If we have a known non-nan operand, place it in the second operand
of fmin/fmax that is returned if either operand is nan.
Differential Revision: https://reviews.llvm.org/D62448
llvm-svn: 361704
The test diffs show improved vector narrowing for integer min/max opcodes because
those were all absent from the list. I'm not sure if we can expose functional diffs
for all of the moved/added opcodes though.
It seems like we are missing an AVX512 opportunity to use 256-bit ops in place of
512-bit ops on some tests/targets, but I think that can be a follow-up.
Preliminary steps to make sure the callers are not misusing these queries:
rL361268
rL361547
Differential Revision: https://reviews.llvm.org/D62191
llvm-svn: 361701
The test based on PR42010:
https://bugs.llvm.org/show_bug.cgi?id=42010
...may show an inaccuracy for PPC's target defs, but we should not
be so aggressive with an assert here. There's no telling what out-of-tree
targets look like.
llvm-svn: 361696
INC/DEC is really a special case of a more generic issue. We should also turn leas into add reg/reg or add reg/imm regardless of the slow lea flags.
This also supports LEA64_32 which has 64 bit input registers and 32 bit output registers. So we need to convert the 64 bit inputs to their 32 bit equivalents to check if they are equal to base reg.
One thing to note, the original code preserved the kill flags by adding operands to the new instruction instead of using addReg. But I think tied operands aren't supposed to have the kill flag set. I dropped the kill flags, but I could probably try to preserve it in the add reg/reg case if we think its important. Not sure which operand its supposed to go on for the LEA64_32r instruction due to the super reg implicit uses. Though I'm also not sure those are needed since they were probably just created by an INSERT_SUBREG from a 32-bit input.
Differential Revision: https://reviews.llvm.org/D61472
llvm-svn: 361691
In a few places in getInstrMapping, we check if use/def instructions for the
instruction we're mapping have floating point constraints.
We can improve this check and reduce the number of copies in GISel-compiled code
if we make a couple observations:
- For a def instruction, it only matters if the def instruction must always
output a value stored on a FPR
- For a use instruction, it only matters if the use instruction must always
only take in values stored in FPRs
This adds two new functions:
- onlyUsesFP
- onlyDefinesFP
Then we can use those when we're checking the uses/defs instead.
Without this patch, the load, unmerge, store, and select in the added test
would have unnecessary copies.
Differential Revision: https://reviews.llvm.org/D62426
llvm-svn: 361679
Summary:dd
This patch implements call lowering for calls without parameters
on AIX as initial support.
Reviewers: sfertile, hubert.reinterpretcast, aheejin, efriedma
Differential Revision: https://reviews.llvm.org/D61948
llvm-svn: 361669
The fcsel and csel instructions differ in only the register banks they work on.
So, they're entirely interchangeable otherwise.
With this in mind, this does two things:
- Teach AArch64RegisterBankInfo to consider the inputs to G_SELECT as well as
the outputs.
- Teach it to choose the best register bank mapping based off the constraints
of the inputs and outputs.
The "best" in this case means the one that requires the smallest number of
copies to properly emit a fcsel/csel.
For example, if the inputs are all already going to be on FPRs, we should
emit a fcsel, even if the output is a GPR. This costs one copy to produce the
result, but saves us from copying the inputs into GPRs.
Also update the regbank-select.mir to check that we end up with the right
select instruction.
Differential Revision: https://reviews.llvm.org/D62267
llvm-svn: 361665
Details: To make instruction selection really divergence driven it is necessary to assign
the correct register classes to the cross block values beforehand. For the divergent targets
same value type requires different register classes dependent on the value divergence.
Reviewers: rampitec, nhaehnle
Differential Revision: https://reviews.llvm.org/D59990
llvm-svn: 361644
For the situation, where we generate the following code:
crxor 8, 8, 8
< Some instructions>
.LBB0_1:
< Some instructions>
cror 1, 8, 8
cror (COPY of CRbit) depends on the result of the crxor instruction.
CR8 is known to be zero as crxor is equivalent to CRUNSET. We can simply use
crxor 1, 1, 1 instead to zero out CR1, which does not have any dependency on
any previous instruction.
This patch will optimize it to:
< Some instructions>
.LBB0_1:
< Some instructions>
cror 1, 1, 1
Patch By: Victor Huang (NeHuang)
Differential Revision: https://reviews.llvm.org/D62044
llvm-svn: 361632
This patch adds the overridable TargetLowering::getTargetConstantFromLoad function which allows targets to return any constant value loaded by a LoadSDNode node - only X86 makes use of this so far but everything should be in place for other targets.
computeKnownBits then uses this function to improve codegen, notably vector code after legalization.
A future commit will do the same for ComputeNumSignBits but computeKnownBits sees the bigger benefit.
This required a couple of fixes:
* SimplifyDemandedBits must early-out for getTargetConstantFromLoad cases to prevent infinite loops of constant regeneration (similar to what we already do for BUILD_VECTOR).
* Fix a DAGCombiner::visitTRUNCATE issue as we had trunc(shl(v8i32),v8i16) <-> shl(trunc(v8i16),v8i32) infinite loops after legalization on AVX512 targets.
Differential Revision: https://reviews.llvm.org/D61887
llvm-svn: 361620
swifterror marks an argument as a register pretending to be a pointer, so we
need a guaranteed mem2reg-like analysis of its uses. Fortunately most of the
infrastructure can be reused from the DAG world.
llvm-svn: 361608
The D45316 introduced the `shouldTransformMulToShiftsAddsSubs` function
to check that breaking down constant multiplications into a series
of shifts, adds, and subs is efficient. Unfortunately, this function
does not check maximum number of steps on all paths of the algorithm.
This patch fixes this bug.
Fix for PR41929.
Differential Revision: https://reviews.llvm.org/D62166
llvm-svn: 361606
When we are scheduling the load and addi, if all other heuristic didn't take effect,
we will try to schedule the addi before the load, to hide the latency, and avoid the
true dependency added by RA. And this only take effects for Power9.
Differential Revision: https://reviews.llvm.org/D61930
llvm-svn: 361600
Summary:
On Windows, X8 may be used to pass in the address of an aggregate that
is returned indirectly. Therefore, it should be forwarded to variadic
musttail calls and preserved in thunks.
Fixes PR41997
Reviewers: mgrang, efriedma
Subscribers: javed.absar, kristof.beyls, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62344
llvm-svn: 361585
Summary:
Appears identical to powerpc64{,le}.
Regenerate test that is being affected by upcoming patch.
Reviewers: RKSimon
Reviewed By: RKSimon
Subscribers: nemanjai, jsji, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62339
llvm-svn: 361543
We were assuming a much larger possible per-wave visible stack
allocation than is possible:
faa3ae5138/src/core/runtime/amd_gpu_agent.cpp (L70)
Based on this, we can assume the high 15 bits of a frame index or sret
are 0. The frame index value is the per-lane offset, so the maximum
frame index value is MAX_WAVE_SCRATCH / wavesize.
Remove the corresponding subtarget feature and option that made
this configurable.
llvm-svn: 361541
Fixes https://bugs.llvm.org/show_bug.cgi?id=40969
The functions findPotentiallyBlockedCopies and buildCopy are currently not
accounting for the presence of debug instructions. In the former this results
in the optimization not being trigerred, and in the latter results in
inconsistent codegen.
This patch enables the optimization to be performed in a debug build and
ensures the codegen is consistent with non-debug builds.
Patch by Chris Dawson.
Differential Revision: https://reviews.llvm.org/D61680
llvm-svn: 361527
Summary:
These features will both be implemented soon, so I thought I would
save time by adding the boilerplate for both of them at the same time.
Reviewers: aheejin
Subscribers: dschuff, sbc100, jgravelle-google, hiraditya, sunfish, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D62047
llvm-svn: 361516
When printing assembly for PtrToInt, AsmPrinter::lowerConstant
incorrectly assumed that if PtrToInt was not converting to an
int with exactly the same number of bits, it must be widening
to a larger int. But this isn't necessarily true; PtrToInt can
also shrink the size, which is useful when you want to produce
a known 32-bit pointer on a 64-bit platform (on x86_64 ELF
this yields a R_X86_64_32 relocation).
