[SimplifyCFG] Defer folding unconditional branches to LateSimplifyCFG if it can destroy canonical loop structure.
Summary:
When simplifying unconditional branches from empty blocks, we pre-test if the
BB belongs to a set of loop headers and keep the block to prevent passes from
destroying canonical loop structure. However, the current algorithm fails if
the destination of the branch is a loop header. Especially when such a loop's
latch block is folded into loop header it results in additional backedges and
LoopSimplify turns it into a nested loop which prevent later optimizations
from being applied (e.g., loop unrolling and loop interleaving).
This patch augments the existing algorithm by further checking if the
destination of the branch belongs to a set of loop headers and defer
eliminating it if yes to LateSimplifyCFG.
[LV] Test once if vector trip count is zero, instead of twice
Generate a single test to decide if there are enough iterations to jump to the
vectorized loop, or else go to the scalar remainder loop. This test compares the
Scalar Trip Count: if STC < VF * UF go to the scalar loop. If
requiresScalarEpilogue() holds, at-least one iteration must remain scalar; the
rest can be used to form vector iterations. So in this case the test checks
instead if (STC - 1) < VF * UF by comparing STC <= VF * UF, and going to the
scalar loop if so. Otherwise the vector loop is entered for at-least one vector
iteration.
This test covers the case where incrementing the backedge-taken count will
overflow leading to an incorrect trip count of zero. In this (rare) case we will
also avoid the vector loop and jump to the scalar loop.
This patch simplifies the existing tests and effectively removes the basic-block
originally named "min.iters.checked", leaving the single test in block
"vector.ph".
Original observation and initial patch by Evgeny Stupachenko.
[PM/LCG] Follow-up fix to r308088 to handle deletion of library
functions.
In the prior commit, we provide ordering to the LCG between functions
and library function definitions that they might begin to call through
transformations. But we still would delete these library functions from
the call graph if they became dead during inlining.
While this immediately crashed, it also exposed a loss of information.
We shouldn't remove definitions of library functions that can still
usefully participate in the LCG-powered CGSCC optimization process. If
new call edges are formed, we want to have definitions to be called.
We can still remove these functions if truly dead using global-dce, etc,
but removing them during the CGSCC walk is premature.
This fixes a crash in the new PM when optimizing some unusual libraries
that end up with "internal" lib functions such as the code in the "R"
language's libraries.
Summary:
This patch adds the following
1. Adds a skeleton scheduler model for AMD Znver1.
2. Introduces the znver1 execution units and pipes.
3. Caters the instructions based on the generic scheduler classes.
4. Further additions to the scheduler model with instruction itineraries will be carried out incrementally based on
a. Instructions types
b. Registers used
5. Since itineraries are not added based on instructions, throughput information are bound to change when incremental changes are added.
6. Scheduler testcases are modified accordingly to suit the new model.
Patch by Ganesh Gopalasubramanian. With minor formatting tweaks from me.
Preserve the actual library name as provided by the user. This is
required to properly replicate link's behaviour about the module import
name handling. This requires an associated change to lld for updating
the tests for the proper behaviour for the import library module name
handling in various cases.
Adrian Prantl [Wed, 19 Jul 2017 00:09:54 +0000 (00:09 +0000)]
Debug Info: Add a file: field to DIImportedEntity.
DIImportedEntity has a line number, but not a file field. To determine
the decl_line/decl_file we combine the line number from the
DIImportedEntity with the file from the DIImportedEntity's scope. This
does not work correctly when the parent scope is a DINamespace or a
DIModule, both of which do not have a source file.
This patch adds a file field to DIImportedEntity to unambiguously
identify the source location of the using/import declaration. Most
testcase updates are mechanical, the interesting one is the removal of
the FIXME in test/DebugInfo/Generic/namespace.ll.
This fixes PR33822. See https://bugs.llvm.org/show_bug.cgi?id=33822
for more context.
Accept and ignore --wide/-W. In GNU readelf this switch is
necessary to get the output format that's consistent between
32-bit and 64-bit targets. llvm-readobj always produces that
output format.
Petr Hosek [Tue, 18 Jul 2017 23:35:22 +0000 (23:35 +0000)]
[llvm-readobj] Accept -S as an alias for --sections
In GNU readelf, the short option for --sections is upper-case -S.
