Currently, if you use a MultiArg<> option, then printing out the help/usage
message will cause an assert. This fixes getOptionHelpName() to work with
MultiArg Options.
llvm-svn: 215770
We were setting the comdat when functions were copied in the initial pass, but
not when they were linked only when we found out that they are needed.
llvm-svn: 215765
The floating-point value positive zero (+0.0) is a valid immedate value
according to isFPImmLegal. As a result AArch64 FastISel went ahead and
used the immediate version of fmov to materialize the constant.
The problem is that the immediate version of fmov cannot encode an imediate for
postive zero. Instead a fmov from the zero register was supposed to be used in
this case.
This fix adds handling for this special case and uses fmov from the zero
register to materialize a positive zero (negative zeroes go to the constant
pool).
There is no test case for this, because this code is currently dead. It will be
enabled in a future commit and I will add a test case in a separate commit
after that.
This fixes <rdar://problem/18027157>.
llvm-svn: 215753
Note: This reapplies r215582 without any modifications. The refactoring wasn't
responsible for the buildbot failures.
Original commit message:
Cleanup and prepare constant materialization code for future commits.
llvm-svn: 215752
MSVC gives this awesome diagnostic:
..\lib\Target\X86\X86ISelLowering.cpp(7085) : error C2971: 'llvm::VariadicFunction1' : template parameter 'Func' : 'isShuffleEquivalentImpl' : a local variable cannot be used as a non-type argument
..\include\llvm/ADT/VariadicFunction.h(153) : see declaration of 'llvm::VariadicFunction1'
..\lib\Target\X86\X86ISelLowering.cpp(7061) : see declaration of 'isShuffleEquivalentImpl'
Using an anonymous namespace makes the problem go away.
llvm-svn: 215744
Ordinarily (shl (add x, c1), c2) -> (add (shl x, c2), c1 << c2)
is only done if the add has one use. If the resulting constant
add can be folded into an addressing mode, force this to happen
for the pointer operand.
This ends up happening a lot because of how LDS objects are allocated.
Since the globals are allocated next to each other, acessing the first
element of the second object is directly indexed by a shifted pointer.
llvm-svn: 215739
As Jim pointed out this assert isn't really needed to test for correctness,
because the code right afterwards does the same check and falls-back to
SelectionDAG - as intended.
llvm-svn: 215735
The default assumes that a 16-bit signed offset is used.
LDS instruction use a 16-bit unsigned offset, so it wasn't
being used in some cases where it was assumed a negative offset
could be used.
More should be done here, but first isLegalAddressingMode needs
to gain an addressing mode argument. For now, copy most of the rest
of the default implementation with the immediate offset change.
llvm-svn: 215732
In a previous iteration of the pass, we would try to compensate for
writeback by updating later instructions and/or inserting a SUBS to
reset the base register if necessary.
Since such a SUBS sets the condition flags it's not generally safe to do
this. For now, only merge LDR/STRs if there is no writeback to the base
register (LDM that loads into the base register) or the base register is
killed by one of the merged instructions. These cases are clear wins
both in terms of instruction count and performance.
Also add three new test cases, and update the existing ones accordingly.
llvm-svn: 215729
This adds some code back that was deleted in r92053. The location of the
last merged memory operation needs to be kept up-to-date since MemOps
may be in a different order to the original instruction stream to
allow merging (since registers need to be in ascending order). Also
simplify the logic to determine BaseKill using findRegisterUseOperandIdx
to use an equivalent function call instead.
llvm-svn: 215728
We actually need to return the register into which we materialized the constant
and not just "true" for success. This code is currently partially dead, that is
why it didn't trigger any failures yet. Once I change the order of the constant
materialization this code will be fully exercised.
llvm-svn: 215727
the new shuffle lowering and an implementation for v4 shuffles.
This allows us to handle non-half-crossing shuffles directly for v4
shuffles, both integer and floating point. This currently misses places
where we could perform the blend via UNPCK instructions, but otherwise
generates equally good or better code for the test cases included to the
existing vector shuffle lowering. There are a few cases that are
entertainingly better. ;]
llvm-svn: 215702
BLENDPS, BLENDPD, and PBLENDW instructions into pretty shuffle comments.
