As discussed on Issue #59217, under certain circumstances the DAG can generate duplicate MUL and MUL_LOHI nodes, often during MULO legalization.
This patch attempts to replace MUL nodes with additional uses of the LO result from the MUL_LOHI node
Differential Revision: https://reviews.llvm.org/D138790
This reverts commit e8b3ffa532b8ebac5dcdf17bb91b47817382c14d.
The AMDGPU/mad_64_32.ll seems to fail on some of the build bots but
passes locally. I'm really confused.
(sra X, BW-1) is either 0 or -1. So the multiply is a conditional
negate of Y.
This pattern shows up when type legalizing wide multiplies involving
a sign extended value.
Fixes PR57549.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D133399
This reverts commit 0148df8157f05ecf3b1064508e6f012aefb87dad.
Getting a lit test failures on AMDGPU but I can't reproduce it so far.
Reverting to investigate.
(sra X, BW-1) is either 0 or -1. So the multiply is a conditional
negate of Y.
This pattern shows up when type legalizing wide multiplies involving
a sign extended value.
Fixes PR57549.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D133399
This stops reporting CostPerUse 1 for `R8`-`R15` and `XMM8`-`XMM31`.
This was previously done because instruction encoding require a REX
prefix when using them resulting in longer instruction encodings. I
found that this regresses the quality of the register allocation as the
costs impose an ordering on eviction candidates. I also feel that there
is a bit of an impedance mismatch as the actual costs occure when
encoding instructions using those registers, but the order of VReg
assignments is not primarily ordered by number of Defs+Uses.
I did extensive measurements with the llvm-test-suite wiht SPEC2006 +
SPEC2017 included, internal services showed similar patterns. Generally
there are a log of improvements but also a lot of regression. But on
average the allocation quality seems to improve at a small code size
regression.
Results for measuring static and dynamic instruction counts:
Dynamic Counts (scaled by execution frequency) / Optimization Remarks:
Spills+FoldedSpills -5.6%
Reloads+FoldedReloads -4.2%
Copies -0.1%
Static / LLVM Statistics:
regalloc.NumSpills mean -1.6%, geomean -2.8%
regalloc.NumReloads mean -1.7%, geomean -3.1%
size..text mean +0.4%, geomean +0.4%
Static / LLVM Statistics:
mean -2.2%, geomean -3.1%) regalloc.NumSpills
mean -2.6%, geomean -3.9%) regalloc.NumReloads
mean +0.6%, geomean +0.6%) size..text
Static / LLVM Statistics:
regalloc.NumSpills mean -3.0%
regalloc.NumReloads mean -3.3%
size..text mean +0.3%, geomean +0.3%
Differential Revision: https://reviews.llvm.org/D133902
With SSE4.1 and above we were using 3 multiply instructions. This
was due to type legalization widening to v4i32 and the low half
being done with pmulld while the high half used two pmuldq/pmuludq.
Instead of that, we can use a single pmuludq/pmuldq to calculate
the full product at once, extract the high and low bits and compare
to check for overflow.
I've restricted SMULO to sse4.1 to get pmuldq. We can probably
do a fixup to pmuludq on earlier targets, but that's for another day.
I was going through my git stash and found an early version of this patch
from a year or two ago so I went ahead and finished it.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D130432
If all the demanded bits of the AND mask covering the inserted subvector 'X' are known to be one, then the mask isn't affecting the subvector at all.
In which case, if the base vector 'C' is undef/constant, then move the AND mask up to just (constant) fold it directly.
Addresses some of the regressions from D129150, particularly the cases where we're attempting to zero the upper elements of a widened vector.
Differential Revision: https://reviews.llvm.org/D129290
Test updates were performed using:
https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34
These are only the test updates where the test passed without
further modification (which is almost all of them, as the backend
is largely pointer-type agnostic).
We already perform some basic folds (add/sub with zero etc.) on scalar types, this patch adds some basic support for constant splats as well in a few cases (we can add more with future test coverage).
In the cases I've enabled, we can handle buildvector implicit truncation as we're not creating new constant nodes from the vector types - we're just returning existing nodes. This allows us to get a number of extra cases in the aarch64 tests.
