Sanjay Patel [Thu, 15 Dec 2016 19:13:37 +0000 (19:13 +0000)]
[InstCombine] add folds for icmp (smin X, Y), X
Min/max canonicalization (r287585) exposes the fact that we're missing combines for min/max patterns.
This patch won't solve the example that was attached to that thread, so something else still needs fixing.
The line between InstCombine and InstSimplify gets blurry here because sometimes the icmp instruction that
we want to fold to already exists, but sometimes it's the swapped form of what we want.
Corresponding changes for smax/umin/umax to follow.
Zachary Turner [Thu, 15 Dec 2016 19:02:43 +0000 (19:02 +0000)]
[StringRef] Add enable-if to StringLiteral.
to prevent StringLiteral from being created with a non-literal
char array, clang has a macro enable_if() that can be used
in such a way as to guarantee that the constructor is disabled
unless the length fo the string can be computed at compile time.
This only works on clang, but at least it should allow bots
to catch abuse of StringLiteral.
Ahmed Bougacha [Thu, 15 Dec 2016 18:45:30 +0000 (18:45 +0000)]
[GlobalISel] Drop workaround for Legalizer member/class sharing a name. NFC.
MachineLegalizer used to be the name of both the class and the member,
causing GCC errors. r276522 fixed that by renaming the member to just
'Legalizer'. The 'class' workaround isn't necessary anymore; drop it.
Teresa Johnson [Thu, 15 Dec 2016 18:21:01 +0000 (18:21 +0000)]
[ThinLTO] Ensure callees get hot threshold when first seen on cold path
This is split out from D27696, since it turned out to be a bug fix and
not part of the NFC efficiency change.
Keep the same adjusted (possibly decayed) threshold in both the worklist
and the ImportList. Otherwise if we encountered it first along a cold
path, the callee would be added to the worklist with a lower decayed
threshold than when it is later encountered along a hot path. But the
logic uses the threshold recorded in the ImportList entry to check if
we should re-add it, and without this patch the threshold recorded there
is the same along both paths so we don't re-add it. Using the
same possibly decayed threshold in the ImportList ensures we re-add it
later with the higher non-decayed hot path threshold.
Chris Bieneman [Thu, 15 Dec 2016 18:17:07 +0000 (18:17 +0000)]
[CMake] Minor change to symlink generation for LLDB
If OUTPUT_DIR is not specified we can assume the symlink is linking to a file in the same directory, so we can use $<TARGET_FILE_NAME:${target}> to create a relative symlink.
In the case of LLDB, when we build a framework, we are creating symlinks in a different directory than the file we're pointing to, and we don't install those links. To make this work in the build directory we can use $<TARGET_FILE:${target}> instead, which uses the full path to the target.
And this happens several times in the diffs. For chips with domain-crossing penalties,
the instruction count and size reduction should usually overcome any potential
domain-crossing penalty due to using an FP op in a sequence of int ops. For chips such
as recent Intel big cores and Atom, there is no domain-crossing penalty for shufps, so
using shufps is a pure win.
So the test case diffs all appear to be improvements except one test in
vector-shuffle-combining.ll where we miss an opportunity to use a shift to generate
zero elements and one test in combine-sra.ll where multiple uses prevent the expected
shuffle combining.
Jacques Pienaar [Thu, 15 Dec 2016 16:56:16 +0000 (16:56 +0000)]
[lanai] Simplify small section check in LowerGlobalAddress and treat ldata sections specially.
Move the check for the code model into isGlobalInSmallSectionImpl and return false (not in small section) for variables placed in sections prefixed with .ldata (workaround for a tool limitation).
Robert Lougher [Thu, 15 Dec 2016 16:17:53 +0000 (16:17 +0000)]
[SimplifyCFG] In sinkLastInstruction correctly set debugloc of "common" inst
Simplify CFG will try to sink the last instruction in a series of basic blocks,
creating a "common" instruction in the successor block (sinkLastInstruction).
When it does this, the debug location of the single instruction should be the
merged debug locations of the commoned instructions.
