The register scavenger maintains a DistanceMap that maps MI pointers to their
distance from the top of the current MBB. The DistanceMap is built
incrementally in forward() and in bulk in findFirstUse(). It is used by
scavengeRegister() to determine which candidate register has the longest
unused interval.
Unfortunately the DistanceMap contents can become outdated. The first time
scavengeRegister() is called, the DistanceMap is filled to cover the MBB. If
then instructions are inserted in the MBB (as they always are following
scavengeRegister()), the recorded distances are too short. This causes bad
behaviour in the included test case where a register use /after/ the current
position is ignored because findFirstUse() thinks is is /before/ the current
position. A "using an undefined register" assertion follows promptly.
The fix is to build a fresh DistanceMap at the top of scavengeRegister(), and
discard it after use. This means that DistanceMap is no longer needed as a
RegScavenger member variable, and forward() doesn't need to update it.
The fix then discloses issue number two in the same test case: The candidate
search in scavengeRegister() finds a CSR that has been saved in the prologue,
but is currently unused. It would be both inefficient and wrong to spill such
a register in the emergency spill slot. In the present case, the emergency
slot restore is placed immediately before the normal epilogue restore, leading
to a "Redefining a live register" assertion.
Fix number two: When scavengerRegister() stumbles upon an unused register that
is overwritten later in the MBB, return that register early. It is important
to verify that the register is defined later in the MBB, otherwise it might be
an unspilled CSR.
llvm-svn: 78650
the overloaded vector types allowed floating-point or integer vector elements.
Most of these operations actually depend on the element type, so bitcasting
was not an option.
If you include the vpadd intrinsics that I updated earlier, this gets rid
of 20 intrinsics.
llvm-svn: 78646
and short. Well, it's kinda short. Definitely nasty and brutish.
The front-end generates the register/unregister calls into the SjLj runtime,
call-site indices and landing pad dispatch. The back end fills in the LSDA
with the call-site information provided by the front end. Catch blocks are
not yet implemented.
Built on Darwin and verified no llvm-core "make check" regressions.
llvm-svn: 78625
This definitely slows down asm output so put it under an -asm-exuberant
flag.
This information is useful when doing static analysis of performance
issues.
llvm-svn: 78567
instead of syntactically as a string. This means that it keeps track of the
segment, section, flags, etc directly and asmprints them in the right format.
This also includes parsing and validation support for llvm-mc and
"attribute(section)", so we should now start getting errors about invalid
section attributes from the compiler instead of the assembler on darwin.
Still todo:
1) Uniquing of darwin mcsections
2) Move all the Darwin stuff out to MCSectionMachO.[cpp|h]
3) there are a few FIXMEs, for example what is the syntax to get the
S_GB_ZEROFILL segment type?
llvm-svn: 78547
2. Move section switch printing to MCSection virtual method which takes a
TAI. This eliminates textual formatting stuff from TLOF.
3. Eliminate SwitchToSectionDirective, getSectionFlagsAsString, and
TLOFELF::AtIsCommentChar.
llvm-svn: 78510
A TAI hook is appropriate in this case because this is just an
asm syntax issue, not a semantic difference. TLOF should model
the semantics of the section.
llvm-svn: 78498
- Part of optimal static profiling patch sequence by Andreas Neustifter.
- Store edge, block, and function information separately for each functions
(instead of in one giant map).
- Return frequencies as double instead of int, and use a sentinel value for
missing information.
llvm-svn: 78477
Handle large integers, x86_fp80, ConstantAggregateZero, and two more ConstantExpr:
GetElementPtr and IntToPtr
Set SHF_MERGE bit for mergeable strings
Avoid zero initialized strings to be classified as a bss symbol
Don't allow common symbols to be classified as STB_WEAK
Add a constant to be used as a global value offset in data relocations
llvm-svn: 78476
Now there is no special treatment of instructions that redefine part of a
super-register. Instead, the super-register is marked with <imp-use,kill> and
<imp-def>. For instance, from LowerSubregs on ARM:
subreg: CONVERTING: %Q1<def> = INSERT_SUBREG %Q1<undef>, %D1<kill>, 5
subreg: %D2<def> = FCPYD %D1<kill>, 14, %reg0, %Q1<imp-def>
subreg: CONVERTING: %Q1<def> = INSERT_SUBREG %Q1, %D0<kill>, 6
subreg: %D3<def> = FCPYD %D0<kill>, 14, %reg0, %Q1<imp-use,kill>, %Q1<imp-def>
llvm-svn: 78466
Verify that early clobber registers and their aliases are not used.
All changes to RegsAvailable are now done as a transaction so the order of
operands makes no difference.
The included test case is from PR4686. It has behaviour that was dependent on the order of operands.
llvm-svn: 78465
as vector shuffles did not work out well. Shuffles that produce double-wide
vectors accurately represent the operation but make it hard to do anything
with the results. I considered splitting them up into 2 shuffles, one to
write each register separately, but there doesn't seem to be a good way to
reunite them for codegen.
llvm-svn: 78437
driven by TAI to being static, driven by tblgen. This means that a
target doesn't get impacted by this stuff at all if it doesn't opt
into it.
llvm-svn: 78427
The use case is if you have a wrapper class:
class Base {
void *Ptr;
public:
Base() : Ptr(0) { }
operator bool() const { return Ptr; }
.....