Note that GNU uses lower-case -s to mean --symbols, while LLVM
uses -s to mean --sections and -t to mean --symbols (-t has yet a
different meaning in GNU). So command-line uses with -S can now
be compatible, but uses with -s or -t are still incompatible.
[asan] Copy arguments passed by value into explicit allocas for ASan
Summary:
ASan determines the stack layout from alloca instructions. Since
arguments marked as "byval" do not have an explicit alloca instruction, ASan
does not produce red zones for them. This commit produces an explicit alloca
instruction and copies the byval argument into the allocated memory so that red
zones are produced.
Submitted on behalf of @morehouse (Matt Morehouse)
Object: rename parameter from DLLName to ImportName
When I originally wrote this code, I neglected the fact that the import
library may be created for executables. This name is not the name of
the DLL, but rather the name for the imported module. It will be
embedded into the IAT/ILT reference. Rename it to make it more obvious.
NFC.
When given an extension as part of the `library` directive in a def
file, the extension is preserved/honoured by link/lib. Behave similarly
when parsing the def file. This requires checking if a native extension
is provided as a keyword parameter. If no extension is present, append
a standard `.dll` or `.exe` extension.
This is best tested via lld, and I will add tests there as a follow up.
Jakub Kuderski [Tue, 18 Jul 2017 20:19:52 +0000 (20:19 +0000)]
[Dominators] Improve error checking in deleteEdge
Summary: This patch improves error detection in deleteEdge. It asserts that the edge doesn't exist in the CFG and that DomTree knew about this edge before.
Nirav Dave [Tue, 18 Jul 2017 20:06:24 +0000 (20:06 +0000)]
[DAG] Improve Aliasing of operations to static alloca
Re-recommiting after landing DAG extension-crash fix.
Recommiting after adding check to avoid miscomputing alias information
on addresses of the same base but different subindices.
Memory accesses offset from frame indices may alias, e.g., we
may merge write from function arguments passed on the stack when they
are contiguous. As a result, when checking aliasing, we consider the
underlying frame index's offset from the stack pointer.
Static allocs are realized as stack objects in SelectionDAG, but its
offset is not set until post-DAG causing DAGCombiner's alias check to
consider access to static allocas to frequently alias. Modify isAlias
to consider access between static allocas and access from other frame
objects to be considered aliasing.
Many test changes are included here. Most are fixes for tests which
indirectly relied on our aliasing ability and needed to be modified to
preserve their original intent.
The remaining tests have minor improvements due to relaxed
ordering. The exception is CodeGen/X86/2011-10-19-widen_vselect.ll
which has a minor degradation dispite though the pre-legalized DAG is
improved.
Nirav Dave [Tue, 18 Jul 2017 19:49:20 +0000 (19:49 +0000)]
[DAG] Reverse node replacement in extension operation. NFCI.
Reorder replacements to be user first in preparation for multi-level
folding to premptively avoid inadvertantly deleting later nodes from
sharing found from replacement.
Brian Gesiak [Tue, 18 Jul 2017 19:25:34 +0000 (19:25 +0000)]
[opt-viewer] Handle file names that contain '#'
Summary:
When using opt-viewer.py with files with '#' in their name, such as
'foo#bar.cpp', opt-viewer.py would generate links such as
'/path/to/foo#bar.cpp.opt.yaml#L42'. In this case, the link is
interpreted by browsers as a link to the file '/path/to/foo', and to the
section within that file with ID 'bar.cpp.opt.yaml#L42'.
To work around this issue, replace '#' with '_' in file names and links
in opt-viewer.py.
Added a feature to the Sparc back-end that replaces the integer multiply and
divide instructions with calls to .mul/.sdiv/.udiv. This is a step towards
having full v7 support.
Patch by: Eric Kedaigle
Differential Revision: https://reviews.llvm.org/D35500
Nirav Dave [Tue, 18 Jul 2017 17:39:15 +0000 (17:39 +0000)]
[DAG] Avoid deleting nodes before combining them.
When replacing a node and it's operand, replacing the operand node may
cause the deletion of the original node leading to an assertion
failure. Case around these replacements to avoid this without relying
on inspecting the DELETED_NODE opcode in various extend
dagcombiner cases.