These will be used in my next commit as part of test cases for AVX
shuffles which can directly use blend in more places.
llvm-svn: 215701
These are system-only instructions for CPUs with virtualization
extensions, allowing a hypervisor easy access to all of the various
different AArch32 registers.
rdar://problem/17861345
llvm-svn: 215700
target-specific shuffl DAG combines.
We were recognizing the paired shuffles backwards. This code needs to be
replaced anyways as we have the same functionality elsewhere, but I'll
do the refactoring in a follow-up, this is the minimal fix to the
behavior.
In addition to fixing miscompiles with the new vector shuffle lowering,
it also causes the canonicalization to kick in much better, selecting
the smaller encoding variants in lots of places in the new AVX path.
This still isn't quite ideal as we don't need both the shufpd and the
punpck instructions, but that'll get fixed in a follow-up patch.
llvm-svn: 215690
broken logic for merging shuffle masks in the face of SM_SentinelZero
mask operands.
While these are '-1' they don't mean 'undef' the way '-1' means in the
pre-legalized shuffle masks. Instead, they mean that the shuffle
operation is forcibly zeroing that lane. Reflect this and explicitly
handle it in a bunch of places. In one place the effect is equivalent
but much more clear. In the rest it was really weirdly broken.
Also, rewrite the entire merging thing to be a more directy operation
with a single loop and just doing math to map the indices through the
various masks.
Also add a bunch of asserts to try to make in extremely clear what the
different masks can possibly look like.
Finally, add some comments to clarify that we're merging shuffle masks
*up* here rather than *down* as we do everywhere else, and thus the
logic is quite confusing.
Thanks to several different people for sending test cases, and for
Robert Khasanov for an initial attempt at fixing.
llvm-svn: 215687
The LDinto_toc pattern has been part of 64-bit PowerPC for a long
time, and represents loading from a memory location into the TOC
register (X2). However, this pattern doesn't explicitly record that
it modifies that register. This patch adds the missing dependency.
It was very surprising to me that this has never shown up as a problem
in the past, and that we only saw this problem recently in a single
scenario when building a self-hosted clang. It turns out that in most
cases we have another dependency present that keeps the LDinto_toc
instruction tied in place. LDinto_toc is used for TOC restore
following a call site, so this is a typical sequence:
BCTRL8 <regmask>, %CTR8<imp-use>, %RM<imp-use>, %X3<imp-use>, %X12<imp-use>, %X1<imp-def>, ...
LDinto_toc 24, %X1
ADJCALLSTACKUP 96, 0, %R1<imp-def>, %R1<imp-use>
Because the LDinto_toc is inserted prior to the ADJCALLSTACKUP, there
is a natural anti-dependency between the two that keeps it in place.
Therefore we don't usually see a problem. However, in one particular
case, one call is followed immediately by another call, and the second
call requires a parameter that is a TOC-relative address. This is the
code sequence:
BCTRL8 <regmask>, %CTR8<imp-use>, %RM<imp-use>, %X3<imp-use>, %X4<imp-use>, %X5<imp-use>, %X12<imp-use>, %X1<imp-def>, ...
LDinto_toc 24, %X1
ADJCALLSTACKUP 96, 0, %R1<imp-def>, %R1<imp-use>
ADJCALLSTACKDOWN 96, %R1<imp-def>, %R1<imp-use>
%vreg39<def> = ADDIStocHA %X2, <ga:@.str>; G8RC_and_G8RC_NOX0:%vreg39
%vreg40<def> = ADDItocL %vreg39<kill>, <ga:@.str>; G8RC:%vreg40 G8RC_and_G8RC_NOX0:%vreg39
Note that the back-to-back stack adjustments are the same size! The
back end is smart enough to recognize this and optimize them away:
BCTRL8 <regmask>, %CTR8<imp-use>, %RM<imp-use>, %X3<imp-use>, %X4<imp-use>, %X5<imp-use>, %X12<imp-use>, %X1<imp-def>, ...