Specifically avoid implicit conversions from/to integral types to
avoid potential errors when changing the underlying type. For example,
a typical initialization of a "full" mask was "LaneMask = ~0u", which
would result in a value of 0x00000000FFFFFFFF if the type was extended
to uint64_t.
Ehsan Amiri [Thu, 15 Dec 2016 12:25:13 +0000 (12:25 +0000)]
[InstCombine] New opportunities for FoldAndOfICmp and FoldXorOfICmp
A number of new patterns for simplifying and/xor of icmp:
(icmp ne %x, 0) ^ (icmp ne %y, 0) => icmp ne %x, %y if the following is true:
1- (%x = and %a, %mask) and (%y = and %b, %mask)
2- %mask is a power of 2.
(icmp eq %x, 0) & (icmp ne %y, 0) => icmp ult %x, %y if the following is true:
1- (%x = and %a, %mask1) and (%y = and %b, %mask2)
2- Let %t be the smallest power of 2 where %mask1 & %t != 0. Then for any
%s that is a power of 2 and %s & %mask2 != 0, we must have %s <= %t.
For example if %mask1 = 24 and %mask2 = 16, setting %s = 16 and %t = 8
violates condition (2) above. So this optimization cannot be applied.
Nemanja Ivanovic [Thu, 15 Dec 2016 11:16:20 +0000 (11:16 +0000)]
[Power9] Allow AnyExt immediates for XXSPLTIB
In some situations, the BUILD_VECTOR node that builds a v18i8 vector by
a splat of an i8 constant will end up with signed 8-bit values and other
situations, it'll end up with unsigned ones. Handle both situations.
Pavel Labath [Thu, 15 Dec 2016 09:40:27 +0000 (09:40 +0000)]
Simplify format member detection in FormatVariadic
Summary:
This replaces the format member search, which was quite complicated, with a more
direct approach to detecting whether a class should be formatted using the
format-member method. Instead we use a special type llvm::format_adapter, which
every adapter must inherit from. Then the search can be simply implemented with
the is_base_of type trait.
Aside from the simplification, I like this way more because it makes it more
explicit that you are supposed to use this type only for adapter-like
formattings, and the other approach (format_provider overloads) should be used
as a default (a mistake I made when first trying to use this library).
The only slight change in behaviour here is that now choose the format-adapter
branch even if the format member invocation will fail to compile (e.g. because it is a
non-const member function and we are passing a const adapter), whereas
previously we would have gone on to search for format_providers for the type.
However, I think that is actually a good thing, as it probably means the
programmer did something wrong.
Sjoerd Meijer [Thu, 15 Dec 2016 09:38:59 +0000 (09:38 +0000)]
[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently
This is essentially a recommit of r285893, but with a correctness fix. The
problem of the original commit was that this:
bic r5, r7, #31
cbz r5, .LBB2_10
got rewritten into:
lsrs r5, r7, #5
beq .LBB2_10
The result in destination register r5 is not the same and this is incorrect
when r5 is not dead. So this fix includes checking the uses of the AND
destination register. And also, compared to the original commit, some regression
tests didn't need changing anymore because of this extra check.
For completeness, this was the original commit message:
For the common pattern (CMPZ (AND x, #bitmask), #0), we can do some more
efficient instruction selection if the bitmask is one consecutive sequence of
set bits (32 - clz(bm) - ctz(bm) == popcount(bm)).
1) If the bitmask touches the LSB, then we can remove all the upper bits and
set the flags by doing one LSLS.
2) If the bitmask touches the MSB, then we can remove all the lower bits and
set the flags with one LSRS.
3) If the bitmask has popcount == 1 (only one set bit), we can shift that bit
into the sign bit with one LSLS and change the condition query from NE/EQ to
MI/PL (we could also implement this by shifting into the carry bit and
branching on BCC/BCS).
4) Otherwise, we can emit a sequence of LSLS+LSRS to remove the upper and lower
zero bits of the mask.
1-3 require only one 16-bit instruction and can elide the CMP. 4 requires two
16-bit instructions but can elide the CMP and doesn't require materializing a
complex immediate, so is also a win.