I haven't enabled support for undefs in buildvector splats, as we're often checking for zero/allones patterns that return the original constant and we shouldn't be returning undef elements in some of these cases - we can enable this later if we're OK with creating new constants.
Differential Revision: https://reviews.llvm.org/D118264
This patch contains following enhancements to SrcRegMap and DstRegMap:
1 In findOnlyInterestingUse not only check if the Reg is two address usage,
but also check after commutation can it be two address usage.
2 If a physical register is clobbered, remove SrcRegMap entries that are
mapped to it.
3 In processTiedPairs, when create a new COPY instruction, add a SrcRegMap
entry only when the COPY instruction is coalescable. (The COPY src is
killed)
With these enhancements isProfitableToCommute can do better commute decision,
and finally more register copies are removed.
Differential Revision: https://reviews.llvm.org/D108731
This library function only exists in compiler-rt not libgcc. So
this would fail to link unless we were linking with compiler-rt.
This is consistent with the recent removal of calls to mulodi4 on
32-bit targets like D108928.
I suppose maybe we could keep the libcalls for platforms like
Darwin that use compiler-rt exclusively?
Reviewed By: nickdesaulniers, MaskRay
Differential Revision: https://reviews.llvm.org/D109385
The motivation is that the update script has at least two deviations
(`<...>@GOT`/`<...>@PLT`/ and not hiding pointer arithmetics) from
what pretty much all the checklines were generated with,
and most of the tests are still not updated, so each time one of the
non-up-to-date tests is updated to see the effect of the code change,
there is a lot of noise. Instead of having to deal with that each
time, let's just deal with everything at once.
This has been done via:
```
cd llvm-project/llvm/test/CodeGen/X86
grep -rl "; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py" | xargs -L1 <...>/llvm-project/llvm/utils/update_llc_test_checks.py --llc-binary <...>/llvm-project/build/bin/llc
```
Not all tests were regenerated, however.
Currently, X86 backend only has a global one-size-fits-all `FeatureFastVariableShuffle` feature,
which controls profitability of both the cross-lane and per-lane variable shuffles.
I guess, this has been fine so far.
But at least on AMD Zen 3, while per-line variable shuffles (e.g. `VPSHUFB`)
are as fast as as shuffles with fixed/immediate mask,
while lane-crossing shuffles, e.g. `VPERMPS` is performing worse.
So to get the benefits of variable-mask shuffles, but not the drawbacks of lane-crossing shuffles,
as suggested by @RKSimon, split the feature flag into two.
Differential Revision: https://reviews.llvm.org/D103274
The default expansion creates a MUL and either a MULHS/MULHU. Each
of those separately expand to sequences that use one or more
PMULLW instructions as well as additional instructions to
extend the types to vXi16. The MULHS/MULHU expansion computes the
whole 16-bit product, but only keeps the high part.
We can improve the lowering of SMULO/UMULO for some cases by using the MULHS/MULHU
expansion, but keep both the high and low parts. And we can use
those parts to calculate the overflow.
For AVX512 we might have vXi1 overflow outputs. We can improve those by using
vpcmpeqw to produce a k register if AVX512BW is enabled. This is a little better
than truncating the high result to use vpcmpeqb. If we don't have avx512bw we
can extend up to v16i32 to use vpcmpeqd to produce a k register.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D97624
For these cases we need to extract the upper or lower elements,
multiply them using 16-bit multiplies and repack them.
Previously we used punpcklbw/punpckhbw+psraw or pmovsxbw+pshudfd to
extract and sign extend so we could use pmullw to compute the 16-bit
product and then shift down the high bits.
We can avoid the need to sign extend if we unpack the bytes into
the high byte of each word and fill the lower byte with 0 using
pxor. This puts the sign bit of each byte into the sign bit of
each word. Since the LHS and RHS have 8 trailing zeros, the full
32-bit product of those 16-bit values will have 16 trailing zeros.
This means the 16-bit product of the original bytes is in the upper
16 bits which we can calculate using pmulhw.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D98587
Extend this to support ComputeNumSignBits of the (used) source vector elements so that we can handle more than just the case where we're sext_in_reg from the source element signbit.
Noticed while investigating the poor codegen in D98587.
