- Change register allocation hint to a pair of unsigned integers. The hint type is zero (which means prefer the register specified as second part of the pair) or entirely target dependent.
- Allow targets to specify alternative register allocation orders based on allocation hint.
Part 2.
- Use the register allocation hint system to implement more aggressive load / store multiple formation.
- Aggressively form LDRD / STRD. These are formed *before* register allocation. It has to be done this way to shorten live interval of base and offset registers. e.g.
v1025 = LDR v1024, 0
v1026 = LDR v1024, 0
=>
v1025,v1026 = LDRD v1024, 0
If this transformation isn't done before allocation, v1024 will overlap v1025 which means it more difficult to allocate a register pair.
- Even with the register allocation hint, it may not be possible to get the desired allocation. In that case, the post-allocation load / store multiple pass must fix the ldrd / strd instructions. They can either become ldm / stm instructions or back to a pair of ldr / str instructions.
This is work in progress, not yet enabled.
llvm-svn: 73381
is that, for functions whose bodies are entirely guarded by an if-statement, it
can be profitable to pull the test out of the callee and into the caller.
This code has had some cursory testing, but still has a number of known issues
on the LLVM test suite.
llvm-svn: 73338
Emission for globals, using the correct data sections
Function alignment can be computed for each target using TargetELFWriterInfo
Some small fixes
llvm-svn: 73201
This changes the IndexedModeAction representation to remove the
limitation on the number of value types in MVT. This limitation
prevents us from specifying AVX types.
Prior to this change IndexedModActions was represented as follows...
uint64_t IndexedModeActions[2][ISD::LAST_INDEXED_MODE];
the first dimension was used to represent loads, then stores. This
imposed a limitation of 32 on the number of value types that could be
handled with this method. The value type was used to shift the two bits
into and out of the approprate bits in the uint64_t.
With this change the array is now represented as ...
uint8_t IndexedModeActions[MVT::LAST_VALUETYPE][2][ISD::LAST_INDEXED_MODE];
Takes more space but removes the limitation on MVT::LAST_VALUETYPE. The
first dimension is now the value_type for the reference. The second
dimension is the load [0] vs. store[1]. The third dimension represents
the various modes for load store. Accesses are now direct, no shifting
or masking.
There are other limitations that need to be removed, so that
MVT::LAST_VALUETYPE can be greater than 32. This is merely the first
step towards that goal.
llvm-svn: 73104
This changes the IndexedModeAction representation to remove the
limitation on the number of value types in MVT. This limitation
prevents us from specifying AVX types.
Prior to this change IndexedModActions was represented as follows...
uint64_t IndexedModeActions[2][ISD::LAST_INDEXED_MODE];
the first dimension was used to represent loads, then stores. This
imposed a limitation of 32 on the number of value types that could be
handled with this method. The value type was used to shift the two bits
into and out of the approprate bits in the uint64_t.
With this change the array is now represented as ...
uint8_t IndexedModeActions[MVT::LAST_VALUETYPE][2][ISD::LAST_INDEXED_MODE];
Takes more space but removes the limitation on MVT::LAST_VALUETYPE. The
first dimension is now the value_type for the reference. The second
dimension is the load [0] vs. store[1]. The third dimension represents
the various modes for load store. Accesses are now direct, no shifting
or masking.
There are other limitations that need to be removed, so that
MVT::LAST_VALUETYPE can be greater than 32. This is merely the first
step towards that goal.
llvm-svn: 73102
other operators. For the rare cases where a list type cannot be
deduced, provide a []<type> syntax, where <type> is the list element
type.
llvm-svn: 73078
Also create isValidElementType for ArrayType, PointerType, StructType and
VectorType.
Make LLParser use them. This closes up some holes like an assertion failure on:
%x = type {label}
but largely doesn't change any semantics. The only thing we accept now which
we didn't before is vectors of opaque type such as "<4 x opaque>". The opaque
can be resolved to an int or float when linking.
llvm-svn: 73016
Update code generator to use this attribute and remove NoImplicitFloat target option.
