This particular DAG combine is designed to kick in when both ConstantFPs will
end up being loaded via a litpool, however those nodes have a semi-legal
status, dictated by isFPImmLegal so in some cases there wouldn't have been a
litpool in the first place. Don't try to be clever in those circumstances.
Picked up while merging some AArch64 tests.
llvm-svn: 206365
Print in decimal for inline immediates, and hex otherwise. Use hex
always for offsets in addressing offsets.
This approximately matches what the shader compiler does.
llvm-svn: 206335
handles Intrinsic::trap if TargetOptions::TrapFuncName is set.
This fixes a bug in which the trap function was not taken into consideration
when a program was compiled without optimization (at -O0).
<rdar://problem/16291933>
llvm-svn: 206323
This patch teaches the backend how to efficiently lower logical and
arithmetic packed shifts on both SSE and AVX/AVX2 machines.
When possible, instead of scalarizing a vector shift, the backend should try
to expand the shift into a sequence of two packed shifts by immedate count
followed by a MOVSS/MOVSD.
Example
(v4i32 (srl A, (build_vector < X, Y, Y, Y>)))
Can be rewritten as:
(v4i32 (MOVSS (srl A, <Y,Y,Y,Y>), (srl A, <X,X,X,X>)))
[with X and Y ConstantInt]
The advantage is that the two new shifts from the example would be lowered into
X86ISD::VSRLI nodes. This is always cheaper than scalarizing the vector into
four scalar shifts plus four pairs of vector insert/extract.
llvm-svn: 206316
Sometimes we need emit the bits that would actually be a MOVN when producing a
relocated MOVZ instruction (don't ask). But not always, a check which ARM64 got
wrong until now.
llvm-svn: 206289
Code is mostly copied directly across, with a slight extension of the
ISelDAGToDAG function so that it can cope with the floating-point constants
being behind a litpool.
llvm-svn: 206285
In rare cases the dead definition elimination pass code can cause illegal cmn
instructions when it replaces dead registers on instructions that use
unmaterialized frame indexes. This patch disables the dead definition
optimization for instructions which include frame index operands.
rdar://16438284
llvm-svn: 206208
Previously, BranchProbabilityInfo::calcLoopBranchHeuristics would determine the weights of basic blocks inside loops even when it didn't have enough information to estimate the branch probabilities correctly. This patch fixes the function to exit early if it doesn't see any exit edges or back edges and let the later heuristics determine the weights.
This fixes PR18705 and <rdar://problem/15991090>.
Differential Revision: http://reviews.llvm.org/D3363
llvm-svn: 206194
Summary:
This was another incorrect use of hasMips64() vs isGP64bit().
Depends on D3344
Reviewers: matheusalmeida, vmedic
Reviewed By: vmedic
Differential Revision: http://reviews.llvm.org/D3347
llvm-svn: 206187
Summary:
Two exceptions to this:
test/CodeGen/Mips/octeon.ll
test/CodeGen/Mips/octeon_popcnt.ll
these test extensions to MIPS64
One test is altered for MIPS-IV:
test/CodeGen/Mips/mips64countleading.ll
Tests dclo/dclz which were added in MIPS64. The MIPS-IV version tests
that dclo/dclz are not emitted.
Four tests fail and are not in this patch:
test/CodeGen/Mips/abicalls.ll
test/CodeGen/Mips/fcopysign-f32-f64.ll
test/CodeGen/Mips/fcopysign.ll
test/CodeGen/Mips/stack-alignment.ll
Depends on D3343
Reviewers: matheusalmeida, vmedic
Reviewed By: vmedic
Differential Revision: http://reviews.llvm.org/D3344
llvm-svn: 206185
Summary:
- Conditional moves acting on 64-bit GPR's should require MIPS-IV rather than MIPS64
- ISD::MUL, and ISD::MULH[US] should be lowered on all 64-bit ISA's
Patch by David Chisnall
His work was sponsored by: DARPA, AFRL
I've added additional testcases to cover as much of the codegen changes
affecting MIPS-IV as I can. Where I've been unable to find an existing
MIPS64 testcase that can be re-used for MIPS-IV (mainly tests covering
ISD::GlobalAddress and similar), I at least agree that MIPS-IV should
behave like MIPS64. Further testcases that are fixed by this patch will follow
in my next commit. The testcases from that commit that fail for MIPS-IV without
this patch are:
LLVM :: CodeGen/Mips/2010-07-20-Switch.ll
LLVM :: CodeGen/Mips/cmov.ll
LLVM :: CodeGen/Mips/eh-dwarf-cfa.ll
LLVM :: CodeGen/Mips/largeimmprinting.ll
LLVM :: CodeGen/Mips/longbranch.ll
LLVM :: CodeGen/Mips/mips64-f128.ll
LLVM :: CodeGen/Mips/mips64directive.ll
LLVM :: CodeGen/Mips/mips64ext.ll
LLVM :: CodeGen/Mips/mips64fpldst.ll
LLVM :: CodeGen/Mips/mips64intldst.ll
LLVM :: CodeGen/Mips/mips64load-store-left-right.ll
LLVM :: CodeGen/Mips/sint-fp-store_pattern.ll
Reviewers: dsanders
Reviewed By: dsanders
CC: matheusalmeida
Differential Revision: http://reviews.llvm.org/D3343
llvm-svn: 206183
There was one definite issue in ARM64 (the off-by-1 check for whether
a shift could be folded in) and one difference that is probably
correct: ARM64 didn't fold nodes with multiple uses into the
arithmetic operations unless optimising for code size.
llvm-svn: 206168
Summary:
Previously loadImmediate() would produce MKMSK instructions with invalid
immediate values such as mkmsk r0, 9. Fix this by checking the mask size
is valid.