A 1-bit smulo overflows is both inputs are -1 since the result
should be +1 which can't be represented in a signed 1 bit value.
We can detect this with an AND and a setcc. The multiply result
can also use the same AND.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D97634
Simon modified the check prefixes in these tests while D97160
was pending review. When D97160 was commited it wasn't updated
it merge cleanly, but didn't comprehend the check prefix changes.
Using ComputeNumSignBits or computeKnownBits we might be able
to determine that overflow is impossible.
This especially helps after type legalization if the type was
promoted from a type with half the bits or more. Type legalization
conservatively creates a promoted smulo/umulo and an overflow
check for the promoted bits. The overflow from the promoted
smulo/umulo is ORed with the result of the promoted bits
overflow check. Proving that the promoted smulo/umulo can never
overflow will leave us with just the promoted bits overflow check.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D97160
We can't store garbage in the unused bits. It possible that something like zextload from i1/i2/i4 is created to read the memory. Those zextloads would be legalized assuming the extra bits are 0.
I'm not sure that the code in lowerStore is executed for the v1i1/v2i1/v4i1 case. It looks like the DAG combine in combineStore may have converted them to v8i1 first. And I think we're missing some cases to avoid going to the stack in the first place. But I don't have time to investigate those things at the moment so I wanted to focus on the correctness issue.
Should fix PR48147.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D91294
KMOVWkr produces VK16, there's no reason to copy it to VK16 again.
Test changes are presumably because we were scheduling based on
the COPY that is no longer there.
If the mask input to getV4X86ShuffleImm8 only refers to a single source element (+ undefs) then canonicalize to a full broadcast.
getV4X86ShuffleImm8 defaults to inline values for undefs, which can be useful for shuffle widening/narrowing but does leave SimplifyDemanded* calls thinking the shuffle depends on unnecessary elements.
I'm still investigating what we should do more generally to avoid these undemanded elements, but broadcast cases was a simpler win.
If we lower a v2i64 shuffle to PSHUFD, we currently clamp undef elements to 0, (elements 0,1 of the v4i32) which can result in the shuffle referencing more elements of the source vector than expected, affecting later shuffle combines and KnownBits/SimplifyDemanded calls.
By ensuring we widen the undef mask element we allow getV4X86ShuffleImm8 to use inline elements as the default, which are more likely to fold.
We have a couple main strategies for legalizing MULH.
-If the vXi16 type is legal, extend to do the full i16 multiply
and then shift and truncate the results.
-Use unpcks to split each 128 bit lane into high and low halves.a
For signed we have an extra case to split a v32i8 to v16i8 and then
use the extending to v16i16 strategy.
This patch proposes to use the unpck strategy instead. Which is
what we already do for unsigned.
This seems to be 1 instruction shorter when the RHS is constant
like the idiv case. It's 1 instruction longer for the smulo case.
But we're trading cross lane shuffles for inlane shuffles and a
shift.
Differential Revision: https://reviews.llvm.org/D79652
This enables the register to be changed from XMM/YMM/ZMM0 to
instead match the other source. This prevents a false
dependency.
I added all the integer unpck instructions, but the tests
only show changes for BW and WD.
Unfortunately, we can have undef on operand 1 or 2 of the AVX
instructions. This breaks the interface with hasUndefRegUpdate
which used to tell which operand to check.
Now we scan the input operands looking for an undef register and
then ask hasUndefRegUpdate if its an instruction we care about
and which operands of that instruction we care about.
I also had to make some changes to the load folding code to
always pass operand 1 to hasUndefRegUpdate. I've updated
hasUndefRegUpdate to return false when ForLoadFold is set for
instructions that are not explicitly blocked for load folding in
isel patterns.
Differential Revision: https://reviews.llvm.org/D79615
If the SimplifyMultipleUseDemandedBits calls BITCASTs that peek through back to the original type then we can remove the BITCASTs entirely.
Differential Revision: https://reviews.llvm.org/D79572
We generate PACK instructions with an undef second source when we are truncating from a 128-bit vector to something narrower and we don't care about the upper bits of the vector register. The register allocation process will always assign untied undef uses to xmm0. This creates a false dependency on xmm0.