Update llc to set this attribute when -no-implicit-float command line option is used.
llvm-svn: 72959
build vectors with i64 elements will only appear on 32b x86 before legalize.
Since vector widening occurs during legalize, and produces i64 build_vector
elements, the dag combiner is never run on these before legalize splits them
into 32b elements.
Teach the build_vector dag combine in x86 back end to recognize consecutive
loads producing the low part of the vector.
Convert the two uses of TLI's consecutive load recognizer to pass LoadSDNodes
since that was required implicitly.
Add a testcase for the transform.
Old:
subl $28, %esp
movl 32(%esp), %eax
movl 4(%eax), %ecx
movl %ecx, 4(%esp)
movl (%eax), %eax
movl %eax, (%esp)
movaps (%esp), %xmm0
pmovzxwd %xmm0, %xmm0
movl 36(%esp), %eax
movaps %xmm0, (%eax)
addl $28, %esp
ret
New:
movl 4(%esp), %eax
pmovzxwd (%eax), %xmm0
movl 8(%esp), %eax
movaps %xmm0, (%eax)
ret
llvm-svn: 72957
integer and floating-point opcodes, introducing
FAdd, FSub, and FMul.
For now, the AsmParser, BitcodeReader, and IRBuilder all preserve
backwards compatability, and the Core LLVM APIs preserve backwards
compatibility for IR producers. Most front-ends won't need to change
immediately.
This implements the first step of the plan outlined here:
http://nondot.org/sabre/LLVMNotes/IntegerOverflow.txt
llvm-svn: 72897
Update code generator to use this attribute and remove DisableRedZone target option.
Update llc to set this attribute when -disable-red-zone command line option is used.
llvm-svn: 72894
Add lto_codegen_set_assembler_path() API which allows the linker to specify the
path to the assembler tool to run. When assembler is used (instead of compiler)
different command line options are used.
Add LTO_API_VERSION #define so clients (linkers) can conditionalize use of new APIs.
llvm-svn: 72823
- This matches llvm-ld.
It took a bit of archeology to figure out what the right thing to do was
(whether this was intentionally added or intentionally removed). My final
conclusion is that Chris added this intentionally here:
http://llvm.org/viewvc/llvm-project?view=rev&revision=16913
but the changes weren't propogated to llvm-ld until here:
http://llvm.org/viewvc/llvm-project?view=rev&revision=34058
which was after lto.cpp had been cloned off (of llvm-ld), here:
http://llvm.org/viewvc/llvm-project?view=rev&revision=29494
From the commit message, it looks like the motivation for running global opt
again is because we ran it prior to inlining. Based on that I updated the
comment and also only run the pass if we actually ran the inliner.
Chris, please review.
llvm-svn: 72811
TargetData pointer. The only thing it's used for are
calls to ConstantFoldCompareInstOperands and
ConstantFoldInstOperands, which both already accept a
null TargetData pointer. This makes
ConstantFoldConstantExpression easier to use in clients
where TargetData is optional.
llvm-svn: 72741
ADDC/ADDE use MVT::i1 (later, whatever it gets legalized to)
instead of MVT::Flag. Remove CARRY_FALSE in favor of 0; adjust
all target-independent code to use this format.
Most targets will still produce a Flag-setting target-dependent
version when selection is done. X86 is converted to use i32
instead, which means TableGen needs to produce different code
in xxxGenDAGISel.inc. This keys off the new supportsHasI1 bit
in xxxInstrInfo, currently set only for X86; in principle this
is temporary and should go away when all other targets have
been converted. All relevant X86 instruction patterns are
modified to represent setting and using EFLAGS explicitly. The
same can be done on other targets.
The immediate behavior change is that an ADC/ADD pair are no
longer tightly coupled in the X86 scheduler; they can be
separated by instructions that don't clobber the flags (MOV).
