it is definitely profitable to tail duplicate indirect branches for x86.
This is likely to be true to various degrees for all modern x86 processors.
llvm-svn: 89865
Make tail duplication of indirect branches much more aggressive (for targets
that indicate that it is profitable), based on further experience with
this transformation. I compiled 3 large applications with and without
this more aggressive tail duplication and measured minimal changes in code
size. ("size" on Darwin seems to round the text size up to the nearest
page boundary, so I can only say that any code size increase was less than
one 4k page.) Radar 7421267.
llvm-svn: 89814
This violates the ABI (that area is "reserved"), and
while it is safe if all code is generated with current
compilers, there is some very old code around that uses
that slot for something else, and breaks if it is stored
into. Adjust testcases looking for current behavior.
I've verified that the stack frame size is right in all
testcases, whether it changed or not. 7311323.
llvm-svn: 89811
than doing the same via constpool:
1. Load from constpool costs 3 cycles on A9, movt/movw pair - just 2.
2. Load from constpool might stall up to 300 cycles due to cache miss.
3. Movt/movw does not use load/store unit.
4. Less constpool entries => better compiler performance.
This is only enabled on ELF systems, since darwin does not have needed
relocations (yet).
llvm-svn: 89720
way for each TargetJITInfo subclass to allocate its own stubs. This
means stubs aren't as exactly-sized anymore, but it lets us get rid of
TargetJITInfo::emitFunctionStubAtAddr(), which lets ARM and PPC
support the eager JIT, fixing http://llvm.org/PR4816.
* Rename the JITEmitter's stub creation functions to describe the kind
of stub they create. So far, all of them create lazy-compilation
stubs, but they sometimes get used when far-call stubs are needed.
Fixing http://llvm.org/PR5201 will involve fixing this.
llvm-svn: 89715
Note that "hasDotLocAndDotFile"-style debug info was already broken;
people wanting this functionality should implement it in the
AsmPrinter/DwarfWriter code.
llvm-svn: 89711
constant pool ranges, as CPEIsInRange() makes conservative assumptions about
the potential alignment changes from branch adjustments. The verification,
on the other hand, runs after those branch adjustments are made, so the
effects on alignment are known and already taken into account. The sanity
check in verify should check the range directly instead.
llvm-svn: 89473
assembly can confuse things utterly, as it's assumed that instructions in
inline assembly are 4 bytes wide. For Thumb mode, that's often not true,
so the calculations for when alignment padding will be present get thrown off,
ultimately leading to out of range constant pool entry references. Making
more conservative assumptions that padding may be necessary when inline asm
is present avoids this situation.
llvm-svn: 89403
fully specified at this level. Subclasses of NLdStLN can specify selective
bit(s) for Inst{7-4}, as is done for VLD[234]LN* and VST[234]LN* inside
ARMInstrNEON.td.
llvm-svn: 89377
contents of the block to be duplicated. Use this for ARM Cortex A8/9 to
be more aggressive tail duplicating indirect branches, since it makes it
much more likely that they will be predicted in the branch target buffer.
Testcase coming soon.
llvm-svn: 89187
This is probably not confined to *just* these two things.
Anyway, the llvm-gcc front-end may look up the structure layout information for
an abstract type. That information will be stored into a table with the FE's
TD. Instruction combine can come along and also ask for information on that
abstract type, but for a separate TD (the one associated with the pass manager).
After the type is refined, the old structure layout information in the pass
manager's TD file is out of date. If a new type is allocated in the same space
as the old-unrefined type, then the structure type information in the pass
manager's TD file will be wrong, but won't know it.
Fix this by making the TD's structure type information an abstract type user.
llvm-svn: 89176
The large code model is documented at
http://www.x86-64.org/documentation/abi.pdf and says that calls should
assume their target doesn't live within the 32-bit pc-relative offset
that fits in the call instruction.
To do this, we turn off the global-address->target-global-address
conversion in X86TargetLowering::LowerCall(). The first attempt at
this broke the lazy JIT because it can separate the movabs(imm->reg)
from the actual call instruction. The lazy JIT receives the address of
the movabs as a relocation and needs to record the return address from
the call; and then when that call happens, it needs to patch the
movabs with the newly-compiled target. We could thread the call
instruction into the relocation and record the movabs<->call mapping
explicitly, but that seems to require at least as much new
complication in the code generator as this change.
To fix this, we make lazy functions _always_ go through a call
stub. You'd think we'd only have to force lazy calls through a stub on
difficult platforms, but that turns out to break indirect calls
through a function pointer. The right fix for that is to distinguish
between calls and address-of operations on uncompiled functions, but
that's complex enough to leave for someone else to do.
