[PPC64] Fix PR20071 (fctiduz generated for targets lacking that
instruction)
PR20071 identifies a problem in PowerPC's fast-isel implementation
for floating-point conversion to integer. The fctiduz instruction
was added in Power ISA 2.06 (i.e., Power7 and later). However, this
instruction is being generated regardless of which 64-bit PowerPC
target is selected.
The intent is for fast-isel to punt to DAG selection when this
instruction is not available. This patch implements that change.
For testing purposes, the existing fast-isel-conversion.ll test adds
a RUN line for -mcpu=970 and tests for the expected code generation.
Additionally, the existing test fast-isel-conversion-p5.ll was found
to be incorrectly expecting the unavailable instruction to be
generated. I've removed these test variants since we have adequate
coverage in fast-isel-conversion.ll.
This is needed to compile clang with debug+asserts on older powerpc64
and ppc970 targets.
Requested by: jhibbits
MFC after: 3 days
Legalizer: Add support for splitting insert_subvectors.
We handle this by spilling the whole thing to the stack and doing the
insertion as a store.
PR19492. This happens in real code because the vectorizer creates
v2i128 when AVX is enabled.
This fixes a "fatal error: error in backend: Do not know how to split
the result of this operator!" message encountered during compilation of
the net-p2p/libtorrent-rasterbar port.
Reported by: Evgeniy <iron@mail.ua>
MFC after: 3 days
Debug info: Support variadic functions.
Variadic functions have an unspecified parameter tag after the last
argument. In IR this is represented as an unspecified parameter in the
subroutine type.
Paired commit with CFE r202185.
rdar://problem/13690847
This re-applies r202184 + a bugfix in DwarfDebug's argument handling.
This merge includes a change to use the LLVM 3.4 API in
lib/CodeGen/AsmPrinter/DwarfCompileUnit.cpp:
DwarfUnit -> CompileUnit
Sponsored by: DARPA, AFRL
ISel: Make VSELECT selection terminate in cases where the condition type has to
be split and the result type widened.
When the condition of a vselect has to be split it makes no sense widening the
vselect and thereby widening the condition. We end up in an endless loop of
widening (vselect result type) and splitting (condition mask type) doing this.
Instead, split both the condition and the vselect and widen the result.
I ran this over the test suite with i686 and mattr=+sse and saw no regressions.
Fixes PR18036.
With this fix the original problem case from the graphics/rawtherapee
port (posted in http://llvm.org/PR18036 ) now compiles within ~97MB RSS.
Reported by: mandree
MFC after: 1 week
Reland "Fix miscompile of MS inline assembly with stack realignment"
This re-lands commit r196876, which was reverted in r196879.
The tests have been fixed to pass on platforms with a stack alignment
larger than 4.
Update to clang side tests will land shortly.
Pull in r196986 from upstream llvm trunk (by Reid Kleckner):
Revert the backend fatal error from r196939
The combination of inline asm, stack realignment, and dynamic allocas
turns out to be too common to reject out of hand.
ASan inserts empy inline asm fragments and uses aligned allocas.
Compiling any trivial function containing a dynamic alloca with ASan is
enough to trigger the check.
XFAIL the test cases that would be miscompiled and add one that uses the
relevant functionality.
Pull in r202930 from upstream llvm trunk (by Hans Wennborg):
Check for dynamic allocas and inline asm that clobbers sp before building
selection dag (PR19012)
In X86SelectionDagInfo::EmitTargetCodeForMemcpy we check with MachineFrameInfo
to make sure that ESI isn't used as a base pointer register before we choose to
emit rep movs (which clobbers esi).
The problem is that MachineFrameInfo wouldn't know about dynamic allocas or
inline asm that clobbers the stack pointer until SelectionDAGBuilder has
encountered them.
This patch fixes the problem by checking for such things when building the
FunctionLoweringInfo.
Differential Revision: http://llvm-reviews.chandlerc.com/D2954
Together, these commits fix the problem encountered in the devel/emacs
port on the i386 architecture, where a combination of stack realignment,
alloca() and memcpy() could incidentally clobber the %esi register,
leading to segfaults in the temacs build-time utility.