Reviewers: robertlytton
Reviewed By: robertlytton
CC: llvm-commits
Differential Revision: http://reviews.llvm.org/D3289
llvm-svn: 206163
We had been using the known-zero values of the operand of the or to construct
the mask for an rlwimi; this is not quite correct, but fine when the mask is
constant. When the mask is constant, then the known zeros of the operand must
be a superset of the zeros in the mask. However, when the mask is not a
constant, then there might be bits in the operand that are not known to be zero
that, at runtime, might be zero in the mask. Therefore, we check that any bits
not known to be zero *are* known to be one in the mask. Otherwise, we can't
fold the mask with the or and shift.
This was revealed as a miscompile of
MultiSource/Benchmarks/BitBench/drop3/drop3 when I started experimenting with
constant hoisting.
llvm-svn: 206136
We had disabled use of TBAA during CodeGen (even when otherwise using AA)
because the ptrtoint/inttoptr used by CGP for address sinking caused BasicAA to
miss basic type punning that it should catch (and, thus, we'd fail to override
TBAA when we should).
However, when AA is in use during CodeGen, CGP now uses normal GEPs and
bitcasts, instead of ptrtoint/inttoptr, when doing address sinking. As a
result, BasicAA should be able to make us do the right thing in the face of
type-punning, and it seems safe to enable use of TBAA again. self-hosting seems
fine on PPC64/Linux on the P7, with TBAA enabled and -misched=shuffle.
Note: We still don't update TBAA when merging stack slots, although because
BasicAA should now catch all such cases, this is no longer a blocking issue.
Nevertheless, I plan to commit code to deal with this properly in the near
future.
llvm-svn: 206093
The current memory-instruction optimization logic in CGP, which sinks parts of
the address computation that can be adsorbed by the addressing mode, does this
by explicitly converting the relevant part of the address computation into
IR-level integer operations (making use of ptrtoint and inttoptr). For most
targets this is currently not a problem, but for targets wishing to make use of
IR-level aliasing analysis during CodeGen, the use of ptrtoint/inttoptr is a
problem for two reasons:
1. BasicAA becomes less powerful in the face of the ptrtoint/inttoptr
2. In cases where type-punning was used, and BasicAA was used
to override TBAA, BasicAA may no longer do so. (this had forced us to disable
all use of TBAA in CodeGen; something which we can now enable again)
This (use of GEPs instead of ptrtoint/inttoptr) is not currently enabled by
default (except for those targets that use AA during CodeGen), and so aside
from some PowerPC subtargets and SystemZ, there should be no change in
behavior. We may be able to switch completely away from the ptrtoint/inttoptr
sinking on all targets, but further testing is required.
I've doubled-up on a number of existing tests that are sensitive to the
address sinking behavior (including some store-merging tests that are
sensitive to the order of the resulting ADD operations at the SDAG level).
llvm-svn: 206092
-fexhaustive-register-search option to allow an exhaustive search during last
chance recoloring.
This is related to PR18747
Patch by MAYUR PANDEY <mayur.p@samsung.com>.
llvm-svn: 206072
The TargetLowering::expandMUL() helper contains lowering code extracted
from the DAGTypeLegalizer and allows the SelectionDAGLegalizer to expand more
ISD::MUL patterns without having to use a library call.
llvm-svn: 206037
This removes the -segmented-stacks command line flag in favor of a
per-function "split-stack" attribute.
Patch by Luqman Aden and Alex Crichton!
llvm-svn: 205997
Refactored stack-protector.ll to use new-style function attributes everywhere
and eliminated unnecessary attributes.
This cleanup is in preparation for an upcoming test change.
llvm-svn: 205996
AVX supports logical operations using an operand from memory. Unfortunately
because integer operations were not added until AVX2 the AVX1 logical
operation's types were preventing the isel from folding the loads. In a limited
number of cases the peephole optimizer would fold the loads, but most were
missed. This patch adds explicit patterns with appropriate casts in order for
these loads to be folded.
The included test cases run on reduced examples and disable the peephole
optimizer to ensure the folds are being pattern matched.
Patch by Louis Gerbarg <lgg@apple.com>
rdar://16355124
llvm-svn: 205938
FoldConstantArithmetic() only knows how to deal with a few target independent
ISD opcodes. Bail early if it sees a target-specific ISD node. These node do
funny things with operand types which may break the assumptions of the code
that follows, and there's no actual folding that can be done anyway. For example,
non-constant 256 bit vector shifts on X86 have a shift-amount operand that's a
128-bit v4i32 vector regardless of what the first operand type is and that breaks
the assumption that the operand types must match.
rdar://16530923
llvm-svn: 205937
In AArch64 i64 to i32 truncate operation is a subregister access.
This allows more opportunities for LSR optmization to eliminate
variables of different types (i32 and i64).
llvm-svn: 205925
sign/zero/any extensions. However a few places were not checking properly the
property of the load and were turning an indexed load into a regular extended
load. Therefore the indexed value was lost during the process and this was
triggering an assertion.
<rdar://problem/16389332>
llvm-svn: 205923
This commit adds intrinsics and codegen support for the surface read/write and texture read instructions that take an explicit sampler parameter. Codegen operates on image handles at the PTX level, but falls back to direct replacement of handles with kernel arguments if image handles are not enabled. Note that image handles are explicitly disabled for all target architectures in this change (to be enabled later).
llvm-svn: 205907
Summary:
They behave in accordance with the Has2008 and ABS2008 configuration bits of the processor which are used to select between the 1985 and 2008 versions of IEEE 754. In 1985 mode, these instructions are arithmetic (i.e. they raise invalid operation exceptions when given NaN), in 2008 mode they are non-arithmetic (i.e. they are copies).
nmadd.[ds], and nmsub.[ds] are still subject to -enable-no-nans-fp-math because the ISA spec does not explicitly state that they obey Has2008 and ABS2008.