Matt Arsenault [Tue, 18 Jul 2017 16:44:58 +0000 (16:44 +0000)]
AMDGPU: Annotate necessity of flat-scratch-init
As an approximation of the existing handling to avoid
regressions. Fixes using too many registers with calls
on subtargets with the SGPR allocation bug.
Matt Arsenault [Tue, 18 Jul 2017 16:44:56 +0000 (16:44 +0000)]
AMDGPU: Figure out private memory regs after lowering
Introduce pseudo-registers for registers needed for stack
access, which are replaced during finalizeLowering.
Note these pseudo-registers are currently only used for the
used register location, and not for determining their
input argument register.
This is better because it avoids the need to try to predict
whether a call will be emitted from the IR, and also
detects stack objects introduced by legalization.
Test changes are from the HasStackObjects check being more
accurate since stack objects introduced during legalization
are now known.
Simon Pilgrim [Tue, 18 Jul 2017 15:55:30 +0000 (15:55 +0000)]
[x86, CGP] increase memcmp() expansion up to 4 load pairs
It should be a win to avoid going out to the system lib for all small memcmp() calls using scalar ops. For x86 32-bit, this means most everything up to 16 bytes. For 64-bit, that doubles because we can do 8-byte loads.
Notes:
Reduced from 4 to 2 loads for -Os behavior, which might not be optimal in all cases. It's effectively a question of how much do we trust the system implementation. Linux and macOS (and Windows I assume, but did not test) have optimized memcmp() code for x86, so it's probably not bad either way? PPC is using 8/4 for defaults on these. We do not expand at all for -Oz.
There are still potential improvements to make for the CGP expansion IR and/or lowering such as avoiding select-of-constants (D34904) and not doing zexts to the max load type before doing a compare.
We have special-case SSE/AVX codegen for (memcmp(x, y, 16/32) == 0) that will no longer be produced after this patch. I've shown the experimental justification for that change in PR33329:
https://bugs.llvm.org/show_bug.cgi?id=33329#c12
TLDR: While the vector code is a likely winner, we can't guarantee that it's a winner in all cases on all CPUs, so I'm willing to sacrifice it for the greater good of expanding all small memcmp(). If we want to resurrect that codegen, it can be done by adjusting the CGP params or poking a hole to let those fall-through the CGP expansion.
[Hexagon] Emit lookup tables in text section based on a flag
The flag "-hexagon-emit-lut-text" (defaulted to false) is added to decide
on where to keep the switch generated lookup table.
Differential Revision: https://reviews.llvm.org/D34818
AMDGPU: Fix crash when folding immediates into multiple uses
Summary:
When an immediate is folded by constant folding, we re-scan the entire
use list for two reasons:
1. The constant folding may have created a new use of the same reg.
2. The constant folding may have removed an additional use in the list
we're currently traversing (e.g., constant folding an S_ADD_I32 c, c).
However, this could previously lead to a crash when an unrelated use was
added twice into the FoldList. Since we re-scan the whole list anyway, we
might as well just clear the FoldList again before we do so.
Using a MIR test to show this because real code seems to trigger the issue
only in connection with some really subtle control flow structures.
Fixes GL45-CTS.shading_language_420pack.binding_images on gfx9.
Simon Pilgrim [Tue, 18 Jul 2017 14:19:34 +0000 (14:19 +0000)]
[X86] Added cmov target to memcmp test
As discussed by @spatel on D35067:
"I added the cmov attribute to the 32-bit codegen test because it removes some noise for that file. I think the intent for the SSE vs no-SSE runs is to show the potential difference for the 16 and 32 byte cases rather than the lack of cmov (which has been available for all CPUs since SSE1, so that's why it shows up automatically with -mattr=sse2)."
[mips] Alter register classes for MSA pseudo f16 instructions
This change introduces additional machine instructions in functions
dealing with the expansion of msa pseudo f16 instructions due to
register classes being inappropriate when checked with machine
verifier.
PSCEV] Create AddRec for Phis in cases of possible integer overflow,
using runtime checks
Extend the SCEVPredicateRewriter to work a bit harder when it encounters an
UnknownSCEV for a Phi node; Try to build an AddRecurrence also for Phi nodes
whose update chain involves casts that can be ignored under the proper runtime
overflow test. This is one step towards addressing PR30654.