See also: http://llvm.org/PR18171 and http://llvm.org/PR19012
Reported by: ashish
PR: ports/183064
MFC after: 1 week
Fix a crash that occurs when PWD is invalid.
MCJIT needs to be able to run in hostile environments, even when PWD
is invalid. There's no need to crash MCJIT in this case.
The obvious fix is to simply leave MCContext's CompilationDir empty
when PWD can't be determined. This way, MCJIT clients,
and other clients that link with LLVM don't need a valid working directory.
If we do want to guarantee valid CompilationDir, that should be done
only for clients of getCompilationDir(). This is as simple as checking
for an empty string.
The only current use of getCompilationDir is EmitGenDwarfInfo, which
won't conceivably run with an invalid working dir. However, in the
purely hypothetically and untestable case that this happens, the
AT_comp_dir will be omitted from the compilation_unit DIE.
This should help fix assertions occurring with ports-mgmt/tinderbox,
when it is using jails, and sometimes invalidates clang's current
working directory.
Reported by: decke
MFC after: 2 weeks
X-MFC-With: r261991
Lower FNEG just like FABS to fneg[ds] and fmov[ds], thus avoiding
expensive libcall. Also, Qp_neg is not implemented on at least
FreeBSD. This is also what gcc is doing.
SPARC: Implement TRAP lowering. Matches what GCC emits.
This lets clang emit "ta 5" for trap instructions on sparc64, instead of
emitting a call to abort(), making it possible to link the kernel.
Implement SPARCv9 atomic_swap_64 with a pseudo.
The SWAP instruction only exists in a 32-bit variant, but the 64-bit
atomic swap can be implemented in terms of CASX, like the other
atomic rmw primitives.
Submitted by: rdivacky
all of the features in the current working draft of the upcoming C++
standard, provisionally named C++1y.
The code generator's performance is greatly increased, and the loop
auto-vectorizer is now enabled at -Os and -O2 in addition to -O3. The
PowerPC backend has made several major improvements to code generation
quality and compile time, and the X86, SPARC, ARM32, Aarch64 and SystemZ
backends have all seen major feature work.
Release notes for llvm and clang can be found here:
<http://llvm.org/releases/3.4/docs/ReleaseNotes.html>
<http://llvm.org/releases/3.4/tools/clang/docs/ReleaseNotes.html>
MFC after: 1 month
Don't use nopl in cpus that don't support it.
Patch by Mikulas Patocka. I added the test. I checked that for cpu names that
gas knows about, it also doesn't generate nopl.
The modified cpus:
i686 - there are i686-class CPUs that don't have nopl: Via c3, Transmeta
Crusoe, Microsoft VirtualBox - see
https://bbs.archlinux.org/viewtopic.php?pid=775414
k6, k6-2, k6-3, winchip-c6, winchip2 - these are 586-class CPUs
via c3 c3-2 - see https://bugs.archlinux.org/task/19733 as a proof that
Via c3 and c3-Nehemiah don't have nopl
PR: bin/185777
MFC after: 3 days
X86: cvtpi2ps is just an SSE instruction with MMX operands. It has no AVX
equivalent.
Give it the right register format so we can also emit it when AVX is enabled.
This should fix a "Cannot select: intrinsic %llvm.x86.sse.cvtpi2ps" fatal error
in clang while building the gnuradio port for amd64.
Reported by: db
MFC after: 3 days
The basic problem is that some mainstream programs cannot deal with the way
clang optimizes tail calls, as in this example:
int foo(void);
int bar(void) {
return foo();
}
where the call is transformed to:
calll .L0$pb
.L0$pb:
popl %eax
.Ltmp0:
addl $_GLOBAL_OFFSET_TABLE_+(.Ltmp0-.L0$pb), %eax
movl foo@GOT(%eax), %eax
popl %ebp
jmpl *%eax # TAILCALL
However, the GOT references must all be resolved at dlopen() time, and so this
approach cannot be used with lazy dynamic linking (e.g. using RTLD_LAZY), which
usually populates the PLT with stubs that perform the actual resolving.
This patch changes X86TargetLowering::LowerCall() to skip tail call
optimization, if the called function is a global or external symbol.