Fixed the issue with the previous version of this patch (r205628). A pre-existing 'let Predicate =' statement was removing some predicates that were necessary for FP64 to behave correctly.
Reviewers: matheusalmeida
Reviewed By: matheusalmeida
Differential Revision: http://llvm-reviews.chandlerc.com/D3274
llvm-svn: 205844
This implements the target-hooks for ARM64 to enable constant hoisting.
This fixes <rdar://problem/14774662> and <rdar://problem/16381500>.
llvm-svn: 205791
Confusingly, the NEON fmla instructions put the accumulator first but the
scalar versions put it at the end (like the fma lib function & LLVM's
intrinsic).
This should fix PR19345, assuming there's only one issue.
llvm-svn: 205758
Moving these patterns from TableGen files to PerformDAGCombine()
should allow us to generate better code by eliminating unnecessary
shifts and extensions earlier.
This also fixes a bug where the MAD pattern was calling
SimplifyDemandedBits with a 24-bit mask on the first operand
even when the full pattern wasn't being matched. This occasionally
resulted in some instructions being incorrectly deleted from the
program.
v2:
- Fix bug with 64-bit mul
llvm-svn: 205731
It affected callee's stack pop in x86. It is one of devergences between cygwin and mingw since mingw-gcc-4.6.
Added testcases to llvm/test/CodeGen/X86/win32_sret.ll for cygwin.
llvm-svn: 205688
gcc inline asm supports specifying "cc" as a clobber of all condition
registers. Add just enough modeling of the full register to make this work.
Fixed PR19326.
llvm-svn: 205630
Summary:
They behave in accordance with the Has2008 and ABS2008 configuration bits of the
processor which are used to select between the 1985 and 2008 versions of IEEE
754. In 1985 mode, these instructions are arithmetic (i.e. they raise invalid
operation exceptions when given NaN), in 2008 mode they are non-arithmetic
(i.e. they are copies).
nmadd.[ds], and nmsub.[ds] are still subject to -enable-no-nans-fp-math because
the ISA spec does not explicitly state that they obey Has2008 and ABS2008.
Reviewers: matheusalmeida
Reviewed By: matheusalmeida
Differential Revision: http://llvm-reviews.chandlerc.com/D3274
llvm-svn: 205628
When LLVM sees something like (v1iN (vselect v1i1, v1iN, v1iN)) it can
decide that the result is OK (v1i64 is legal on AArch64, for example)
but it still need scalarising because of that v1i1. There was no code
to do this though.
AArch64 and ARM64 have DAG combines to produce efficient code and
prevent that occuring in *most* such situations, but there are edge
cases that they miss. This adds a legalization to cope with that.
llvm-svn: 205626
There were several overlapping problems here, and this solution is
closely inspired by the one adopted in AArch64 in r201381.
Firstly, scalarisation of v1i1 setcc operations simply fails if the
input types are legal. This is fixed in LegalizeVectorTypes.cpp this
time, and allows AArch64 code to be simplified slightly.
Second, vselect with such a setcc feeding into it ends up in
ScalarizeVectorOperand, where it's not handled. I experimented with an
implementation, but found that whatever DAG came out was rather
horrific. I think Hao's DAG combine approach is a good one for
quality, though there are edge cases it won't catch (to be fixed
separately).
Should fix PR19335.
llvm-svn: 205625
The previous patterns directly inserted FMOV or INS instructions into
the DAG for scalar_to_vector & bitconvert patterns. This is horribly
inefficient and can generated lots more GPR <-> FPR register traffic
than necessary.
It's much better to emit instructions the register allocator
understands so it can coalesce the copies when appropriate.
It led to at least one ISelLowering hack to avoid the problems, which
was incorrect for v1i64 (FPR64 has no dsub). It can now be removed
entirely.
This should also fix PR19331.
llvm-svn: 205616
Without this change, the llvm_unreachable kicked in. The code pattern
being spotted is rather non-canonical for 128-bit MLAs, but it can
happen and there's no point in generating sub-optimal code for it just
because it looks odd.
Should fix PR19332.
llvm-svn: 205615
recoloring cut-offs are encountered and register allocation failed.
This is related to PR18747
Patch by MAYUR PANDEY <mayur.p@samsung.com>.
llvm-svn: 205601
Removes unnecessary casts from non-generic address spaces to the generic address
space for certain code patterns.
Patch by Jingyue Wu.
llvm-svn: 205571
When rematerializing through truncates, the coalescer may produce instructions
with dead defs, but live implicit-defs of subregs:
E.g.
%X1<def,dead> = MOVi64imm 2, %W1<imp-def>; %X1:GPR64, %W1:GPR32
These instructions are live, and their definitions should not be rewritten.
Fixes <rdar://problem/16492408>
llvm-svn: 205565
Acording to AMD documentation, the correct opcode for
BFE_INT is 0x5, not 0x4
Fixes Arithm/Absdiff.Mat/3 OpenCV test
Patch by: Bruno Jiménez
llvm-svn: 205562
More updating of tests to be explicit about the target triple rather than
relying on the default target triple supporting ARM mode.
Indicate to lit that object emission is not yet available for Windows on ARM.
llvm-svn: 205545
This changes the tests that were targeting ARM EABI to explicitly specify the
environment rather than relying on the default. This breaks with the new
Windows on ARM support when running the tests on Windows where the default
environment is no longer EABI.
Take the opportunity to avoid a pointless redirect (helps when trying to debug
with providing a command line invocation which can be copy and pasted) and
removing a few greps in favour of FileCheck.
llvm-svn: 205541
Implementing this via ComputeMaskedBits has two advantages:
+ It actually works. DAGISel doesn't deal with the chains properly
in the previous pattern-based solution, so they never trigger.