Prakhar Bahuguna [Thu, 15 Dec 2016 07:59:15 +0000 (07:59 +0000)]
Allow ELF section flags to be specified numerically
Summary:
GAS already allows flags for sections to be specified directly as a
numeric value. This functionality is particularly useful for setting
processor or application-specific values that may not be directly
supported or understood by LLVM. This patch allows LLVM to use numeric
section flag values verbatim if specified by the assembly file.
Prakhar Bahuguna [Thu, 15 Dec 2016 07:59:08 +0000 (07:59 +0000)]
[ARM] Implement execute-only support in CodeGen
This implements execute-only support for ARM code generation, which
prevents the compiler from generating data accesses to code sections.
The following changes are involved:
* Add the CodeGen option "-arm-execute-only" to the ARM code generator.
* Add the clang flag "-mexecute-only" as well as the GCC-compatible
alias "-mpure-code" to enable this option.
* When enabled, literal pools are replaced with MOVW/MOVT instructions,
with VMOV used in addition for floating-point literals. As the MOVT
instruction is required, execute-only support is only available in
Thumb mode for targets supporting ARMv8-M baseline or Thumb2.
* Jump tables are placed in data sections when in execute-only mode.
* The execute-only text section is assigned section ID 0, and is
marked as unreadable with the SHF_ARM_PURECODE flag with symbol 'y'.
This also overrides selection of ELF sections for globals.
Hal Finkel [Thu, 15 Dec 2016 05:50:45 +0000 (05:50 +0000)]
Revert part of r289765 that is not necessary
CS.doesNotAccessMemory(ArgNo) and CS.onlyReadsMemory(ArgNo) calls
dataOperandHasImpliedAttr, so revert this part of r289765 because
it should not be necessary.
Sanjoy Das [Thu, 15 Dec 2016 05:08:57 +0000 (05:08 +0000)]
[MachineBlockPlacement] Don't make blocks "uneditable"
Summary:
This fixes an issue with MachineBlockPlacement due to a badly timed call
to `analyzeBranch` with `AllowModify` set to true. The timeline is as
follows:
1. `MachineBlockPlacement::maybeTailDuplicateBlock` calls
`TailDup.shouldTailDuplicate` on its argument, which in turn calls
`analyzeBranch` with `AllowModify` set to true.
2. This `analyzeBranch` call edits the terminator sequence of the block
based on the physical layout of the machine function, turning an
unanalyzable non-fallthrough block to a unanalyzable fallthrough
block. Normally MBP bails out of rearranging such blocks, but this
block was unanalyzable non-fallthrough (and thus rearrangeable) the
first time MBP looked at it, and so it goes ahead and decides where
it should be placed in the function.
3. When placing this block MBP fails to analyze and thus update the
block in keeping with the new physical layout.
Concretely, before (1) we have something like:
```
LBL0:
< unknown terminator op that may branch to LBL1 >
jmp LBL1
LBL1:
... A
LBL2:
... B
```
In (2), analyze branch simplifies this to
```
LBL0:
< unknown terminator op that may branch to LBL2 >
;; jmp LBL1 <- redundant jump removed
LBL1:
... A
LBL2:
... B
```
In (3), MachineBlockPlacement goes ahead with its plan of putting LBL2
after the first block since that is profitable.
```
LBL0:
< unknown terminator op that may branch to LBL2 >
;; jmp LBL1 <- redundant jump
LBL2:
... B
LBL1:
... A
```
and the program now has incorrect behavior (we no longer fall-through
from `LBL0` to `LBL1`) because MBP can no longer edit LBL0.
There are several possible solutions, but I went with removing the teeth
off of the `analyzeBranch` calls in TailDuplicator. That makes thinking
about the result of these calls easier, and breaks nothing in the lit
test suite.
I've also added some bookkeeping to the MachineBlockPlacement pass and
used that to write an assert that would have caught this.