}
and sub-wrappers that have exactly the same size:
class Sub : public Base {
public:
....
static bool classof(const Base*);
}
and in the code you would do:
void f(Base b) {
Sub sub = dyn_cast<Sub>(b);
if (sub) {
....
}
}
llvm-svn: 78424
http://llvm.org/viewvc/llvm-project?view=rev&revision=78127, I'm changing the
ExecutionEngine's global mappings to hold AssertingVH<const GlobalValue>. That
way, if unregistering a mapping fails to actually unregister it, we'll get an
assert. Running the jit nightly tests didn't uncover any actual instances of
the problem.
This also uncovered the fact that AssertingVH<const X> didn't work, so I fixed
that too.
llvm-svn: 78400
LoopDependenceAnalysis::getLoops is currently O(N*M) for a loop-nest of
depth N and a compound SCEV of M atomic SCEVs. As both N and M will
typically be very small, this should not be a problem. If it turns out
to be one, rewriting getLoops as SCEVVisitor will reduce complexity to
O(M).
llvm-svn: 78394
creation activity into the target-specific subclasses of TLOF.
Before this, globals with explicit sections could be created by
the base class.
1. make getOrCreateSection protected, add a new getExplicitSectionGlobal
pure virtual method to assign sections to globals with a specified
section.
2. eliminate getSpecialCasedSectionGlobals, which is now PIC specific.
3. eliminate the getKindForNamedSection virtual method, which is
now just a static method for ELF.
4. Add implementions of getExplicitSectionGlobal for ELF/PECOFF/Darwin/PIC16.
They are now all detangled and understandable, woo! :)
llvm-svn: 78319
- start support for new PEI w/reg alloc, allow running RS from emit{Pro,Epi}logue() target hooks.
- fix minor issue with recursion detection.
llvm-svn: 78318
just argv[0]. And remove the code for searching the current
working directory and for searching PATH; the point of FindExecutable
is not to find whatever version of the executable can be found by
searching around, but to find an executable that accompanies the
current executable.
Update the tools to use sys::Program::FindProgramByName when they
want PATH searching.
llvm-svn: 78240
a dirty hack and isn't need anymore since the last x86 code emitter patch)
- Add a target-dependent modifier to addend calculation
- Use R_X86_64_32S relocation for X86::reloc_absolute_word_sext
- Use getELFSectionFlags whenever possible
- fix getTextSection to use TLOF and emit the right text section
- Handle global emission for static ctors, dtors and Type::PointerTyID
- Some minor fixes
llvm-svn: 78176
Instead of awkwardly encoding calling-convention information with ISD::CALL,
ISD::FORMAL_ARGUMENTS, ISD::RET, and ISD::ARG_FLAGS nodes, TargetLowering
provides three virtual functions for targets to override:
LowerFormalArguments, LowerCall, and LowerRet, which replace the custom
lowering done on the special nodes. They provide the same information, but
in a more immediately usable format.
This also reworks much of the target-independent tail call logic. The
decision of whether or not to perform a tail call is now cleanly split
between target-independent portions, and the target dependent portion
in IsEligibleForTailCallOptimization.
This also synchronizes all in-tree targets, to help enable future
refactoring and feature work.
llvm-svn: 78142
for ELF to work.
2) RIP addressing: Use SIB bytes for absolute relocations where RegBase=0,
IndexReg=0.
3) The JIT can get the real address of cstpools and jmptables during
code emission, fix that for object code emission
llvm-svn: 78129
This is not just a matter of passing in the target triple from the module;
currently backends are making decisions based on the build and host
architecture. The goal is to migrate to making these decisions based off of the
triple (in conjunction with the feature string). Thus most clients pass in the
target triple, or the host triple if that is empty.
This has one important change in the way behavior of the JIT and llc.
For the JIT, it was previously selecting the Target based on the host
(naturally), but it was setting the target machine features based on the triple
from the module. Now it is setting the target machine features based on the
triple of the host.
For LLC, -march was previously only used to select the target, the target
machine features were initialized from the module's triple (which may have been
empty). Now the target triple is taken from the module, or the host's triple is
used if that is empty. Then the triple is adjusted to match -march.
The take away is that -march for llc is now used in conjunction with the host
triple to initialize the subtarget. If users want more deterministic behavior
from llc, they should use -mtriple, or set the triple in the input module.
llvm-svn: 77946
__builtin_bfin_ones does the same as ctpop, so it can be implemented in the front-end.
__builtin_bfin_loadbytes loads from an unaligned pointer with the disalignexcpt instruction. It does the same as loading from a pointer with the low bits masked. It is better if the front-end creates a masked load. We can always instruction select the masked to disalignexcpt+load.