+ The information can be used in other DAG combines, as well as the
trivial "get rid of truncs". For example if the trunc is in a
different basic block.
rdar://problem/16227836
llvm-svn: 205540
The terminal barrier of a cmpxchg expansion will be either Acquire or
SequentiallyConsistent. In either case it can be skipped if the
operation has Monotonic requirements on failure.
rdar://problem/15996804
llvm-svn: 205535
The previous situation where ATOMIC_LOAD_WHATEVER nodes were expanded
at MachineInstr emission time had grown to be extremely large and
involved, to account for the subtly different code needed for the
various flavours (8/16/32/64 bit, cmpxchg/add/minmax).
Moving this transformation into the IR clears up the code
substantially, and makes future optimisations much easier:
1. an atomicrmw followed by using the *new* value can be more
efficient. As an IR pass, simple CSE could handle this
efficiently.
2. Making use of cmpxchg success/failure orderings only has to be done
in one (simpler) place.
3. The common "cmpxchg; did we store?" idiom can be exposed to
optimisation.
I intend to gradually improve this situation within the ARM backend
and make sure there are no hidden issues before moving the code out
into CodeGen to be shared with (at least ARM64/AArch64, though I think
PPC & Mips could benefit too).
llvm-svn: 205525
add operation since extract_vector_elt can perform an extend operation. Get the input lane
type from the vector on which we're performing the vpaddl operation on and extend or
truncate it to the output type of the original add node.
llvm-svn: 205523
This changes the tests that were targeting ARM EABI to explicitly specify the
environment rather than relying on the default. This breaks with the new
Windows on ARM support when running the tests on Windows where the default
environment is no longer EABI.
Take the opportunity to avoid a pointless redirect (helps when trying to debug
with providing a command line invocation which can be copy and pasted) and
removing a few greps in favour of FileCheck.
llvm-svn: 205465
Unlike other v6+ processors, cortex-m0 never supports unaligned accesses.
From the v6m ARM ARM:
"A3.2 Alignment support: ARMv6-M always generates a fault when an unaligned
access occurs."
rdar://16491560
llvm-svn: 205452
Adds the instructions ext/ext32/cins/cins32.
It also changes pop/dpop to accept the two operand version and
adds a simple pattern to generate baddu.
Tests for the two operand versions (including baddu/dmul/dpop/pop)
and the code generation pattern for baddu are included.
Reviewed by: Daniel.Sanders@imgtec.com
llvm-svn: 205449
Weak symbols cannot use the small code model's usual ADRP sequences since the
instruction simply may not be able to encode a value of 0.
This redirects them to use the GOT, which hopefully linkers are able to cope
with even in the static relocation model.
llvm-svn: 205426
Again, coalescing and other optimisations swiftly made the MachineInstrs
consistent again, but when compiled at -O0 a bad INSERT_SUBREGISTER was
produced.
llvm-svn: 205423
The previous attempt was fine with optimisations, but was actually rather
cavalier with its types. When compiled at -O0, it produced invalid COPY
MachineInstrs.
llvm-svn: 205422
ARM specific optimiztion, finding places in ARM machine code where 2 dmbs
follow one another, and eliminating one of them.
Patch by Reinoud Elhorst.
llvm-svn: 205409
Identical to Win32 method except the GS segment register is used for TLS
instead of FS and pvArbitrary is at TEB offset 0x28 instead of 0x14.
llvm-svn: 205342
The Cyclone CPU is similar to swift for most LLVM purposes, but does have two
preferred instructions for zeroing a VFP register. This teaches LLVM about
them.
llvm-svn: 205309
This moves one case of raw text checking down into the MCStreamer
interfaces in the form of a virtual function, even if we ultimately end
up consolidating on the one-or-many line tables issue one day, this is
nicer in the interim. This just generally streamlines a bunch of use
cases into a common code path.
llvm-svn: 205287
This commit updates the stackmap format to version 1 to indicate the
reorganizaion of several fields. This was done in order to align stackmap
entries to their natural alignment and to minimize padding.
Fixes <rdar://problem/16005902>
llvm-svn: 205254
If we have two unique values for a v2i64 build vector, this will always result
in two vector loads if we expand using shuffles. Only one is necessary.
llvm-svn: 205231
This is a more thorough fix for the issue than r203483. An IR pass will run
before NVPTX codegen to make sure there are no invalid symbol names that can't
be consumed by the ptxas assembler.
llvm-svn: 205212
When the loop vectorizer vectorizes code that uses the loop induction variable,
we often end up with IR like this:
%b1 = insertelement <2 x i32> undef, i32 %v, i32 0
%b2 = shufflevector <2 x i32> %b1, <2 x i32> undef, <2 x i32> zeroinitializer
%i = add <2 x i32> %b2, <i32 2, i32 3>
If the add in this example is not legal (as is the case on PPC with VSX), it
will be scalarized, and we'll end up with a number of extract_vector_elt nodes
with the vector shuffle as the input operand, and that vector shuffle is fed by
one or more build_vector nodes. By the time that vector operations are
expanded, visitEXTRACT_VECTOR_ELT will not create new extract_vector_elt by
looking through the vector shuffle (to make sure that no illegal operations are
created), and so the extract_vector_elt -> vector shuffle -> build_vector is
never simplified to an operand of the build vector.