Hal Finkel [Thu, 15 Dec 2016 03:02:15 +0000 (03:02 +0000)]
Remove the AssumptionCache
After r289755, the AssumptionCache is no longer needed. Variables affected by
assumptions are now found by using the new operand-bundle-based scheme. This
new scheme is more computationally efficient, and also we need much less
code...
Hal Finkel [Thu, 15 Dec 2016 02:53:42 +0000 (02:53 +0000)]
Make processing @llvm.assume more efficient by using operand bundles
There was an efficiency problem with how we processed @llvm.assume in
ValueTracking (and other places). The AssumptionCache tracked all of the
assumptions in a given function. In order to find assumptions relevant to
computing known bits, etc. we searched every assumption in the function. For
ValueTracking, that means that we did O(#assumes * #values) work in InstCombine
and other passes (with a constant factor that can be quite large because we'd
repeat this search at every level of recursion of the analysis).
Several of us discussed this situation at the last developers' meeting, and
this implements the discussed solution: Make the values that an assume might
affect operands of the assume itself. To avoid exposing this detail to
frontends and passes that need not worry about it, I've used the new
operand-bundle feature to add these extra call "operands" in a way that does
not affect the intrinsic's signature. I think this solution is relatively
clean. InstCombine adds these extra operands based on what ValueTracking, LVI,
etc. will need and then those passes need only search the users of the values
under consideration. This should fix the computational-complexity problem.
At this point, no passes depend on the AssumptionCache, and so I'll remove
that as a follow-up change.
Use PIC relocation model as default for PowerPC64 ELF.
Most of the PowerPC64 code generation for the ELF ABI is already PIC.
There are four main exceptions:
(1) Constant pointer arrays etc. should in writeable sections.
(2) The TOC restoration NOP after a call is needed for all global
symbols. While GNU ld has a workaround for questionable GCC self-calls,
we trigger the checks for calls from COMDAT sections as they cross input
sections and are therefore not considered self-calls. The current
decision is questionable and suboptimal, but outside the scope of the
change.
(3) TLS access can not use the initial-exec model.
(4) Jump tables should use relative addresses. Note that the current
encoding doesn't work for the large code model, but it is more compact
than the default for any non-trivial jump table. Improving this is again
beyond the scope of this change.
At least (1) and (3) are assumptions made in target-independent code and
introducing additional hooks is a bit messy. Testing with clang shows
that a -fPIC binary is 600KB smaller than the corresponding -fno-pic
build. Separate testing from improved jump table encodings would explain
only about 100KB or so. The rest is expected to be a result of more
aggressive immediate forming for -fno-pic, where the -fPIC binary just
uses TOC entries.
This change brings the LLVM output in line with the GCC output, other
PPC64 compilers like XLC on AIX are known to produce PIC by default
as well. The relocation model can still be provided explicitly, i.e.
when using MCJIT.
One test case for case (1) is included, other test cases with relocation
mode sensitive behavior are wired to static for now. They will be
reviewed and adjusted separately.
Justin Lebar [Wed, 14 Dec 2016 23:20:40 +0000 (23:20 +0000)]
[NVPTX] Remove dead code.
I've chosen to remove NVPTXInstrInfo::CanTailMerge but not
NVPTXInstrInfo::isLoadInstr and isStoreInstr (which are also dead)
because while the latter two are reasonably useful utilities, the former
cannot be used safely: It relies on successful address space inference
to identify writes to shared memory, but addrspace inference is a
best-effort thing.
Sanjay Patel [Wed, 14 Dec 2016 22:59:14 +0000 (22:59 +0000)]
[DAG] allow more select folding for targets that have 'and not' (PR31175)
The original motivation for this patch comes from wanting to canonicalize
more IR to selects and also canonicalizing min/max.
If we're going to do that, we need more backend fixups to undo select codegen
when simpler ops will do. I chose AArch64 for the tests because that shows the
difference in the simplest way. This should fix:
https://llvm.org/bugs/show_bug.cgi?id=31175
Greg Clayton [Wed, 14 Dec 2016 22:38:08 +0000 (22:38 +0000)]
Add the ability to get attribute values as Optional<T>
When getting attributes it is sometimes nicer to use Optional<T> some of the time instead of magic values. I tried to cut over to only using the Optional values but it made many of the call sites very messy, so it makes sense the leave in the calls that can return a default value. Otherwise code that looks like this:
Justin Lebar [Wed, 14 Dec 2016 22:32:44 +0000 (22:32 +0000)]
[NVPTX] Remove string constants from NVPTXBaseInfo.h.