I will soon add some peephole optimizations based on using
other instructions that set the flags to feed into ADC.
llvm-svn: 72707
e.g.
orl $65536, 8(%rax)
=>
orb $1, 10(%rax)
Since narrowing is not always a win, e.g. i32 -> i16 is a loss on x86, dag combiner consults with the target before performing the optimization.
llvm-svn: 72507
entries as there are basic blocks in the function. LiveVariables::getVarInfo
creates a VarInfo struct for every register in the function, leading to
quadratic space use. This patch changes the BitVector to a SparseBitVector,
which doesn't help the worst-case memory use but does reduce the actual use in
very long functions with short-lived variables.
llvm-svn: 72426
in the case where a loop exit value cannot be computed, instead of only in
some cases while using SCEVCouldNotCompute in others. This simplifies
getSCEVAtScope's callers.
llvm-svn: 72375
sending SCEVUnknowns to expandAddToGEP. This avoids the need for
expandAddToGEP to bend the rules and peek into SCEVUnknown
expressions.
Factor out the code for testing whether a SCEV can be factored by
a constant for use in a GEP index. This allows it to handle
SCEVAddRecExprs, by recursing.
As a result, SCEVExpander can now put more things in GEP indices,
so it emits fewer explicit mul instructions.
llvm-svn: 72366
Fix by clearing the rewriter cache before deleting the trivially dead
instructions.
Also make InsertedExpressions use an AssertingVH to catch these
bugs easier.
llvm-svn: 72364
and it wasn't generating calls through @PLT for these functions.
hasLocalLinkage() is now false for available_externally,
I attempted to fix the inliner and dce to handle available_externally properly.
It passed make check.
llvm-svn: 72328
will allow simplifying LegalizeDAG to eliminate type legalization. (I
have a patch to do that, but it's not quite finished; I'll commit it
once it's finished and I've fixed any review comments for this patch.)
See the comment at the beginning of
lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp for more details on the
motivation for this patch.
llvm-svn: 72325
code in preparation for code generation. The main thing it does
is handle the case when eh.exception calls (and, in a future
patch, eh.selector calls) are far away from landing pads. Right
now in practice you only find eh.exception calls close to landing
pads: either in a landing pad (the common case) or in a landing
pad successor, due to loop passes shifting them about. However
future exception handling improvements will result in calls far
from landing pads:
(1) Inlining of rewinds. Consider the following case:
In function @f:
...
invoke @g to label %normal unwind label %unwinds
...
unwinds:
%ex = call i8* @llvm.eh.exception()
...
In function @g:
...
invoke @something to label %continue unwind label %handler
...
handler:
%ex = call i8* @llvm.eh.exception()
... perform cleanups ...
"rethrow exception"
Now inline @g into @f. Currently this is turned into:
In function @f:
...
invoke @something to label %continue unwind label %handler
...
handler:
%ex = call i8* @llvm.eh.exception()
... perform cleanups ...
invoke "rethrow exception" to label %normal unwind label %unwinds
unwinds:
%ex = call i8* @llvm.eh.exception()
...
However we would like to simplify invoke of "rethrow exception" into
a branch to the %unwinds label. Then %unwinds is no longer a landing
pad, and the eh.exception call there is then far away from any landing
pads.
(2) Using the unwind instruction for cleanups.
It would be nice to have codegen handle the following case:
invoke @something to label %continue unwind label %run_cleanups
...
handler:
... perform cleanups ...
unwind
This requires turning "unwind" into a library call, which
necessarily takes a pointer to the exception as an argument
(this patch also does this unwind lowering). But that means
you are using eh.exception again far from a landing pad.
(3) Bugpoint simplifications. When bugpoint is simplifying
exception handling code it often generates eh.exception calls
far from a landing pad, which then causes codegen to assert.
Bugpoint then latches on to this assertion and loses sight
of the original problem.
