Normally this would be cleaned up by promoting the condition operand next. But in the attached case we promoted the result from v2i48 to v2i64 and the condition from v2i1 to v2i48. Then we tried to "promote" the v2i48 condition back to v2i1 because that's what the SetCC result type for v2i64 is on X86 with VLX. But promote is either a NOP or SIGN_EXTEND and this would need a truncation.
With the change here we now get the SetCC type of v2i1 when we're handling the result promotion and the operand no longer needs to be promoted itself.
Fixes PR35272.
llvm-svn: 318706
MachineSink attempts to place instructions near the basic blocks where
they are needed. Once an instruction has been sunk, its location
relative to other instructions is no longer consistent with the
original source code. In order to ensure correct single-stepping and
profiling, the debug location for sunk instructions is either merged
with the insertion point or erased if the target successor block is
empty.
Patch by Matthew Voss!
Differential Revision: https://reviews.llvm.org/D39933
llvm-svn: 318679
PR34553 has gone, adding tests to ensure it doesn't come back.
vselect_packss_v16i64 still has some awful codegen on AVX512 targets....
llvm-svn: 318599
This makes sure that functions that only clobber xmm registers
(on win64) also get the right cfi directives, if dwarf exceptions
are enabled.
Differential Revision: https://reviews.llvm.org/D40191
llvm-svn: 318591
Only do this pre-legalize in case we're using the sign extend to legalize for KNL.
This recovers all of the tests that changed when I stopped SelectionDAGBuilder from deleting sign extends.
There's more work that could be done here particularly to fix the i8->i64 test case that experienced split.
llvm-svn: 318468
Previously SelectionDAGBuilder would remove this sign extend leading to a failure during isel.
The codegen here isn't very nice as we ended up triggering a split.
llvm-svn: 318467
The sign extend might be from an i16 or i8 type and was inserted by InstCombine to match the pointer width. X86 gather legalization isn't currently detecting this to reinsert a sign extend to make things legal.
It's a bit weird for the SelectionDAGBuilder to do this kind of optimization in the first place. With this removed we can at least lean on InstCombine somewhat to ensure the index is i32 or i64.
I'll work on trying to recover some of the test cases by removing sign extends in the backend when its safe to do so with an understanding of the current legalizer capabilities.
This should fix PR30690.
llvm-svn: 318466
The wider element type will normally cause legalize to try to split and scalarize the gather/scatter, but we can't handle that. Instead, truncate the index early so the gather/scatter node is insulated from the legalization.
This really shouldn't happen in practice since InstCombine will normalize index types to the same size as pointers.
llvm-svn: 318452
The type legalizer will try to scalarize these operations if it sees them, but there is no handling for scalarizing them. This leads to a fatal error. With this change they will now be scalarized by the mem intrinsic scalarizing pass before SelectionDAG.
llvm-svn: 318380
Some CPUs are already overriding these sign extension instructions but we should be able to use the WriteALU schedule class by default.
Differential Revision: https://reviews.llvm.org/D39899
llvm-svn: 318308
The test was doing -stop-after=isel, but that pass is actually the
AMDGPUDAGToDAGISel pass, which might not be built when targeting x86_64.
This changes the test to -stop-after=expand-isel-pseudos instead.
Follow-up to r318202.
llvm-svn: 318220
artifacts along with DCE
Legalization Artifacts are all those insts that are there to make the
type system happy. Currently, the target needs to say all combinations
of extends and truncs are legal and there's no way of verifying that
post legalization, we only have *truly* legal instructions. This patch
changes roughly the legalization algorithm to process all illegal insts
at one go, and then process all truncs/extends that were added to
satisfy the type constraints separately trying to combine trivial cases
until they converge. This has the added benefit that, the target
legalizerinfo can only say which truncs and extends are okay and the
artifact combiner would combine away other exts and truncs.
Updated legalization algorithm to roughly the following pseudo code.
WorkList Insts, Artifacts;
collect_all_insts_and_artifacts(Insts, Artifacts);
do {
for (Inst in Insts)
legalizeInstrStep(Inst, Insts, Artifacts);
for (Artifact in Artifacts)
tryCombineArtifact(Artifact, Insts, Artifacts);
} while(!Insts.empty());
Also, wrote a simple wrapper equivalent to SetVector, except for
erasing, it avoids moving all elements over by one and instead just
nulls them out.
llvm-svn: 318210
This patch peels off the top case in switch statement into a branch if the
probability exceeds a threshold. This will help the branch prediction and
avoids the extra compares when lowering into chain of branches.