Summary:
Previously they were defined as a 2D char array in a header file. This
is kind of overkill -- we can let the linker lay out these strings
however it pleases. While we're at it, we might as well just inline
these constants where they're used, as each of them is used only once.
Also move NVPTXUtilities.{h,cpp} into namespace llvm.
Dehao Chen [Wed, 14 Dec 2016 22:06:49 +0000 (22:06 +0000)]
Only sets profile summary when it was not preset.
Summary: SampleProfileLoader pass may be invoked twice by LTO. The 2nd pass should not append more summary info as it is already preset by the 1st pass.
Davide Italiano [Wed, 14 Dec 2016 21:57:04 +0000 (21:57 +0000)]
[LTO] Reject modules without datalayout.
Also, udpate the ~60 failing tests in the tree which did
not contain a valid datalayout.
This fixes PR31123. lld will be updated in a following patch,
immediately after this is committed.
Dehao Chen [Wed, 14 Dec 2016 21:40:47 +0000 (21:40 +0000)]
Create SampleProfileLoader pass in llvm instead of clang
Summary: We used to create SampleProfileLoader pass in clang. This makes LTO/ThinLTO unable to add this pass in the linker plugin. This patch moves the SampleProfileLoader pass creation from clang to llvm pass manager builder.
Eli Friedman [Wed, 14 Dec 2016 20:44:38 +0000 (20:44 +0000)]
[ARM] Split 128-bit vectors in BUILD_VECTOR lowering
Given that INSERT_VECTOR_ELT operates on D registers anyway, combining
64-bit vectors into a 128-bit vector is basically free. Therefore, try
to split BUILD_VECTOR nodes before giving up and lowering them to a series
of INSERT_VECTOR_ELT instructions. Sometimes this allows dramatically
better lowerings; see testcases for examples. Inspired by similar code
in the x86 backend for AVX.
Robert Lougher [Wed, 14 Dec 2016 20:27:22 +0000 (20:27 +0000)]
[InstCombine] Folding of a compare with RHS const should merge debug locations
If all the operands to a phi node are compares that have a RHS constant,
instcombine will try to pull them through the phi node, combining them into
a single operation. When it does this, the debug location of the new op
should be the merged debug locations of the phi node arguments.
Patch 8 of 8 for D26256. Folding of a compare that has a RHS constant.
Eli Friedman [Wed, 14 Dec 2016 20:25:26 +0000 (20:25 +0000)]
[ARM] Add ARMISD::VLD1DUP to match vld1_dup more consistently.
Currently, there are substantial problems forming vld1_dup even if the
VDUP survives legalization. The lack of an actual node
leads to terrible results: not only can we not form post-increment vld1_dup
instructions, but we form scalar pre-increment and post-increment
loads which force the loaded value into a GPR. This patch fixes that
by combining the vdup+load into an ARMISD node before DAGCombine
messes it up.
Also includes a crash fix for vld2_dup (see testcase @vld2dupi8_postinc_variable).
Amjad Aboud [Wed, 14 Dec 2016 20:24:54 +0000 (20:24 +0000)]
[DebugInfo] Changed DIBuilder::createCompileUnit() to take DIFile instead of FileName and Directory.
This way it will be easier to expand DIFile (e.g., to contain checksum) without the need to modify the createCompileUnit() API.
Robert Lougher [Wed, 14 Dec 2016 20:07:49 +0000 (20:07 +0000)]
[InstCombine] Folding of a binop with RHS const should merge the debug locations
If all the operands to a phi node are a binop with a RHS constant, instcombine
will try to pull them through the phi node, combining them into a single
operation. When it does this, the debug location of the new op should be the
merged debug locations of the phi node arguments.