Coverage hooks that take less-than-64-bit-integers as parameters need the
zeroext parameter attribute (http://llvm.org/docs/LangRef.html#paramattrs)
to make sure they are properly extended by the x86_64 ABI.
[LoopInterchange] Split up interchange.ll test case (NFC).
Summary:
Currently most tests for the loop interchange pass are in
test/Transforms/LoopInterchange/interchange.ll. This patch splits up the
large test file in smaller pieces, which makes debugging test failures
easier.
Revert part of r308100 since the cause (r308025) was also reverted.
The commit r308100 updated WebAssembly tests for r308025. In one case it
merely made the test more resilient but in another case it made
a substantive update. Because r308025 was reverted in r308271, these
changes to the test also need to be reverted. They should be folded into
the recommit of r308025 when it is ready.
[x86] Add a missing triple, without which the CPU won't parse.
Notably, this is failing on our PPC build bots:
http://lab.llvm.org:8011/builders/clang-ppc64le-linux/builds/8338/steps/ninja%20check%201/logs/FAIL%3A%20LLVM%3A%3Apr33772.ll
[X86] Prevent an assertion failure if a gather intrinsic is passed a non-constant scale value.
This isn't legal code, but we shouldn't crash on it. Now we just don't convert the gather intrinsic if the scale isn't constant and let it go through to isel where we'll report an isel failure.
[CGP] Cleanup - remove redundant code in OptimizeMemoryInst. NFC
optimizeMemoryInst contains a vector AddrModeInsts.
The only use of this vector is to check that all instructions are in the same
block as memory instruction. This check is guarded by PhiSeen flag,
so if we traversed through phi node then we do not need to keep information
in AddrModeInsts. AddModeInsts is set first time we found some addressing mode
and updated if we found new one later.
We can find next addressing mode only if we traverse phi node so all code
related to update of AddModeInsts can be safely removed.
Max Kazantsev [Tue, 18 Jul 2017 04:53:48 +0000 (04:53 +0000)]
[IRCE] Recognize loops with ne/eq latch conditions
In some particular cases eq/ne conditions can be turned into equivalent
slt/sgt conditions. This patch teaches parseLoopStructure to handle some
of these cases.
[Analysis] RemoveTotalMemInst counting in InstCount to avoid reading back other Statistic variables
Summary:
Previously, we counted TotalMemInst by reading certain instruction counters before and after calling visit and then finding the difference. But that wouldn't be thread safe if this same pass was being ran on multiple threads.
This list of "memory instructions" doesn't make sense to me as it includes call/invoke and is missing atomics.
Daniel Neilson [Tue, 18 Jul 2017 01:06:54 +0000 (01:06 +0000)]
Add element-atomic mem intrinsic canary tests for Memory Sanitizer.
Summary:
Add canary tests to verify that MSAN currently does nothing with the element atomic memory intrinsics for memcpy, memmove, and memset.
Placeholder tests that will fail once element atomic @llvm.mem[cpy|move|set] instrinsics have been added to the MemIntrinsic class hierarchy. These will act as a reminder to verify that MSAN handles these intrinsics properly once they have been added to that class hierarchy.
Daniel Neilson [Tue, 18 Jul 2017 01:06:53 +0000 (01:06 +0000)]
Add element-atomic mem intrinsic canary tests for Efficiency Sanitizer.
Summary:
Add canary tests to verify that ESAN currently does nothing with the element atomic memory intrinsics for memcpy, memmove, and memset.
Placeholder tests that will fail once element atomic @llvm.mem[cpy|move|set] instrinsics have been added to the MemIntrinsic class hierarchy. These will act as a reminder to verify that ESAN handles these intrinsics properly once they have been added to that class hierarchy.
Daniel Neilson [Tue, 18 Jul 2017 01:06:52 +0000 (01:06 +0000)]
Add element-atomic mem intrinsic canary tests for Dataflow Sanitizer.
Summary:
Add canary tests to verify that DFSAN currently does nothing with the element atomic memory intrinsics for memcpy, memmove, and memset.
Placeholder tests that will fail once @llvm.mem[cpy|move|set] instrinsics have been added to the MemIntrinsic class hierarchy. These will act as a reminder to verify that DFSAN handles these intrinsics properly once they have been added to that class hierarchy.