We keep csync/ssync/idle. These intrinsics represent instructions that need workarounds for some silicon revisions. We may even want to convert inline assembler to intrinsics to enable the workarounds.
llvm-svn: 77917
Allow imp-def and imp-use of anything in the scavenger asserts, just like the machine code verifier.
Allow redefinition of a sub-register of a live register.
llvm-svn: 77904
the only real caller (GetFunctionSizeInBytes) uses it.
The custom ARM implementation of this is basically reimplementing
an assembler poorly for negligible gain. It should be removed
IMNSHO, but I'll leave that to ARMish folks to decide.
llvm-svn: 77877
getLSDASection() to be more specific. This makes it pretty obvious
that the ELF LSDA section is being specified wrong in PIC mode. We're
probably getting a lot of startup-time relocations to a readonly page,
which is expensive and bad.
Someone who cares about ELF C++ should investigate this.
llvm-svn: 77847
compute it based on what it knows. As part of this, rename getSectionForMergeableConstant
to getSectionForConstant because it works for non-mergable constants also.
The only functionality change from this is that Xcore will start dropping
its jump tables into readonly section instead of data section in -static mode.
This should be fine as the linker resolves the relocations. If this is a
problem, let me know and we'll come up with another solution.
llvm-svn: 77833
should have no state that is specific to particular globals in the
section. In this case, it means the removal of the "isWeak" and
"ExplicitSection" bits. MCSection uses the new form of SectionKind.
To handle isWeak, I introduced a new SectionInfo class, which is
SectionKind + isWeak, and it is used by the part of the code generator
that does classification of a specific global.
The ExplicitSection disappears. It is moved onto MCSection as a new
"IsDirective" bit. Since the Name of a section is either a section
or directive, it makes sense to keep this bit in MCSection. Ultimately
the creator of MCSection should canonicalize (e.g.) .text to whatever
the actual section is.
llvm-svn: 77803
padding is disabled, tabs get replaced by spaces except in the case of
the first operand, where the tab is output to line up the operands after
the mnemonics.
Add some better comments and eliminate redundant code.
Fix some testcases to not assume tabs.
llvm-svn: 77740
- Uses MCAsmToken::getIdentifier which returns the (sub)string representing the
meaningfull contents a string or identifier token.
- Directives aren't done yet.
llvm-svn: 77739
thing is #if0'd out anyway. Just simplify the code by reducing the interface.
Not deleting this is essential for Bill's continuing happiness.
llvm-svn: 77736
shouldn't do AU.setPreservesCFG(), because even though CodeGen passes
don't modify the LLVM IR CFG, they may modify the MachineFunction CFG,
and passes like MachineLoop are registered with isCFGOnly set to true.
llvm-svn: 77691
failures when building assorted projects with clang.
--- Reverse-merging r77654 into '.':
U include/llvm/CodeGen/Passes.h
U include/llvm/CodeGen/MachineFunctionPass.h
U include/llvm/CodeGen/MachineFunction.h
U include/llvm/CodeGen/LazyLiveness.h
U include/llvm/CodeGen/SelectionDAGISel.h
D include/llvm/CodeGen/MachineFunctionAnalysis.h
U include/llvm/Function.h
U lib/Target/CellSPU/SPUISelDAGToDAG.cpp
U lib/Target/PowerPC/PPCISelDAGToDAG.cpp
U lib/CodeGen/LLVMTargetMachine.cpp
U lib/CodeGen/MachineVerifier.cpp
U lib/CodeGen/MachineFunction.cpp
U lib/CodeGen/PrologEpilogInserter.cpp
U lib/CodeGen/MachineLoopInfo.cpp
U lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
D lib/CodeGen/MachineFunctionAnalysis.cpp
D lib/CodeGen/MachineFunctionPass.cpp
U lib/CodeGen/LiveVariables.cpp
llvm-svn: 77661
This is currently used by X86 to handle atomic_load_add when the output of the node is not used. I believe there is a better solution. But I find MERGE_VALUES useful for selecting multi-output node when the dead output can be selected as a IMPLICIT_DEF.
llvm-svn: 77583
and convert code to using it, instead of having lots of things
poke the isLookupPtrRegClass() method directly.
2. Make PointerLikeRegClass contain a 'kind' int, and store it in
the existing regclass field of TargetOperandInfo when the
isLookupPtrRegClass() predicate is set. Make getRegClass pass
this into TargetRegisterInfo::getPointerRegClass(), allowing
targets to have multiple ptr_rc things.
llvm-svn: 77504
wide vectors. Likewise, change VSTn intrinsics to take separate arguments
for each vector in a multi-vector struct. Adjust tests accordingly.
llvm-svn: 77468
- Provides static constructors for doing number to string conversions without
using temporaries.
- There are several ways to do this, I think given the Twine constraints this
is the simplest one.
- One FIXME for fast number -> hex conversion.
- Added another comment on one last major bit of perf work Twines need, which
is to make raw_svector_ostream more efficient.
llvm-svn: 77445
