Summary:
In rare cases, loads that don't get prefetched that were marked as
strided loads could cause a crash if they occurred in a loop with other
colliding loads.
Reviewers: mcrosier
Subscribers: aemerson, rengolin, javed.absar, kristof.beyls
Differential Revision: https://reviews.llvm.org/D38261
llvm-svn: 314252
Summary:
This addresses a correctness bug for LD[1234]*_POST opcodes that have
the prefetcher fix applied to them: the base register was not being
written back from the temp after being incremented, so it would appear
to never be incremented.
Also, fix some opcode tag computations based on some updated HW details
to get better tag avoidance and thus better prefetcher performance.
Reviewers: mcrosier
Subscribers: aemerson, rengolin, javed.absar, kristof.beyls
Differential Revision: https://reviews.llvm.org/D38256
llvm-svn: 314251
In the past while, I've committed a number of patches in the PowerPC back end
aimed at eliminating comparison instructions. However, this causes some failures
in proprietary source and these issues are not observed in SPEC or any open
source packages I've been able to run.
As a result, I'm pulling the entire series and will refactor it to:
- Have a single entry point for easy control
- Have fine-grained control over which patterns we transform
A side-effect of this is that test cases for these patches (and modified by
them) are XFAIL-ed. This is a temporary measure as it is counter-productive
to remove/modify these test cases and then have to modify them again when
the refactored patch is recommitted.
The failure will be investigated in parallel to the refactoring effort and
the recommit will either have a fix for it or will leave this transformation
off by default until the problem is resolved.
llvm-svn: 314244
This patch expands the support of lowerInterleavedStore to {8|16|32}x8i stride 3.
LLVM creates suboptimal shuffle code-gen for AVX2. In overall, this patch is a specific fix for the pattern (Strid=3 VF={8|16|32}) .
This patch is part two of two patches and it covers the store (interlevaed) side.
The patch goal is to optimize the following sequence:
a0 a1 a2 a3 a4 a5 a6 a7
b0 b1 b2 b3 b4 b5 b6 b7
c0 c1 c2 c3 c4 c5 c6 c7
into
a0 b0 c0 a1 b1 c1 a2 b2
c2 a3 b3 c3 a4 b4 c4 a5
b5 c5 a6 b6 c6 a7 b7 c7
Reviewers:
zvi
guyblank
dorit
Ayal
Differential Revision: https://reviews.llvm.org/D37117
Change-Id: I56ced8bcbea809a37654060771911ade20246ccc
llvm-svn: 314234
Removing X86 broadcast(f/i)32x2 intrinsics from llvm.
Adding autoUpgrade support.
Moving matching tests from avx512dq-intrinsics.ll to avx512dq-intrinsics-upgrade.ll and from avx512dqvl-intrinsics.ll to avx512dqvl-intrinsics-upgrade.ll.
Differential Revision: https://reviews.llvm.org/D38220
llvm-svn: 314195
It leads to some improvements, but also a regression for the simple
case, so it's not clearly a good idea.
test/CodeGen/ARM/vcvt.ll now has test coverage to show the difference.
Ultimately, the right solution is probably to custom-lower fp-to-int
conversions, to something like ARMISD::VCVT_F32_S32 plus a bitcast.
It's hard to do the right thing when the implicit bitcast isn't visible
to DAG transforms.
llvm-svn: 314169
R12 is used for the SwiftError parameter. It is no longer a CSR as it
is used for transfer the SwiftError, and the caller must preserve it if
they need to.
llvm-svn: 314165
As far as I know SUBREG_TO_REG is stating that the upper bits are 0. But if we are just converting the GR32 with no checks, then we have no reason to say the upper bits are 0.
I don't really know how to test this today since I can't find anything that looks that closely at SUBREG_TO_REG. The test changes here seems to be some perturbance of register allocation.
Differential Revision: https://reviews.llvm.org/D38001
llvm-svn: 314152
This teach simplifyDemandedBits to handle constant splat vector shifts.
This required changing some uses of getZExtValue to getLimitedValue since we can't rely on legalization using getShiftAmountTy for the shift amount.
I believe there may have been a bug in the ((X << C1) >>u ShAmt) handling where we didn't check if the inner shift was too large. I've fixed that here.
I had to add new patterns to ARM because the zext/sext the patterns were trying to look for got turned into an any_extend with this patch. Happy to split that out too, but not sure how to test without this change.
Differential Revision: https://reviews.llvm.org/D37665
llvm-svn: 314139
Summary:
This code iterates the 'Orders' vector in parallel with the DbgValue
list, emitting all DBG_VALUEs that occurred between the last IR order
insertion point and the next insertion point. This assumes the
SDDbgValue list is sorted in IR order, which it usually is. However, it
is not sorted when a node with a debug value is replaced with another
one. When this happens, TransferDbgValues is called, and the new value
is added to the end of the list.
The problem can be solved by stably sorting the list by IR order.
Reviewers: aprantl, Ka-Ka
Reviewed By: aprantl
Subscribers: MatzeB, hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D38197
llvm-svn: 314114
This patch expands the support of lowerInterleavedStore to 8x8i stride 4.
LLVM creates suboptimal shuffle code-gen for AVX2.
In overall, this patch is a specific fix for the pattern (Strid=4 VF=8) and we plan to include more patterns in the future.
