This was the slowest target custom pass and was spending 80%
of the time in getMinimalPhysRegClass which was called
for every register operand.
Try to use the statically known register class when possible from
the instruction's MCOperandInfo. There are a few pseudo instructions
which are not well behaved with unknown register classes which still
require the expensive physical register class search.
There are a few other possibilities for making this even faster,
such as not inspecting implicit operands. For now those are checked
because it is technically possible to have a scalar load into
exec or vcc which can be implicitly used.
llvm-svn: 249079
We emit denormalized tables, where every range of invokes in the same
state gets a complete list of EH action entries. This is significantly
simpler than trying to infer the correct nested scoping structure from
the MI. Fortunately, for SEH, the nesting structure is really just a
size optimization.
With this, some basic __try / __except examples work.
llvm-svn: 249078
Summary:
Instead of asserting when the kernel metadata is different than we expect,
we should just skip lowering that function. This fixes assertion
failures with OpenCL argument metadata from older LLVM releases.
Reviewers: arsenm
Subscribers: arsenm, llvm-commits
Differential Revision: http://reviews.llvm.org/D13356
llvm-svn: 249073
Catchret transfers control from a catch funclet to an earlier funclet.
However, it is not completely clear which funclet the catchret target is
part of. Make this clear by stapling the catchret target's funclet
membership onto the CATCHRET SDAG node.
llvm-svn: 249052
Support for pairing unscaled loads and stores has been enabled since the
original ARM64 port. This feature is no longer experimental, AFAICT.
llvm-svn: 249049
Add generic instructions for load complement, load negative and load positive
for fp32 and fp64, and let isel prefer them. They do not clobber CC, and so
give scheduler more freedom. SystemZElimCompare pass will convert them when it
can to the CC-setting variants.
Regression tests updated to expect the new opcodes in places where the old ones
where used. New test case SystemZ/fp-cmp-05.ll checks that
SystemZCompareElim.cpp can handle the new opcodes.
README.txt updated (bullet removed).
Note that fp128 is not yet handled, because it is relatively rare, and is a
bit trickier, because of the fact that l.dfr would operate on the sign bit of
one of the subregisters of a fp128, but we would not want to copy the other
sub-reg in case src and dst regs are not the same.
Reviewed by Ulrich Weigand.
llvm-svn: 249046
v2: Add test (Matt).
Fix capitalization of isEOP (Matt).
Move pattern to class parameter (Matt).
Make the instruction available to Cayman (Matt).
Change name from MEM_RAT WRITE_TYPED to MEM_RAT STORE_TYPED.
Patch by: Zoltan Gilian
llvm-svn: 249042
Summary:
Some passes may open up opportunities for optimizations, leaving empty
lifetime start/end ranges. For example, with the following code:
void foo(char *, char *);
void bar(int Size, bool flag) {
for (int i = 0; i < Size; ++i) {
char text[1];
char buff[1];
if (flag)
foo(text, buff); // BBFoo
}
}
the loop unswitch pass will create 2 versions of the loop, one with
flag==true, and the other one with flag==false, but always leaving
the BBFoo basic block, with lifetime ranges covering the scope of the for
loop. Simplify CFG will then remove BBFoo in the case where flag==false,
but will leave the lifetime markers.
This patch teaches InstCombine to remove trivially empty lifetime marker
ranges, that is ranges ending right after they were started (ignoring
debug info or other lifetime markers in the range).
This fixes PR24598: excessive compile time after r234581.
Reviewers: reames, chandlerc
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D13305
llvm-svn: 249018
Summary:
The instructions SeenExprs records may be deleted during rewriting.
FindClosestMatchingDominator should ignore these deleted instructions.
Fixes PR24301.
Reviewers: grosser
Subscribers: grosser, llvm-commits
Differential Revision: http://reviews.llvm.org/D13315
llvm-svn: 248983
Summary:
Without this patch, the memory manager would call `mprotect` on every memory
region it ever allocated whenever it wanted to finalize memory (i.e. not just
the ones it just allocated). This caused terrible performance problems for
long running memory managers. In one particular compile heavy julia benchmark,
we were spending 50% of time in `mprotect` if running under MCJIT.