LDinto_toc 24, %X1
%vreg39<def> = ADDIStocHA %X2, <ga:@.str>; G8RC_and_G8RC_NOX0:%vreg39
%vreg40<def> = ADDItocL %vreg39<kill>, <ga:@.str>; G8RC:%vreg40 G8RC_and_G8RC_NOX0:%vreg39
Now there is nothing to prevent the ADDIStocHA instruction from moving
ahead of the LDinto_toc instruction, and because of the longest-path
heuristic, this is what happens.
With the accompanying patch, %X2 is represented as an implicit def:
BCTRL8 <regmask>, %CTR8<imp-use>, %RM<imp-use>, %X3<imp-use>, %X4<imp-use>, %X5<imp-use>, %X12<imp-use>, %X1<imp-def>, ...
LDinto_toc 24, %X1, %X2<imp-def,dead>
ADJCALLSTACKUP 96, 0, %R1<imp-def,dead>, %R1<imp-use>
ADJCALLSTACKDOWN 96, %R1<imp-def,dead>, %R1<imp-use>
%vreg39<def> = ADDIStocHA %X2, <ga:@.str>; G8RC_and_G8RC_NOX0:%vreg39
%vreg40<def> = ADDItocL %vreg39<kill>, <ga:@.str>; G8RC:%vreg40 G8RC_and_G8RC_NOX0:%vreg39
So now when the two stack adjustments are removed, ADDIStocHA is
prevented from being moved above LDinto_toc.
I have not yet created a test case for this, because the original
failure occurs on a relatively large function that needs reduction.
However, this is a fairly serious bug, despite its infrequency, and I
wanted to get this patch onto the list as soon as possible so that it
can be considered for a 3.5 backport. I'll work on whittling down a
test case.
Have we missed the boat for 3.5 at this point?
Thanks,
Bill
llvm-svn: 215685
FastEmit_i won't always succeed to materialize an i32 constant and just fail.
This would trigger a fall-back to SelectionDAG, which is really not necessary.
This fix will first fall-back to a constant pool load to materialize the constant
before giving up for good.
This fixes <rdar://problem/18022633>.
llvm-svn: 215682
When a call site with noalias metadata is inlined, that metadata can be
propagated directly to the inlined instructions (only those that might access
memory because it is not useful on the others). Prior to inlining, the noalias
metadata could express that a call would not alias with some other memory
access, which implies that no instruction within that called function would
alias. By propagating the metadata to the inlined instructions, we preserve
that knowledge.
This should complete the enhancements requested in PR20500.
llvm-svn: 215676
This reverts:
r215595 "[FastISel][X86] Add large code model support for materializing floating-point constants."
r215594 "[FastISel][X86] Use XOR to materialize the "0" value."
r215593 "[FastISel][X86] Emit more efficient instructions for integer constant materialization."
r215591 "[FastISel][AArch64] Make use of the zero register when possible."
r215588 "[FastISel] Let the target decide first if it wants to materialize a constant."
r215582 "[FastISel][AArch64] Cleanup constant materialization code. NFCI."
llvm-svn: 215673
No functional change. This will be used by the new FMA intrinsic lowering
code.
We can probably add NO_EXC here as well, I am just not too familiar with this
part of AVX512 yet. We can add that later.
llvm-svn: 215662
This change further evolves the base class AVX512_masking in order to make it
suitable for the masking variants of the FMA instructions.
Besides AVX512_masking there is now a new base class that instructions
including FMAs can use: AVX512_masking_3src. With three-source (destructive)
instructions one of the sources is already tied to the destination. This
difference from AVX512_masking is captured by this new class. The common bits
between _masking and _masking_3src are broken out into a new super class
called AVX512_masking_common.
As with valign, there is some corresponding restructuring of the underlying
format classes. The idea is the same we want to derive from two classes
essentially: one providing the format bits and another format-independent
multiclass supplying the various masking and non-masking instruction variants.
Existing fma tests in avx512-fma*.ll provide coverage here for the non-masking
variants. For masking, the next patches in the series will add intrinsics and
intrinsic tests.