The old behavior of falling through to the widening case for a
narrowing PtrToInt yields bogus assembly code like this, which
fails to assemble because the no-op bit and it accidentally
creates is not a valid relocation:
```
.long a&-1
```
The fix is to treat a narrowing PtrToInt exactly the same as
it already treats Trunc: just emit the expression and let
the assembler deal with truncating it in the appropriate way.
Patch by Mat Hostetter <mjh@fb.com>.
Differential Revision: https://reviews.llvm.org/D61325
llvm-svn: 361508
The previous patch added a member set to store instructions that we
could allow to wrap. But this wasn't cleared between searches meaning
that they could get promoted, incorrectly, during the promotion of a
separate valid chain.
Differential Revision: https://reviews.llvm.org/D62254
llvm-svn: 361462
Summary:
In this patch, `ISD::RETURNADDR` is lowered on the emscripten target
to the new Emscripten runtime function `emscripten_return_address`, which
implements the functionality.
Patch by Guanzhong Chen
Reviewers: tlively, aheejin
Reviewed By: tlively
Subscribers: dschuff, sbc100, jgravelle-google, hiraditya, sunfish, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62210
llvm-svn: 361454
In general dynamic/local dynamic TLS models, with -fno-plt,
* x86: emit `calll *___tls_get_addr@GOT(%ebx)` instead of `calll ___tls_get_addr@PLT`
Note, on x86, if we can get rid of %ebx as the PIC register,
it may be better to use a register not preserved across function calls.
* x86_64: emit `callq *__tls_get_addr@GOTPCREL(%rip)` instead of `callq __tls_get_addr@PLT`
Reorganize the code by separating 32-bit and 64-bit.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D62106
llvm-svn: 361453
We effectively had a second set of isel patterns that tried to use a
regular store instruction and an extract_subreg instruction. Or a masked move
and an extract_subreg. These patterns were intended to override the
matching of VEXTRACT instructions by taking advantage of the priority
of the explicit immediate 0 for the index.
This patch instaed just disables the immediate 0 matchin the VEXTRACT
patterns. This each of the component pieces of the larger patterns will
match by themselves.
This found a bug of sorts were we didn't use 128-bit store for 512->128
extract on KNL. Its unclear what the right thing here should be.
Using the vextract avoids constraining the register allocator to use
xmm0-15. But it always results in a longer encoding if the register
allocator ends up choosing xmm0-15 anyway.
llvm-svn: 361431
We were turning roundss/sd/ps/pd intrinsics with immediates of 1 or 2 into
llvm.floor/ceil. The llvm.ceil/floor intrinsics are supposed to correspond
to the libm functions. For the libm functions we need to disable the
precision exception so the llvm.floor/ceil functions should always map to
encodings 0x9 and 0xA.
We had a mix of isel patterns where some used 0x9 and 0xA and others used
0x1 and 0x2. We need to be consistent and always use 0x9 and 0xA.
Since we have no way in isel of knowing where the llvm.ceil/floor came
from, we can't map X86 specific intrinsics with encodings 1 or 2 to it.
We could map 0x9 and 0xA to llvm.ceil/floor instead, but I'd really like
to see a use case and optimization advantage first.
I've left the backend test cases to show the blend we now emit without
the extra isel patterns. But I've removed the InstCombine tests completely.
llvm-svn: 361425
This fix is for the problem from https://bugs.llvm.org/show_bug.cgi?id=38714.
Specifically, Simple Register Coalescing creates following conversion :
undef %0.sub_32:gpr64 = ORRWrs $wzr, %3:gpr32common, 0, debug-location !24;
It copies 32-bit value from gpr32 into gpr64. But Live DEBUG_VALUE analysis
is not able to create debug location record for that instruction. So the problem
is in that debug info for argc variable is incorrect. The fix is
to write custom isCopyInstrImpl() which would recognize the ORRWrs instr.
llvm-svn: 361417
Keep it optional in cases this is ever needed in some global
context. Currently it's only used for getting an upper bound inline
asm code size.
For AMDGPU, gfx10 increases the maximum instruction size to
20-bytes. This avoids penalizing older subtargets when estimating code
size, and making some annoying branch relaxation test adjustments.
llvm-svn: 361405
This is a minimal start to correcting a problem most directly discussed in PR38086:
https://bugs.llvm.org/show_bug.cgi?id=38086
We have been hacking around a limitation for FP select patterns by using the
fast-math-flags on the condition of the select rather than the select itself.
This patch just allows FMF to appear with the 'select' opcode. No changes are
needed to "FPMathOperator" because it already includes select-of-FP because
that definition is based on the (return) value type.
Once we have this ability, we can start correcting and adding IR transforms
to use the FMF on a 'select' instruction. The instcombine and vectorizer test
diffs only show that the IRBuilder change is behaving as expected by applying
an FMF guard value to 'select'.
For reference:
rL241901 - allowed FMF with fcmp
rL255555 - allowed FMF with FP calls
Differential Revision: https://reviews.llvm.org/D61917
llvm-svn: 361401
Does not affect update_llc_test_checks, or the actual output,
but is not accepted by the actual FileCheck.
Sorry, i should have noticed this before committing,
not the very next second after..
llvm-svn: 361398
Summary:
Another target that prefers to use `-march` in tests
```
llvm/test/CodeGen/SPARC$ grep -ri mtriple | wc -l
25
llvm/test/CodeGen/SPARC$ grep -ri march | wc -l
165
```
This test is being affected by a further patch,
so regenerate it to better visualize the changes
Reviewers: RKSimon, dcederman, gberry
Reviewed By: RKSimon
Subscribers: jyknight, fedor.sergeev, jrtc27, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62242
llvm-svn: 361381
This is the second part of the commit fixing PR38917 (hoisting
partitially redundant machine instruction). Most of PRE (partitial
redundancy elimination) and CSE work is done on LLVM IR, but some of
redundancy arises during DAG legalization. Machine CSE is not enough
to deal with it. This simple PRE implementation works a little bit
intricately: it passes before CSE, looking for partitial redundancy
and transforming it to fully redundancy, anticipating that the next
CSE step will eliminate this created redundancy. If CSE doesn't
eliminate this, than created instruction will remain dead and eliminated
later by Remove Dead Machine Instructions pass.
The third part of the commit is supposed to refactor MachineCSE,
to make it more clear and to merge MachinePRE with MachineCSE,
so one need no rely on further Remove Dead pass to clear instrs
not eliminated by CSE.
First step: https://reviews.llvm.org/D54839
Fixes llvm.org/PR38917
llvm-svn: 361356
Summary:
For big-endian powerpc64, the default ABI is ELFv1. OpenPower ABI ELFv2 is supported when -mabi=elfv2 is specified. FreeBSD support for PowerPC64 ELFv2 ABI with LLVM is in progress[1]. This patch adds an alternative way to specify ELFv2 ABI on target triple [2].
The following results are expected:
ELFv1 when using:
-target powerpc64-unknown-freebsd12.0
-target powerpc64-unknown-freebsd12.0 -mabi=elfv1
-target powerpc64-unknown-freebsd12.0-elfv1
ELFv2 when using:
-target powerpc64-unknown-freebsd12.0 -mabi=elfv2
-target powerpc64-unknown-freebsd12.0-elfv2
[1] https://wiki.freebsd.org/powerpc/llvm-elfv2
[2] https://clang.llvm.org/docs/CrossCompilation.html
Patch by Alfredo Dal'Ava Júnior!
