allow target to correctly compute latency for cases where static scheduling
itineraries isn't sufficient. e.g. variable_ops instructions such as
ARM::ldm.
This also allows target without scheduling itineraries to compute operand
latencies. e.g. X86 can return (approximated) latencies for high latency
instructions such as division.
- Compute operand latencies for those defined by load multiple instructions,
e.g. ldm and those used by store multiple instructions, e.g. stm.
llvm-svn: 115755
expand to an initializeMyPass() function (in additional to the extant static ctors). Eventually, these will be called
from a big InitializeAllPasses() function, and the PassInfo's they create (which would be leaked if this code were used
at the moment) will be handed off to a PassRegistry for ownership.
llvm-svn: 115703
it in with the SSSE3 instructions.
Steward! Could you place this chair by the aft sun deck? I'm trying to get away
from the Astors. They are such boors!
llvm-svn: 115552
1) Changed ValidateDwarfFileNumber() to isValidDwarfFileNumber() to be better
named. Since it is just a predicate and isn't actually changing any state.
2) Added a missing return in the comments for setCurrentDwarfLoc() in
include/llvm/MC/MCContext.h for fix formatting.
3) Changed clearDwarfLocSeen() to ClearDwarfLocSeen() since it does change
state.
4) Simplified the last test in isValidDwarfFileNumber() to just a one line
boolean test of MCDwarfFiles[FileNumber] != 0 for the final return statement.
llvm-svn: 115551
is partly because this attribute caused trouble in the past (the
SmallVector one had to be changed from aligned to aligned(8) due
to causing crashes on i386 for example; in theory the same might
be needed in the Allocator case...). But it's mostly because
there seems to be no point in special casing gcc here. Using the
same implementation for all compilers results in better testing.
llvm-svn: 115462
LiveInterval::MergeValueNumberInto instead of trying to extend LiveRanges and
getting it wrong.
This fixed PR8249 where a valno with a multi-segment live range was defined by
an identity copy created by RemoveCopyByCommutingDef. Some of the live
segments disappeared.
llvm-svn: 115385
stick with a constant estimate of 90% (branch predictors are good!), but we might find that we want to provide
more nuanced estimates in the future.
llvm-svn: 115364
The x86_mmx type is used for MMX intrinsics, parameters and
return values where these use MMX registers, and is also
supported in load, store, and bitcast.
Only the above operations generate MMX instructions, and optimizations
do not operate on or produce MMX intrinsics.
MMX-sized vectors <2 x i32> etc. are lowered to XMM or split into
smaller pieces. Optimizations may occur on these forms and the
result casted back to x86_mmx, provided the result feeds into a
previous existing x86_mmx operation.
The point of all this is prevent optimizations from introducing
MMX operations, which is unsafe due to the EMMS problem.
llvm-svn: 115243
resolved or not. Different object files have different restrictions and
different native assemblers have different idiosyncrasies we want to emulate
for now.
Move the existing MachO logic to the new place and implement an ELF one that
gets fixups to globals right.
llvm-svn: 115131
or not. TableGen needs to generate the printInstruction() function as taking
an MCInstr* or a MachineInstr*, depending. Default to the old non-MC
version so that everything not yet using MC continues to just work without
fidding.
llvm-svn: 115126
hide jump threading opportunities by turning control flow into data flow. Run an early JumpThreading pass
(adds approximately an additional 1% to optimization time on SPEC), allowing it to get a shot at these cases
first. Fixes <rdar://problem/8447345>.
llvm-svn: 115099
Rather than having arbitrary cutoffs, actually try to cost model the conversion.
For now, the constants are tuned to more or less match our existing behavior, but these will be
changed to reflect realistic values as this work proceeds.
llvm-svn: 114973
Allocator instances can now be created by calling createPBQPRegisterAllocator.
Tidied up use of CoalescerPair as per Jakob's suggestions.
Made the new PBQPBuilder based construction process the default. The internal construction process
remains in-place and available via -pbqp-builder=false for now. It will be removed shortly if the new
process doesn't cause any regressions.
llvm-svn: 114626
new VariantKind to the MCSymbolExpr seems like overkill, but I'm not sure
there's a more straightforward way to get the printing difference captured.
(i.e., x86 uses @PLT, ARM uses (PLT)).
llvm-svn: 114613
that complex patterns are matched after the entire pattern has
a structural match, therefore the NodeStack isn't in a useful
state when the actual call to the matcher happens.
llvm-svn: 114489
passed the root of the match, even though only a few patterns
actually needed this (one in X86, several in ARM [which should
be refactored anyway], and some in CellSPU that I don't feel
like detangling). Instead of requiring all ComplexPatterns to
take the dead root, have targets opt into getting the root by
putting SDNPWantRoot on the ComplexPattern.
llvm-svn: 114471
I think I've audited all uses, so it should be dependable for address spaces,
and the pointer+offset info should also be accurate when there.
llvm-svn: 114464
instead of calling lower_bound or upper_bound directly.