By looking at build_vectors through a shuffle we fix this particular situation,
preventing a vector from being built, only to be deconstructed again (for the
scalarized add) -- an expensive proposition when this all needs to be done via
the stack. We probably want a more comprehensive fix here where we look back
recursively through any shuffles to any build_vectors or scalar_to_vectors,
etc. but that can come later.
llvm-svn: 205179
While reviewing r204163, I noticed that the MIPS16 test only checked for a .ent
directive and didn't actually check the code emitted. Fixed this and added a
check for llvm.bswap.i32 on MIPS64 at the same time.
llvm-svn: 205177
is not a pattern to lower this with clever instructions that zero the
register, so restrict the zero immediate legality special case to f64
and f32 (the only two sizes which fmov seems to directly support). Fixes
backend errors when building code such as libxml.
llvm-svn: 205161
When expanding EXTRACT_VECTOR_ELT and EXTRACT_SUBVECTOR using
SelectionDAGLegalize::ExpandExtractFromVectorThroughStack, we store the entire
vector and then load the piece we want. This is fine in isolation, but
generating a new store (and corresponding stack slot) for each extraction ends
up producing code of poor quality. When we scalarize a vector operation (using
SelectionDAG::UnrollVectorOp for example) we generate one EXTRACT_VECTOR_ELT
for each element in the vector. This used to generate one stored copy of the
vector for each element in the vector. Now we search the uses of the vector for
a suitable store before generating a new one, which results in much more
efficient scalarization code.
llvm-svn: 205153
sitofp from v2i32 to v2f64 ends up generating a SIGN_EXTEND_INREG v2i64 node
(and similarly for v2i16 and v2i8). Even though there are no sign-extension (or
algebraic shifts) for v2i64 types, we can handle v2i32 sign extensions by
converting two and from v2i64. The small trick necessary here is to shift the
i32 elements into the right lanes before the i32 -> f64 step. This is because
of the big Endian nature of the system, we need the i32 portion in the high
word of the i64 elements.
For v2i16 and v2i8 we can do the same, but we first use the default Altivec
shift-based expansion from v2i16 or v2i8 to v2i32 (by casting to v4i32) and
then apply the above procedure.
llvm-svn: 205146
v2i64 is a legal type under VSX, however we don't have native vector
comparisons. We can handle eq/ne by casting it to an Altivec type, but
everything else must be expanded.
llvm-svn: 205106
This adds a second implementation of the AArch64 architecture to LLVM,
accessible in parallel via the "arm64" triple. The plan over the
coming weeks & months is to merge the two into a single backend,
during which time thorough code review should naturally occur.
Everything will be easier with the target in-tree though, hence this
commit.
llvm-svn: 205090
We had stored both f64 values and v2f64, etc. values in the VSX registers. This
worked, but was suboptimal because we would always spill 16-byte values even
through we almost always had scalar 8-byte values. This resulted in an
increase in stack-size use, extra memory bandwidth, etc. To fix this, I've
added 64-bit subregisters of the Altivec registers, and combined those with the
existing scalar floating-point registers to form a class of VSX scalar
floating-point registers. The ABI code has also been enhanced to use this
register class and some other necessary improvements have been made.
llvm-svn: 205075
Emit 32-bit register names instead of 64-bit register names if the target does
not have 64-bit general purpose registers.
<rdar://problem/14653996>
llvm-svn: 205067
WinCOFF cannot form PC relative relocations to support absolute
MCValues. We should reenable this once WinCOFF supports emission of
IMAGE_REL_I386_REL32 relocations.
This fixes PR19272.
llvm-svn: 205058
v2[fi]64 values need to be explicitly passed in VSX registers. This is because
the code in TRI that finds the minimal register class given a register and a
value type will assert if given an Altivec register and a non-Altivec type.
llvm-svn: 205041
As explained in r204976, because of how the allocation of VSX registers
interacts with the call-lowering code, we sometimes end up generating self VSX
copies. Specifically, things like this:
%VSL2<def> = COPY %F2, %VSL2<imp-use,kill>
(where %F2 is really a sub-register of %VSL2, and so this copy is a nop)
This adds a small cleanup pass to remove these prior to post-RA scheduling.
llvm-svn: 204980
First, v2f64 vector extract had not been declared legal (and so the existing
patterns were not being used). Second, the patterns for that, and for
scalar_to_vector, should really be a regclass copy, not a subregister
operation, because the VSX registers directly hold both the vector and scalar data.
llvm-svn: 204971
These operations need to be expanded during legalization so that isel does not
crash. In theory, we might be able to custom lower some of these. That,
however, would need to be follow-up work.
llvm-svn: 204963
This adds back r204781.
Original message:
Aliases are just another name for a position in a file. As such, the
regular symbol resolutions are not applied. For example, given
define void @my_func() {
ret void
}
@my_alias = alias weak void ()* @my_func
@my_alias2 = alias void ()* @my_alias
We produce without this patch:
.weak my_alias
my_alias = my_func
.globl my_alias2
my_alias2 = my_alias
That is, in the resulting ELF file my_alias, my_func and my_alias are
just 3 names pointing to offset 0 of .text. That is *not* the
semantics of IR linking. For example, linking in a
@my_alias = alias void ()* @other_func
would require the strong my_alias to override the weak one and
my_alias2 would end up pointing to other_func.
There is no way to represent that with aliases being just another
name, so the best solution seems to be to just disallow it, converting
a miscompile into an error.
llvm-svn: 204934
These instructions have access to the complete VSX register file. In addition,
they "swap" the order of the elements so that element 0 (the scalar part) comes
first in memory and element 1 follows at a higher address.
llvm-svn: 204838
In some cases it is possible for CGP to attempt to reuse a base address from
another basic block. In those cases we have to be sure that all the address
math was either done at the same bit width, or that none of it overflowed
before it was extended.
Patch by Louis Gerbarg <lgg@apple.com>
rdar://16307442
llvm-svn: 204833
> For functions where esi is used as base pointer, we would previously fall ba
> from lowering memcpy with "rep movs" because that clobbers esi.
>
> With this patch, we just store esi in another physical register, and restore
> it afterwards. This adds a little bit of register preassure, but the more
> efficient memcpy should be worth it.