This fixes problems with loading X.org driver modules, which could occur
when X.org was compiled on i386 with tailcall optimization on, for which
ports r312583 was committed as a workaround. After this change, the
workaround can be removed.
MFC after: 3 days
CaptureTracking: Plug a loophole in the "too many uses" heuristic.
The heuristic was added to avoid spending too much compile time in a
specially crafted test case (PR17461, PR16474) with many uses on a
select or bitcast instruction can still trigger the slow case. Add a
check for that case.
This only affects compile time, don't have a good way to test it.
This fixes the excessive compile time spent on a specific file of the
graphics/rawtherapee port.
Reported by: mandree
MFC after: 3 days
X86: Don't fold spills into SSE operations if the stack is unaligned.
Regalloc can emit unaligned spills nowadays, but we can't fold the
spills into SSE ops if we can't guarantee alignment. PR12250.
This fixes unaligned SSE accesses (leading to a SIGBUS) which could
occur in the ffmpeg ports.
Approved by: re (kib)
Reported by: tijl
MFC after: 3 days
Add ms_abi and sysv_abi attribute handling.
Based on a patch by Benno Rice!
This will help to develop EFI support.
Approved by: re (kib)
Verified by: benno
MFC after: 1 week
Remove invalid assert in DAGTypeLegalizer::RemapValue
There is a comment at the top of DAGTypeLegalizer::PerformExpensiveChecks
which, in part, says:
// Note that these invariants may not hold momentarily when processing a node:
// the node being processed may be put in a map before being marked Processed.
Unfortunately, this assert would be valid only if the above-mentioned invariant
held unconditionally. This was causing llc to assert when, in fact,
everything was fine.
Thanks to Richard Sandiford for investigating this issue!
Fixes PR16562.
This fixes assertions which could occur in the multimedia/ffmpeg1 and
multimedia/ffmpeg2 ports.
Approved by: re (hrs)
Reported by: Matthias Apitz <guru@unixarea.de>
MFC after: 3 days
The X86FixupLEAs pass for Intel Atom must not call
convertToThreeAddress on ADD16rr opcodes, if src1 != src, since that
would cause convertToThreeAddress to try to create a virtual register.
This is not permitted after register allocation, which is when the
X86FixupLEAs pass runs.
This patch fixes PR16785.
Pull in r191715 from upstream llvm trunk:
Forgot to add a break statement.
This should enable building the x11-toolskits/libXaw port with
CPUTYPE=atom.
Approved by: re (gjb)
Reported by: Kenta Suzumoto <kentas@hush.com>
MFC after: 3 days
ISelDAG: spot chain cycles involving MachineNodes
Previously, the DAGISel function WalkChainUsers was spotting that it
had entered already-selected territory by whether a node was a
MachineNode (amongst other things). Since it's fairly common practice
to insert MachineNodes during ISelLowering, this was not the correct
check.
Looking around, it seems that other nodes get their NodeId set to -1
upon selection, so this makes sure the same thing happens to all
MachineNodes and uses that characteristic to determine whether we
should stop looking for a loop during selection.
This should fix PR15840.
Specifically, this fixes the long-standing assertion failure when
compiling the multimedia/gstreamer port on i386. Thanks to Tijl
Coosemans for his help in getting upstream to fix it.
Approved by: re (marius)
InstCombine: Check for zero shift amounts before subtracting one
causing integer overflow.
PR17026. Also avoid undefined shifts and shift amounts larger than 64
bits (those are always undef because we can't represent integer types
that large).
This should fix assertion failures when building the emulators/xmame
port.
Reported by: bapt
Author: Daniel Malea <daniel.malea@intel.com>
Date: Thu Aug 1 21:18:16 2013 +0000
Fixed the Intel-syntax X86 disassembler to respect the (existing)
option for hexadecimal immediates, to match AT&T syntax. This also
brings a new option for C-vs-MASM-style hex.
Patch by Richard Mitton
Reviewed: http://llvm-reviews.chandlerc.com/D1243
FastISel can only append to basic blocks.
Compute the insertion point from the end of the basic block instead of
skipping labels from the front.
This caused failures in landing pads when live-in copies where inserted
before instruction selection.