This cleans up the search logic a bit because {lower,upper}_bound compare
LR->start by default, and it is usually simpler to search LR->end.
Funnelling all searches through one function also makes it possible to replace
the search algorithm with something faster than binary search.
llvm-svn: 114448
into OptimizeCompareInstr.
This necessitates the passing of CmpValue around,
so widen the virtual functions to accomodate.
No functionality changes.
llvm-svn: 114428
"getFixedStack" on the MachinePointerInfo class. While
this isn't the problem I'm setting out to solve, it is the
right way to eliminate PseudoSourceValue, so lets go with it.
llvm-svn: 114406
MachinePointerInfo struct, no functionality change.
This also adds an assert to MachineMemOperand::MachineMemOperand
that verifies that the Value* is either null or is an IR pointer type.
llvm-svn: 114389
For now the allocator still uses the old (internal) construction mechanism by default. This will be phased out soon assuming
no issues with the builder system come up.
To invoke the new construction mechanism just pass '-regalloc=pbqp -pbqp-builder' to llc. To provide custom constraints a
Target just needs to extend PBQPBuilder and pass an instance of their derived builder to the RegAllocPBQP constructor.
llvm-svn: 114272
more careful not to call SimplifyInstructionsInBlock() on an unreachable block, the issue has been fixed at a higher level. Add
a big warning to SimplifyInstructionsInBlock() to hopefully prevent this in the future.
llvm-svn: 114117
that all the setup of this class currently happens at static initialization time, this misses the fact
that some later events can cause mutation of the PassRegistrationListeners list, and thus cause race issues.
llvm-svn: 114036
The ELF implementation now creates text, data and bss to match the gnu as
behavior.
The text streamer still has the old MachO specific behavior since
the testsuite checks that it will error when a directive is given
before a setting the current section for example.
A nice benefit is that -n is not required anymore when producing
ELF files.
llvm-svn: 114027
VFP instructions use it for loading some constants, so implement that
handling.
Not thrilled with adding a member to MCOperand, but not sure there's much of
a better option that's not pretty fragile (like putting a double in the
union instead and just assuming that's good enough). Suggestions welcome...
llvm-svn: 113996
The problem was that the test for whether a compiler supports it or not was
inaccurate, but has to be accurate: LLVM_LOCAL_VISIBILITY is an optimization
and not needed for correctness, so wrongly thinking a compiler doesn't support
it is not a big deal, but LLVM_GLOBAL_VISIBILITY is for correctness, and not
an optimization: getting it wrong is fatal: it needs to be set based on a
configure test not testing the gcc version. Since dragonegg has moved to a
different scheme, and it was the only user of LLVM_GLOBAL_VISIBILITY, just
remove this macro.
llvm-svn: 113959
isn't a good level of abstraction for memdep. Instead, generalize
AliasAnalysis::alias and related interfaces with a new Location
class for describing a memory location. For now, this is the same
Pointer and Size as before, plus an additional field for a TBAA tag.
Also, introduce a fixed MD_tbaa metadata tag kind.
llvm-svn: 113858
the end of the line on a parser error, allowing skipping to happen
for syntactic errors but not for semantic errors. Before we would
miss emitting a diagnostic about the second line, because we skipped
it due to the semantic error on the first line:
foo %eax
bar %al
This fixes rdar://8414033 - llvm-mc ignores lines after an invalid instruction mnemonic errors
llvm-svn: 113688
iterator when an optimization took place. This allows us to do more insane
things with the code than just remove an instruction or two.
llvm-svn: 113640
take multiple cycles to decode.
For the current if-converter clients (actually only ARM), the instructions that
are predicated on false are not nops. They would still take machine cycles to
decode. Micro-coded instructions such as LDM / STM can potentially take multiple
cycles to decode. If-converter should take treat them as non-micro-coded
simple instructions.
llvm-svn: 113570
is different from what the code now uses in a two ways: NamedMDNodes
were considered Values and included in the numbering, and the
function-local metadata counter wasn't reset between functions.