Fix this by splitting allocated memory blocks into those on which memory
permissions have been set and those on which they haven't and only running
`mprotect` on the latter.
Reviewers: lhames
Subscribers: reames, llvm-commits
Differential Revision: http://reviews.llvm.org/D13156
llvm-svn: 248981
The custom code produces incorrect results if later reassociated.
Since r221657, on x86, vNi32 uitofp is lowered using an optimized
sequence:
movdqa LCPI0_0(%rip), %xmm1 ## xmm1 = [65535, ...]
pand %xmm0, %xmm1
por LCPI0_1(%rip), %xmm1 ## [0x4b000000, ...]
psrld $16, %xmm0
por LCPI0_2(%rip), %xmm0 ## [0x53000000, ...]
addps LCPI0_3(%rip), %xmm0 ## [float -5.497642e+11, ...]
addps %xmm1, %xmm0
Since r240361, the machine combiner opportunistically reassociates
2-instruction sequences (with -ffast-math). In the new code sequence,
the ADDPS' are eligible. In isolation, for simple examples (without
reassociable users), this makes no performance difference (the goal
being to enable reassociation of longer chains).
In the trivial example (just one uitofp), the reassociation doesn't
happen, because (I think) it would require the emission of a separate
movaps for a constantpool load (instead of folding it into addps).
However, when we have multiple uitofp sequences, and the constantpool
loads are CSE'd earlier, the machine combiner can do the reassociation.
When the ADDPS' are reassociated, the resulting sequence isn't correct
anymore, as we'd be adding large (2**39) constants with comparatively
smaller values (~2**23). Given that two of the three inputs are powers
of 2 larger than 2**16, and that ulp(2**39) == 2**(39-24) == 2**15,
the reassociated chain will produce 0 for any input in [0, 2**14[.
In my testing, it also produces wrong results for 99.5% of [0, 2**32[.
Avoid this by disabling the new lowering when -ffast-math. It does
mean that we'll get slower code than without it, but at least we
won't get egregiously incorrect code.
One might argue that, considering -ffast-math is all but meaningless,
uitofp producing wrong results isn't a compiler bug. But it really is.
Fixes PR24512.
...though this is really more of a workaround.
Ideally, we'd have some sort of Machine FMF, but that's a problem
that's not worth tackling until we do more with machine IR.
llvm-svn: 248965
The Win64 unwinder disassembles forwards from each PC to try to
determine if this PC is in an epilogue. If so, it skips calling the EH
personality function for that frame. Typically, this means you cannot
catch an exception in the same frame that you threw it, because 'throw'
calls a noreturn runtime function.
Previously we avoided this problem with the TrapUnreachable
TargetOption, but that's a much bigger hammer than we need. All we need
is a 1 byte non-epilogue instruction right after the call. Instead,
what we got was an unconditional branch to a shared block containing the
ud2, potentially 7 bytes instead of 1. So, this reverts r206684, which
added TrapUnreachable, and replaces it with something better.
The new code pattern matches for invoke/call followed by unreachable and
inserts an int3 into the DAG. To be 100% watertight, we would need to
insert SEH_Epilogue instructions into all basic blocks ending in a call
with no terminators or successors, but in practice this is unlikely to
come up.
llvm-svn: 248959
Summary:
Given an array of i2 elements, 4 consecutive scalar loads will be lowered to
i8-sized loads and thus will access 4 consecutive bytes in memory. If we
vectorize these loads into a single <4 x i2> load, it'll access only 1 byte in
memory. Hence, we should prohibit vectorization in such cases.
PS: Initial patch was proposed by Arnold.
