When dealing with complex<float>, and similar structures with two
single-precision floating-point numbers, especially when such things are being
passed around by value, we'll sometimes end up loading both float values by
extracting them from one 64-bit integer load. It looks like this:
t13: i64,ch = load<LD8[%ref.tmp]> t0, t6, undef:i64
t16: i64 = srl t13, Constant:i32<32>
t17: i32 = truncate t16
t18: f32 = bitcast t17
t19: i32 = truncate t13
t20: f32 = bitcast t19
The problem, especially before the P8 where those bitcasts aren't legal (and
get expanded via the stack), is that it would have been better to use two
floating-point loads directly. Here we add a target-specific DAGCombine to do
just that. In short, we turn:
ld 3, 0(5)
stw 3, -8(1)
rldicl 3, 3, 32, 32
stw 3, -4(1)
lfs 3, -4(1)
lfs 0, -8(1)
into:
lfs 3, 4(5)
lfs 0, 0(5)
llvm-svn: 264988
The test case was defining and using a function 'notExported()', but
the FileCheck checks were checking for the name 'not_exported'. This
changes the test to use 'notExported' across the board. Also, the test
defined a function 'not_defined()', but doesn't have any checks related
to it. For consistency, this name is changed to 'notDefined'. A later
commit will add checks for 'notDefined'.
Patch by Warren Ristow!
llvm-svn: 264984
Summary:
As discussed on llvm-dev[1].
This change adds the basic boilerplate code around having this intrinsic
in LLVM:
- Changes in Intrinsics.td, and the IR Verifier
- A lowering pass to lower @llvm.experimental.guard to normal
control flow
- Inliner support
[1]: http://lists.llvm.org/pipermail/llvm-dev/2016-February/095523.html
Reviewers: reames, atrick, chandlerc, rnk, JosephTremoulet, echristo
Subscribers: mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D18527
llvm-svn: 264976
The size savings are significant, and from what I can tell, both ICC and GCC do this.
Differential Revision: http://reviews.llvm.org/D18573
llvm-svn: 264966
r264884 introduced a helper to escape the backslashes in the source file
path, but I since discovered an existing mechanism to escape strings.
llvm-svn: 264936
Commit r260791 contained an error in that it would introduce a cross-module
reference in the old module. It also introduced O(N^2) complexity in the
module cloner by requiring the entire module to be visited for each function.
Fix both of these problems by avoiding use of the CloneDebugInfoMetadata
function (which is only designed to do intra-module cloning) and cloning
function-attached metadata in the same way that we clone all other metadata.
Differential Revision: http://reviews.llvm.org/D18583
llvm-svn: 264935
For the same reason as the corresponding load change.
Note that ExpandStore is completely broken for non-byte sized element
vector stores, but preserve the current broken behavior which has tests
for it. The behavior should be the same, but now introduces a new typed
store that is incorrectly split later rather than doing it directly.
llvm-svn: 264928
On AMDGPU we want to be able to promote i64/f64 loads to v2i32.
If the access is unaligned, this would conclude that since i64 is legal,
it would convert it back to i64 and there is an endless legalization
loop.
Extract the logic for scalarizing the load into a new TargetLowering
function, where this can also replace the custom function AMDGPU
has for this.
llvm-svn: 264927
Widening a PHI requires us to insert a trunc.
The logical place for this trunc is in the same BB as the PHI.
This is not possible if the BB is terminated by a catchswitch.
This fixes PR27133.
llvm-svn: 264926
Fix for issue introduced D17297, where we were breaking early from the loop detecting consecutive loads which could leave us thinking a consecutive load with zeros was possible.
llvm-svn: 264922
Summary:
Currently it's a module pass. Make it a function pass so that we can
move it to PassManagerBuilder's EP_EarlyAsPossible extension point,
which only accepts function passes.
Reviewers: rnk
Subscribers: tra, llvm-commits, jholewinski
Differential Revision: http://reviews.llvm.org/D18615
llvm-svn: 264919
Summary:
This gives callers flexibility to pass lambdas with captures, which lets
callers avoid the C-style void*-ptr closure style. (Currently, callers
in clang store state in the PassManagerBuilderBase arg.)