Note that it is currently rare for this pass to actually do
anything. And in fact it normally shouldn't do anything at
all given the code coming out of llvm-gcc! But it does fire
a few times in the testsuite. As far as I can see this is
almost always due to the LoopStrengthReduce codegen pass
introducing pointless loop preheader blocks which are landing
pads and only contain a branch to another block. This other
block contains an eh.exception call. So probably by tweaking
LoopStrengthReduce a bit this can be avoided.
llvm-svn: 72276
If this causes any new assertion failures that I didn't catch in
testing, the fix is usually to change "&v[0]" to "v.data()" for some
SmallVector v.
llvm-svn: 72221
type as a target independent constant expression. I confess
that I didn't check that this method works as intended (though
I did test the equivalent hand-written IR a little). But what
could possibly go wrong!
llvm-svn: 72213
mutex support. LLVM_MULTITHREADED indicates (or will indicate) the ability to run LLVM itself across multiple threads, and requires atomics support.
llvm-svn: 72140
instructions. It attempts to create high-level multi-operand GEPs,
though in cases where this isn't possible it falls back to casting
the pointer to i8* and emitting a GEP with that. Using GEP instructions
instead of ptrtoint+arithmetic+inttoptr helps pointer analyses that
don't use ScalarEvolution, such as BasicAliasAnalysis.
Also, make the AddrModeMatcher more aggressive in handling GEPs.
Previously it assumed that operand 0 of a GEP would require a register
in almost all cases. It now does extra checking and can do more
matching if operand 0 of the GEP is foldable. This fixes a problem
that was exposed by SCEVExpander using GEPs.
llvm-svn: 72093
Introduce a new class (MachineCodeInfo) that the JIT can fill in with details. Right now, just the address and the size of the machine code are reported.
Patch by Evan Phoenix!
llvm-svn: 72040
The following is checked:
* Operand counts: All explicit operands must be present.
* Register classes: All physical and virtual register operands must be
compatible with the register class required by the instruction descriptor.
* Register live intervals: Registers must be defined only once, and must be
defined before use.
The machine code verifier is enabled with the command-line option
'-verify-machineinstrs', or by defining the environment variable
LLVM_VERIFY_MACHINEINSTRS to the name of a file that will receive all the
verifier errors.
llvm-svn: 71918
to low-level sync operations.
The only one present at the moment is MemoryFence(), and only for the platforms
for which I could easily discern the proper way to do it. If your favorite platform
isn't represented, patches are welcome!
llvm-svn: 71770
llvm.eh.sjlj.* for better clarity as to their purpose and scope. Add
a description of llvm.eh.sjlj.setjmp to ExceptionHandling.html.
(llvm.eh.sjlj.longjmp documentation coming when that implementation is
added).
llvm-svn: 71758
of exception handling builtin sjlj targets in functions turns out not to
be necessary. Marking the intrinsic implementation in the .td file as
defining all registers is sufficient to get the context saved properly by
the containing function.
llvm-svn: 71743
booleans. This gives a better indication of what the "addReg()" is
doing. Remembering what all of those booleans mean isn't easy, especially if you
aren't spending all of your time in that code.
I took Jakob's suggestion and made it illegal to pass in "true" for the
flag. This should hopefully prevent any unintended misuse of this (by reverting
to the old way of using addReg()).
llvm-svn: 71722
getNoopOrSignExtend, and getTruncateOrNoop. These are similar
to getTruncateOrZeroExtend etc., except that they assert that
the conversion is either not widening or narrowing, as
appropriate. These will be used in some upcoming fixes.
llvm-svn: 71632
without one. Use it where we were using abs on
int64_t objects.