For AVX512_masking_3src to work, the (ins ...) dag has to be passed *without*
the leading source operand that is tied to dst ($src1). This is necessary to
properly construct the (ins ...) for the different variants. For the record,
I did check that if $src is mistakenly included, you do get a fairly intuitive
error message from the tablegen backend.
Part of <rdar://problem/17688758>
llvm-svn: 215660
When preserving noalias function parameter attributes by adding noalias
metadata in the inliner, we should do this for general function calls (not just
memory intrinsics). The logic is very similar to what already existed (except
that we want to add this metadata even for functions taking no relevant
parameters). This metadata can be used by ModRef queries in the caller after
inlining.
This addresses the first part of PR20500. Adding noalias metadata during
inlining is still turned off by default.
llvm-svn: 215657
This patch allows a vector fneg of a bitcasted integer value to be optimized in the same way that we already optimize a scalar fneg. If the integer variable is a constant, we can precompute the result and not require any logic ops.
This patch is very similar to a fabs patch committed at r214892.
Differential Revision: http://reviews.llvm.org/D4852
llvm-svn: 215646
Summary:
This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected.
Contains work done by Matheus Almeida.
Reviewers: matheusalmeida, dsanders
Reviewed By: dsanders
Subscribers: tomatabacu
Differential Revision: http://reviews.llvm.org/D4227
llvm-svn: 215640
lowering scheme.
Currently, this just directly bails to the fallback path of splitting
the 256-bit vector into two 128-bit vectors, operating there, and then
joining the results back together. While the results are far from
perfect, they are *shockingly* good for what we're doing here. I'll be
layering the rest of the functionality on top of this piece by piece and
updating tests as I go.
Note that 256-bit vectors in this mode are still somewhat WIP. While
I think the code paths that I'm adding here are clean and good-to-go,
there are still a lot of 128-bit assumptions that I'll need to stomp out
as I march through the functional spread here.
llvm-svn: 215637
Summary:
This pseudo-instruction allows the programmer to load an address from a symbolic expression into a register.
Patch by David Chisnall.
His work was sponsored by: DARPA, AFRL
I've made some minor changes to the original, such as improving the formatting and adding some comments, and I've also added a test case.
Reviewers: dsanders
Reviewed By: dsanders
Differential Revision: http://reviews.llvm.org/D4808
llvm-svn: 215630
Summary:
getCanHaveModuleDir() is renamed to isModuleDirectiveAllowed(), and
setCanHaveModuleDir() is renamed to forbidModuleDirective() since it is only
ever given a false argument.
Reviewers: vmedic
Reviewed By: vmedic
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D4885
llvm-svn: 215628
input node after manually adding it to the worklist and using CombineTo.
Once we use CombineTo the input node may have been deleted. Despite this
being *completely confusing* and somewhat broken, the only way to
"correctly" return from a DAG combine after potentially deleting the
input node is to return *that exact node*....
But really, this code should just never have used CombineTo. It won't do
what it wants (returning the node as mentioned above just causes the
combine to infloop). The correct way to combine away a casted load to
a load of the correct type is to RAUW the chain directly and then return
the loaded value to replace the actual value node.
I managed to find this with the vector shuffle fuzzer even though it
clearly has nothing at all to do with vector shuffles and rather those
happen to trigger a load of a constant pool that hits this combine *just
right*. I've included the test as it is small and a nice stress test
that the infrastructure isn't asserting.
llvm-svn: 215622
As X86MCAsmInfoDarwin uses '##' as CommentString although a single '#' starts a
comment a workaround for this special case is added.
Fixes divisions in constant expressions for the AArch64 assembler and other
targets which use '//' as CommentString.
Patch by Janne Grunau!
llvm-svn: 215615
combining by replacing it with something else but not re-process the
node afterward to remove it.
In a truly remarkable stroke of bad luck, this would (in the test case
attached) end up getting some other node combined into it without ever
getting re-processed. By adding it back on to the worklist, in addition
to deleting the dead nodes more quickly we also ensure that if it
*stops* being dead for any reason it makes it back through the
legalizer. Without this, the test case will end up failing during
instruction selection due to an and node with a type we don't have an
instruction pattern for.