No functional change, and the new API is backwards-compatible.
Reviewers: chandlerc
Subscribers: joker.eph, cfe-commits
Differential Revision: http://reviews.llvm.org/D18613
llvm-svn: 264918
The TailDup transform was removed in r138841 in 2011, along with most
of the tests for it. This test, however, was missed. Probably because
it had already been XFAIL'd for 3 years at that point (since r52243!)
and continued to fail when the opt flag for -tailduplicate stopped
being valid.
llvm-svn: 264916
There is code under review that requires StringMap to have a copy constructor,
and this makes StringMap more consistent with our other containers (like
DenseMap) that have copy constructors.
Differential Revision: http://reviews.llvm.org/D18506
llvm-svn: 264906
This change prevents the loop vectorizer from vectorizing when all of the vector
types it generates will be scalarized. I've run into this problem on the PPC's QPX
vector ISA, which only holds floating-point vector types. The loop vectorizer
will, however, happily vectorize loops with purely integer computation. Here's
an example:
LV: The Smallest and Widest types: 32 / 32 bits.
LV: The Widest register is: 256 bits.
LV: Found an estimated cost of 0 for VF 1 For instruction: %indvars.iv25 = phi i64 [ 0, %entry ], [ %indvars.iv.next26, %for.body ]
LV: Found an estimated cost of 0 for VF 1 For instruction: %arrayidx = getelementptr inbounds [1600 x i32], [1600 x i32]* %a, i64 0, i64 %indvars.iv25
LV: Found an estimated cost of 0 for VF 1 For instruction: %2 = trunc i64 %indvars.iv25 to i32
LV: Found an estimated cost of 1 for VF 1 For instruction: store i32 %2, i32* %arrayidx, align 4
LV: Found an estimated cost of 1 for VF 1 For instruction: %indvars.iv.next26 = add nuw nsw i64 %indvars.iv25, 1
LV: Found an estimated cost of 1 for VF 1 For instruction: %exitcond27 = icmp eq i64 %indvars.iv.next26, 1600
LV: Found an estimated cost of 0 for VF 1 For instruction: br i1 %exitcond27, label %for.cond.cleanup, label %for.body
LV: Scalar loop costs: 3.
LV: Found an estimated cost of 0 for VF 2 For instruction: %indvars.iv25 = phi i64 [ 0, %entry ], [ %indvars.iv.next26, %for.body ]
LV: Found an estimated cost of 0 for VF 2 For instruction: %arrayidx = getelementptr inbounds [1600 x i32], [1600 x i32]* %a, i64 0, i64 %indvars.iv25
LV: Found an estimated cost of 0 for VF 2 For instruction: %2 = trunc i64 %indvars.iv25 to i32
LV: Found an estimated cost of 2 for VF 2 For instruction: store i32 %2, i32* %arrayidx, align 4
LV: Found an estimated cost of 1 for VF 2 For instruction: %indvars.iv.next26 = add nuw nsw i64 %indvars.iv25, 1
LV: Found an estimated cost of 1 for VF 2 For instruction: %exitcond27 = icmp eq i64 %indvars.iv.next26, 1600
LV: Found an estimated cost of 0 for VF 2 For instruction: br i1 %exitcond27, label %for.cond.cleanup, label %for.body
LV: Vector loop of width 2 costs: 2.
LV: Found an estimated cost of 0 for VF 4 For instruction: %indvars.iv25 = phi i64 [ 0, %entry ], [ %indvars.iv.next26, %for.body ]
LV: Found an estimated cost of 0 for VF 4 For instruction: %arrayidx = getelementptr inbounds [1600 x i32], [1600 x i32]* %a, i64 0, i64 %indvars.iv25
LV: Found an estimated cost of 0 for VF 4 For instruction: %2 = trunc i64 %indvars.iv25 to i32
LV: Found an estimated cost of 4 for VF 4 For instruction: store i32 %2, i32* %arrayidx, align 4
LV: Found an estimated cost of 1 for VF 4 For instruction: %indvars.iv.next26 = add nuw nsw i64 %indvars.iv25, 1
LV: Found an estimated cost of 1 for VF 4 For instruction: %exitcond27 = icmp eq i64 %indvars.iv.next26, 1600
LV: Found an estimated cost of 0 for VF 4 For instruction: br i1 %exitcond27, label %for.cond.cleanup, label %for.body
LV: Vector loop of width 4 costs: 1.