The later problem breaks lazy deserialization, so instead of trying
to emulate the old numbering, just drop the old metadata. The only
in-tree use case is debug info with LTO, where the QOI loss is
considered acceptable.
llvm-svn: 113557
not unrolling loops that contain calls that would be better off getting inlined. This mostly
comes up when an interleaved devirtualization pass has devirtualized a call which the inliner
will inline on a future pass. Thus, rather than blocking all loops containing calls, add
a metric for "inline candidate calls" and block loops containing those instead.
llvm-svn: 113535
instruction in the class would be decoded to. Or zero if the number of
uOPs must be determined dynamically.
This will be used to determine the cost-effectiveness of predicating a
micro-coded instruction.
llvm-svn: 113513
don't use any InlineCostAnalyzer state, and are useful for other clients who don't necessarily want to use
all of InlineCostAnalyzer's logic, some of which is fairly inlining-specific.
No intended functionality change.
llvm-svn: 113499
AliasAnalysis, and some code for implementing the new query on top of
existing implementations by making standard alias and getModRefInfo
queries.
llvm-svn: 113329
switch to using a ManagedStatic for the global PassRegistry instead of a
ManagedCleanup, and fix a destruction ordering bug this exposed.
llvm-svn: 113283
Fix zeroExtend and signExtend to support empty sets, and to return the smallest
possible result set which contains the extension of each element in their
inputs. For example zext i8 [100, 10) to i16 is now [0, 256), not i16 [100, 10)
which contains 63446 members.
llvm-svn: 113187
Loop::hasLoopInvariantOperands method. Remove
a useless and confusing Loop::isLoopInvariant(Instruction)
method, which didn't do what you thought it did.
No functionality change.
llvm-svn: 113133
Since mem2reg isn't run at -O0, we get a ton of reloads from the stack,
for example, before, this code:
int foo(int x, int y, int z) {
return x+y+z;
}
used to compile into:
_foo: ## @foo
subq $12, %rsp
movl %edi, 8(%rsp)
movl %esi, 4(%rsp)
movl %edx, (%rsp)
movl 8(%rsp), %edx
movl 4(%rsp), %esi
addl %edx, %esi
movl (%rsp), %edx
addl %esi, %edx
movl %edx, %eax
addq $12, %rsp
ret
Now we produce:
_foo: ## @foo
subq $12, %rsp
movl %edi, 8(%rsp)
movl %esi, 4(%rsp)
movl %edx, (%rsp)
movl 8(%rsp), %edx
addl 4(%rsp), %edx ## Folded load
addl (%rsp), %edx ## Folded load
movl %edx, %eax
addq $12, %rsp
ret
Fewer instructions and less register use = faster compiles.
llvm-svn: 113102
vabd intrinsic and add and/or zext operations. In the case of vaba, this
also avoids the need for a DAG combine pattern to combine vabd with add.
Update tests. Auto-upgrade the old intrinsics.
llvm-svn: 112941
- Add patterns to match the following MMX builtins:
* __builtin_ia32_vec_init_v8qi
* __builtin_ia32_vec_init_v4hi
* __builtin_ia32_vec_init_v2si
* __builtin_ia32_vec_ext_v2si
These builtins do not correspond to a single MMX instruction. They will have
to be lowered -- most likely in the back-end.
llvm-svn: 112881
I'm sure it is harmless. Original commit message:
If PrototypeValue is erased in the middle of using the SSAUpdator
then the SSAUpdator may access freed memory. Instead, simply pass
in the type and name explicitly, which is all that was used anyway.
llvm-svn: 112810
add, and subtract operations with zero-extended or sign-extended vectors.
Update tests. Add auto-upgrade support for the old intrinsics.
llvm-svn: 112773
moment, as there's a testcase that uses it and expects it
to be subject to optimizations; we won't be doing that.
Some adjustments based on feedback from Bill.
llvm-svn: 112754
of a base class.
This makes it possible to unregister the file from FilesToRemove when
the file is done. Also, this eliminates the need for
formatted_tool_output_file.
llvm-svn: 112706
available in normal llvm operators. We aren't going to
use those for MMX any more because it's unsafe for the
optimizers to synthesize new MMX instructions.
llvm-svn: 112685
and output the dwarf line number tables. This takes the current loc info after
an instruction is assembled and saves the needed info into an object that has
vector and for each section. These objects will be used for the final patch to
build and emit the encoded dwarf line number tables. Again for now this is only
in the Mach-O streamer but at some point will move to a more generic place.
llvm-svn: 112668
any more. I plan to reimplement alloca promotion using SSAUpdater later.
It looks like Bill's URoR logic really always needs domtree, so the pass
now always asks for domtree info.
llvm-svn: 112597
assertingvh so we get a violent explosion if the pointer dangles.