Differential Revision: http://reviews.llvm.org/D39262
llvm-svn: 318202
Clang implements the -finstrument-functions flag inherited from GCC, which
inserts calls to __cyg_profile_func_{enter,exit} on function entry and exit.
This is useful for getting a trace of how the functions in a program are
executed. Normally, the calls remain even if a function is inlined into another
function, but it is useful to be able to turn this off for users who are
interested in a lower-level trace, i.e. one that reflects what functions are
called post-inlining. (We use this to generate link order files for Chromium.)
LLVM already has a pass for inserting similar instrumentation calls to
mcount(), which it does after inlining. This patch renames and extends that
pass to handle calls both to mcount and the cygprofile functions, before and/or
after inlining as controlled by function attributes.
Differential Revision: https://reviews.llvm.org/D39287
llvm-svn: 318195
Summary:
Bypass of slow divs based on operand values is currently disabled for
-Os. Do the same when profile summary is available and the working set
size of the application is huge. This is similar to how loop peeling is
guarded by hasHugeWorkingSetSize. In the div bypass case, the generated
extra code (and the extra branch) tendss to outweigh the benefits of the
bypass. This results in noticeable performance improvement on an
internal application.
Reviewers: davidxl
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D39992
llvm-svn: 318179
If the base of our gather corresponds to something contained in X86ISD::Wrapper we should be able to fold it into the address.
This patch refactors some of the address matching to more fully use the X86ISelAddressMode struct and the getAddressOperands helper. A new helper function matchVectorAddress is added to call matchWrapper or fall back to matchAddressBase.
We should also be able to support constant offsets from a wrapper, but I'll look into that in a future patch. We may even be able to completely reuse matchAddress here, but I wanted to start simple and work up to it.
Differential Revision: https://reviews.llvm.org/D39927
llvm-svn: 318057
Remove builtins from llvm and add AutoUpgrade support.
Also add fast-isel tests for the TEST and TESTN instructions.
Differential Revision: https://reviews.llvm.org/D38736
llvm-svn: 318036
This patch, together with a matching clang patch (https://reviews.llvm.org/D38672), implements the lowering of X86 shuffle i/f intrinsics to IR.
Differential Revision: https://reviews.llvm.org/D38671
Change-Id: I1e7d359a74743e995ec356237a85214ce55d3661
llvm-svn: 318026
Updated the scheduling information of the SKX subtarget in the file X86SchedSkylakeServer.td under lib/Target/X86 to:
1. add regular opcodes in addition to the suffixed "_Int" opcodes
2. add the (V)MAXCPD/MAXCPS/MAXCSD/MAXCSS/MINCPD/MINCPS/MINCSD/MINCSS
instructions that are equivalent to their counterparts without the 'C' as they are part of a hack to
make floating point min/max commutable under fast math.
Reviewers: zvi, RKSimon, craig.topper
Differential Revision: https://reviews.llvm.org/D39833
Change-Id: Ie13702a5ce1b1a08af91ca637a52b6962881e7d6
llvm-svn: 318024
The VRNDSCALE instructions implement a superset of the (V)ROUND instructions. They are equivalent if the upper 4-bits of the immediate are 0.
This patch lowers the legacy intrinsics to the VRNDSCALE ISD node and masks the upper bits of the immediate to 0. This allows us to take advantage of the larger register encoding space.
We should maybe consider converting VRNDSCALE back to VROUND in the EVEX to VEX pass if the extended registers are not being used.
I notice some load folding opportunities being missed for the VRNDSCALESS/SD instructions that I'll try to fix in future patches.
llvm-svn: 318008
This is consistent with out normal implementation of scalar instructions.
While there disable load folding for the patterns with IMPLICIT_DEF unless optimizing for size which is also our standard practice.
llvm-svn: 317977
Currently we can only get a uniform base from a simple GEP with 2 operands. This causes us to miss address folding opportunities for simple global array accesses as the test case shows.
This patch adds support for larger GEPs if the other indices are 0 since those don't require any additional computations to be inserted.
We may also want to handle constant splats of zero here, but I'm leaving that for future work when I have a real world example.
Differential Revision: https://reviews.llvm.org/D39911
llvm-svn: 317947
r317453 added new ISD nodes without rounding modes that were added to an existing if/else chain. But all the previous nodes handled there included a rounding mode. The final code after this if/else chain expected an extra operand that isn't present for the new nodes.
llvm-svn: 317748
This reverts r317579, originally committed as r317100.
There is a design issue with marking CFI instructions duplicatable. Not
all targets support the CFIInstrInserter pass, and targets like Darwin
can't cope with duplicated prologue setup CFI instructions. The compact
unwind info emission fails.