...
LV: Selecting VF: 8.
LV: The target has 32 registers
LV(REG): Calculating max register usage:
LV(REG): At #0 Interval # 0
LV(REG): At #1 Interval # 1
LV(REG): At #2 Interval # 2
LV(REG): At #4 Interval # 1
LV(REG): At #5 Interval # 1
LV(REG): VF = 8
The problem is that the cost model here is not wrong, exactly. Since all of
these operations are scalarized, their cost (aside from the uniform ones) are
indeed VF*(scalar cost), just as the model suggests. In fact, the larger the VF
picked, the lower the relative overhead from the loop itself (and the
induction-variable update and check), and so in a sense, picking the largest VF
here is the right thing to do.
The problem is that vectorizing like this, where all of the vectors will be
scalarized in the backend, isn't really vectorizing, but rather interleaving.
By itself, this would be okay, but then the vectorizer itself also interleaves,
and that's where the problem manifests itself. There's aren't actually enough
scalar registers to support the normal interleave factor multiplied by a factor
of VF (8 in this example). In other words, the problem with this is that our
register-pressure heuristic does not account for scalarization.
While we might want to improve our register-pressure heuristic, I don't think
this is the right motivating case for that work. Here we have a more-basic
problem: The job of the vectorizer is to vectorize things (interleaving aside),
and if the IR it generates won't generate any actual vector code, then
something is wrong. Thus, if every type looks like it will be scalarized (i.e.
will be split into VF or more parts), then don't consider that VF.
This is not a problem specific to PPC/QPX, however. The problem comes up under
SSE on x86 too, and as such, this change fixes PR26837 too. I've added Sanjay's
reduced test case from PR26837 to this commit.
Differential Revision: http://reviews.llvm.org/D18537
llvm-svn: 264904
PGOFuncNames are used as the key to retrieve the Function definition from the
MD5 stored in the profile. For internal linkage function, we prefix the source
file name to the PGOFuncNames. LTO's internalization privatizes many global linkage
symbols. This happens after value profile annotation, but those internal
linkage functions should not have a source prefix. To differentiate compiler
generated internal symbols from original ones, PGOFuncName meta data are
created and attached to the original internal symbols in the value profile
annotation step. If a symbol does not have the meta data, its original linkage
must be non-internal.
Also add a new map that maps PGOFuncName's MD5 value to the function definition.
Differential Revision: http://reviews.llvm.org/D17895
llvm-svn: 264902
What we are really trying to do here is to figure out if we are using
the 2015 STL. Unfortunately, so far as I know the MSVC STL does not
define a version macro that we can check directly. Instead I wrote a
check to see if char16_t works.
llvm-svn: 264881
Using ArrayRef in annotateValueSite's parameter instead of using an array
and it's size.
Differential Revision: http://reviews.llvm.org/D18568
llvm-svn: 264879
Summary:
This results in higher register usage, but should make it easier for
the compiler to hide latency.
This pass is a prerequisite for some more scheduler improvements, and I
think the increase register usage with this patch is acceptable, because
when combined with the scheduler improvements, the total register usage
will decrease.
shader-db stats:
2382 shaders in 478 tests
Totals:
SGPRS: 48672 -> 49088 (0.85 %)
VGPRS: 34148 -> 34847 (2.05 %)
Code Size: 1285816 -> 1289128 (0.26 %) bytes
LDS: 28 -> 28 (0.00 %) blocks
Scratch: 492544 -> 573440 (16.42 %) bytes per wave
Max Waves: 6856 -> 6846 (-0.15 %)
Wait states: 0 -> 0 (0.00 %)
Depends on D18451
Reviewers: nhaehnle, arsenm
Subscribers: arsenm, llvm-commits
Differential Revision: http://reviews.llvm.org/D18452
llvm-svn: 264876
For compatability with GAS, nop and nopr are recognized as alises for
bc and bcr, respectively. A mask of 0 turns these instructions
effectively into no-operations.