Differential Revision: https://reviews.llvm.org/D61950
llvm-svn: 361355
Fix for https://bugs.llvm.org/show_bug.cgi?id=41971. Make the
combineVectorSizedSetCCEquality() transform more conservative by
checking that the bitcast to the vector type will be cheap/free
for both operands. I'm considering it cheap if it's a constant,
a load or already a vector. I've dropped the explicit check for
f128 because it should fall out naturally (in the cases where
it'd be detrimental).
Differential Revision: https://reviews.llvm.org/D62220
llvm-svn: 361352
CET-IBT enabled
Return-twice functions will indirectly jump after the caller's position.
So when CET-IBT is enable, we should make sure these is endbr*
instructions follow these Return-twice function caller. Like GCC does.
Patch by Xiang Zhang (xiangzhangllvm)
Differential Revision: https://reviews.llvm.org/D61881
llvm-svn: 361342
r360889 added new llround builtin functions. This patch adds their
signatures for the WebAssembly backend.
It also adds wasm32 support to utils/update_llc_test_checks.py, since
that's the script other targets are using for their testcases for this
feature.
Differential Revision: https://reviews.llvm.org/D62207
llvm-svn: 361327
Register coalescer fails for the test in the patch with the assertion in
JoinVals::ConflictResolution `DefMI != nullptr'. It attempts to join
live intervals for two adjacent instructions and erase the copy:
%2:vreg_256 = COPY %1
%3:vreg_256 = COPY killed %1
The LI needs to be adjusted to kill subrange for the erased instruction
and extend the subrange of the original def. That was done for the main
interval only but not for the subrange. As a result subrange had a VNI
pointing to the erased slot resulting in the above failure.
Differential Revision: https://reviews.llvm.org/D62162
llvm-svn: 361293
Add an intrinsic that takes 2 signed integers with the scale of them provided
as the third argument and performs fixed point multiplication on them. The
result is saturated and clamped between the largest and smallest representable
values of the first 2 operands.
This is a part of implementing fixed point arithmetic in clang where some of
the more complex operations will be implemented as intrinsics.
Differential Revision: https://reviews.llvm.org/D55720
llvm-svn: 361289
Meaning if we were to produce 'neg' in dagcombine, we will get an
endless cycle; some inverse transform would need to be guarded somehow.
Also, the 'and (sub 0, x), 31' variant is sticky,
doesn't get optimized in any way.
https://bugs.llvm.org/show_bug.cgi?id=41952
llvm-svn: 361254
Some checks in isShuffleMaskLegal expect an even number of elements,
e.g. isTRN_v_undef_Mask or isUZP_v_undef_Mask, otherwise they access
invalid elements and crash. This patch adds checks to the impacted
functions.
Fixes PR41951
Reviewers: t.p.northover, dmgreen, samparker
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D60690
llvm-svn: 361235
PrepareConstants step converts add/sub with 'negative' immediates to
sub/add with a 'positive' imm to make promotion more simple. nuw
already states that the add shouldn't cause an unsigned wrap, so
it shouldn't need any tweaking. Plus, we also don't allow a sub with
a 'negative' immediate to be safe wrap, so this functionality has
been removed. The PrepareConstants step now just handles the add
instructions that we've determined would be safe if they wrap around
zero.
Differential Revision: https://reviews.llvm.org/D62057
llvm-svn: 361227
Summary:
The endianess used in the calling convention does not always match the
endianess of the target on all architectures, namely AVR.
When an argument is too large to be legalised by the architecture and is
split for the ABI, a new hook TargetLoweringInfo::shouldSplitFunctionArgumentsAsLittleEndian
is queried to find the endianess that function arguments must be laid
out in.
This approach was recommended by Eli Friedman.
Originally reported in https://github.com/avr-rust/rust/issues/129.
Patch by Carl Peto.
Reviewers: bogner, t.p.northover, RKSimon, niravd, efriedma
Reviewed By: efriedma
Subscribers: JDevlieghere, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62003
llvm-svn: 361222
Unfortunately the way SIInsertSkips works is backwards, and is
required for correctness. r338235 added handling of some special cases
where skipping is mandatory to avoid side effects if no lanes are
active. It conservatively handled asm correctly, but the same logic
needs to apply to calls.
Usually the call sequence code is larger than the skip threshold,
although the way the count is computed is really broken, so I'm not
sure if anything was likely to really hit this.
llvm-svn: 361202
That commit makes sure we flush PendingExports in SelectDAGBuilder
before we create INLINEASM_BR. Unfortunatley, I haven't yet found
a CodeGen failure without that change.
This commit uses the debug output from SelectionDAG to at least
ensure we build the DAG correctly.
llvm-svn: 361179
Fixes issue reported by aemerson on D57348. Vector op legalization
support is added for uaddo, usubo, saddo and ssubo (umulo and smulo
were already supported). As usual, by extracting TargetLowering methods
and calling them from vector op legalization.
Vector op legalization doesn't really deal with multiple result nodes,
so I'm explicitly performing a recursive legalization call on the
result value that is not being legalized.
There are some existing test changes because expansion happens
earlier, so we don't get a DAG combiner run in between anymore.
Differential Revision: https://reviews.llvm.org/D61692
llvm-svn: 361166
The code did not match the example in the comment, and was checking
the undef flag on the copy dest instead of source. The existing tests
were only hitting the > 2 operands case.
llvm-svn: 361156
The scalar start/accumulator value of the fadd- and fmul reduction
should match the result type of the reduction, as well as the vector
element-type of the input vector. Although this was not explicitly
specified in the LangRef, it was taken for granted in code implementing
the reductions. The patch also fixes the LangRef by adding this
constraint.
Reviewed By: aemerson, nikic
Differential Revision: https://reviews.llvm.org/D60260
llvm-svn: 361133
Summary:
Avoid introducing hazard mitigation when lgkmcnt is reduced to 0.
Clarify code comments to explain assumptions made for this hazard
mitigation. Expand and correct test cases to cover variants of
s_waitcnt.
Reviewers: nhaehnle, rampitec
Subscribers: arsenm, kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62058
llvm-svn: 361124
Summary:
Was looking into supporting `(srl (shl x, c1), c2)` with c1 != c2 in dagcombiner,
this test changes, but makes `update_llc_test_checks.py` unhappy.
**Many** AMDGPU tests specify `-march`, not `-mtriple`, which results in `update_llc_test_checks.py`
defaulting to x86 asm function detection heuristics, which don't work here.
I propose to fix this by adding an infrastructure to map from `-march` to `-mtriple`,
in the UpdateTestChecks tooling.
Reviewers: RKSimon, MaskRay, arsenm
Reviewed By: arsenm
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62099
llvm-svn: 361101
Summary:
Was looking into supporting `(srl (shl x, c1), c2)` with c1 != c2 in dagcombiner,
this test changes, but makes `update_llc_test_checks.py` unhappy
Reviewers: RKSimon
Reviewed By: RKSimon
Subscribers: javed.absar, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62097
llvm-svn: 361100
This is ported from the custom AMDGPU DAG implementation. I think this
is a better default expansion than what the DAG currently uses, at
least if the target has CTLZ.
This implements the signed version in terms of the unsigned
conversion, which is implemented with bit operations. SelectionDAG has
several other implementations that should eventually be ported
depending on what instructions are legal.
llvm-svn: 361081
Same as what we do for vector reductions in combineHorizontalPredicateResult, use movmsk+cmp for scalar (and(extract(x,0),extract(x,1)) reduction patterns.
llvm-svn: 361052
Summary:
That check claims that the transform is illegal otherwise.
That isn't true:
1. For `ISD::ADD`, we only process `ISD::SHL` outer shift => sign bit does not matter
https://rise4fun.com/Alive/K4A
2. For `ISD::AND`, there is no restriction on constants:
https://rise4fun.com/Alive/Wy3
3. For `ISD::OR`, there is no restriction on constants:
https://rise4fun.com/Alive/GOH
3. For `ISD::XOR`, there is no restriction on constants:
https://rise4fun.com/Alive/ml6
So, why is it there then?
This changes the testcase that was touched by @spatel in rL347478,
but i'm not sure that test tests anything particular?