(I strongly suspect the casts to unsigned in the
fragments in LoopStrengthReduce are not doing whatever
the original intent was, but the obvious change to
uint64_t doesn't work. Maybe later.)
llvm-svn: 71612
a supporting preliminary patch for GCC-compatible SjLJ exception handling. Note that these intrinsics are not designed to be invoked directly by the user, but
rather used by the front-end as target hooks for exception handling.
llvm-svn: 71610
and generalize it so that it can be used by IndVarSimplify. Implement the
base IndVarSimplify transformation code using IVUsers. This removes
TestOrigIVForWrap and associated code, as ScalarEvolution now has enough
builtin overflow detection and folding logic to handle all the same cases,
and more. Run "opt -iv-users -analyze -disable-output" on your favorite
loop for an example of what IVUsers does.
This lets IndVarSimplify eliminate IV casts and compute trip counts in
more cases. Also, this happens to finally fix the remaining testcases
in PR1301.
Now that IndVarSimplify is being more aggressive, it occasionally runs
into the problem where ScalarEvolutionExpander's code for avoiding
duplicate expansions makes it difficult to ensure that all expanded
instructions dominate all the instructions that will use them. As a
temporary measure, IndVarSimplify now uses a FixUsesBeforeDefs function
to fix up instructions inserted by SCEVExpander. Fortunately, this code
is contained, and can be easily removed once a more comprehensive
solution is available.
llvm-svn: 71535
- reduces _static_ callee saved register spills
and restores similar to Chow's original algorithm.
- iterative implementation with simple heuristic
limits to mitigate compile time impact.
- handles placing spills/restores for multi-entry,
multi-exit regions in the Machine CFG without
splitting edges.
- passes test-suite in LLCBETA mode.
Added contains() method to ADT/SparseBitVector.
llvm-svn: 71438
which are not analyzed with SCEV techniques, which can require
brute-forcing through a large number of instructions. This
fixes a massive compile-time issue on 400.perlbench (in
particular, the loop in MD5Transform).
llvm-svn: 71259
checking for bcopy... no
checking for getc_unlocked... Assertion failed: (0 && "Unknown SCEV kind!"), function operator(), file /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/src/lib/Analysis/ScalarEvolution.cpp, line 511.
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~obj/src/libdecnumber/decUtility.c:360: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[4]: *** [decUtility.o] Error 1
make[4]: *** Waiting for unfinished jobs....
Assertion failed: (0 && "Unknown SCEV kind!"), function operator(), file /Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmCore.roots/llvmCore~obj/src/lib/Analysis/ScalarEvolution.cpp, line 511.
/Volumes/Sandbox/Buildbot/llvm/full-llvm/build/llvmgcc42.roots/llvmgcc42~obj/src/libdecnumber/decNumber.c:5591: internal compiler error: Abort trap
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL:http://developer.apple.com/bugreporter> for instructions.
make[4]: *** [decNumber.o] Error 1
make[3]: *** [all-stage2-libdecnumber] Error 2
make[3]: *** Waiting for unfinished jobs....
llvm-svn: 71165
CallbackVH, with fixes. allUsesReplacedWith need to
walk the def-use chains and invalidate all users of a
value that is replaced. SCEVs of users need to be
recalcualted even if the new value is equivalent. Also,
make forgetLoopPHIs walk def-use chains, since any
SCEV that depends on a PHI should be recalculated when
more information about that PHI becomes available.
llvm-svn: 70927
shows up in -print-machineinstrs. This doesn't appear to affect anything, but it was
weird for some DBG_LABELs to have DebugLocs but not all of them.
llvm-svn: 70921
VirtRegMap keeps track of allocations so it knows what's not used. As a horrible hack, the stack coloring can color spill slots with *free* registers. That is, it replace reload and spills with copies from and to the free register. It unfold instructions that load and store the spill slot and replace them with register using variants.
Not yet enabled. This is part 1. More coming.
llvm-svn: 70787
makes ScalarEvolution::deleteValueFromRecords, and it's code that
subtly needed to be called before ReplaceAllUsesWith, unnecessary.