2) Fix AliasSetTracker::deleteValue to remove call sites with
by-pointer comparisons instead of by-alias queries. Using
findAliasSetForCallSite can cause alias sets to get merged
when they shouldn't, and can also miss alias sets when the
call is readonly.
#2 fixes PR6889, which only repros with a .c file :(
llvm-svn: 112452
since none of them use it. With this, we now only run
domfrontier (an N^2 analysis) 3 times at clang -O3: once for
"early" per-function cleanup, once at the start of the
per-function pipeline to support SRoA, and once late because
the EHPrepare class uses it.
EHPrepare needs to stop using it, this is silly and wasteful.
llvm-svn: 112420
fix: add a flag to MapValue and friends which indicates whether
any module-level mappings are being made. In the common case of
inlining, no module-level mappings are needed, so MapValue doesn't
need to examine non-function-local metadata, which can be very
expensive in the case of a large module with really deep metadata
(e.g. a large C++ program compiled with -g).
This flag is a little awkward; perhaps eventually it can be moved
into the ClonedCodeInfo class.
llvm-svn: 112190
I think there are good reasons to change this, but in the interests
of short-term stability, make SmallVector<...,0> reserve non-zero
capacity in its constructors. This means that SmallVector<...,0>
uses more memory than SmallVector<...,1> and should really only be
used (unless/until this workaround is removed) by clients that
care about using SmallVector with an incomplete type.
llvm-svn: 112147
expanding: e.g. <2 x float> -> <4 x float> instead of -> 2 floats. This
affects two places in the code: handling cross block values and handling
function return and arguments. Since vectors are already widened by
legalizetypes, this gives us much better code and unblocks x86-64 abi
and SPU abi work.
For example, this (which is a silly example of a cross-block value):
define <4 x float> @test2(<4 x float> %A) nounwind {
%B = shufflevector <4 x float> %A, <4 x float> undef, <2 x i32> <i32 0, i32 1>
%C = fadd <2 x float> %B, %B
br label %BB
BB:
%D = fadd <2 x float> %C, %C
%E = shufflevector <2 x float> %D, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
ret <4 x float> %E
}
Now compiles into:
_test2: ## @test2
## BB#0:
addps %xmm0, %xmm0
addps %xmm0, %xmm0
ret
previously it compiled into:
_test2: ## @test2
## BB#0:
addps %xmm0, %xmm0
pshufd $1, %xmm0, %xmm1
## kill: XMM0<def> XMM0<kill> XMM0<def>
insertps $0, %xmm0, %xmm0
insertps $16, %xmm1, %xmm0
addps %xmm0, %xmm0
ret
This implements rdar://8230384
llvm-svn: 112101
needed parsing for the .loc directive and saves the current info from that
into the context. The next patch will take the current loc info after an
instruction is assembled and save that info into a vector for each section for
use to build the line number tables. The patch after that will encode the info
from those vectors into the output file as the dwarf line tables.
llvm-svn: 111956
which does the same thing. This eliminates redundant code and
handles MDNodes better. MDNode linking still doesn't fully
work yet though.
llvm-svn: 111941
- Cache used characters in a bitset to reduce memory overhead to just 32 bytes.
- On my core2 this code is faster except when the checked string was very short
(smaller than the list of delimiters).
llvm-svn: 111817
general idea here is to have a group of x86 target specific nodes which are
going to be selected during lowering and then directly matched in isel.
The commit includes the addition of those specific nodes and a *bunch* of
patterns, and incrementally we're going to switch between them and what we
have right now. Both the patterns and target specific nodes can change as
we move forward with this work.
llvm-svn: 111691
It's similar to "linker_private_weak", but it's known that the address of the
object is not taken. For instance, functions that had an inline definition, but
the compiler decided not to inline it. Note, unlike linker_private and
linker_private_weak, linker_private_weak_def_auto may have only default
visibility. The symbols are removed by the linker from the final linked image
(executable or dynamic library).
llvm-svn: 111684
not part of the IR, are not uniqued, and may be safely RAUW'd.
This replaces a variety of alternate mechanisms for achieving
the same effect.
llvm-svn: 111681
functionality that most command-line tools need: ensuring that the
output file gets deleted if the tool is interrupted or encounters an
error.
llvm-svn: 111595
base registers were required. This will allow for slightly better packing
of the locals when alignment padding is necessary after callee saved registers.
llvm-svn: 111508
constructed with an output filename of "-". In particular, allow the
file descriptor to be closed, and close the file descriptor in the
destructor if it hasn't been explicitly closed already, to ensure
that any write errors are detected.
llvm-svn: 111436