When the following code is compiled for arm64 on Mac at -O3, the CFI
instructions end up getting tail duplicated, which causes compact unwind
info emission to fail:
int a, c, d, e, f, g, h, i, j, k, l, m;
void n(int o, int *b) {
if (g)
f = 0;
for (; f < o; f++) {
m = a;
if (l > j * k > i)
j = i = k = d;
h = b[c] - e;
}
}
We get assembly that looks like this:
; BB#1: ; %if.then
Lloh3:
adrp x9, _f@GOTPAGE
Lloh4:
ldr x9, [x9, _f@GOTPAGEOFF]
mov w8, wzr
Lloh5:
str wzr, [x9]
stp x20, x19, [sp, #-16]! ; 8-byte Folded Spill
.cfi_def_cfa_offset 16
.cfi_offset w19, -8
.cfi_offset w20, -16
cmp w8, w0
b.lt LBB0_3
b LBB0_7
LBB0_2: ; %entry.if.end_crit_edge
Lloh6:
adrp x8, _f@GOTPAGE
Lloh7:
ldr x8, [x8, _f@GOTPAGEOFF]
Lloh8:
ldr w8, [x8]
stp x20, x19, [sp, #-16]! ; 8-byte Folded Spill
.cfi_def_cfa_offset 16
.cfi_offset w19, -8
.cfi_offset w20, -16
cmp w8, w0
b.ge LBB0_7
LBB0_3: ; %for.body.lr.ph
Note the multiple .cfi_def* directives. Compact unwind info emission
can't handle that.
llvm-svn: 317726
These will be using inline asm to ensure we have coverage that we're unlikely to get from lowering of basic ir.
Currently waiting for D39728 to land to add support for scheduler comments for inline asm.
llvm-svn: 317698
Reland r317100 with minor fix regarding ComputeCommonTailLength function in
BranchFolding.cpp. Skipping top CFI instructions block needs to executed on
several more return points in ComputeCommonTailLength().
Original r317100 message:
"Correct dwarf unwind information in function epilogue for X86"
This patch aims to provide correct dwarf unwind information in function
epilogue for X86.
It consists of two parts. The first part inserts CFI instructions that set
appropriate cfa offset and cfa register in emitEpilogue() in
X86FrameLowering. This part is X86 specific.
The second part is platform independent and ensures that:
- CFI instructions do not affect code generation
- Unwind information remains correct when a function is modified by
different passes. This is done in a late pass by analyzing information
about cfa offset and cfa register in BBs and inserting additional CFI
directives where necessary.
Changed CFI instructions so that they:
- are duplicable
- are not counted as instructions when tail duplicating or tail merging
- can be compared as equal
Added CFIInstrInserter pass:
- analyzes each basic block to determine cfa offset and register valid at
its entry and exit
- verifies that outgoing cfa offset and register of predecessor blocks match
incoming values of their successors
- inserts additional CFI directives at basic block beginning to correct the
rule for calculating CFA
Having CFI instructions in function epilogue can cause incorrect CFA
calculation rule for some basic blocks. This can happen if, due to basic
block reordering, or the existence of multiple epilogue blocks, some of the
blocks have wrong cfa offset and register values set by the epilogue block
above them.
CFIInstrInserter is currently run only on X86, but can be used by any target
that implements support for adding CFI instructions in epilogue.
Patch by Violeta Vukobrat.
llvm-svn: 317579
Summary:
Calls using invoke in funclet based functions are assumed to clobber
all registers, which causes the stack adjustment using pops to consider
all registers not defined by the call to be undefined, which can
unfortunately include the base pointer, if one is needed.
To prevent this (and possibly other hazards), skip reserved registers
when looking for candidate registers.
This fixes issue #45034 in the Rust compiler.
Reviewers: mkuper
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D39636
llvm-svn: 317551
The EVEX to VEX pass is already assuming this is true under AVX512VL. We had special patterns to use zmm instructions if VLX and F16C weren't available.
Instead just make AVX512 imply F16C to make the EVEX to VEX behavior explicitly legal and remove the extra patterns.
All known CPUs with AVX512 have F16C so this should safe for now.
llvm-svn: 317521
Summary:
Print %subreg.<subregidxname> instead of just the subregister
index when printing immediate operands corresponding to subreg
indices in INSERT_SUBREG, EXTRACT_SUBREG, SUBREG_TO_REG and
REG_SEQUENCE.