Reviewers: RKSimon, spatel, craig.topper, jojo, rengolin
Reviewed By: spatel
Subscribers: javed.absar, llvm-commits, spatel
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61918
llvm-svn: 361044
With a fix for PR41917: The predecessor list was changing under our feet.
- for (BasicBlock *Pred : predecessors(EntryBlock_)) {
+ while (!pred_empty(EntryBlock_)) {
+ BasicBlock* const Pred = *pred_begin(EntryBlock_);
llvm-svn: 361009
Summary:
In order to combine memory operations efficiently, the load/store
optimizer might move some instructions around. It's usually safe
to move instructions down past the merged instruction because the
pass checks if memory operations can be re-ordered.
Though, the current logic doesn't handle Write-after-Write hazards.
This fixes a reflection issue with Monster Hunter World and DXVK.
v2: - rebased on top of master
- clean up the test case
- handle WaW hazards correctly
Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=40130
Original patch by Samuel Pitoiset.
Reviewers: tpr, arsenm, nhaehnle
Reviewed By: nhaehnle
Subscribers: ronlieb, arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye
Differential Revision: https://reviews.llvm.org/D61313
llvm-svn: 361008
Make sure to not unroll a vector division/remainder (with a constant splat
divisor) after type legalization, since the scalar type may then be illegal.
Review: Ulrich Weigand
https://reviews.llvm.org/D62036
llvm-svn: 360965
The recent introduction of v3i32 etc as an MVT, and its use in AMDGPU
3-dword memory instructions, caused a de-optimization problem for code
with such a load that then bitcasts via vector of i8, because v12i8 is
not an MVT so it legalizes the bitcast by widening it.
This commit adds the ability to widen a bitcast using extract_subvector
on the result, so the value does not need to go via memory.
Differential Revision: https://reviews.llvm.org/D60457
Change-Id: Ie4abb7760547e54a2445961992eafc78e80d4b64
llvm-svn: 360942
This suppresses exceptions which is what we should be doing for ceil and floor. We already use the correct immediate
in patterns without masking.
llvm-svn: 360915
This is the conservatively correct default. It is always safe to
assume xnack is enabled, but not the converse.
Introduce a feature to blacklist targets where xnack can never be
meaningfully enabled. I'm not sure the targets this is applied to is
100% correct.
llvm-svn: 360903
This patch add the ISD::LROUND and ISD::LLROUND along with new
intrinsics. The changes are straightforward as for other
floating-point rounding functions, with just some adjustments
required to handle the return value being an interger.
The idea is to optimize lround/llround generation for AArch64
in a subsequent patch. Current semantic is just route it to libm
symbol.
llvm-svn: 360889
Trace through multiple COPYs when looking for a physreg source. Add
hinting for vregs that will be copied into physregs (we only hinted
for vregs getting copied to a physreg previously). Give hinted a
register a bonus when deciding which value to spill. This is part of
my rewrite regallocfast series. In fact this one doesn't even have an
effect unless you also flip the allocation to happen from back to
front of a basic block. Nonetheless it helps to split this up to ease
review of D52010
Patch by Matthias Braun
llvm-svn: 360887
If we're trying to match an LEA, its possible the LEA match will be deemed unprofitable. In which case the negation we created in matchAddress would be left dangling in the SelectionDAG. This could artificially increase use counts for other nodes in the DAG. Though I don't have an example of that. But it just seems like bad form to have dangling nodes in isel.
Differential Revision: https://reviews.llvm.org/D61047
llvm-svn: 360823
Before this change, they were erroneously constructed with the EH_LABEL
SDNode opcode, which caused other passes to interact with them in
incorrect ways. See the FIXME about fastisel that this addresses in the
existing test case.
Fixes PR41890
llvm-svn: 360818
Summary:
Otherwise, we emit directives for CFI without any actual CFI opcodes to
go with them, which causes tools to malfunction. The technique is
similar to what the x86 backend already does.
Fixes https://bugs.llvm.org/show_bug.cgi?id=40876
Patch by: froydnj (Nathan Froyd)
Reviewers: mstorsjo, eli.friedman, rnk, mgrang, ssijaric
Reviewed By: rnk
Subscribers: javed.absar, kristof.beyls, llvm-commits, dmajor
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61960
llvm-svn: 360816
Summary:
The emitError path allows the program to continue, unlike report_fatal_error.
This is friendlier to use cases where LLVM is embedded in a larger program,
because the caller may be able to deal with the error somewhat gracefully.
Change the number of requested NOP bytes in the AArch64 and PowerPC
test cases to avoid triggering an unrelated assertion. The compilation
still fails, as verified by the test.
Change-Id: Iafb9ca341002a597b82e59ddc7a1f13c78758e3d
Reviewers: arsenm, MatzeB
Subscribers: qcolombet, nemanjai, wdng, javed.absar, kristof.beyls, kbarton, jsji, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61489
llvm-svn: 360786
Summary:
SGPR in CC can be either hw initialized or set by other chained shaders
and so this increases the SGPR count availalbe to CC to 105.
Change-Id: I3dfadc750fe4a3e2bd07117a2899fd13f3e2fef3
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61261
llvm-svn: 360778
Instead of patching the original blocks, we now generate new blocks and
delete the old blocks. This results in simpler code with a less twisted
control flow (see the change in `entry-block-shuffled.ll`).
This will make https://reviews.llvm.org/D60318 simpler by making it more
obvious where control flow created and deleted.
Reviewers: gchatelet
Subscribers: hiraditya, llvm-commits, spatel
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61736
llvm-svn: 360771
The new cortex-m schedule in rL360768 helps performance, but can increase the
amount of high-registers used. This, on average, ends up increasing the
codesize by a fair amount (because less instructions are converted from T2 to
T1). On cortex-m at -Oz, where we are quite size-paranoid, it is better to use
the existing DAG scheduler with the RegPressure scheduling preference (at least
until the issues around T2 vs T1 instructions can be improved).
I have also made sure that the Sched::RegPressure dag scheduler is always
chosen for MinSize.
The test shows one case where we increase the number of registers used.
Differential Revision: https://reviews.llvm.org/D61882
llvm-svn: 360769
This patch adds a simple Cortex-M4 schedule, renaming the existing M3
schedule to M4 and filling in the latencies as-per the Cortex-M4 TRM:
https://developer.arm.com/docs/ddi0439/latest
Most of these are 1, with the important exception being loads taking 2
cycles. A few others are also higher, but I don't believe they make a
large difference. I've repurposed the M3 schedule as the latencies are
mostly the same between the two cores, with the M4 having more FP and
DSP instructions. We also turn on MISched and UseAA for the cores that
now use this.
It also adds some schedule Write's to various instruction to make things
simpler.
Differential Revision: https://reviews.llvm.org/D54142
llvm-svn: 360768
They encode the same way, but OR32mi8Locked sets hasUnmodeledSideEffects set
which should be stronger than the mayLoad/mayStore on LOCK_OR32mi8. I think
this makes sense since we are using it as a fence.
This also seems to hide the operation from the speculative load hardening pass
so I've reverted r360511.
llvm-svn: 360747
The 3-field form was introduced by D3499 in 2014 and the legacy 2-field
form was planned to be removed in LLVM 4.0
For the textual format, this patch migrates the existing 2-field form to
use the 3-field form and deletes the compatibility code.
test/Verifier/global-ctors-2.ll checks we have a friendly error message.
For bitcode, lib/IR/AutoUpgrade UpgradeGlobalVariables will upgrade the
2-field form (add i8* null as the third field).
Reviewed By: rnk, dexonsmith
Differential Revision: https://reviews.llvm.org/D61547
llvm-svn: 360742
This was the portion split off D58632 so that it could follow the redzone API cleanup. Note that I changed the offset preferred from -8 to -64. The difference should be very minor, but I thought it might help address one concern which had been previously raised.
Differential Revision: https://reviews.llvm.org/D61862
llvm-svn: 360719
Unlike instcombine, we currently don't turn and+shift into shift+and.