>
> Differential Revision: http://llvm-reviews.chandlerc.com/D2968
This didn't work. I was ending up with code like this:
lea edi,[esi+38h]
mov ecx,0Fh
mov edx,esi
mov esi,ebx
rep movs dword ptr es:[edi],dword ptr [esi]
lea ecx,[esi+74h] <-- Ooops, we're now using esi before restoring it from edx.
add ebx,3Ch
mov esi,edx
I guess if we want to do this we need stronger glue or something, or doing the expansion
much later.
llvm-svn: 204829
v2i64 needs to be a legal VSX type because it is the SetCC result type from
v2f64 comparisons. We need to expand all non-arithmetic v2i64 operations.
This fixes the lowering for v2f64 VSELECT.
llvm-svn: 204828
We've already got versions without the barriers, so this just adds IR-level
support for generating the new v8 ones.
rdar://problem/16227836
llvm-svn: 204813
Implementing the LLVM part of the call to __builtin___clear_cache
which translates into an intrinsic @llvm.clear_cache and is lowered
by each target, either to a call to __clear_cache or nothing at all
incase the caches are unified.
Updating LangRef and adding some tests for the implemented architectures.
Other archs will have to implement the method in case this builtin
has to be compiled for it, since the default behaviour is to bail
unimplemented.
A Clang patch is required for the builtin to be lowered into the
llvm intrinsic. This will be done next.
llvm-svn: 204802
With VSX there is a real vector select instruction, and so we should use it.
Note that VSELECT will still scalarize for v2f64 because the corresponding
SetCC result type (v2i64) is not currently a legal type.
llvm-svn: 204801
This reverts commit r204781.
I will follow up to with msan folks to see what is what they
were trying to do with aliases to weak aliases.
llvm-svn: 204784
Aliases are just another name for a position in a file. As such, the
regular symbol resolutions are not applied. For example, given
define void @my_func() {
ret void
}
@my_alias = alias weak void ()* @my_func
@my_alias2 = alias void ()* @my_alias
We produce without this patch:
.weak my_alias
my_alias = my_func
.globl my_alias2
my_alias2 = my_alias
That is, in the resulting ELF file my_alias, my_func and my_alias are
just 3 names pointing to offset 0 of .text. That is *not* the
semantics of IR linking. For example, linking in a
@my_alias = alias void ()* @other_func
would require the strong my_alias to override the weak one and
my_alias2 would end up pointing to other_func.
There is no way to represent that with aliases being just another
name, so the best solution seems to be to just disallow it, converting
a miscompile into an error.
llvm-svn: 204781
Adds the different broadcast instructions to the ReplaceableInstrsAVX2 table.
That way the ExeDepsFix pass can take better decisions when AVX2 broadcasts are
across domain (int <-> float).
In particular, prior to this patch we were generating:
vpbroadcastd LCPI1_0(%rip), %ymm2
vpand %ymm2, %ymm0, %ymm0
vmaxps %ymm1, %ymm0, %ymm0 ## <- domain change penalty
Now, we generate the following nice sequence where everything is in the float
domain:
vbroadcastss LCPI1_0(%rip), %ymm2
vandps %ymm2, %ymm0, %ymm0
vmaxps %ymm1, %ymm0, %ymm0
<rdar://problem/16354675>
llvm-svn: 204770
The VSX instruction set has two types of FMA instructions: A-type (where the
addend is taken from the output register) and M-type (where one of the product
operands is taken from the output register). This adds a small pass that runs
just after MI scheduling (and, thus, just before register allocation) that
mutates A-type instructions (that are created during isel) into M-type
instructions when:
1. This will eliminate an otherwise-necessary copy of the addend
2. One of the product operands is killed by the instruction
The "right" moment to make this decision is in between scheduling and register
allocation, because only there do we know whether or not one of the product
operands is killed by any particular instruction. Unfortunately, this also
makes the implementation somewhat complicated, because the MIs are not in SSA
form and we need to preserve the LiveIntervals analysis.
As a simple example, if we have:
%vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
%vreg5<def,tied1> = XSMADDADP %vreg5<tied0>, %vreg17, %vreg16,
%RM<imp-use>; VSLRC:%vreg5,%vreg17,%vreg16
...
%vreg9<def,tied1> = XSMADDADP %vreg9<tied0>, %vreg17, %vreg19,
%RM<imp-use>; VSLRC:%vreg9,%vreg17,%vreg19
...
We can eliminate the copy by changing from the A-type to the
M-type instruction. This means:
%vreg5<def,tied1> = XSMADDADP %vreg5<tied0>, %vreg17, %vreg16,
%RM<imp-use>; VSLRC:%vreg5,%vreg17,%vreg16
is replaced by:
%vreg16<def,tied1> = XSMADDMDP %vreg16<tied0>, %vreg18, %vreg9,
%RM<imp-use>; VSLRC:%vreg16,%vreg18,%vreg9
and we remove: %vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
llvm-svn: 204768
This used to resort to splitting the 256-bit operation into two 128-bit
shuffles and then recombining the results.
Fixes <rdar://problem/16167303>
llvm-svn: 204735
This is a pretty straight forward translation for COFF, we just need to
stick the data in a COMDAT section marked as
IMAGE_COMDAT_SELECT_NODUPLICATES.
N.B. We must be careful to avoid sticking entities with private linkage
in COMDAT groups. COFF is pretty hostile to the renaming of entities so
we must be careful to disallow GlobalVariables with unstable names.
llvm-svn: 204703
When register allocator's stage is RS_Spill, we choose spill over using the CSR
for the first time, if the spill cost is lower than CSRCost.
When register allocator's stage is < RS_Split, we choose pre-splitting over
using the CSR for the first time, if the cost of splitting is lower than
CSRCost.
CSRCost is set with command-line option "regalloc-csr-first-time-cost". The
default value is 0 to generate the same codes as before this commit.
