This came up when a change in block placement formed a cmov and slowed down a
hot loop by 50%:
ucomisd (%rdi), %xmm0
cmovbel %edx, %esi
cmov is a really bad choice in this context because it doesn't get branch
prediction. If we emit it as a branch, an out-of-order CPU can do a better job
(if the branch is predicted right) and avoid waiting for the slow load+compare
instruction to finish. Of course it won't help if the branch is unpredictable,
but those are really rare in practice.
This patch uses a dumb conservative heuristic, it turns all cmovs that have one
use and a direct memory operand into branches. cmovs usually save some code
size, so we disable the transform in -Os mode. In-Order architectures are
unlikely to benefit as well, those are included in the
"predictableSelectIsExpensive" flag.
It would be better to reuse branch probability info here, but BPI doesn't
support select instructions currently. It would make sense to use the same
heuristics as the if-converter pass, which does the opposite direction of this
transform.
Test suite shows a small improvement here and there on corei7-level machines,
but the actual results depend a lot on the used microarchitecture. The
transformation is currently disabled by default and available by passing the
-enable-cgp-select2branch flag to the code generator.
Thanks to Chandler for the initial test case to him and Evan Cheng for providing
me with comments and test-suite numbers that were more stable than mine :)
llvm-svn: 156234
This will be used to determine whether it's profitable to turn a select into a
branch when the branch is likely to be predicted.
Currently enabled for everything but Atom on X86 and Cortex-A9 devices on ARM.
I'm not entirely happy with the name of this flag, suggestions welcome ;)
llvm-svn: 156233
We want the representative register class to contain the largest
super-registers available. This makes the function less sensitive to the
register class numbering.
llvm-svn: 156220
In file included from ../lib/Target/NVPTX/VectorElementize.cpp:53:
../lib/Target/NVPTX/NVPTX.h:44:3: warning: default label in switch which covers all enumeration values [-Wcovered-switch-default]
default: assert(0 && "Unknown condition code");
^
1 warning generated.
The prevailing pattern in LLVM is to not use a default label, and instead to
use llvm_unreachable to denote that the switch in fact covers all return paths
from the function.
llvm-svn: 156209
add a new Region::block_iterator which actually iterates over the basic
blocks of the region.
The old iterator, now call 'block_node_iterator' iterates over
RegionNodes which contain a single basic block. This works well with the
GraphTraits-based iterator design, however most users actually want an
iterator over the BasicBlocks inside these RegionNodes. Now the
'block_iterator' is a wrapper which exposes exactly this interface.
Internally it uses the block_node_iterator to walk all nodes which are
single basic blocks, but transparently unwraps the basic block to make
user code simpler.
While this patch is a bit of a wash, most of the updates are to internal
users, not external users of the RegionInfo. I have an accompanying
patch to Polly that is a strict simplification of every user of this
interface, and I'm working on a pass that also wants the same simplified
interface.
This patch alone should have no functional impact.
llvm-svn: 156202
The new target machines are:
nvptx (old ptx32) => 32-bit PTX
nvptx64 (old ptx64) => 64-bit PTX
The sources are based on the internal NVIDIA NVPTX back-end, and
contain more functionality than the current PTX back-end currently
provides.
NV_CONTRIB
llvm-svn: 156196
of the CodeExtractor utility. This allows speculatively computing input
and output sets to measure the likely size impact of the code
extraction.
These sets cannot be reused sadly -- we mutate the function prior to
forming the final sets used by the actual extraction.
The interface has been revamped slightly to make it easier to use
correctly by making the interface const and sinking the computation of
the number of exit blocks into the full extraction function and away
from the rest of this logic which just computed two output parameters.
llvm-svn: 156168
and expose it as a utility class rather than as free function wrappers.
The simple free-function interface works well for the bugpoint-specific
pass's uses of code extraction, but in an upcoming patch for more
advanced code extraction, they simply don't expose a rich enough
interface. I need to expose various stages of the process of doing the
code extraction and query information to decide whether or not to
actually complete the extraction or give up.
Rather than build up a new predicate model and pass that into these
functions, just take the class that was actually implementing the
functions and lift it up into a proper interface that can be used to
perform code extraction. The interface is cleaned up and re-documented
to work better in a header. It also is now setup to accept the blocks to
be extracted in the constructor rather than in a method.