We probably should, likely unconditionally.
While i'm adding only all-ones (potentially shifted) mask,
this obviously isn't limited to any particular mask pattern:
https://rise4fun.com/Alive/kmX
Related to https://bugs.llvm.org/show_bug.cgi?id=41874
llvm-svn: 360706
Tighten conditions on SMEM WAR hazard unit tests to ensure rejection
of workaround insertion where a s_waitcnt is present in dependency
chain. The current workaround code already conforms to these revise
tests.
llvm-svn: 360686
D61068 handled vector shifts, this patch does the same for scalars where there are similar number of pipes for shifts as bit ops - this is true almost entirely for AMD targets where the scalar ALUs are well balanced.
This combine avoids AND immediate mask which usually means we reduce encoding size.
Some tests show use of (slow, scaled) LEA instead of SHL in some cases, but thats due to particular shift immediates - shift+mask generate these just as easily.
Differential Revision: https://reviews.llvm.org/D61830
llvm-svn: 360684
For known CRBit spills, CRSET/CRUNSET, it is more efficient to load and spill
the known value instead of extracting the bit.
eg. This sequence is currently used to spill a CRUNSET:
crclr 4*cr5+lt
mfocrf r3,4
rlwinm r3,r3,20,0,0
stw r3,132(r1)
This patch custom lower it to:
li r3,0
stw r3,132(r1)
Differential Revision: https://reviews.llvm.org/D61754
llvm-svn: 360677
When breaking up loads and stores of aggregates, the IRTranslator uses
LLT::scalar(64) for the index type of the G_GEP instructions that
compute the addresses. This is unnecessarily large for 32-bit targets.
Use the int ptr type provided by the DataLayout instead.
Note that we're already doing the right thing when translating
getelementptr instructions from the IR. This is just an oversight when
generating new ones while translating loads/stores.
Both x86 and AArch64 already have tests confirming that the old
behaviour is preserved for 64-bit targets.
Differential Revision: https://reviews.llvm.org/D61852
llvm-svn: 360656
This is a follow on to D58632, with the same logic. Given a memory operation which needs ordering, but doesn't need to modify any particular address, prefer to use a locked stack op over an mfence.
Differential Revision: https://reviews.llvm.org/D61863
llvm-svn: 360649
X32 can refer to a 64-bit ABI that uses 32-bit ints, longs, and pointers.
I plan to add gnux32 command lines to this test so this prepares for that.
Also remove some check lines that have a prefix that is not in any run lines.
llvm-svn: 360642
Usually this will abort fast-isel at the instruction using the
non-legal result, but if the only use is in a different basic block,
we'll incorrectly assume that the zext/sext is to i32 (rather than
i128 in this case).
Differential Revision: https://reviews.llvm.org/D61823
llvm-svn: 360616
Summary:
X86TargetLowering::LowerAsmOperandForConstraint had better support than
TargetLowering::LowerAsmOperandForConstraint for arbitrary depth
getelementpointers for "i", "n", and "s" extended inline assembly
constraints. Hoist its support from the derived class into the base
class.
Link: https://github.com/ClangBuiltLinux/linux/issues/469
Reviewers: echristo, t.p.northover
Reviewed By: t.p.northover
Subscribers: t.p.northover, E5ten, kees, jyknight, nemanjai, javed.absar, eraman, hiraditya, jsji, llvm-commits, void, craig.topper, nathanchance, srhines
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61560
llvm-svn: 360604
Fixes the regression noted in D61782 where a VZEXT_MOVL was being inserted because we weren't discriminating between 'zeroable' and 'all undef' for the upper elts.
Differential Revision: https://reviews.llvm.org/D61782
llvm-svn: 360596
Now that we can use HADD/SUB for scalar additions from any pair of extracted elements (D61263), we can relax the one use limit as we will be able to merge multiple uses into using the same HADD/SUB op.
This exposes a couple of missed opportunities in LowerBuildVectorv4x32 which will be committed separately.
Differential Revision: https://reviews.llvm.org/D61782
llvm-svn: 360594
We catch most of these patterns (on x86 at least) by matching
a concat vectors opcode early in combining, but the pattern may
emerge later using insert subvector instead.
The AVX1 diffs for add/sub overflow show another missed narrowing
pattern. That one may be falling though the cracks because of
combine ordering and multiple uses.
llvm-svn: 360585
The new fptrunc and fpext intrinsics are constrained versions of the
regular fptrunc and fpext instructions.
Reviewed by: Andrew Kaylor, Craig Topper, Cameron McInally, Conner Abbot
Approved by: Craig Topper
Differential Revision: https://reviews.llvm.org/D55897
llvm-svn: 360581
This adds the FPC (floating-point control register) as a reserved
physical register and models its use by SystemZ instructions.
Note that only the current rounding modes and the IEEE exception
masks are modeled. *Changes* of the FPC due to exceptions (in
particular the IEEE exception flags and the DXC) are not modeled.
At this point, this patch is mostly NFC, but it will prevent
scheduling of floating-point instructions across SPFC/LFPC etc.
llvm-svn: 360570
When deciding the safety of generating smlad, we checked for any
writes within the block that may alias with any of the loads that
need to be widened. This is overly conservative because it only
matters when there's a potential aliasing write to a location
accessed by a pair of loads.
Now we check for aliasing writes only once, during setup. If two
loads are found to have an aliasing write between them, we don't add
these loads to LoadPairs. This means that later during the transform,
we can safely widened a pair without worrying about aliasing.
However, to maintain correctness, we also need to change the way that
wide loads are inserted because the order is now important.
The MatchSMLAD method has also been changed, absorbing
MatchReductions and AddMACCandidate to hopefully improve readability.
Differential Revision: https://reviews.llvm.org/D6102
llvm-svn: 360567
Summary:
When we know for sure whether two addresses do or do not alias, we
should immediately return from DAGCombiner::isAlias().
I think this comes from a bad copy/paste, Sorry for not catching that during the
code review.
Fixes PR41855.
Reviewers: niravd, gchatelet, EricWF
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61846
llvm-svn: 360566
Currently, without -g, BTF sections may still be emitted with
data sections, e.g., for linux kernel bpf selftest
test_tcp_check_syncookie_kern.c issue discovered by Martin
as shown below.
-bash-4.4$ bpftool btf dump file test_tcp_check_syncookie_kern.o
[1] VAR 'results' type_id=0, linkage=global-alloc
[2] VAR '_license' type_id=0, linkage=global-alloc
[3] DATASEC 'license' size=0 vlen=1
type_id=2 offset=0 size=4
[4] DATASEC 'maps' size=0 vlen=1
type_id=1 offset=0 size=28
Let disable BTF generation if no debuginfo, which is
the original design.
Signed-off-by: Yonghong Song <yhs@fb.com>
Differential Revision: https://reviews.llvm.org/D61826
llvm-svn: 360556
I've included a new fix in X86RegisterInfo to prevent PR41619 without
reintroducing r359392. We might be able to improve that in the base class
implementation of shouldRewriteCopySrc somehow. But this hopefully enables
forward progress on SimplifyDemandedBits improvements for now.
Original commit message:
This patch adds support for BigBitWidth -> SmallBitWidth bitcasts, splitting the DemandedBits/Elts accordingly.
The AMDGPU backend needed an extra (srl (and x, c1 << c2), c2) -> (and (srl(x, c2), c1) combine to encourage BFE creation, I investigated putting this in DAGComb
but it caused a lot of noise on other targets - some improvements, some regressions.
The X86 changes are all definite wins.
llvm-svn: 360552
I noticed that we were failing to narrow an x86 ymm math op in a case similar
to the 'madd' test diff. That is because a bitcast is sitting between the math
and the extract subvector and thwarting our pattern matching for narrowing:
t56: v8i32 = add t59, t58
t68: v4i64 = bitcast t56
t73: v2i64 = extract_subvector t68, Constant:i64<2>
t96: v4i32 = bitcast t73
There are a few wins and neutral diffs in the other tests.