The patch goal is to optimize the following sequence:
At the end of the computation, we have xmm2, xmm0, xmm12 and xmm3 holding
each 8 chars:
c0, c1, , c7
m0, m1, , m7
y0, y1, , y7
k0, k1, ., k7
And these need to be transposed/interleaved and stored like so:
c0 m0 y0 k0 c1 m1 y1 k1 c2 m2 y2 k2 c3 m3 y3 k3 ....
Reviewers
DavidKreitzer
Farhana
zvi
igorb
guyblank
RKSimon
Ayal
Differential Revision: https://reviews.llvm.org/D36058
Change-Id: I3cc5c2ca5d6318901c192a4428493b99ef424c32
llvm-svn: 314109
As mentioned in https://reviews.llvm.org/D33718, this simply adds another
pattern to the compare elimination sequence and is committed without a
differential review.
llvm-svn: 314106
This required changing the ISD opcode for these instructions to have the commutable operands first and the addend last. This way tablegen can autogenerate the additional patterns for us.
llvm-svn: 314083
This patch acts as a reverse to combineBitcastvxi1 - bitcasting a scalar integer to a boolean vector and extending it 'in place' to the requested legal type.
Currently this doesn't handle AVX512 at all - but the current mask register approach is lacking for some cases.
Differential Revision: https://reviews.llvm.org/D35320
llvm-svn: 314076
As mentioned in https://reviews.llvm.org/D33718, this simply adds another
pattern to the compare elimination sequence and is committed without a
differential review.
llvm-svn: 314073
We use a v16i32/v16f32 compare instead and truncate the result. We already did this for the unmasked version, but were missing the version with 'and'.
llvm-svn: 314072
This is a follow-up from D38181 (r314023). We have to put 64-bit
constants into a register using a separate instruction, so we
should try harder to avoid that.
From what I see, we're not likely to encounter this pattern in the
DAG because the upstream setcc combines from this don't (usually?)
produce this pattern. If we fix that, then this will become more
relevant. Since the cost of handling this case is just loosening
the predicate of the existing fold, we might as well do it now.
llvm-svn: 314064
As mentioned in https://reviews.llvm.org/D33718, this simply adds another
pattern to the compare elimination sequence and is committed without a
differential revision.
llvm-svn: 314062
As mentioned in https://reviews.llvm.org/D33718, this simply adds another
pattern to the compare elimination sequence and is committed without a
differential revision.
llvm-svn: 314060
As mentioned in https://reviews.llvm.org/D33718, this simply adds another
pattern to the compare elimination sequence and is committed without a
differential revision.
llvm-svn: 314055
The (non-)obvious win comes from saving 3 bytes by using the 0x83 'and' opcode variant instead of 0x81.
There are also better improvements based on known-bits that allow us to eliminate the mask entirely.
As noted, this could be extended. There are potentially other wins from always shifting first, but doing
that reveals a tangle of problems in other pattern matching. We do this transform generically in
instcombine, but we often have icmp IR that doesn't match that pattern, so we must account for this
in the backend.
Differential Revision: https://reviews.llvm.org/D38181
llvm-svn: 314023
Summary: Conditional returns were not taken into consideration at all. Implement them by turning them into jumps and normal returns. This means there is a slightly higher performance penalty for conditional returns, but this is the best we can do, and it still disturbs little of the rest.
Reviewers: dberris, echristo
Subscribers: sanjoy, nemanjai, hiraditya, kbarton, llvm-commits
Differential Revision: https://reviews.llvm.org/D38102
llvm-svn: 314005
If the two instructions being compared for equivalence have corresponding operands
that are integer constants, then check their values to determine equivalence.
Patch by Suyog Sarda!
llvm-svn: 313993
Combine CMOV[i16]<-[SIGN,ZERO,ANY]_EXTEND to [i32,i64] into CMOV[i32,i64].
One example of where it is useful is:
before (20 bytes)
<foo>:
test $0x1,%dil
mov $0x307e,%ax
mov $0xffff,%cx
cmovne %ax,%cx
movzwl %cx,%eax
retq
after (18 bytes)
<foo>:
test $0x1,%dil
mov $0x307e,%ecx
mov $0xffff,%eax
cmovne %ecx,%eax
retq
Reviewers: craig.topper, aaboud, spatel, RKSimon, zvi
Reviewed By: spatel
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D36711
llvm-svn: 313982
This patch re-commits the patch that was pulled out due to a
problem it caused, but with a fix for the problem. The fix
was reviewed separately by Eric Christopher and Hal Finkel.
Differential Revision: https://reviews.llvm.org/D38054
llvm-svn: 313978
For the following function:
double fn1(double d0, double d1, double d2) {
double a = -d0 - d1 * d2;
return a;
}
on ARM, LLVM generates code along the lines of
vneg.f64 d0, d0
vmls.f64 d0, d1, d2
i.e., a negate and a multiply-subtract.
The attached patch adds instruction selection patterns to allow it to generate the single instruction
vnmla.f64 d0, d1, d2
(multiply-add with negation) instead, like GCC does.
Committed on behalf of @gergo- (Gergö Barany)
Differential Revision: https://reviews.llvm.org/D35911
llvm-svn: 313972
Craig Topper