated we now skip that requirement
as well.
Differential revision: https://reviews.llvm.org/D27742
llvm-svn: 291609
1.Fix pessimized case in FIXME.
2.Add tests for it.
3.The canonicalisation on shifts results in different sequence for
tests of machine-licm.Correct some check lines.
Differential Revision: https://reviews.llvm.org/D27916
llvm-svn: 290410
This is essentially a recommit of r285893, but with a correctness fix. The
problem of the original commit was that this:
bic r5, r7, #31
cbz r5, .LBB2_10
got rewritten into:
lsrs r5, r7, #5
beq .LBB2_10
The result in destination register r5 is not the same and this is incorrect
when r5 is not dead. So this fix includes checking the uses of the AND
destination register. And also, compared to the original commit, some regression
tests didn't need changing anymore because of this extra check.
For completeness, this was the original commit message:
For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more
efficient instruction selection if the bitmask is one consecutive sequence of
set bits (32 - clz(bm) - ctz(bm) == popcount(bm)).
1) If the bitmask touches the LSB, then we can remove all the upper bits and
set the flags by doing one LSLS.
2) If the bitmask touches the MSB, then we can remove all the lower bits and
set the flags with one LSRS.
3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit
into the sign bit with one LSLS and change the condition query from NE/EQ to
MI/PL (we could also implement this by shifting into the carry bit and
branching on BCC/BCS).
4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower
zero bits of the mask.
1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two
16-bit instructions but can elide the CMP and doesn't require materializing a
complex immediate, so is also a win.
Differential Revision: https://reviews.llvm.org/D27761
llvm-svn: 289794
This recommits r281323, which was backed out for two reasons. One, a selfhost failure, and two, it apparently caused Chromium failures. Actually, the latter was a red herring. The log has expired from the former, but I suspect that was a red herring too (actually caused by another problematic patch of mine). Therefore reapplying, and will watch the bots like a hawk.
For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more efficient instruction selection if the bitmask is one consecutive sequence of set bits (32 - clz(bm) - ctz(bm) == popcount(bm)).
1) If the bitmask touches the LSB, then we can remove all the upper bits and set the flags by doing one LSLS.
2) If the bitmask touches the MSB, then we can remove all the lower bits and set the flags with one LSRS.
3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit into the sign bit with one LSLS and change the condition query from NE/EQ to MI/PL (we could also implement this by shifting into the carry bit and branching on BCC/BCS).
4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower zero bits of the mask.
1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two 16-bit instructions but can elide the CMP and doesn't require materializing a complex immediate, so is also a win.
llvm-svn: 285893
[Reapplying r284580 and r285917 with fix and testing to ensure emitted jump tables for Thumb-1 have 4-byte alignment]
The TBB and TBH instructions in Thumb-2 allow jump tables to be compressed into sequences of bytes or shorts respectively. These instructions do not exist in Thumb-1, however it is possible to synthesize them out of a sequence of other instructions.
It turns out this sequence is so short that it's almost never a lose for performance and is ALWAYS a significant win for code size.
TBB example:
Before: lsls r0, r0, #2 After: add r0, pc
adr r1, .LJTI0_0 ldrb r0, [r0, #6]
ldr r0, [r0, r1] lsls r0, r0, #1
mov pc, r0 add pc, r0
=> No change in prologue code size or dynamic instruction count. Jump table shrunk by a factor of 4.
The only case that can increase dynamic instruction count is the TBH case:
Before: lsls r0, r4, #2 After: lsls r4, r4, #1
adr r1, .LJTI0_0 add r4, pc
ldr r0, [r0, r1] ldrh r4, [r4, #6]
mov pc, r0 lsls r4, r4, #1
add pc, r4
=> 1 more instruction in prologue. Jump table shrunk by a factor of 2.
So there is an argument that this should be disabled when optimizing for performance (and a TBH needs to be generated). I'm not so sure about that in practice, because on small cores with Thumb-1 performance is often tied to code size. But I'm willing to turn it off when optimizing for performance if people want (also note that TBHs are fairly rare in practice!)
llvm-svn: 285690
Sjoerd Meijer