Differential Revision: https://reviews.llvm.org/D61806
llvm-svn: 360541
See if we can simplify the demanded vector elts from the extraction before trying to simplify the demanded bits.
This helps us with target shuffles and hops in particular.
llvm-svn: 360535
Summary: Skip over prefetches when assigning debug info to instructions with memory operands. This way, the debug info is stable after instrumenting a binary with prefetches, allowing for iterative profiling and instrumentation.
Reviewers: davidxl
Reviewed By: davidxl
Subscribers: aprantl, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61789
llvm-svn: 360471
Split out from D61692 per RKSimon's suggestion. Vector op
legalization will automatically recursively legalize the returned
SDValue, but we need to take care of the other results ourselves.
Otherwise it will end up getting legalized only during op
legalization, by which point it might be too late (though I'm not
aware of any specific cases right now).
There are codegen differences because expansion occurs earlier now
and we don't get a DAGCombiner run in between.
Differential Revision: https://reviews.llvm.org/D61744
llvm-svn: 360470
This fix allows the scheduler to take into account the number of instances of
each ProcResource specified. Previously a declaration in a scheduler of
ProcResource<1> would be treated identically to a declaration of
ProcResource<2>. Now the hazard recognizer would report a hazard only after all
of the resource instances are busy.
Patch by Jackson Woodruff and Momchil Velikov.
Differential Revision: https://reviews.llvm.org/D51160
llvm-svn: 360441
Fixes https://bugs.llvm.org/show_bug.cgi?id=40969
The functions findPotentiallyBlockedCopies and buildCopy are currently not
accounting for the presence of debug instructions. In the former this results
in the optimization not being trigerred, and in the latter results in
inconsistent codegen.
This patch enables the optimization to be performed in a debug build and
ensures the codegen is consistent with non-debug builds.
Patch by Chris Dawson.
Differential Revision: https://reviews.llvm.org/D61680
llvm-svn: 360436
If we only use the lower xmm of a ymm hop, then extract the xmm's (for free), perform the xmm hop and then insert back into a ymm (for free).
Fixes some of the regressions noted in D61782
llvm-svn: 360435
The current PIC model for WebAssembly is more like ELF in that it
allows symbol interposition.
This means that more functions end up being addressed via the GOT
and fewer directly added to the wasm table.
One effect is a reduction in the number of wasm table entries similar
to the previous attempt in https://reviews.llvm.org/D61539 which was
reverted.
Differential Revision: https://reviews.llvm.org/D61772
llvm-svn: 360402
This also allows three op patterns to use increased constant bus
limit of GFX10.
Differential Revision: https://reviews.llvm.org/D61763
llvm-svn: 360395
The current lowering uses an mfence. mfences are substaintially higher latency than the locked operations originally requested, but we do want to avoid contention on the original cache line. As such, use a locked instruction on a cache line assumed to be thread local.
Differential Revision: https://reviews.llvm.org/D58632
llvm-svn: 360393
As reported on PR39920, "slow horizontal ops" targets tend to internally expand to 2*shuffle+add/sub - so if we can reduce 2*shuffle+add/sub to a hadd/sub then we should do it - similar port usage but reduced instruction count.
This works out in most cases, although the "PR22377" regression in vector-shuffle-combining.ll is annoying - going from 2*shuffle+add+shuffle to hadd+2*shuffle - I've opened PR41813 to cover this.
Differential Revision: https://reviews.llvm.org/D61308
llvm-svn: 360360
Add an Argument that has the SExtAttr attached, as well as SIToFP
instructions, as values that generate sign bits. SIToFP doesn't
strictly do this and could be treated as a sink to be sign-extended.
Differential Revision: https://reviews.llvm.org/D61381
llvm-svn: 360331
This patch allows for expansion of ADDCARRY and SUBCARRY when the target does not support it.
Differential Revision: https://reviews.llvm.org/D61411
llvm-svn: 360303
This is extracted from the original draft of D61419 with some additional tests.
We don't currently get this in IR (it's conservatively turned into a NaN),
but presumably that'll get updated as we add real IR support for 'fneg'
rather than 'fsub -0.0, x'.
The x86-32 run shows the following, and I haven't looked further to see why,
but that seems to be independent:
Legalizing: t1: f32 = undef
Trying to expand node
Creating fp constant: t4: f32 = ConstantFP<0.000000e+00>
Differential Revision: https://reviews.llvm.org/D61516
llvm-svn: 360296
The VOP3 form should always be the preferred selection, to be shrunk
later. This should only be an optimization issue, but this partially
works around a problem from clobbering VCC when SIFixSGPRCopies
rewrites an SCC defining operation directly to VCC.
3 of the testcases are regressions from failing to fold the immediate
in cases it should. These can be avoided by improving the VCC liveness
handling in SIFoldOperands. Simply increasing the threshold to
computeRegisterLiveness works, although this is common enough that VCC
liveness should probably be tracked throughout the pass. The hack of
leaving behind an implicit_def instruction to avoid breaking iterator
wastes instruction count, which inhibits finding the VCC def in long
chains of adds. Doing this however exposes different, worse looking
regressions from poor scheduling behavior. This could probably be
avoided around by forcing the shrink of the addc here, but the
scheduler should probably be fixed.
The r600 add test needs to be split out because it asserts on the
arguments in the new test during the calling convention lowering.
llvm-svn: 360293
When assigning the definitions of an instruction we were updating
the available registers while walking the definitions. Some of
those definitions may be from physical registers and thus, they are
not available for other definitions to take, but by the time we see
that we may have already assign these registers to another
virtual register.
Fix that by walking through all the definitions and mark as unavailable
the physical register definitions, then do the virtual register assignments.
PR41790
llvm-svn: 360278
This patch adds support for calling selectFNeg for FNeg instructions in addition to the fsub idiom
Differential Revision: https://reviews.llvm.org/D61624
llvm-svn: 360273
This was committed in rL358887 but reverted in rL360066 due to a x86 regression, really it should be have been pre-committed instead of being part of the SimplifyDemandedBits bitcast patch.
llvm-svn: 360263
Using SP in this position is unpredictable in ARMv7. CMP and CMN are not
affected, and of course v8 relaxes this requirement, but that's handled
elsewhere.
llvm-svn: 360242
Summary:
A COFF stub indirects the reference to a symbol through memory. A
.refptr.$sym global variable pointer is created to refer to $sym.
Typically mingw uses these for external global variable declarations,
but we can use them for weak function declarations as well.
Updates the dso_local classification to add a special case for
extern_weak symbols on COFF in both clang and LLVM.
Fixes PR37598
Reviewers: smeenai, mstorsjo
Subscribers: hiraditya, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D61615
llvm-svn: 360207
Summary: GCNHazardRecognizer fails to identify hazards that are in and around bundles. This patch allows the hazard recognizer to consider bundled instructions in both scheduler and hazard recognizer mode. We ignore “bundledness” for the purpose of detecting hazards and examine the instructions individually.
Reviewers: arsenm, msearles, rampitec
Reviewed By: rampitec
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61564
llvm-svn: 360199
build-vector-tests.ll is a huge testcase, it is hard to maintain: eg:
any fundamental changes might need to update hundreds of lines. We should
leverage the script to maintain it.
This patch simply run utils/update_llc_test_checks.py on it. There
should be no missing test points.
llvm-svn: 360175
The single-constant algorithm produces infinities on a lot of denormal values.
The precision of the two-constant algorithm is actually sufficient across the
range of denormals. We will switch to that algorithm for now to avoid the
infinities on denormals. In the future, we will re-evaluate the algorithm to
find the optimal one for PowerPC.
Differential revision: https://reviews.llvm.org/D60037
llvm-svn: 360144
This fixes the https://bugs.llvm.org/show_bug.cgi?id=41355.
Previously with -r we printed relocation section name instead of the target section name.