Reviewed by Ulrich Weigand.
llvm-svn: 264875
These checks are redundant and can be removed
Reviewers: hans
Subscribers: llvm-commits, mzolotukhin
Differential Revision: http://reviews.llvm.org/D18564
llvm-svn: 264872
This reverts commit r264869. I am seeing Windows bot failures due to the
"\" in the path being mishandled at some point (seems to be interpreted
wrongly at some point and llvm-as | llvm-dis is yielding some junk
characters). Need to investigate.
llvm-svn: 264871
XOP's VPPERM has some great 'permute operations' that it can do as well as part of shuffling the bytes of a 128-bit vector - in this case we use it to perform BITREVERSE in a single instruction.
llvm-svn: 264870
Summary:
This change serializes out and in the SourceFileName to LLVM assembly
so that it is preserved through "llvm-dis | llvm-as". This is
necessary to ensure that the global identifiers created for local values
in the module summary index are the same even if the bitcode is
streamed out and read back from LLVM assembly.
Serializing the summary itself to LLVM assembly is in progress.
Reviewers: joker.eph
Subscribers: llvm-commits, joker.eph
Differential Revision: http://reviews.llvm.org/D18588
llvm-svn: 264869
We are currently doing a REALLY bad job of packing results of vector comparisons into the legalized <X x i1> result equivalents - a mixture of PACKSS/PMOVMSKB would be much better here.
llvm-svn: 264867
We already try not to truncate PHIs in computeMinimalBitwidths. LoopVectorize can't handle it and we really don't need to, because both induction and reduction PHIs are truncated by other means.
However, we weren't bailing out in all the places we should have, and we ended up by returning a PHI to be truncated, which has caused PR27018.
This fixes PR17018.
llvm-svn: 264852
operations.
Specifically, we had code that tried to badly approximate reconstructing
all of the possible variations on addressing modes in two x86
instructions based on those in one pseudo instruction. This is not the
first bug uncovered with doing this, so stop doing it altogether.
Instead generically and pedantically copy every operand from the address
over to both new instructions, and strip kill flags from any register
operands.
This fixes a subtle bug seen in the wild where we would mysteriously
drop parts of the addressing mode, causing for example the index
argument in the added test case to just be completely ignored.
Hypothetically, this was an extremely bad miscompile because it actually
caused a predictable and leveragable write of a 64bit quantity to an
unintended offset (the first element of the array intead of whatever
other element was intended). As a consequence, in theory this could even
have introduced security vulnerabilities.
However, this was only something that could happen with an atomic
floating point add. No other operation could trigger this bug, so it
seems extremely unlikely to have occured widely in the wild.
But it did in fact occur, and frequently in scientific applications
which were using relaxed atomic updates of a floating point value after
adding a delta. Those would end up being quite badly miscompiled by
LLVM, which is how we found this. Of course, this often looks like
a race condition in the code, but it was actually a miscompile.
I suspect that this whole RELEASE_FADD thing was a complete mistake.
There is no such operation, and I worry that anything other than add
will get remarkably worse codegeneration. But that's not for this
change....
llvm-svn: 264845
Prior to this patch, the MemorySSA caching visitor would cache all
calls that it visited. When paired with phi optimization, this can be
problematic. Consider:
define void @foo() {
; 1 = MemoryDef(liveOnEntry)
call void @clobberFunction()
br i1 undef, label %if.end, label %if.then
if.then:
; MemoryUse(??)
call void @readOnlyFunction()
; 2 = MemoryDef(1)
call void @clobberFunction()
br label %if.end
if.end:
; 3 = MemoryPhi(...)
; MemoryUse(?)
call void @readOnlyFunction()
ret void
}
When optimizing MemoryUse(?), we visit defs 1 and 2, so we note to
cache them later. We ultimately end up not being able to optimize
passed the Phi, so we set MemoryUse(?) to point to the Phi. We then
cache the clobbering call for def 1 to be the Phi.
This commit changes this behavior so that we wipe out any calls
added to VisistedCalls while visiting the defs of a phi we couldn't
optimize.
Aside: With this patch, we now can bootstrap clang/LLVM without a
single MemorySSA verifier failure. Woohoo. :)
llvm-svn: 264820
Hal Finkel