Reviewers: aschwaighofer, nadav, hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D13277
llvm-svn: 248943
Previously, the index was constrained to the size of the memory operation for
no apparent reason. This change removes that constraint so that we can form
pre-index instructions with any valid offset.
llvm-svn: 248931
Same strategy as simplifyInstructionsInBlock. ~1/3 less time
on my test suite. This pass doesn't have many in-tree users,
but getting rid of an O(N^2) worst case and making it cleaner
should at least make it a viable alternative to ADCE, since
it's now consistently somewhat faster.
llvm-svn: 248927
As Richard Barton observed at http://reviews.llvm.org/D12937#inline-107121
TargetParser in LLVM has insufficient support for ARMv6Z and ARMv6ZK.
In particular, there were no tests for TrustZone being supported in these
architectures.
The patch clears a FIXME: left by Saleem Abdulrasool in r201471, and fixes
his test case which hadn't really been testing what it was claiming to test.
Differential Revision: http://reviews.llvm.org/D13236
llvm-svn: 248921
Usually large blocks are not a problem. But if a large block (> 10k instructions)
contains many (potential) chains of vector instructions, and those chains are
spread over a wide range of instructions, then scheduling becomes a compile time problem.
This change introduces a limit for the accumulate scheduling region size of a block.
For real-world functions this limit will never be exceeded (it's about 10x larger than
the maximum value seen in the test-suite and external test suite).
llvm-svn: 248917
This patch teaches InstCombiner how to convert a SSSE3/AVX2 byte shuffle to a
builtin shuffle if the mask is constant.
Converting byte shuffle intrinsic calls to builtin shuffles can help finding
more opportunities for combining shuffles later on in selection dag.
We may end up with byte shuffles with constant masks as the result of inlining.
Differential Revision: http://reviews.llvm.org/D13252
llvm-svn: 248913
This commit changes the interface of the vld[1234], vld[234]lane, and vst[1234],
vst[234]lane ARM neon intrinsics and associates an address space with the
pointer that these intrinsics take. This changes, e.g.,
<2 x i32> @llvm.arm.neon.vld1.v2i32(i8*, i32)
to
<2 x i32> @llvm.arm.neon.vld1.v2i32.p0i8(i8*, i32)
This change ensures that address spaces are fully taken into account in the ARM
target during lowering of interleaved loads and stores.
Differential Revision: http://reviews.llvm.org/D12985
llvm-svn: 248887
The XOP shifts just have logical/arithmetic versions and the left/right shifts are controlled by whether the value is positive/negative. Because of this I've added new X86ISD nodes instead of trying to force them to use the existing shift nodes.
Additionally Excavator cores (bdver4) support XOP and AVX2 - meaning that it should use the AVX2 shifts when it can and fall back to XOP in other cases.
Differential Revision: http://reviews.llvm.org/D8690
llvm-svn: 248878
to prevent setting a huge stride, because DATA_FORMAT has a different
meaning if ADD_TID_ENABLE is set.
This is a candidate for stable llvm 3.7.
Tested-and-Reviewed-by: Christian König <christian.koenig@amd.com>
llvm-svn: 248858
The x64 ABI requires that epilogues do not contain code other than stack
adjustments and some limited control flow. However, we'd insert code to
initialize the return address after stack adjustments. Instead, insert
EAX/RAX with the current value before we create the stack adjustments in
the epilogue.
llvm-svn: 248839
Add support to the indexed instrprof reader and writer for the format
that will be used for value profiling.
Patch by Betul Buyukkurt, with minor modifications.
llvm-svn: 248833
HHVM calling convention, hhvmcc, is used by HHVM JIT for
functions in translated cache. We currently support LLVM back end to
generate code for X86-64 and may support other architectures in the
future.
In HHVM calling convention any GP register could be used to pass and
return values, with the exception of R12 which is reserved for
thread-local area and is callee-saved. Other than R12, we always
pass RBX and RBP as args, which are our virtual machine's stack pointer
and frame pointer respectively.