It was like this: "RELOCATION RECORDS FOR [.rel.text]"
Now it is: "RELOCATION RECORDS FOR [.text]"
Also when relocation target section has more than one relocation section,
we did not combine the output. Now we do.
Differential revision: https://reviews.llvm.org/D61312
llvm-svn: 360143
Basic "revectorization" combine, we can probably do more opcodes here but it can be a tricky cost-benefit depending on where the subvectors came from - but this case helps shuffle combining.
llvm-svn: 360134
Summary:
No test case because I don't know of a way to trigger this, but I
accidentally caused this to fail while working on a different change.
Change-Id: I8015aa447fe27163cc4e4902205a203bd44bf7e3
Reviewers: arsenm, rampitec
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61490
llvm-svn: 360123
Summary:
If fneg lowering for fsub -0.0, x fails we currently fall back to treating it as an fsub. This has different behavior for nans than the xor with sign bit trick we normally try to do. On X86, the xor trick for double fails fast-isel in 32-bit mode with sse2 due to 64 bit integer types not being available. With -O2 we would always use an xorpd for this case. If we use subsd, this creates an observable behavior difference between -O0 and -O2. So fall back to SelectionDAG if we can't fast-isel it, that way SelectionDAG will use the xorpd.
I believe this patch is restoring the behavior prior to r345295 from last October. This was missed then because our fast isel case in 32-bit mode aborted fast-isel earlier for another reason. But I've added new tests to cover that.
Reviewers: andrew.w.kaylor, cameron.mcinally, spatel, efriedma
Reviewed By: cameron.mcinally
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61622
llvm-svn: 360111
Summary:
When there are multiple instances of a forward decl record type, only the first one is emitted with a type index, because
the type is added to a map with a null type index. Avoid this by reordering so that forward decl types aren't added to the map.
Reviewers: rnk
Subscribers: aprantl, hiraditya, arphaman, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61460
llvm-svn: 360101
This generally follows what other targets do. I don't completely
understand why the special case for tail calls existed in the first
place; even when the code was committed in r105413, call lowering didn't
work in the way described in the comments.
Stack protector lowering breaks if the register copies are not glued to
a tail call: we have to insert the stack protector check before the tail
call, and we choose the location based on the assumption that all
physical register dependencies of a tail call are adjacent to the tail
call. (See FindSplitPointForStackProtector.) This is sort of fragile,
but I don't see any reason to break that assumption.
I'm guessing nobody has seen this before just because it's hard to
convince the scheduler to actually schedule the code in a way that
breaks; even without the glue, the only computation that could actually
be scheduled after the register copies is the computation of the call
address, and the scheduler usually prefers to schedule that before the
copies anyway.
Fixes https://bugs.llvm.org/show_bug.cgi?id=41417
Differential Revision: https://reviews.llvm.org/D60427
llvm-svn: 360099
The problem was that we were creating a CMOV64rr <TargetFrameIndex>, <TargetFrameIndex>. The entire point of a TFI is that address code is not generated, so there's no way to legalize/lower this. Instead, simply prevent it's creation.
Arguably, we shouldn't be using *Target*FrameIndices in StatepointLowering at all, but that's a much deeper change.
llvm-svn: 360090
The fneg double case is falling back to a subsd in 32-bit mode if you write a test that doesn't trigger a fast-isel abort on the return value.
The subsd lowering has different behavior with respect to nans than using an xor. This is inconsisent with what we would do in SelectionDAG
and can lead to differences between -O0 and -O2.
llvm-svn: 360088
This reverts r357452 (git commit 21eb771dcb).
This was causing strange optimization-related test failures on an internal test. Will followup with more details offline.
llvm-svn: 360086
We require d/q suffixes on the memory form of these instructions to disambiguate the memory size.
We don't require it on the register forms, but need to support parsing both with and without it.
Previously we always printed the d/q suffix on the register forms, but it's redundant and
inconsistent with gcc and objdump.
After this patch we should support the d/q for parsing, but not print it when its unneeded.
llvm-svn: 360085
It's possible to use the 'y' mmx constraint with a type narrower than 64-bits.
This patch supports this by bitcasting the mmx type to 64-bits and then
truncating to the desired type.
There are probably other missing type combinations we need to support, but this
is the case we have a bug report for.
Fixes PR41748.
Differential Revision: https://reviews.llvm.org/D61582
llvm-svn: 360069
After support for dealing with types that need to be extended in some way was
added in r358032 we didn't correctly handle <1 x T> return types. These types
don't have a GISel direct representation, instead we just see them as scalars.
When we need to pad them into <2 x T> types however we need to use a
G_BUILD_VECTOR instead of trying to do a G_CONCAT_VECTOR.
This fixes PR41738.
llvm-svn: 360068
Reverts "[X86] Remove (V)MOV64toSDrr/m and (V)MOVDI2SSrr/m. Use 128-bit result MOVD/MOVQ and COPY_TO_REGCLASS instead"
Reverts "[TargetLowering][AMDGPU][X86] Improve SimplifyDemandedBits bitcast handling"
Eric Christopher and Jorge Gorbe Moya reported some issues with these patches to me off list.
Removing the CodeGenOnly instructions has changed how fneg is handled during fast-isel with sse/sse2. We're now emitting fsub -0.0, x instead
moving to the integer domain(in a GPR), xoring the sign bit, and then moving back to xmm. This is because the fast isel table no longer
contains an entry for (f32/f64 bitcast (i32/i64)) so the target independent fneg code fails. The use of fsub changes the behavior of nan with
respect to -O2 codegen which will always use a pxor. NOTE: We still have a difference with double with -m32 since the move to GPR doesn't work
there. I'll file a separate PR for that and add test cases.
Since removing the CodeGenOnly instructions was fixing PR41619, I'm reverting r358887 which exposed that PR. Though I wouldn't be surprised
if that bug can still be hit independent of that.
This should hopefully get Google back to green. I'll work with Simon and other X86 folks to figure out how to move forward again.
llvm-svn: 360066
This addresses one half of https://bugs.llvm.org/show_bug.cgi?id=41635
by combining a VECREDUCE_AND/OR into VECREDUCE_UMIN/UMAX (if latter is
legal but former is not) for zero-or-all-ones boolean reductions (which
are detected based on sign bits).
Differential Revision: https://reviews.llvm.org/D61398
llvm-svn: 360054
A condition for exiting the legalization of v4i32 conversion to v2f64 through
extract/convert/build erroneously checks for the extract having type i32.
This is not adequate as smaller extracts are actually legalized to i32 as well.
Furthermore, an early exit is missing which means that we only check that
both extracts are from the same vector if that check fails.
As a result, both cases in the included test case fail - the first gets a
select error and the second generates incorrect code.
The culprit commit is r274535.
llvm-svn: 360043
This is a subset of the original commit from rL359879
which was reverted because it could crash when using the 'RemovedInstructions'
structure that enables delayed deletion of dead instructions. The motivating
compile-time win does not require that change though. We should get most of
that win from this change alone.
Using/updating a dominator tree to match math overflow patterns may be very
expensive in compile-time (because of the way CGP uses a DT), so just handle
the single-block case.
See post-commit thread for rL354298 for more details:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190422/646276.html
Differential Revision: https://reviews.llvm.org/D61075
llvm-svn: 359969
This saves us some unnecessary copies.
If the inputs to a G_SELECT are floating point, we should use fcsel rather than
csel.
Changes here are...
- Teach selectCopy about s1-to-s1 copies across register banks.
- AArch64RegisterBankInfo about G_SELECT in general.
- Teach the instruction selector about the FCSEL instructions.
Also add two tests:
- select-select.mir to show that we get the expected FCSEL
- regbank-select.mir (unfortunately named) to show the register banks on
G_SELECT are properly preserved
And update fast-isel-select.ll to show that we do the same thing as other
instruction selectors in these cases.
llvm-svn: 359940
The VOP3 form should always be the preferred selection form to be
shrunk later.