Note that there could be some trickiness with these element-atomic intrinsics for the dataflow sanitizer in racy multithreaded programs. The data flow sanitizer inserts additional lib calls to mirror the memory intrinsic's action, so it is possible (very likely, even) that the dfsan buffers will not be in sync with the original buffers. Furthermore, implementation of the dfsan buffer updates for the element atomic intrinsics will have to also use unordered atomic instructions. If we can assume that dfsan is never run on racy multithreaded programs, then the element atomic memory intrinsics can pretty much be treated the same as the regular memory intrinsics.
Daniel Neilson [Tue, 18 Jul 2017 01:06:52 +0000 (01:06 +0000)]
Add element-atomic mem intrinsic canary tests for Address Sanitizer.
Summary:
Add canary tests to verify that ASAN currently does nothing with the element atomic memory intrinsics for memcpy, memmove, and memset.
Placeholder tests that will fail once element atomic @llvm.mem[cpy|move|set] instrinsics have been added to the MemIntrinsic class hierarchy. These will act as a reminder to verify that ASAN handles these intrinsics properly once they have been added to that class hierarchy.
Daniel Neilson [Tue, 18 Jul 2017 01:06:47 +0000 (01:06 +0000)]
Add element-atomic mem intrinsic canary tests for InstCombine.
Summary:
Add canary tests to verify that InstCombine currently does nothing with the element atomic memory intrinsics for memmove and memset.
Placeholder tests that will fail once element atomic @llvm.mem[move|set] instrinsics have been added to the MemIntrinsic class hierarchy. These will act as a reminder to verify that inst combine handles these intrinsics properly once they have been added to that class hierarchy.
[DWARF] Modification of code for the verification of .debug_info section.
Summary:
This patch modifies the handleDebugInfo() function so that we verify the contents of each unit
in the .debug_info section only if its header has been successfully verified.
This change will allow for more/different verification checks depending on the type of the unit since from
dwarf5, the .debug_info section may consist of different types of units.
Summary:
This removes the CVTypeVisitor updater and verifier classes. They were
made dead by the minimal type dumping refactoring. Replace them with a
single function that takes a type record and produces a hash. Call this
from the minimal type dumper and compare the hash.
I also noticed that the microsoft-pdb reference repository uses a basic
CRC32 for records that aren't special. We already have an implementation
of that CRC ready to use, because it's used in COFF for ICF.
I'll make LLD call this hashing utility in a follow-up change. We might
also consider using this same hash in type stream merging, so that we
don't have to hash our records twice.
[PDB] Merge in types and items from type servers (/Zi)
Summary:
Object files compiled with /Zi emit type information into a type server
PDB. The .debug$S section will contain a single TypeServer2Record with
the absolute path and GUID of the type server. LLD needs to load the
type server PDB and merge all types and items it finds in it into the
destination PDB.
[codeview] Fix YAML for LF_TYPESERVER2 by hoisting PDB_UniqueId
Summary:
We were treating the GUIDs in TypeServer2Record as strings, and the
non-ASCII bytes in the GUID would not round-trip through YAML.
We already had the PDB_UniqueId type portably represent a Windows GUID,
but we need to hoist that up to the DebugInfo/CodeView library so that
we can use it in the TypeServer2Record as well as in PDB parsing code.
[codeview] Don't use the type visitor to merge types
Summary:
This didn't do much to speed things up, but it implements a FIXME, and I
think it's a nice simplification. We don't need the record kind switch.
We're doing that ourselves.
[codeview] Remove TypeServerHandler and PDBTypeServerHandler
Summary:
Instead of wiring these through the CVTypeVisitor interface, clients
should inspect the CVTypeArray before visiting it and potentially load
up the type server's TPI stream if they need it.
No tests relied on this functionality because LLD was the only client.
Teresa Johnson [Mon, 17 Jul 2017 19:25:38 +0000 (19:25 +0000)]
Revert "Restore with fix "[ThinLTO] Ensure we always select the same function copy to import""
This reverts commit r308114 (and follow on fixes to test).
There is a linking failure in a ThinLTO bot:
http://green.lab.llvm.org/green/job/clang-stage2-configure-Rthinlto_build/3663/
(and undefined reference). It seems like it must be a second order
effect of the heuristic change I made, and may take some time to try
to reproduce locally and track down. Therefore, reverting for now.
projects / llvm / log