When we enter translation cache via hhvmcc function, we expect
the stack to be aligned at 16 bytes, i.e. skewed by 8 bytes as opposed
to standard ABI alignment. This affects stack object alignment and stack
adjustments for function calls.
One extra calling convention, hhvm_ccc, is used to call C++ helpers from
HHVM's translation cache. It is almost identical to standard C calling
convention with an exception of first argument which is passed in RBP
(before we use RDI, RSI, etc.)
Differential Revision: http://reviews.llvm.org/D12681
llvm-svn: 248832
Summary:
Funclets have been turned into functions by the time they hit the object
file. Make sure that they have decent names for the symbol table and
CFI directives explaining how to reason about their prologues.
Differential Revision: http://reviews.llvm.org/D13261
llvm-svn: 248824
The immediate in the load/store should be scaled by the size of the memory
operation, not the size of the register being loaded/stored. This change gets
us one step closer to forming LDPSW instructions. This change also enables
pre- and post-indexing for halfword and byte loads and stores.
llvm-svn: 248804
On some of our benchmarks this change shows about 50% compile time improvement without any noticeable performance difference.
Differential Revision: http://reviews.llvm.org/D13248
llvm-svn: 248801
If a PHI starts at a non-negative constant, monotonically increases
(only adds of a constant are supported at the moment) and that add
does not wrap, then the PHI is known never to be zero.
llvm-svn: 248796
alignment requirements, for example in the case of vectors.
These requirements are exploited by the code generator by using
move instructions that have similar alignment requirements, e.g.,
movaps on x86.
Although the code generator properly aligns the arguments with
respect to the displacement of the stack pointer it computes,
the displacement itself may cause misalignment. For example if
we have
%3 = load <16 x float>, <16 x float>* %1, align 64
call void @bar(<16 x float> %3, i32 0)
the x86 back-end emits:
movaps 32(%ecx), %xmm2
movaps (%ecx), %xmm0
movaps 16(%ecx), %xmm1
movaps 48(%ecx), %xmm3
subl $20, %esp <-- if %esp was 16-byte aligned before this instruction, it no longer will be afterwards
movaps %xmm3, (%esp) <-- movaps requires 16-byte alignment, while %esp is not aligned as such.
movl $0, 16(%esp)
calll __bar
To solve this, we need to make sure that the computed value with which
the stack pointer is changed is a multiple af the maximal alignment seen
during its computation. With this change we get proper alignment:
subl $32, %esp
movaps %xmm3, (%esp)
Differential Revision: http://reviews.llvm.org/D12337
llvm-svn: 248786
Currently SimplifyDemandedVectorElts can only peek through bitcasts if the vectors have the same number of elements.
This patch fixes and enables some existing (disabled) code to support bitcasting to vectors with more/fewer elements. It currently only accepts cases when vectors alias cleanly (i.e. number of elements are an exact multiple of the other vector).
This was added to improve the demanded vector elements support for SSE vector shifts which require the __m128i (<2 x i64>) argument type to be bitcast to the vector type for the builtin shift. I've added extra tests for various additional bitcasts.
Differential Revision: http://reviews.llvm.org/D12935
llvm-svn: 248784
Summary: This patch adds block frequency analysis to LoopUnswitch pass to recognize hot/cold regions. For cold regions the pass only performs trivial unswitches since they do not increase code size, and for hot regions everything works as before. This helps to minimize code growth in cold regions and be more aggressive in hot regions. Currently the default cold regions are blocks with frequencies below 20% of function entry frequency, and it can be adjusted via -loop-unswitch-cold-block-frequency flag. The entire feature is controlled via -loop-unswitch-with-block-frequency flag and it is off by default.
Reviewers: broune, silvas, dnovillo, reames
Subscribers: davidxl, llvm-commits
Differential Revision: http://reviews.llvm.org/D11605
llvm-svn: 248777
Place new and update dbg.declare calls immediately after the
corresponding alloca.