In passing this essentially reverts my previous commit here exposing
a block-level query for eligibility of extraction. That is no longer
necessary with the more rich interface as clients can query the
extraction object for eligibility directly. This will reduce the number
of walks of the input basic block sequence by quite a bit which is
useful if this enters the normal optimization pipeline.
llvm-svn: 156163
This moves the logic for selecting a TLS model to a single place,
instead of the previous three (ARM, Mips, and X86 which already
uses this function).
llvm-svn: 156162
The masks returned by SuperRegClassIterator are computed automatically
by TableGen. This is better than depending on the manually specified
SuperRegClasses.
llvm-svn: 156147
This iterator class provides a more abstract interface to the (Idx,
Mask) lists of super-registers for a register class. The layout of the
tables shouldn't be exposed to clients.
llvm-svn: 156144
minor behavior changes with this, but nothing I have seen evidence of in
the wild or expect to be meaningful. The real goal is unifying our logic
and simplifying the interfaces. A summary of the changes follows:
- Make 'callIsSmall' actually accept a callsite so it can handle
intrinsics, and simplify callers appropriately.
- Nuke a completely bogus declaration of 'callIsSmall' that was still
lurking in InlineCost.h... No idea how this got missed.
- Teach the 'isInstructionFree' about the various more intelligent
'free' heuristics that got added to the inline cost analysis during
review and testing. This mostly surrounds int->ptr and ptr->int casts.
- Switch most of the interesting parts of the inline cost analysis that
were essentially computing 'is this instruction free?' to use the code
metrics routine instead. This way we won't keep duplicating logic.
All of this is motivated by the desire to allow other passes to compute
a roughly equivalent 'cost' metric for a particular basic block as the
inline cost analysis. Sadly, re-using the same analysis for both is
really messy because only the actual inline cost analysis is ever going
to go to the contortions required for simplification, SROA analysis,
etc.
llvm-svn: 156140
for the assembler and disassembler. Which were not being set/read correctly
for offsets greater than 22 bits in some cases.
Changes to lib/Target/ARM/ARMAsmBackend.cpp from Gideon Myles!
llvm-svn: 156118
extraction into a public interface. Also clean it up and apply it more
consistently such that we check for landing pads *anywhere* in the
extracted code, not just in single-block extraction.
This will be used to guide decisions in passes that are planning to
eventually perform a round of code extraction.
llvm-svn: 156114
to catch cases like:
%reg1024<def> = MOV r1
%reg1025<def> = MOV r0
%reg1026<def> = ADD %reg1024, %reg1025
r0 = MOV %reg1026
By commuting ADD, it let coalescer eliminate all of the copies. However, there
was a bug in the heuristics where it ended up commuting the ADD in:
%reg1024<def> = MOV r0
%reg1025<def> = MOV 0
%reg1026<def> = ADD %reg1024, %reg1025
r0 = MOV %reg1026
That did no benefit but rather ensure the last MOV would not be coalesced.
rdar://11355268
llvm-svn: 156048
The ensures that virtual registers always belong to an allocatable class.
If your target attempts to create a vreg for an operand that has no
allocatable register subclass, you will crash quickly.
This ensures that targets define register classes as intended.
llvm-svn: 156046
Expressions for movw/movt don't always have an :upper16: or :lower16:
on them and that's ok. When they don't, it's just a plain [0-65536]
immediate result, effectively the same as a :lower16: variant kind.
rdar://10550147
llvm-svn: 155941
in order to avoid assertion failures in the register scavenger. The assertion
failures were “Bad machine code: Using an undefined physical register” and
“Bad machine code: MBB exits via unconditional fall-through but its successor
differs from its CFG successor!”.
llvm-svn: 155930
Previously, an unsupported/unknown assembler directive issued a warning.
That's generally unsafe, and inconsistent with the behaviour of pretty
much every system assembler. Now that the MC assemblers are mature
enough to be the default on multiple targets, it's reasonable to
issue errors for these.