With a value of 15 (1 << 14 is the entry frequency), I measured performance
gain of 3% on 253.perlbmk and 1.7% on 197.parser, with instrumented PGO,
on an arm device.
rdar://16162005
llvm-svn: 204690
Try to match scalar and first like the other instructions.
Expand 64-bit ands to a pair of 32-bit ands since that is not
available on the VALU.
llvm-svn: 204660
Those patterns are used when the load cannot be folded into the related broadcast
during the select phase.
This happens when the load gets additional uses that were not anticipated during
the previous lowering phases (constant vector to constant load, then constant
load reused) or when selection DAG is not able to prove that folding the load
will not create a cycle in the DAG.
<rdar://problem/16074331>
llvm-svn: 204631
This is a pretty straight forward translation for COFF, we just need to
stick the function in a COMDAT section marked as
IMAGE_COMDAT_SELECT_NODUPLICATES.
llvm-svn: 204565
When VSX is available, these instructions should be used in preference to the
older variants that only have access to the scalar floating-point registers.
llvm-svn: 204559
This patch renames method 'isConstantSplat' as 'getConstantSplatValue'
(mainly for consistency reasons), and rewrites its logic to ensure
that we always perform a legal 'cast<ConstantSDNode>'.
Added test shift-combine-crash.ll to verify that DAGCombiner no longer crashes with an assertion failure in the attempt to simplify a vector shift by a vector of all undef counts.
llvm-svn: 204536
We make sure a spill is not hoisted to a hotter outer loop by adding
a condition. Hoist a spill to outer loop if there are multiple dependents
(it can be beneficial if more than one dependents are hoisted) or
if DepSV (the hoisting source) is hotter than SV (the hoisting destination).
rdar://16268194
llvm-svn: 204522
Previously, only regular AArch64 instructions were annotated with SchedRW lists.
This patch does the same for NEON enabling these instructions to be scheduled by
the MIScheduler. Additionally, store operations are now modeled and a few
SchedRW lists were updated for bug fixes (e.g. multiple def operands).
Reviewers: apazos, mcrosier, atrick
Patch by Dave Estes <cestes@codeaurora.org>!
llvm-svn: 204505
Summary:
VECTOR_SHUFFLE concatenates the vectors in an vectorwise fashion.
<0b00, 0b01> + <0b10, 0b11> -> <0b00, 0b01, 0b10, 0b11>
VSHF concatenates the vectors in a bitwise fashion:
<0b00, 0b01> + <0b10, 0b11> ->
0b0100 + 0b1110 -> 0b01001110
<0b10, 0b11, 0b00, 0b01>
We must therefore swap the operands to get the correct result.
The test case that discovered the issue was MultiSource/Benchmarks/nbench.
Reviewers: matheusalmeida
Reviewed By: matheusalmeida
Differential Revision: http://llvm-reviews.chandlerc.com/D3142
llvm-svn: 204480
Extend the target hook to take also the operand index into account when
calculating the cost of the constant materialization.
Related to <rdar://problem/16381500>
llvm-svn: 204435
.data_region is only used in Darwin, so it shouldn't be generated
for other OS. Currently AArch64 doesn't support darwin yet, so
I removed it from AArch64. When Darwin is supported someday, we can
add it back and associate it with Darwin.
llvm-svn: 204424
Sicne MBB->computeRegisterLivenes() returns Dead for sub regs like s0,
d0 is used in vpop instead of updating sp, which causes s0 dead before
its use.
This patch checks the liveness of each subreg to make sure the reg is
actually dead.
llvm-svn: 204411
This commit extends the coverage of the constant hoisting pass, adds additonal
debug output and updates the function names according to the style guide.
Related to <rdar://problem/16381500>
llvm-svn: 204389
The Octeon cpu from Cavium Networks is mips64r2 based and has an extended
instruction set. In order to utilize this with LLVM, a new cpu feature "octeon"
and a subtarget feature "cnmips" is added. A small set of new instructions
(baddu, dmul, pop, dpop, seq, sne) is also added. LLVM generates dmul, pop and
dpop instructions with option -mcpu=octeon or -mattr=+cnmips.
llvm-svn: 204337
It isn't actually used now, and probably never will be, plus it makes
tests less annoying. I also think SC prints GDS instructions as a
separate instruction name.
llvm-svn: 204270
The "noduplicate" function attribute exists to prevent certain optimizations
from duplicating calls to the function. This is important on platforms where
certain function call duplications are unsafe (for example execution barriers
for CUDA and OpenCL).
This patch makes it possible to specify intrinsics as "noduplicate" and
translates that to the appropriate function attribute.
llvm-svn: 204200
For functions where esi is used as base pointer, we would previously fall back
from lowering memcpy with "rep movs" because that clobbers esi.
With this patch, we just store esi in another physical register, and restore
it afterwards. This adds a little bit of register preassure, but the more
efficient memcpy should be worth it.
Differential Revision: http://llvm-reviews.chandlerc.com/D2968
llvm-svn: 204174
Summary:
SLP Vectorization of intrinsics (r203707) has exposed cases where the
expansion of vector bswap is failing (PR19151).
Reviewers: hfinkel
CC: chandlerc
Differential Revision: http://llvm-reviews.chandlerc.com/D3104
llvm-svn: 204163
When converting a signed 32-bit integer to double-precision floating point on
hardware without a lfiwax instruction, we have to instead use a lfd followed
by fcfid. We were erroneously offsetting the address by 4 bytes in
preparation for either a lfiwax or lfiwzx when generating the lfd. This fixes
that silly error.