Current code in replaceDbgDeclareForAlloca puts the new dbg.declare
at the end of the basic block. LLVM codegen has problems emitting
debug info in a situation when dbg.declare appears after all uses of
the variable. This usually kinda works for inlining and ASan (two
users of this function) but not for SafeStack (see the pending change
in http://reviews.llvm.org/D13178).
llvm-svn: 248769
`ScalarEvolution::isImpliedCondOperandsViaNoOverflow` tries to cast the
operand type of the comparison it is given to an `IntegerType`. This is
incorrect because it could actually be simplifying a comparison between
two pointers. Switch it to using `getTypeSizeInBits` instead, which
does the right thing for both pointers and integers.
Fixed PR24956.
llvm-svn: 248743
The splitting of > 4 dword SMRD instructions
if using an offset in an SGPR instead of an immediate
was not setting the destination register,
resulting an an instruction missing an operand
which would assert later.
Test will be included in a following commit
which fixes a related issue.
llvm-svn: 248739
Patch by Jake VanAdrighem!
Summary:
Fix the way we sort the llvm.used and llvm.compiler.used members.
This bug seems to have been introduced in rL183756 through a set of improper casts to GlobalValue*. In subsequent patches this problem was missed and transformed into a getName call on a ConstantExpr.
Reviewers: silvas
Subscribers: silvas, llvm-commits
Differential Revision: http://reviews.llvm.org/D12851
llvm-svn: 248728
1. Use a worklist, not a recursive approach, to avoid needless
revisitation and being repeatedly forced to jump back to the
start of the BB if a handle is invalidated.
2. Only insert operands to the worklist if they become unused
after a dead instruction is removed, so we don’t have to
visit them again in most cases.
3. Use a SmallSetVector to track the worklist.
4. Instead of pre-initting the SmallSetVector like in
DeadCodeEliminationPass, only put things into the worklist
if they have to be revisited after the first run-through.
This minimizes how much the actual SmallSetVector gets used,
which saves a lot of time.
llvm-svn: 248727
Summary:
The P5600 is an out-of-order, superscalar implementation of the MIPS32R5
architecture.
The scheduler has a few missing details (see the 'Tricky Instructions'
section and some quirks of the P5600 are deliberately omitted due to
implementation difficulty and low chance of significant benefit (e.g. the
predicate on P5600WriteEitherALU). However, testing on SingleSource is
showing significant performance benefits on some apps (seven in the 10-30%
range) and only one significant regression (12%) when
-pre-RA-sched=linearize is given. Without -pre-RA-sched=linearize the
results are more variable. Some do even better (up to 55% improvement) but
increased numbers of copies are slowing others down (up to 12%).
Overall, the scheduler as it currently stands is a 2.4% win with
-pre-RA-sched=linearize and a 2.7% win without -pre-RA-sched=linearize.
I'm sure we can improve on this further.
For completeness, the FPGA this was tested on shows some failures with and
without the P5600 scheduler. These appear to be scheduling related since
the two test runs have fairly different sets of failing tests even after
accounting for other factors (e.g. spurious connection failures) however
it's not P5600 specific since we also get some for the generic scheduler.
Reviewers: vkalintiris
Subscribers: mpf, llvm-commits, atrick, vkalintiris
Differential Revision: http://reviews.llvm.org/D12193
llvm-svn: 248725
This was split off of http://reviews.llvm.org/D13040 to make it easier to test the correctness of the implication logic. For the moment, this only handles a single easy case which shows up when eliminating and combining range checks. In the (near) future, I plan to extend this for other cases which show up in range checks, but I wanted to make those changes incrementally once the framework was in place.
At the moment, the implication logic will be used by three places. One in InstSimplify (this review) and two in SimplifyCFG (http://reviews.llvm.org/D13040 & http://reviews.llvm.org/D13070). Can anyone think of other locations this style of reasoning would make sense?
Differential Revision: http://reviews.llvm.org/D13074
llvm-svn: 248719
Originally, debug intrinsics and annotation intrinsics may prevent
the loop to be rerolled, now they are ignored.