Patch 7 of 8 for D26256. Folding of a binop with RHS constant.
Geoff Berry [Wed, 14 Dec 2016 19:38:22 +0000 (19:38 +0000)]
[GVNHoist] Move GVNHoist to function simplification part of pipeline.
Summary:
Move GVNHoist to later in the optimization pipeline, specifically, to
the function simplification part of the pipeline. The new pipeline
location allows GVNHoist to run on a function after its callees have
been inlined but before the function has been considered for inlining
into its callers, exposing more opportunities for hoisting.
Robert Lougher [Wed, 14 Dec 2016 19:24:01 +0000 (19:24 +0000)]
[InstCombine] When folding casts through a phi node merge the debug locations
If all the operands to a phi node are a cast, instcombine will try to pull
them through the phi node, combining them into a single cast. When it does
this, the debug location of the new cast should be the merged debug locations
of the phi node arguments.
Patch 6 of 8 for D26256. Folding of a cast operation.
Sean Callanan [Wed, 14 Dec 2016 19:09:43 +0000 (19:09 +0000)]
Prepare PrettyStackTrace for LLDB adoption
This patch fixes the linkage for __crashtracer_info__, making it have the proper mangling (extern "C") and linkage (private extern).
It also adds a new PrettyStackTrace type, allowing LLDB to adopt this instead of Host::SetCrashDescriptionWithFormat().
Without this patch, CrashTracer on macOS won't pick up pretty stack traces from any LLVM client.
An LLDB commit adopting this API will follow shortly.
Robert Lougher [Wed, 14 Dec 2016 19:02:14 +0000 (19:02 +0000)]
[InstCombine] Folding loads through a phi node should merge the debug locations
If all the operands to a phi node are a load, instcombine will try to pull
them through the phi node, combining them into a single load. When it does
this, the debug location of the new load should be the merged debug locations
of the phi node arguments.
Patch 5 of 8 for D26256. Folding of a load operation.
Robert Lougher [Wed, 14 Dec 2016 18:37:50 +0000 (18:37 +0000)]
[InstCombine] When folding GEP through a phi node merge the debug locations
If all the operands to a phi node are getelementptr, instcombine
will try to pull them through the phi node, combining them into a single
operation. When it does this, the debug location of the new getelementptr
should be the merged debug locations of the phi node arguments.
Patch 4 of 8 for D26256. Folding of a getelementptr operation.
Eric Christopher [Wed, 14 Dec 2016 18:29:39 +0000 (18:29 +0000)]
This change does two things:
Adds a "Discriminator" field to struct DILineInfo, which defaults to 0.
Fills out the "Discriminator" field in DILineInfo in DWARFDebugLine::LineTable::getFileLineInfoForAddress().
in order to have a slightly nicer interface in getFileLineInfoForAddress.
Robert Lougher [Wed, 14 Dec 2016 18:14:57 +0000 (18:14 +0000)]
[InstCombine] Merge debug locations when folding through a phi node
If all the operands to a phi node are of the same operation, instcombine
will try to pull them through the phi node, combining them into a single
operation. When it does this, the debug location of the operation should
be the merged debug locations of the phi node arguments.
Patch 3 of 8 for D26256. Folding of a compare operation.
Robert Lougher [Wed, 14 Dec 2016 17:49:19 +0000 (17:49 +0000)]
[InstCombine] Merge debug locations when folding through a phi node
If all the operands to a phi node are of the same operation, instcombine
will try to pull them through the phi node, combining them into a single
operation. When it does this, the debug location of the operation should
be the merged debug locations of the phi node arguments.
Patch 2 of 8 for D26256. Folding of a binary operation.
Dehao Chen [Wed, 14 Dec 2016 16:49:28 +0000 (16:49 +0000)]
Create SampleProfileLoader pass in llvm instead of clang
Summary: We used to create SampleProfileLoader pass in clang. This makes LTO/ThinLTO unable to add this pass in the linker plugin. This patch moves the SampleProfileLoader pass creation from clang to llvm pass manager builder.
projects / llvm / log