It took many million runs of the shuffle fuzz tester to find this.
llvm-svn: 215611
Certain functions such as objc_autoreleaseReturnValue have to be called as
tail-calls even at -O0. Since normal fast-isel doesn't emit calls as tail calls,
we have to fall back to SelectionDAG to select calls that are marked as tail.
<rdar://problem/17991614>
llvm-svn: 215600
FastISel didn't take much advantage of the different addressing modes available
to it on AArch64. This commit allows the ComputeAddress method to recognize more
addressing modes that allows shifts and sign-/zero-extensions to be folded into
the memory operation itself.
For Example:
lsl x1, x1, #3 --> ldr x0, [x0, x1, lsl #3]
ldr x0, [x0, x1]
sxtw x1, w1
lsl x1, x1, #3 --> ldr x0, [x0, x1, sxtw #3]
ldr x0, [x0, x1]
llvm-svn: 215597
In the large code model for X86 floating-point constants are placed in the
constant pool and materialized by loading from it. Since the constant pool
could be far away, a PC relative load might not work. Therefore we first
materialize the address of the constant pool with a movabsq and then load
from there the floating-point value.
Fixes <rdar://problem/17674628>.
llvm-svn: 215595
This mostly affects the i64 value type, which always resulted in an 15byte
mobavsq instruction to materialize any constant. The custom code checks the
value of the immediate and tries to use a different and smaller mov
instruction when possible.
This fixes <rdar://problem/17420988>.
llvm-svn: 215593
This change materializes now the value "0" from the zero register.
The zero register can be folded by several instruction, so no
materialization is need at all.
Fixes <rdar://problem/17924413>.
llvm-svn: 215591
This changes the order in which FastISel tries to materialize a constant.
Originally it would try to use a simple target-independent approach, which
can lead to the generation of inefficient code.
On X86 this would result in the use of movabsq to materialize any 64bit
integer constant - even for simple and small values such as 0 and 1. Also
some very funny floating-point materialization could be observed too.
On AArch64 it would materialize the constant 0 in a register even the
architecture has an actual "zero" register.
On ARM it would generate unnecessary mov instructions or not use mvn.
This change simply changes the order and always asks the target first if it
likes to materialize the constant. This doesn't fix all the issues
mentioned above, but it enables the targets to implement such
optimizations.
Related to <rdar://problem/17420988>.
llvm-svn: 215588
This is a cleaner solution to the problem described in r215431.
When instructions are combined a dangling DBG_VALUE is removed.
This resolves bug 20598.
llvm-svn: 215587
Split the constant materialization code into three separate helper functions for
Integer-, Floating-Point-, and GlobalValue-Constants.
llvm-svn: 215586
This change is also in preparation for a future change to make sure that
the constant materialization uses MOVT/MOVW when available and not a load
from the constant pool.
llvm-svn: 215584
getRegClassFor returns the incorrect register class when in Thumb2 mode.
This fix simply manually selects the register class as in the code just a few
lines above.
There is no test case for this code, because the code is currently
unreachable. This will be changed in a future commit and existing test
cases will exercise this code.
llvm-svn: 215583
New function to erase a machine instruction and mark DBG_VALUE
for removal. A DBG_VALUE is marked for removal when it references
an operand defined in the instruction.
Use the new function to cleanup code in dead machine instruction
removal pass.
llvm-svn: 215580
critical edge has been split. The MachineDominatorTree will when lazy update the
underlying dominance properties when require.
** Context **
This is a follow-up of r215410.
Each time a critical edge is split this invalidates the dominator tree
information. Thus, subsequent queries of that interface will be slow until the
underlying information is actually recomputed (costly).
** Problem **
Prior to this patch, splitting a critical edge needed to query the dominator
tree to update the dominator information.
Therefore, splitting a bunch of critical edges will likely produce poor
performance as each query to the dominator tree will use the slow query path.
This happens a lot in passes like MachineSink and PHIElimination.
** Proposed Solution **
Splitting a critical edge is a local modification of the CFG. Moreover, as soon
as a critical edge is split, it is not critical anymore and thus cannot be a
candidate for critical edge splitting anymore. In other words, the predecessor
and successor of a basic block inserted on a critical edge cannot be inserted by
critical edge splitting.