This was not caught in the test suite since the conversion tests were run with
-mcpu=pwr7, which implies availability of lfiwax. I've added another test
case for older hardware that checks the code we expect in the absence of
lfiwax and other flavors of fcfid. There are fewer tests in this test case
because we punt to DAG selection in more cases on older hardware. (We must
generate complex fiddly sequences in those cases, and there is marginal
benefit in duplicating that logic in fast-isel.)
llvm-svn: 204155
Rather than LegalizeAction::Expand, this needs LegalizeAction::Promote to get
promoted to fp_to_sint v8f32->v8i32. This is a legal operation on AVX.
For that to work properly, we also need to teach the legalizer about the
specific promotion required here. The default vector promotion uses
bitcasting to a vector type of the same total size. We want to promote the
vector element type, effectively widening the operation and then truncating
the result. This is analogous to the current logic of how int_to_fp is
promoted.
The change also factors out some code from the int_to_fp promotion code to
ValueType::widenIntegerVectorElementType. This is now shared between
int_to_fp and fp_to_int.
There is no longer need for the custom lowering of fp_to_sint f32->v8i16 in
X86. It can now go through the new target-independent fp_to_*int promotion
logic.
I also checked that no other target uses Promote for these ops yet, so there
shouldn't be any unexpected change in behavior.
Fixes <rdar://problem/16202247>
llvm-svn: 204058
The type of the immediates should not matter as long as the encoding is
equivalent to the encoding of one of the legal inline constants.
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
llvm-svn: 204056
This instructions writes to an 32-bit SGPR. This change required adding
the 32-bit VCC_LO and VCC_HI registers, because the full VCC register
is 64 bits.
This fixes verifier errors on several of the indirect addressing piglit
tests.
Tested-by: Michel Dänzer <michel.daenzer@amd.com>
llvm-svn: 204055
- Adds support for inserting vzerouppers before tail-calls.
This is enabled implicitly by having MachineInstr::copyImplicitOps preserve
regmask operands, which allows VZeroUpperInserter to see where tail-calls use
vector registers.
- Fixes a bug that caused the previous version of this optimization to miss some
vzeroupper insertion points in loops. (Loops-with-vector-code that followed
loops-without-vector-code were mistakenly overlooked by the previous version).
- New algorithm never revisits instructions.
Fixes <rdar://problem/16228798>
llvm-svn: 204021
Commit r181723 introduced code to avoid placing initialized variables
needing relocations into the .rodata section, which avoid copy relocs
that do not work as expected on ppc64 function references.
The same treatment is also needed for *named* .rodata.XXX sections.
This patch changes PPC64LinuxTargetObjectFile::SelectSectionForGlobal
to modify "Kind" *before* calling the default SelectSectionForGlobal
routine, instead of first calling the default routine and then just
checking for the (main) .rodata section afterwards.
llvm-svn: 203921
These linkages were introduced some time ago, but it was never very
clear what exactly their semantics were or what they should be used
for. Some investigation found these uses:
* utf-16 strings in clang.
* non-unnamed_addr strings produced by the sanitizers.
It turns out they were just working around a more fundamental problem.
For some sections a MachO linker needs a symbol in order to split the
section into atoms, and llvm had no idea that was the case. I fixed
that in r201700 and it is now safe to use the private linkage. When
the object ends up in a section that requires symbols, llvm will use a
'l' prefix instead of a 'L' prefix and things just work.
With that, these linkages were already dead, but there was a potential
future user in the objc metadata information. I am still looking at
CGObjcMac.cpp, but at this point I am convinced that linker_private
and linker_private_weak are not what they need.
The objc uses are currently split in
* Regular symbols (no '\01' prefix). LLVM already directly provides
whatever semantics they need.
* Uses of a private name (start with "\01L" or "\01l") and private
linkage. We can drop the "\01L" and "\01l" prefixes as soon as llvm
agrees with clang on L being ok or not for a given section. I have two
patches in code review for this.
* Uses of private name and weak linkage.
The last case is the one that one could think would fit one of these
linkages. That is not the case. The semantics are
* the linker will merge these symbol by *name*.
* the linker will hide them in the final DSO.
Given that the merging is done by name, any of the private (or
internal) linkages would be a bad match. They allow llvm to rename the
symbols, and that is really not what we want. From the llvm point of
view, these objects should really be (linkonce|weak)(_odr)?.
For now, just keeping the "\01l" prefix is probably the best for these
symbols. If we one day want to have a more direct support in llvm,
IMHO what we should add is not a linkage, it is just a hidden_symbol
attribute. It would be applicable to multiple linkages. For example,
on weak it would produce the current behavior we have for objc
metadata. On internal, it would be equivalent to private (and we
should then remove private).
llvm-svn: 203866
This patch fixes the bug in peephole optimization that folds a load which defines one vreg into the one and only use of that vreg. With debug info, a DBG_VALUE that referenced the vreg considered to be a use, preventing the optimization. The fix is to ignore DBG_VALUE's during the optimization, and undef a DBG_VALUE that references a vreg that gets removed.
Patch by Trevor Smigiel!
llvm-svn: 203829
LDS instructions are pseudo instructions which model
the OQAP defs and uses within a single instruction.
This fixes a hang in the opencv MedianFilter tests.
llvm-svn: 203818
This option enables LowerInvoke's obsolete SJLJ EH support, but the
target used in this test (ARM Darwin) no longer uses the LowerInvoke
pass, so the option has no effect here. This target currently uses
the newer SjLjEHPrepare pass instead.
This cleanup will help with removing "-enable-correct-eh-support".
Differential Revision: http://llvm-reviews.chandlerc.com/D3064
llvm-svn: 203810
This is a follow-up to r203635. Saleem pointed out that since symbolic register
names are much easier to read, it would be good if we could turn them off only
when we really need to because we're using an external assembler.
Differential Revision: http://llvm-reviews.chandlerc.com/D3056
llvm-svn: 203806
Tim Northover