For target or platform directives that need to stay warnings, we
should add explicit handlers for them in, e.g., ELFAsmParser.cpp,
DarwinAsmParser.cpp, et. al., and issue the warning there.
rdar://9246275
llvm-svn: 155926
The caller is already responsible for eating any additional input on the
line. Putting an additional EatToEndOfStatement() in ParseStatement()
causes an entire extra statement to be consumed when treating warnings
as errors. For example, test/MC/macros.s will assert() because the
.endmacro directive is missed as a result.
rdar://11355843
llvm-svn: 155925
This patch will optimize the following cases on X86
(a > b) ? (a-b) : 0
(a >= b) ? (a-b) : 0
(b < a) ? (a-b) : 0
(b <= a) ? (a-b) : 0
FROM
movl %edi, %ecx
subl %esi, %ecx
cmpl %edi, %esi
movl $0, %eax
cmovll %ecx, %eax
TO
xorl %eax, %eax
subl %esi, %edi
cmovll %eax, %edi
movl %edi, %eax
rdar: 10734411
llvm-svn: 155919
- Improved parameter names for clarity
- Added comments
- emitCommonSymbols should return void because its return value is not being
used anywhere
- Attempt to reduce the usage of the RelocationValueRef type. Restricts it
for a single goal and may serve as a step for eventual removal.
llvm-svn: 155908
The TargetPassManager's default constructor wants to initialize the PassManager
to 'null'. But it's illegal to bind a null reference to a null l-value. Make the
ivar a pointer instead.
PR12468
llvm-svn: 155902
- There's no point having a different type for the local and global symbol
tables.
- Renamed SymbolTable to GlobalSymbolTable to clarify the intention
- Improved const correctness where relevant
llvm-svn: 155898
<rdar://problem/11291436>.
This is a second attempt at a fix for this, the first was r155468. Thanks
to Chandler, Bob and others for the feedback that helped me improve this.
llvm-svn: 155866
Replace some assert() calls w/ actual diagnostics. In a perfect world,
there'd be range checks on these values long before things ever reached
this code. For now, though, issuing a better-late-than-never diagnostic
is still a big improvement over assert().
rdar://11347287
llvm-svn: 155851
This was exposed by SingleSource/UnitTests/Vector/constpool.c.
The computed size of a basic block isn't always a multiple of its known
alignment, and that can introduce extra alignment padding after the
block.
<rdar://problem/11347135>
llvm-svn: 155845
On x86-32, structure return via sret lets the callee pop the hidden
pointer argument off the stack, which the caller then re-pushes.
However if the calling convention is fastcc, then a register is used
instead, and the caller should not adjust the stack. This is
implemented with a check of IsTailCallConvention
X86TargetLowering::LowerCall but is now checked properly in
X86FastISel::DoSelectCall.
(this time, actually commit what was reviewed!)
llvm-svn: 155825
ARM BUILD_VECTORs created after type legalization cannot use i8 or i16
operands, since those types are not legal. Instead use i32 operands, which
will be implicitly truncated by the BUILD_VECTOR to match the element type.
llvm-svn: 155824
relocations are resolved. It's much more reasonable to do this decision when
relocations are just being added - we have all the information at that point.
Also a bit of renaming and extra comments to clarify extensions.
llvm-svn: 155819
Allow the "SplitCriticalEdge" function to split the edge to a landing pad. If
the pass is *sure* that it thinks it knows what it's doing, then it may go ahead
and specify that the landing pad can have its critical edge split. The loop
unswitch pass is one of these passes. It will split the critical edges of all
edges coming from a loop to a landing pad not within the loop. Doing so will
retain important loop analysis information, such as loop simplify.
llvm-svn: 155817
- Add comments
- Change field names to be more reasonable
- Fix indentation and naming to conform to coding conventions
- Remove unnecessary includes / replace them by forward declatations
llvm-svn: 155815
The code could search past the end of the basic block when there was
already a constant pool entry after the block.
Test case with giant basic block in SingleSource/UnitTests/Vector/constpool.c
llvm-svn: 155753
This time, also fix the caller of AddGlue to properly handle
incomplete chains. AddGlue had failure modes, but shamefully hid them
from its caller. It's luck ran out.
Fixes rdar://11314175: BuildSchedUnits assert.
llvm-svn: 155749
Make sure when parsing the Thumb1 sp+register ADD instruction that
the source and destination operands match. In thumb2, just use the
wide encoding if they don't. In Thumb1, issue a diagnostic.
rdar://11219154
llvm-svn: 155748
On x86-32, structure return via sret lets the callee pop the hidden
pointer argument off the stack, which the caller then re-pushes.