Reviewers: qcolombet, MatzeB
Reviewed By: MatzeB
Subscribers: nhaehnle, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D39696
llvm-svn: 317513
Added TESTM and TESTNM to the list of instructions that already zeroing unused upper bits
and does not need the redundant shift left and shift right instructions afterwards.
Added a pattern for TESTM and TESTNM in iselLowering, so now icmp(neq,and(X,Y), 0) goes folds into TESTM
and icmp(eq,and(X,Y), 0) goes folds into TESTNM
This commit is a preparation for lowering the test and testn X86 intrinsics to IR.
Differential Revision: https://reviews.llvm.org/D38732
llvm-svn: 317465
Summary:
Try to lower a BUILD_VECTOR composed of extract-extract chains that can be
reasoned to be a permutation of a vector by indices in a non-constant vector.
We saw this pattern created by ISPC, which resolts to creating it due to the
requirement that shufflevector's mask operand be a *constant* vector.
I didn't check this but we could possibly use this pattern for lowering the X86 permute
C-instrinsics instead of llvm.x86 instrinsics.
This change can be followed by more improvements:
1. Handle vectors with undef elements.
2. Utilize pshufb and zero-mask-blending to support more effiecient
construction of vectors with constant-0 elements.
3. Use smaller-element vectors of same width, and "interpolate" the indices,
when no native operation available.
Reviewers: RKSimon, craig.topper
Reviewed By: RKSimon
Subscribers: chandlerc, DavidKreitzer
Differential Revision: https://reviews.llvm.org/D39126
llvm-svn: 317463
This patch, together with a matching clang patch (https://reviews.llvm.org/D38683), implements the lowering of X86 broadcastm intrinsics to IR.
Differential Revision: https://reviews.llvm.org/D38684
Change-Id: I709ac0b34641095397e994c8ff7e15d1315b3540
llvm-svn: 317458
Use feature names instead of CPU names.
A future commit will add avx512vl command lines to demonstrate missed use of EVEX instructions.
llvm-svn: 317451
Summary:
AVX512 added RCP14 and RSQRT instructions which improve accuracy over the legacy RCP and RSQRT instruction, but not enough accuracy to remove the need for a Newton Raphson refinement.
Currently we use these new instructions for the legacy packed SSE instrinics, but not the scalar instrinsics. And we use it for fast math optimization of division and reciprocal sqrt.
I think switching the legacy instrinsics maybe surprising to the user since it changes the answer based on which processor you're using regardless of any fastmath settings. It's also weird that we did something different between scalar and packed.
As far at the reciprocal estimation, I think it creates unnecessary deltas in our output behavior (and prevents EVEX->VEX). A little playing around with gcc and icc and godbolt suggest they don't change which instructions they use here.
This patch adds new X86ISD nodes for the RCP14/RSQRT14 and uses those for the new intrinsics. Leaving the old intrinsics to use the old instructions.
Going forward I think our focus should be on
-Supporting 512-bit vectors, which will have to use the RCP14/RSQRT14.
-Using RSQRT28/RCP28 to remove the Newton Raphson step on processors with AVX512ER
-Supporting double precision.
Reviewers: zvi, DavidKreitzer, RKSimon
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D39583
llvm-svn: 317413
This allows masked operations to be used and allows the register allocator to use YMM16-31 if necessary.
As a follow up I'll look into teaching EVEX->VEX how to turn this back into PERM2X128 if any of the additional features don't work out.
llvm-svn: 317403
Similar to the existing code to lower to PACKSS, we can use PACKUS if the input vector's leading zero bits extend all the way to the packed/truncated value.
We have to account for pre-SSE41 targets not supporting PACKUSDW
llvm-svn: 317315
undefined reference to `llvm::TargetPassConfig::ID' on
clang-ppc64le-linux-multistage
This reverts commit eea333c33fa73ad225ef28607795984829f65688.
llvm-svn: 317213
Summary:
This is mostly a noop (most of the test diffs are renamed blocks).
There are a few temporary register renames (eax<->ecx) and a few blocks are
shuffled around.
See the discussion in PR33325 for more details.
Reviewers: spatel
Subscribers: mgorny
Differential Revision: https://reviews.llvm.org/D39456
llvm-svn: 317211
When splitting a large load to smaller legally-typed loads, the last load should be padded to reach the size of the previous one so a CONCAT_VECTORS node could reunite them again.
The code currently pads the last load to reach the size of the first load (instead of the previous).
Differential Revision: https://reviews.llvm.org/D38495
Change-Id: Ib60b55ed26ce901fabf68108daf52683fbd5013f
llvm-svn: 317206
Similar to the existing code to lower to PACKSS, we can use PACKUS if the input vector's leading zero bits extend all the way to the packed/truncated value.