Differential Revision: http://reviews.llvm.org/D13150
llvm-svn: 248718
supportsTailCall() has two callers. Both of them double-check isThumb1Only(),
and refuse to proceed with tail-calling in that case.
Therefore, it makes sense to move this check to
ARMSubtarget::initSubtargetFeatures, where SupportsTailCall is initialized;
and to eliminate the extra checks at the call sites.
Following a review comment, added an "assert(supportsTailCall())"
in IsEligibleForTailCall.
NFC.
llvm-svn: 248703
When AA is being used, non-aliasing stores are canonicalized to use the same
chain, and DAGCombiner::getStoreMergeAndAliasCandidates can take advantage of
this by looking only as users of a store's chain operand. However, user
iteration is not result-number specific, we need to check that the use is as a
chain operand, and not via some other operand. It is certainly possible to have
another potentially-aliasing store, which shares the first's base pointer, and
uses the first's chain's node via some other operand.
Failure to catch this situation caused, at least in the included test case, an
assert later because the relative sequence-number ordering caused later
replacement to create a cycle in the DAG.
llvm-svn: 248698
When llvm declarations have argument names, it's helpful to actually
print those names when debugging. Arguably, it'd be nice to print them
all the time, but that would mean the IR we output wouldn't round trip
through bitcode, which doesn't store the names.
Make the varous print() methods in AsmWriter optionally print "for
debug" and set that flag in the dump() methods. The only thing this
does differently for now is print the argument names in declarations.
llvm-svn: 248692
Before this change `HasSameValue` would return true for distinct
`alloca` instructions if they happened to be allocating the same
type (`alloca` instructions are not specified as reading memory). This
change adds an explicit whitelist of instruction types for which
"identical" instructions compute the same value.
Fixes PR24952.
llvm-svn: 248690
This is one step towards solving PR24766:
https://llvm.org/bugs/show_bug.cgi?id=24766
We were not producing the same IR for these two C functions because the store
to the temp bool causes extra zexts:
#include <stdbool.h>
bool switchy(char x1, char x2, char condition) {
bool conditionMet = false;
switch (condition) {
case 0: conditionMet = (x1 == x2); break;
case 1: conditionMet = (x1 <= x2); break;
}
return conditionMet;
}
bool switchy2(char x1, char x2, char condition) {
switch (condition) {
case 0: return (x1 == x2);
case 1: return (x1 <= x2);
}
return false;
}
As noted in the code comments, this test case manages to avoid the more general existing
phi optimizations where there are only 2 phi inputs or where there are no constant phi
args mixed in with the casts ops. It seems like a corner case, but if we don't catch it,
then I don't think we can get SimplifyCFG to further optimize towards the canonical form
for this function shown in the bug report.
Differential Revision: http://reviews.llvm.org/D12866
llvm-svn: 248689
Summary:
Factor the code that rewrites invokes to calls and rewrites WinEH
terminators to their "unwind to caller" equivalents into a helper in
Utils/Local, and use it in the three places I'm aware of that need to do
this.
Reviewers: andrew.w.kaylor, majnemer, rnk
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D13152
llvm-svn: 248677
llvm::format compiles down to snprintf which has no defined rounding for
floating point arguments, and MSVC has implemented it differently from
what the BSD libcs and glibc do. Try to emulate the glibc rounding
behavior to avoid changing tests.
While there simplify code a bit and move trivial methods inline.
llvm-svn: 248665
These require multiple mov instructions to copy,
but the default value is that 1 instruction is needed.
I'm not sure if this actually changes anything.
llvm-svn: 248651
Summary:
This is the second part of fixing bug 24848 https://llvm.org/bugs/show_bug.cgi?id=24848.
If both operands of a comparison have range metadata, they should be used to constant fold the comparison.
Reviewers: sanjoy, hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D13177
llvm-svn: 248650
Matt Arsenault