By adding these instructions to hasUndefRegUpdate, we can get the BreakFalseDeps pass to reassign the source to match the other input. Normally this interface is used for instructions that might need an xor inserted to break the dependency. But the pass also has a heuristic that tries to use the same register as other sources. That should always be possible for these instructions so we'll never trigger the xor dependency break.
Differential Revision: https://reviews.llvm.org/D79032
We had previously limited the shuffle(HORIZOP,HORIZOP) combine to binary shuffles, but we can often merge unary shuffles just as well, folding in UNDEF/ZERO values into the 64-bit half lanes.
For the (P)HADD/HSUB cases this is limited to fast-horizontal cases but PACKSS/PACKUS combines under all cases.
This patch attempts to peek through vectors based on the demanded bits/elt of a particular ISD::EXTRACT_VECTOR_ELT node, allowing us to avoid dependencies on ops that have no impact on the extract.
In particular this helps remove some unnecessary scalar->vector->scalar patterns.
The wasm shift patterns are annoying - @tlively has indicated that the wasm vector shift codegen are to be refactored in the near-term and isn't considered a major issue.
Reapplied after reversion at rL368660 due to PR42982 which was fixed at rGca7fdd41bda0.
Differential Revision: https://reviews.llvm.org/D65887
Summary:
After bugfix the undef value case here, we used more operations to implement inserting vxi1 sub vector into vXi1 vector, I optimize it by use less operations.
The history information at https://reviews.llvm.org/D68311
Reviewers: craig.topper, LuoYuanke, yubing, annita.zhang, pengfei, LiuChen3, RKSimon
Reviewed By: craig.topper
Subscribers: hiraditya, llvm-commits
Patch by Xiang Zhang (xiangzhangllvm)
Differential Revision: https://reviews.llvm.org/D71917
If we don't demand all elements, then attempt to combine to a simpler shuffle.
At the moment we can only do this if Depth == 0 as combineX86ShufflesRecursively uses Depth to track whether the shuffle has really changed or not - we'll need to change this before we can properly start merging combineX86ShufflesRecursively into SimplifyDemandedVectorElts (see D66004).
This reapplies rL368307 (reverted at rL369167) after the fix for the infinite loop reported at PR43024 was applied at rG3f087e38a2e7b87a5adaaac1c1b61e51220e7ff3
The previous code tried to do a trick where we would extract the subvector from the location we were inserting. Then xor that with the new value. Take the xored value and clear out the bits above the subvector size. Then shift that xored subvector to the insert location. And finally xor that with the original vector. Since the old subvector was used in both xors, this would leave just the new subvector at the inserted location. Since the surrounding bits had been zeroed no other bits of the original vector would be modified.
Unfortunately, if the old subvector came from undef we might aggressively propagate the undef. Then we end up with the XORs not cancelling because they aren't using the same value for the two uses of the old subvector. @bkramer gave me a case that demonstrated this, but we haven't reduced it enough to make it easily readable to see what's happening.
This patch uses a safer, but more costly approach. It isolate the bits above the insertion and bits below the insert point and ORs those together leaving 0 for the insertion location. Then widens the subvector with 0s in the upper bits, shifts it into position with 0s in the lower bits. Then we do another OR.
Differential Revision: https://reviews.llvm.org/D68311
llvm-svn: 373495
I also had to add a new combine to X86's combineExtractSubvector to prevent a regression.
This helps our vXi1 code see the full concat operation and allow it optimize undef to a zero if there is already a zero in the concat. This helped us use a movzx instead of an AND in some of the tests. In those tests, one concat comes from SelectionDAGBuilder and the second comes from type legalization of v4i1->i4 bitcasts which uses an additional concat. Though these changes weren't my original motivation.
I'm looking at making X86ISelLowering's narrowShuffle emit a concat_vectors instead of an insert_subvector since concat_vectors is more canonical during early DAG combine. This patch helps prevent a regression from my experiments with that.
Differential Revision: https://reviews.llvm.org/D66456
llvm-svn: 369459
This reverts r368662 (git commit 1a8d790cf5f89c1df718844f13e934e39bef6ef5)
The compile-time regression repro is in https://bugs.llvm.org/show_bug.cgi?id=43024
llvm-svn: 369167
If we don't demand all elements, then attempt to combine to a simpler shuffle.