We have to account for pre-SSE41 targets not supporting PACKUSDW
llvm-svn: 317128
This will enable us to prefer VALIGND/Q during shuffle lowering in order to get the extended register encoding space when BWI isn't available. But if we end up not using the extended registers we can switch VPALIGNR for the shorter VEX encoding.
Differential Revision: https://reviews.llvm.org/D39401
llvm-svn: 317122
This patch moves the check for opt size and hasPartialRegUpdate into the lower level implementation of foldMemoryOperandImpl to catch the entry point that fast isel uses.
We're still folding undef register instructions in AVX that we should also probably disable, but that's a problem for another patch.
Unfortunately, this requires reordering a bunch of functions which is why the diff is so large. I can do the function reordering separately if we want.
Differential Revision: https://reviews.llvm.org/D39402
llvm-svn: 317112
Summary:
[X86] Teach fast isel to handle i64 sitofp with AVX.
For some reason we only handled i32 sitofp with AVX. But with SSE only we support i64 so we should do the same with AVX.
Also add i686 command lines for the 32-bit tests. 64-bit tests are in a separate file to avoid a fast-isel abort failure in 32-bit mode.
Reviewers: RKSimon, zvi
Reviewed By: RKSimon
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D39450
llvm-svn: 317102
This patch aims to provide correct dwarf unwind information in function
epilogue for X86.
It consists of two parts. The first part inserts CFI instructions that set
appropriate cfa offset and cfa register in emitEpilogue() in
X86FrameLowering. This part is X86 specific.
The second part is platform independent and ensures that:
- CFI instructions do not affect code generation
- Unwind information remains correct when a function is modified by
different passes. This is done in a late pass by analyzing information
about cfa offset and cfa register in BBs and inserting additional CFI
directives where necessary.
Changed CFI instructions so that they:
- are duplicable
- are not counted as instructions when tail duplicating or tail merging
- can be compared as equal
Added CFIInstrInserter pass:
- analyzes each basic block to determine cfa offset and register valid at
its entry and exit
- verifies that outgoing cfa offset and register of predecessor blocks match
incoming values of their successors
- inserts additional CFI directives at basic block beginning to correct the
rule for calculating CFA
Having CFI instructions in function epilogue can cause incorrect CFA
calculation rule for some basic blocks. This can happen if, due to basic
block reordering, or the existence of multiple epilogue blocks, some of the
blocks have wrong cfa offset and register values set by the epilogue block
above them.
CFIInstrInserter is currently run only on X86, but can be used by any target
that implements support for adding CFI instructions in epilogue.
Patch by Violeta Vukobrat.
Differential Revision: https://reviews.llvm.org/D35844
llvm-svn: 317100
So far we've only been using PACKSS truncations with 'all-bits or zero-bits' patterns (vector comparison results etc.). When really we can safely use it for any case as long as the number of sign bits reach down to the last 16-bits (or 8-bits if we're truncating to bytes).
The next steps after this is add the equivalent support for PACKUS and to support packing to sub-128 bit vectors for truncating stores etc.
Differential Revision: https://reviews.llvm.org/D39476
llvm-svn: 317086
Unfortunately, this weakens our ability to do domain fixing when AVX512DQ is not enabled, but it is consistent with our 256-bit behavior.
Maybe we should add custom handling to domain fixing to allow EVEX integer XOR/AND/OR/ANDN to switch to VEX encoded fp instructions if the high registers aren't being used?
llvm-svn: 316978
Summary:
INC/DEC don't update the carry flag so we need to make sure we don't try to use it.
This patch introduces new X86ISD opcodes for locked INC/DEC. Teaches lowerAtomicArithWithLOCK to emit these nodes if INC/DEC is not slow or the function is being optimized for size. An additional flag is added that allows the INC/DEC to be disabled if the caller determines that the carry flag is being requested.
The test_sub_1_cmp_1_setcc_ugt test is currently showing this bug. The other test case changes are recovering cases that were regressed in r316860.
This should fully fix PR35068 finishing the fix started in r316860.
Reviewers: RKSimon, zvi, spatel
Reviewed By: zvi
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D39411
llvm-svn: 316913
- Targets that want to support memcmp expansions now return the list of
supported load sizes.
- Expansion codegen does not assume that all power-of-two load sizes
smaller than the max load size are valid. For examples, this is not the
case for x86(32bit)+sse2.
Fixes PR34887.
llvm-svn: 316905
Craig Topper