However if the calling convention is fastcc, then a register is used
instead, and the caller should not adjust the stack. This is
implemented with a check of IsTailCallConvention
X86TargetLowering::LowerCall but is now checked properly in
X86FastISel::DoSelectCall.
llvm-svn: 155745
Previously, ARMConstantIslandPass would conservatively compute the
address of an aligned basic block as:
RoundUpToAlignment(Offset + UnknownPadding)
This worked fine for the layout algorithm itself, but it could fool the
verify() function because it accounts for alignment padding twice: Once
when adding the worst case UnknownPadding, and again by rounding up the
fictional block offset. This meant that when optimizeThumb2Instructions
would shrink an instruction, the conservative distance estimate could
grow. That shouldn't be possible since the woorst case alignment padding
wss already included.
This patch drops the use of RoundUpToAlignment, and depends only on
worst case padding to compute conservative block offsets. This has the
weird effect that the computed offset for an aligned block may not be
aligned.
The important difference is that shrinking an instruction can never
cause the estimated distance between two instructions to grow. The
estimated distance is always larger than the real distance that only the
assembler knows.
<rdar://problem/11339352>
llvm-svn: 155744
x == -y --> x+y == 0
x != -y --> x+y != 0
On x86, the generated code goes from
negl %esi
cmpl %esi, %edi
je .LBB0_2
to
addl %esi, %edi
je .L4
This case is correctly handled for ARM with "cmn".
Patch by Manman Ren.
rdar://11245199
PR12545
llvm-svn: 155739
Target specific types should not be vectorized. As a practical matter,
these types are already register matched (at least in the x86 case),
and codegen does not always work correctly (at least in the ppc case,
and this is not worth fixing because ppc_fp128 is currently broken and
will probably go away soon).
llvm-svn: 155729
* Model FPSW (the FPU status word) as a register.
* Add ISel patterns for the FUCOM*, FNSTSW and SAHF instructions.
* During Legalize/Lowering, build a node sequence to transfer the comparison
result from FPSW into EFLAGS. If you're wondering about the right-shift: That's
an implicit sub-register extraction (%ax -> %ah) which is handled later on by
the instruction selector.
Fixes PR6679. Patch by Christoph Erhardt!
llvm-svn: 155704
instead of getAggregateElement. This has the advantage of being
more consistent and allowing higher-level constant folding to
procede even if an inner extract element cannot be folded.
Make ConstantFoldInstruction call ConstantFoldConstantExpression
on the instruction's operands, making it more consistent with
ConstantFoldConstantExpression itself. This makes sure that
ConstantExprs get TargetData-aware folding before being handed
off as operands for further folding.
This causes more expressions to be folded, but due to a known
shortcoming in constant folding, this currently has the side effect
of stripping a few more nuw and inbounds flags in the non-targetdata
side of constant-fold-gep.ll. This is mostly harmless.
This fixes rdar://11324230.
llvm-svn: 155682
The required checks are moved to ChainInstruction() itself and the
policy decisions are moved to IVChain::isProfitableInc().
Also cache the ExprBase in IVChain to avoid frequent recomputations.
No functional change intended.
llvm-svn: 155676
DAGCombine strangeness may result in multiple loads from the same
offset. They both may try to glue themselves to another load. We could
insist that the redundant loads glue themselves to each other, but the
beter fix is to bail out from bad gluing at the time we detect it.
Fixes rdar://11314175: BuildSchedUnits assert.
llvm-svn: 155668
The base address for the PC-relative load is Align(PC,4), so it's the
address of the word containing the 16-bit instruction, not the address
of the instruction itself. Ugh.
rdar://11314619
llvm-svn: 155659
On some cores it's a bad idea for performance to mix VFP and NEON instructions
and since these patterns are NEON anyway, the NEON load should be used.
llvm-svn: 155630
elements to minimize the number of multiplies required to compute the
final result. This uses a heuristic to attempt to form near-optimal
binary exponentiation-style multiply chains. While there are some cases
it misses, it seems to at least a decent job on a very diverse range of
inputs.
Initial benchmarks show no interesting regressions, and an 8%
improvement on SPASS. Let me know if any other interesting results (in
either direction) crop up!
Credit to Richard Smith for the core algorithm, and helping code the
patch itself.
llvm-svn: 155616
Benjamin Kramer