The r600 sub test needs to be split out because it asserts on the
arguments in the new test during the calling convention lowering.
llvm-svn: 359899
This was broken if the original operand was killed. The kill flag
would appear on both instructions, and fail the verifier. Keep the
kill flag, but remove the operands from the old instruction. This has
an added benefit of really reducing the use count for future folds.
Ideally the pass would be structured more like what PeepholeOptimizer
does to avoid this hack to avoid breaking instruction iterators.
llvm-svn: 359891
When a fold of an immediate into a sub/subrev required shrinking the
instruction, the wrong VOP2 opcode was used. This was using the VOP2
equivalent of the original instruction, not the commuted instruction
with the inverted opcode.
llvm-svn: 359883
Using/updating a dominator tree to match math overflow patterns may be very
expensive in compile-time (because of the way CGP uses a DT), so just handle
the single-block case.
Also, we were restarting the iterator loops when doing the overflow intrinsic
transforms by marking the dominator tree for update. That was done to prevent
iterating over a removed instruction. But we can postpone the deletion using
the existing "RemovedInsts" structure, and that means we don't need to update
the DT.
See post-commit thread for rL354298 for more details:
http://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20190422/646276.html
Differential Revision: https://reviews.llvm.org/D61075
llvm-svn: 359879
This is the second part of the commit fixing PR38917 (hoisting
partitially redundant machine instruction). Most of PRE (partitial
redundancy elimination) and CSE work is done on LLVM IR, but some of
redundancy arises during DAG legalization. Machine CSE is not enough
to deal with it. This simple PRE implementation works a little bit
intricately: it passes before CSE, looking for partitial redundancy
and transforming it to fully redundancy, anticipating that the next
CSE step will eliminate this created redundancy. If CSE doesn't
eliminate this, than created instruction will remain dead and eliminated
later by Remove Dead Machine Instructions pass.
The third part of the commit is supposed to refactor MachineCSE,
to make it more clear and to merge MachinePRE with MachineCSE,
so one need no rely on further Remove Dead pass to clear instrs
not eliminated by CSE.
First step: https://reviews.llvm.org/D54839
Fixes llvm.org/PR38917
Reviewers: RKSimon
Subscribers: hfinkel, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D56772
llvm-svn: 359870
We use to incorrectly use the store size instead of the alloc size when
creating the stack slot for allocas.
On aarch64 this can be demonstrated by allocating weirdly sized types.
For instance, in the added test case, we use an alloca for i19. We used
to allocate a slot of size 24-bit (19 rounded up to the next byte),
whereas we really want to use a full 32-bit slot for this type.
llvm-svn: 359856
The primary fix here is to WinException.cpp: we need to exclude jump
tables when computing the length of a function, or else we fail to
correctly compute the length. (We can only compute the number of bytes
consumed by certain assembler directives after the entire file is
parsed. ".p2align" is one of those directives, and is used by jump table
generation.)
The secondary fix, to MCWin64EH, is to make sure we don't silently
miscompile if we hit a similar situation in the future.
It's possible we could extend ARM64EmitUnwindInfo so it allows function
bodies that contain assembler directives, but that's a lot more
complicated; see the FIXME in MCWin64EH.cpp.
Fixes https://bugs.llvm.org/show_bug.cgi?id=41581 .
Differential Revision: https://reviews.llvm.org/D61095
llvm-svn: 359849
The default impementation in the base class for TargetLowering::getRegForInlineAsmConstraint doesn't work for mask registers when the VT is a scalar type integer types since the only legal mask types are vXi1. So we end up just getting whatever the first register class that contains the register. Currently this appears to be VK1, but its really dependent on the order tablegen outputs the register classes.
Some code in the caller ends up looking up the type for this register class and find v1i1 then generates a copyfromreg from the physical k-register with the v1i1 type. Then it generates an any_extend from v1i1 to the scalar VT which isn't legal. This bad any_extend sticks around until isel where it selects a MOVZX32rr8 with a v1i1 input or maybe a i8 input. Not sure but eventually we pick up a copy from VK1 to GR8 in MachineIR which isn't supported. This leads to a failure in physical register copying.
This patch uses the scalar type to find a VK class of the right size. In the attached test case this will be VK16. This causes a bitcast from vk16 to i16 to be generated instead of an any_extend. This will be properly iseled to a VK16 to GR32 copy and a GR32->GR16 extract_subreg.
Fixes PR41678
Differential Revision: https://reviews.llvm.org/D61453
llvm-svn: 359837
The original patch was committed at rL359398 and reverted at rL359695 because of
infinite looping.
This includes a fix to check for a vector splat of "1.0" to avoid the infinite loop.
Original commit message:
This was originally part of D61028, but it's an independent diff.
If we try the repeated divisor reciprocal transform before producing an estimate sequence,
then we have an opportunity to use scalar fdiv. On x86, the trade-off is 1 divss vs. 5
vector FP ops in the default estimate sequence. On recent chips (Skylake, Ryzen), the
full-precision division is only 3 cycle throughput, so that's probably the better perf
default option and avoids problems from x86's inaccurate estimates.
The last 2 tests show that users still have the option to override the defaults by using
the function attributes for reciprocal estimates, but those patterns are potentially made
faster by converting the vector ops (including ymm ops) to scalar math.
Differential Revision: https://reviews.llvm.org/D61149
llvm-svn: 359793
We don't have FP exception limits in the IR constant folder for the binops (apart from strict ops),
so it does not make sense to have them here in the DAG either. Nothing else in the backend tries
to preserve exceptions (again outside of strict ops), so I don't see how this could have ever
worked for real code that cares about FP exceptions.
There are still cases (examples: unary opcodes in SDAG, FMA in IR) where we are trying (at least
partially) to preserve exceptions without even asking if the target supports FP exceptions. Those
should be corrected in subsequent patches.
Real support for FP exceptions requires several changes to handle the constrained/strict FP ops.
Differential Revision: https://reviews.llvm.org/D61331
llvm-svn: 359791
Limiting scalar hadd/hsub generation to the lowest xmm looks to be unnecessary - we will be extracting one upper xmm whatever, and we can remove a shuffle by using the hop which is inline with what shouldUseHorizontalOp expects to happen anyway.
Testing on btver2 (the main target for fast-hops) shows this is beneficial even for float ops where we have a 'shuffle' to extract the float result:
https://godbolt.org/z/0R-U-K
Differential Revision: https://reviews.llvm.org/D61426
llvm-svn: 359786
Select G_SEXT and G_ZEXT with destination types smaller than 32 bits in
the exact same way as 32 bits. This overwrites the higher bits, but that
should be ok since all legal users of types smaller than 32 bits ignore
those bits anyway.
llvm-svn: 359768
Accidentally put a fast-isel-abort=2 instead of the GISel abort line.
This test doesn't actually fall back at all for GISel though, so remove the
fallback checks entirely.
llvm-svn: 359737
This adds support for using fmov rather than a standard mov to materialize
G_FCONSTANT when it's safe to do so.
Update arm64-fast-isel-materialize.ll and select-constant.mir to show that the
selection is correct.
llvm-svn: 359734
This is a slightly reduced version of the test from D61384.
Adding this as a preliminary step, so I can update D61149 with
the proposed fix.
llvm-svn: 359709
We already perform horizontal add/sub if we extract from elements 0 and 1, this patch extends it to non-0/1 element extraction indices (as long as they are from the lowest 128-bit vector).
Differential Revision: https://reviews.llvm.org/D61263
llvm-svn: 359707
This is an alternative to D59669 which more aggressively extracts i1 elements from vXi1 bool vectors using a MOVMSK.
Differential Revision: https://reviews.llvm.org/D61189
llvm-svn: 359666
-t is --symbols in llvm-readobj but --section-details (unimplemented) in readelf.
The confusing option should not be used since we aim for improving
compatibility.
Keep just one llvm-readobj -t use case in test/tools/llvm-readobj/symbols.test
llvm-svn: 359661
David Green