At the moment we can only do this if Depth == 0 as combineX86ShufflesRecursively uses Depth to track whether the shuffle has really changed or not - we'll need to change this before we can properly start merging combineX86ShufflesRecursively into SimplifyDemandedVectorElts.
The insertps-combine.ll regression is because XFormVExtractWithShuffleIntoLoad can't see through shuffles of different widths - this will be fixed in a follow-up commit.
Reapplying this as rL368307 had to be reverted as part of rL368660 to revert rL368276
llvm-svn: 368662
This introduced a false positive MemorySanitizer warning about use of
uninitialized memory in a vectorized crc function in Chromium. That suggests
maybe something is not right with this transformation. See
https://crbug.com/992853#c7 for a reproducer.
This also reverts the follow-up commits r368307 and r368308 which
depended on this.
> This patch attempts to peek through vectors based on the demanded bits/elt of a particular ISD::EXTRACT_VECTOR_ELT node, allowing us to avoid dependencies on ops that have no impact on the extract.
>
> In particular this helps remove some unnecessary scalar->vector->scalar patterns.
>
> The wasm shift patterns are annoying - @tlively has indicated that the wasm vector shift codegen are to be refactored in the near-term and isn't considered a major issue.
>
> Differential Revision: https://reviews.llvm.org/D65887
llvm-svn: 368660
If we don't demand all elements, then attempt to combine to a simpler shuffle.
At the moment we can only do this if Depth == 0 as combineX86ShufflesRecursively uses Depth to track whether the shuffle has really changed or not - we'll need to change this before we can properly start merging combineX86ShufflesRecursively into SimplifyDemandedVectorElts.
The insertps-combine.ll regression is because XFormVExtractWithShuffleIntoLoad can't see through shuffles of different widths - this will be fixed in a follow-up commit.
llvm-svn: 368307
This patch attempts to peek through vectors based on the demanded bits/elt of a particular ISD::EXTRACT_VECTOR_ELT node, allowing us to avoid dependencies on ops that have no impact on the extract.
In particular this helps remove some unnecessary scalar->vector->scalar patterns.
The wasm shift patterns are annoying - @tlively has indicated that the wasm vector shift codegen are to be refactored in the near-term and isn't considered a major issue.
Differential Revision: https://reviews.llvm.org/D65887
llvm-svn: 368276
The assert that caused this to be reverted should be fixed now.
Original commit message:
This patch changes our defualt legalization behavior for 16, 32, and
64 bit vectors with i8/i16/i32/i64 scalar types from promotion to
widening. For example, v8i8 will now be widened to v16i8 instead of
promoted to v8i16. This keeps the elements widths the same and pads
with undef elements. We believe this is a better legalization strategy.
But it carries some issues due to the fragmented vector ISA. For
example, i8 shifts and multiplies get widened and then later have
to be promoted/split into vXi16 vectors.
This has the potential to cause regressions so we wanted to get
it in early in the 10.0 cycle so we have plenty of time to
address them.
Next steps will be to merge tests that explicitly test the command
line option. And then we can remove the option and its associated
code.
llvm-svn: 368183
This reverts commit 3de33245d2c992c9e0af60372043540b60f3a810.
This commit broke the MSan buildbots. See
https://reviews.llvm.org/rL367901 for more information.
llvm-svn: 368107
This patch changes our defualt legalization behavior for 16, 32, and
64 bit vectors with i8/i16/i32/i64 scalar types from promotion to
widening. For example, v8i8 will now be widened to v16i8 instead of
promoted to v8i16. This keeps the elements widths the same and pads
with undef elements. We believe this is a better legalization strategy.
But it carries some issues due to the fragmented vector ISA. For
example, i8 shifts and multiplies get widened and then later have
to be promoted/split into vXi16 vectors.
This has the potential to cause regressions so we wanted to get
it in early in the 10.0 cycle so we have plenty of time to
address them.
Next steps will be to merge tests that explicitly test the command
line option. And then we can remove the option and its associated
code.
llvm-svn: 367901