I missed this change in r252720; without it, certain compilation flags
can cause exception labels to not be generated, but still referenced,
leading to link errors.
Reported by: zeising
MFC after: 3 days
Add MachineBasicBlock::addLiveIn().
This function adds a live-in physical register to an MBB and ensures
that it is copied to a virtual register immediately.
Pull in r185615 from llvm trunk:
Live-in copies go *after* EH_LABELs.
This will soon be tested by exception handling working at all.
Pull in r185617 from llvm trunk:
Simplify landing pad lowering.
Stop using the ISD::EXCEPTIONADDR and ISD::EHSELECTION when lowering
landing pad arguments. These nodes were previously legalized into
CopyFromReg nodes, but that never worked properly because the
CopyFromReg node weren't guaranteed to be scheduled at the top of the
basic block.
This meant the exception pointer and selector registers could be
clobbered before being copied to a virtual register.
This patch copies the two physical registers to virtual registers at
the beginning of the basic block, and lowers the landingpad instruction
directly to two CopyFromReg nodes reading the *virtual* registers. This
is safe because virtual registers don't get clobbered.
A future patch will remove the ISD::EXCEPTIONADDR and ISD::EHSELECTION
nodes.
Together, these changes fix llvm PR 16038 ('qt4 webcore file results in
"Bad machine code: Using an undefined physical register"'), and should
make it possible again to compile the www/qt4-webkit port again on the
i386 arch, without using a CPUTYPE=i686 or higher setting.
the stack in a leaf function that uses TLS.
The issue is, when using TLS, the function is no longer a leaf as it calls
__aeabi_read_tp. With statically linked programs this is not an issue as
it doesn't make use of the stack, however with dynamically linked
applications we enter rtld which does use the stack and makes assumptions
about it's alignment.
This is only a temporary fix until a better patch can be made and submitted
upstream.
Make PrologEpilogInserter save/restore all callee saved registers in
functions which call __builtin_unwind_init()
__builtin_unwind_init() is an undocumented gcc intrinsic which has
this effect, and is used in libgcc_eh.
Goes part of the way toward fixing PR8541.
This obsoletes the ugly hack to libgcc's unwind code from r245272, and
should also work for other arches, so revert the hack too.
[ms-inline asm] Fix a crasher when we fail on a direct match.
The issue was that the MatchingInlineAsm and VariantID args to the
MatchInstructionImpl function weren't being set properly. Specifically, when
parsing intel syntax, the parser thought it was parsing inline assembly in the
at&t dialect; that will never be the case.
The crash was caused when the emitter tried to emit the instruction, but the
operands weren't set. When parsing inline assembly we only set the opcode, not
the operands, which is used to lookup the instruction descriptor.
rdar://13854391 and PR15945
Also, this commit reverts r176036. Now that we're correctly parsing the intel
syntax the pushad/popad don't match properly. I've reimplemented that fix using
a MnemonicAlias.
Pull in r183907 from llvm trunk:
X86: Make the cmov aliases work with intel syntax too.
These commits make a number of Intel-style inline assembly mnemonics
aliases (occurring in several ports) work properly, which could cause
assertions otherwise.
Reported by: kwm, bapt
PR15662: Optimized debug info produces out of order function
parameters
When a function is inlined we lazily construct the variables
representing the function's parameters. After that, we add any
remaining unused parameters.
If the function doesn't use all the parameters, or uses them out of
order, then the DWARF would produce them in that order, producing a
parameter order that doesn't match the source.
This fix causes us to always keep the arg variables at the start of
the variable list & in the original order from the source.
Reported by: avg
MFC after: 1 week
LoopVectorize: LoopSimplify can't canonicalize loops with an
indirectbr in it, don't assert on those cases.
Fixes PR16139.
This should fix clang assertion failures when optimizing at -O3, similar
to:
Assertion failed: (TheLoop->getLoopPreheader() && "No preheader!!"),
function canVectorize, file
contrib/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp, line 2171.