Another attempt at this defined a new CALL64i pseudo-instruction,
which expanded to a 2-instruction sequence in the assembly output and
was special-cased in the X86CodeEmitter's emitInstruction()
function. That broke indirect calls in the same way as above.
This patch also removes a hack forcing Darwin to the small code model.
Without far-call-stubs, the small code model requires things of the
JITMemoryManager that the DefaultJITMemoryManager can't provide.
Thanks to echristo for lots of testing!
llvm-svn: 88984
- This is an initial step towards -march=native support in Clang, and towards
eliminating host dependencies in the targets. See PR5389.
- Patch by Roman Divacky!
llvm-svn: 88768
Provide special isLoadFromStackSlotPostFE and isStoreToStackSlotPostFE
interfaces to explicitly request checking for post-frame ptr elimination
operands. This uses a heuristic so it isn't reliable for correctness.
llvm-svn: 87047
machine instruction loads or stores from/to a stack slot. Unlike
isLoadFromStackSlot and isStoreFromStackSlot, the instruction may be
something other than a pure load/store (e.g. it may be an arithmetic
operation with a memory operand). This helps AsmPrinter determine when
to print a spill/reload comment.
This is only a hint since we may not be able to figure this out in all
cases. As such, it should not be relied upon for correctness.
Implement for X86. Return false by default for other architectures.
llvm-svn: 87026
slots. The AsmPrinter will use this information to determine whether to
print a spill/reload comment.
Remove default argument values. It's too easy to pass a wrong argument
value when multiple arguments have default values. Make everything
explicit to trap bugs early.
Update all targets to adhere to the new interfaces..
llvm-svn: 87022
can only branch forward. To best take advantage of them, we'd like to adjust
the basic blocks around a bit when reasonable. This patch puts basics in place
to do that, with a super-simple algorithm for backwards jump table targets that
creates a new branch after the jump table which branches backwards. Real
heuristics for reordering blocks or other modifications rather than inserting
branches will follow.
llvm-svn: 86791
generates a sequence similar to this:
__Z4funci:
LFB2:
mflr r0
LCFI0:
stmw r30,-8(r1)
LCFI1:
stw r0,8(r1)
LCFI2:
stwu r1,-80(r1)
LCFI3:
mr r30,r1
LCFI4:
where LCFI3 and LCFI4 are used by the FDE to indicate what the FP, LR, and other
things are. We generated something more like this:
Leh_func_begin1:
mflr r0
stw r31, 20(r1)
stw r0, 8(r1)
Llabel1:
stwu r1, -80(r1)
Llabel2:
mr r31, r1
Note that we are missing the "mr" instruction. This patch makes it more like the
GCC output.
llvm-svn: 86729
- Force NDEBUG on in any Release build. This drops the compile time to ~100s
from ~600s, in Release mode.
- This may just be a temporary workaround, I don't know the true nature of the
gcc-4.2 compile time performance problem.
llvm-svn: 86695
This patch forbids implicit conversion of DenseMap::const_iterator to
DenseMap::iterator which was possible because DenseMapIterator inherited
(publicly) from DenseMapConstIterator. Conversion the other way around is now
allowed as one may expect.
The template DenseMapConstIterator is removed and the template parameter
IsConst which specifies whether the iterator is constant is added to
DenseMapIterator.
Actually IsConst parameter is not necessary since the constness can be
determined from KeyT but this is not relevant to the fix and can be addressed
later.
Patch by Victor Zverovich!
llvm-svn: 86636
was generated. This caused code like this:
## The asm code for the function
.section __TEXT,__const
.align 2
lJTI11_0:
LJTI11_0:
.long LBB11_16
.long LBB11_4
.long LBB11_5
.long LBB11_6
.long LBB11_7
.long LBB11_8
.long LBB11_9
.long LBB11_10
.long LBB11_11
.long LBB11_12
.long LBB11_13
.long LBB11_14
Leh_func_end11: ## <---now in the wrong section!
The `Leh_func_end11' would then end up in the wrong section, causing the
resulting EH frame information to be wrong:
__ZL11CheckRightsjPKcbRbRP6NSData.eh:
.set Lset500eh,Leh_frame_end11-Leh_frame_begin11
.long Lset500eh ; Length of Frame Information Entry
Leh_frame_begin11:
.long Leh_frame_begin11-Leh_frame_common
.long Leh_func_begin11-.
.set Lset501eh,Leh_func_end11-Leh_func_begin11
.long Lset501eh ; FDE address range
`Lset501eh' is now something huge instead of the real value.
The X86 back-end generates the jump table after the EH information is
emitted. Do the same here.
llvm-svn: 86588
Bob Wilson