It also makes ValueDeletionListener unnecessary.
llvm-svn: 70645
of returning a list of pointers to Values that are deleted. This was
unsafe, because the pointers in the list are, by nature of what
RecursivelyDeleteDeadInstructions does, always dangling. Replace this
with a simple callback mechanism. This may eventually be removed if
all clients can reasonably be expected to use CallbackVH.
Use this to factor out the dead-phi-cycle-elimination code from LSR
utility function, and generalize it to use the
RecursivelyDeleteTriviallyDeadInstructions utility function.
This makes LSR more aggressive about eliminating dead PHI cycles;
adjust tests to either be less trivial or to simply expect fewer
instructions.
llvm-svn: 70636
-Replace DebugLocTuple's Source ID with CompileUnit's GlobalVariable*
-Remove DwarfWriter::getOrCreateSourceID
-Make necessary changes for the above (fix callsites, etc.)
llvm-svn: 70520
compute an upper-bound value for the trip count, in addition to
the actual trip count. Use this to allow getZeroExtendExpr and
getSignExtendExpr to fold casts in more cases.
This may eventually morph into a more general value-range
analysis capability; there are certainly plenty of places where
more complete value-range information would allow more folding.
llvm-svn: 70509
Massive check in. This changes the "-fast" flag to "-O#" in llc. If you want to
use the old behavior, the flag is -O0. This change allows for finer-grained
control over which optimizations are run at different -O levels.
Most of this work was pretty mechanical. The majority of the fixes came from
verifying that a "fast" variable wasn't used anymore. The JIT still uses a
"Fast" flag. I'll change the JIT with a follow-up patch.
llvm-svn: 70343
use the old behavior, the flag is -O0. This change allows for finer-grained
control over which optimizations are run at different -O levels.
Most of this work was pretty mechanical. The majority of the fixes came from
verifying that a "fast" variable wasn't used anymore. The JIT still uses a
"Fast" flag. I'm not 100% sure if it's necessary to change it there...
llvm-svn: 70270
the comparison operators were not only unnecessary in the presence of the
implicit conversion; they caused ambiguous overload errors. So I deleted them.
llvm-svn: 70243
PR2957
ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask. A value of -1 represents UNDEF.
In addition to eliminating the creation of illegal BUILD_VECTORS just to
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.
llvm-svn: 70225
BLOCKNAME and SETRECORDNAME. This allows a bitcode
file to be self describing with pretty names for
records and blocks in addition to numbers. This
enhances llvm-bcanalyzer to use this to print prettily.
llvm-svn: 70165
state out of the BitstreamReader class into a BitstreamCursor class.
Doing this allows the client to have multiple cursors into the same
file, each with potentially different live block stacks and
abbreviation records.
llvm-svn: 70157
true), and casts make me nervous and are verbose anyway, so here's a
ConstantInt::getSigned(Ty, int64_t) method. Just overloading
ConstantInt::get() to take an int64_t too would cause ambiguous
overload errors."
Patch by Jeffrey Yasskin!
llvm-svn: 69958
ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask. A value of -1 represents UNDEF.
In addition to eliminating the creation of illegal BUILD_VECTORS just to
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.
A clean up of x86 shuffle code, and some canonicalizing in DAGCombiner is next.
llvm-svn: 69952
with the persistent insertion point, and change IndVars to make
use of it. This fixes a bug where IndVars was holding on to a
stale insertion point and forcing the SCEVExpander to continue to
use it.
This fixes PR4038.
llvm-svn: 69892
This fixes a very subtle bug. vr defined by an implicit_def is allowed overlap with any register since it doesn't actually modify anything. However, if it's used as a two-address use, its live range can be extended and it can be spilled. The spiller must take care not to emit a reload for the vn number that's defined by the implicit_def. This is both a correctness and performance issue.
llvm-svn: 69743
in the MachineFunction class, renaming it to addLiveIn for consistency with
the same method in MachineBasicBlock. Thanks for Anton for suggesting this.
llvm-svn: 69615