Reported by: O. Hartmann <ohartman@zedat.fu-berlin.de>
PR: ports/178332, ports/178977
MFC after: 3 days
LoopVectorize: getConsecutiveVector must respect signed arithmetic
We were passing an i32 to ConstantInt::get where an i64 was needed and we must
also pass the sign if we pass negatives numbers. The start index passed to
getConsecutiveVector must also be signed.
Should fix PR15882.
This should fix Firefox crashes some people have been reporting, when it
is compiled with -O3.
LoopVectorizer: Fix 15830. When scalarizing and unrolling stores make
sure that the order in which the elements are scalarized is the same
as the original order.
This fixes a miscompilation in FreeBSD's regex library.
This should fix lib/libc/regex/regcomp.c at -O3 with clang 3.3 r178860
on CPUs with SSE. Before this change, the vectorizer could incorrectly
rearrange the second loop in computejumps(), leading to possibly invalid
entries in the re_gets::charjump table.
The net result was that for example "sed s/@CC@/foo/" failed to work
correctly, leading to trouble with many configure scripts.
upcoming 3.3 release (branching and freezing expected in a few weeks).
Preliminary release notes can be found at the usual location:
<http://llvm.org/docs/ReleaseNotes.html>
An MFC is planned once the actual 3.3 release is finished.
X86: Disable cmov-memory patterns on subtargets without cmov.
Fixes PR15115.
For the i386 arch, this should enable cmov instructions only on
-march=pentiumpro and higher. Since our default CPU is i486, cmov
instructions will now be disabled by default.
MFC after: 1 week
MCParser: Reject .balign with non-pow2 alignments.
GNU as rejects them and there are configure scripts in the wild that
check if the assembler rejects ".align 3" to determine whether the
alignment is in bytes or powers of two.
MFC after: 3 days
X86: Disable generation of rep;movsl when %esi is used as a base pointer.
This happens when there is both stack realignment and a dynamic alloca in the
function. If we overwrite %esi (rep;movsl uses fixed registers) we'll lose the
base pointer and the next register spill will write into oblivion.
Fixes PR15249 and unbreaks firefox on i386/freebsd. Mozilla uses dynamic allocas
and freebsd a 4 byte stack alignment.
MFC after: 1 week
Fix another SROA crasher, PR14601.
This was a silly oversight, we weren't pruning allocas which were used
by variable-length memory intrinsics from the set that could be widened
and promoted as integers. Fix that.
This should fix the following assertion failure:
Assertion failed: (CanSROA), function visitUsers, file
/usr/src/lib/clang/libllvmscalaropts/../../../contrib/llvm/lib/Transforms/Scalar/SROA.cpp,
line 2395.
Reported by: gerald
X86: fcmov doesn't handle all possible EFLAGS, fall back to a branch
for the others.
Otherwise it will try to use SSE patterns and fail horribly if sse is
disabled.
Fixes PR14035.
This should fix the following assertion failure:
Assertion failed: (Reg >= X86::FP0 && Reg <= X86::FP6 && "Expected FP
register!"), function getFPReg, file
contrib/llvm/lib/Target/X86/X86FloatingPoint.cpp, line 330.
which can show up when compiling contrib/compiler-rt, using -march=i686
through -march=pentium3 (CPU's which do support fcmov, but don't support
SSE2).
MFC after: 1 week
Make sure always-inline functions get inlined. <rdar://problem/12423986>
Without this change, when the estimated cost for inlining a function with
an "alwaysinline" attribute was lower than the inlining threshold, the
getInlineCost function was returning that estimated cost rather than the
special InlineCost::AlwaysInlineCost value. That is fine in the normal
inlining case, but it can fail when the inliner considers the opportunity
cost of inlining into an internal or linkonce-odr function. It may decide
not to inline the always-inline function in that case. The fix here is just
to make getInlineCost always return the special value for always-inline
functions. I ran into this building clang with libc++. Tablegen failed to
link because of an always-inline function that was not inlined. I have been
unable to reduce the testcase down to a reasonable size.
This should fix the link errors that were reported when atf-run was
compiled with clang -stdlib=libc++. In this case, at -O3 optimization,
some calls to basic_ios::clear() were not inlined, even when the
function was marked __always_inline__.
Reported by: Jan Beich <jbeich@tormail.org>
MFC after: 1 week