Using these observations, we can pile up the splitting of critical edge and
apply then at once before updating the DT information.
The core of this patch moves the update of the MachineDominatorTree information
from MachineBasicBlock::SplitCriticalEdge to a lazy MachineDominatorTree.
** Performance **
Thanks to this patch, the motivating example compiles in 4- minutes instead of
6+ minutes. No test case added as the motivating example as nothing special but
being huge!
The binaries are strictly identical for all the llvm test-suite + SPECs with and
without this patch for both Os and O3.
Regarding compile time, I observed only noise, although on average I saw a
small improvement.
<rdar://problem/17894619>
llvm-svn: 215576
v2: continue iterating through the rest of the bb
use for loop
v3: initialize FlattenCFG pass in ScalarOps
add test
v4: split off initializing flattencfg to a separate patch
add comment
Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 215574
This for some reason fixes v1i64 kernel arguments on pre-SI. This
currently breaks some other cases in the kernel-args.ll test for R600,
but I'm not particularly confident in the new output. VTX_READ_* are not
used for some of the scalarized cases, and the code reading from the
constant buffer doesn't make much sense to me.
llvm-svn: 215564
Add header guards to files that were missing guards. Remove #endif comments
as they don't seem common in LLVM (we can easily add them back if we decide
they're useful)
Changes made by clang-tidy with minor tweaks.
llvm-svn: 215558
This patch improves the existing algorithm in DAGCombiner that
attempts to fold shuffles according to rule:
shuffle(shuffle(x, y, M1), undef, M2) -> shuffle(y, undef, M3)
Before this change, there were cases where the DAGCombiner conservatively
avoided folding shuffles even if the resulting mask would have been legal.
That is because the algorithm wrongly assumed that commuting
an illegal shuffle mask would always produce an illegal mask.
With this change, we now correctly compute the commuted shuffle mask before
calling method 'isShuffleMaskLegal' on it.
On X86, this improves for example the codegen for the following function:
define <4 x i32> @test(<4 x i32> %A, <4 x i32> %B) {
%1 = shufflevector <4 x i32> %B, <4 x i32> %A, <4 x i32> <i32 1, i32 2, i32 6, i32 7>
%2 = shufflevector <4 x i32> %1, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
ret <4 x i32> %2
}
Before this change the X86 backend (-mcpu=corei7) generated
the following assembly code for function @test:
shufps $-23, %xmm0, %xmm1 # xmm1 = xmm1[1,2],xmm0[2,3]
movhlps %xmm1, %xmm1 # xmm1 = xmm1[1,1]
movaps %xmm1, %xmm0
Now we produce:
movhlps %xmm0, %xmm0 # xmm0 = xmm0[1,1]
Added extra test cases in combine-vec-shuffle-2.ll to verify that we correctly
fold according to the above-mentioned rule.
llvm-svn: 215555
Summary:
Moved some calls to setCanHaveModuleDir to the MipsTargetStreamer base class and removed the resulting empty functions from the MipsTargetELFStreamer class.
Also fixed a missing call to setCanHaveModuleDir in MipsTargetELFStreamer::emitDirectiveSetMicroMips.
Reviewers: dsanders
Reviewed By: dsanders
Subscribers: tomatabacu
Differential Revision: http://reviews.llvm.org/D4781
llvm-svn: 215542
Especially with blends and large tree heights there was a problem with
the fuzzer where it would end up with enough undef shuffle elements in
enough parts of the tree that in a birthday-attack kind of way we ended
up regularly having large numbers of undef elements in the result. I was
seeing reasonably frequent cases of *all* results being undef which
prevents us from doing any correctness checking at all. While having
undef lanes is important, this was too much.
So I've tried to apply some math to the probabilities of having an undef
lane and balance them against the tree height. Please be gentle, I'm
really terrible at math. I probably made a bunch of amateur mistakes
here. Fixes, etc. are quite welcome. =D At least in running it some, it
seems to be producing more interesting (for correctness testing)
results.
llvm-svn: 215540
Nick Kledzik