Original log entry:
Refactor getActionType and getTypeToTransformTo ; place all of the 'decision'
code in one place.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131536 91177308-0d34-0410-b5e6-96231b3b80d8
than either the primitive size or the element primitive size (in the case
of vectors), simplify the vector logic. No functionality change. There
is some distracting churn in the patch because I lined up comments better
while there - sorry about that.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131533 91177308-0d34-0410-b5e6-96231b3b80d8
happily accept things like "sext <2 x i32> to <999 x i64>". It would
also accept "sext <2 x i32> to i64", though the verifier would catch
that later. Fixed by having castIsValid check that vector lengths match
except when doing a bitcast. (2) When creating a cast instruction, check
that the cast is valid (this was already done when creating constexpr
casts). While there, replace getScalarSizeInBits (used to allow more
vector casts) with getPrimitiveSizeInBits in getCastOpcode and isCastable
since vector to vector casts are now handled explicitly by passing to the
element types; i.e. this bit should result in no functional change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131532 91177308-0d34-0410-b5e6-96231b3b80d8
can be used to turn a <4 x i64> into a <4 x i32> but getCastOpcode would assert
if you passed these types to it. Note that this strictly extends the previous
functionality: if getCastOpcode previously accepted two vector types (i.e. didn't
assert) then it still will and returns the same opcode (BitCast). That's because
before it would only accept vectors with the same bitwidth, and the new code only
touches vectors with the same length. However if two vectors have both the same
bitwidth and the same length then their element types have the same bitwidth, so
the new logic will return BitCast as before.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131530 91177308-0d34-0410-b5e6-96231b3b80d8
GAS has no such directives (not even mingw-w64 GAS has them), so I took
creative license with their names in assembly. I prefixed them all with
"w64_" to avoid namespace collisions, for example. If I discover that GAS
has taken a different approach, I'll change ours to match.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131525 91177308-0d34-0410-b5e6-96231b3b80d8
the purposes of the Win64 EH tables, I realized we had no way to tell where
the function ends. (MASM bounds functions with PROC and ENDP keywords.)
Add a directive to delimit the end of the function, and rename the 'frame'
directive to more accurately reflect its duality with the new directive.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131522 91177308-0d34-0410-b5e6-96231b3b80d8
LiveInterval::shrinkToUses recomputes the live range from scratch instead of
removing snippets. This should avoid the problem with dangling live ranges.
Leave physreg identity copies alone. They can be created when joining a virtreg
with a physreg. They don't affect register allocation, and they will be removed
by the rewriter.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131521 91177308-0d34-0410-b5e6-96231b3b80d8
As an example, the change to InstCombineCalls catches a common case where a call to a bitcast of a function is rewritten.
Chris, does this approach look reasonable?
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131516 91177308-0d34-0410-b5e6-96231b3b80d8
The greedy register allocator has live range splitting and register class
inflation, so it can actually fully undo this join, including restoring the
original register classes.
We still don't want to do this for long live ranges, mostly because of the high
register pressure of there are many constrained live ranges overlapping.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131466 91177308-0d34-0410-b5e6-96231b3b80d8
When instructions are deleted, they leave tombstone SlotIndex entries.
The isZeroLength method should ignore these null indexes.
This causes RABasic to sometimes spill a callee-saved register in the
abi-isel.ll test, so don't run that test with -regalloc=basic. Prioritizing
register allocation according to spill weight can cause more registers to be
used.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131436 91177308-0d34-0410-b5e6-96231b3b80d8
("T is 1 if the target symbol S has type STT_FUNC and the
symbol addresses a Thumb instruction ;it is 0 otherwise."
from "ELF for the ARM Architecture" 4.7.1.2)
Patch by Koan-Sin Tan!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131406 91177308-0d34-0410-b5e6-96231b3b80d8
As an ExecutionEngine class function, its definition
really belongs in ExecutionEngine.cpp, not JIT.cpp.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131320 91177308-0d34-0410-b5e6-96231b3b80d8
In particular, into EngineBuilder. This should only impact
the private API between the EE and EB classes, not external
clients, since JITCtor and MCJITCtor are both protected members.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131317 91177308-0d34-0410-b5e6-96231b3b80d8
This prepares for making JITCtor/MCJITCtor take a
TargetMachine* directly from clients like EngineBuilder.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131316 91177308-0d34-0410-b5e6-96231b3b80d8
actually takes rather than how much memory was allocated for it. This
is more accurate and should help the manager pack things more effectively.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131305 91177308-0d34-0410-b5e6-96231b3b80d8
by non-CMP expressions. The executable test case (129821) would test
this as well, if we had an "-O0 -disable-arm-fast-isel" LLVM-GCC
tester. Alas, the ARM assembly would be very difficult to check with
FileCheck.
The thumb2-cbnz.ll test is affected; it generates larger code (tst.w
vs. cmp #0), but I believe the new version is correct.
rdar://problem/9298790
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131261 91177308-0d34-0410-b5e6-96231b3b80d8
erroring out completely. Some modules produce sections that aren't referenced,
so it's friendlier to clients like LLDB to just skip them, at least for now.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131243 91177308-0d34-0410-b5e6-96231b3b80d8
markers. In some cases a register def is dead on one path, but not on
another.
This is passing Clang self-hosting.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131214 91177308-0d34-0410-b5e6-96231b3b80d8
Interleave IV simplifications. Currently involves EliminateComparison
and EliminateRemainder. Next I'll add EliminateExtend.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131210 91177308-0d34-0410-b5e6-96231b3b80d8
LLVM and binutils.
With this patch, there are no functional differences between the .o
produced directly from LLVM versus the .s to .o via GNU as, for relocation tags
at least, for both PIC and non-PIC modes.
Because some non-PIC reloc tags are used (legally) on PIC, so IsPCRel flag is
necessary but not sufficient to determine whether the overall codegen mode is
PIC or not. Why is this necessary? There is an incompatibility of how relocs
are emitted in the .rodata section. Binutils PIC likes to emit certain relocs
as section relative offsets. Non-PIC does not do this.
So I added a hidden switch on the ELFObjectwriter "-arm-elf-force-pic" which
forces the objectwriter to pretend that all relocs are for PIC mode.
Todo: Activate ForceARMElfPIC to true if -relocation-model=pic is selected
on llc.
Todo: There are probably more issues for PIC mode on ARM/MC/ELF...
Todo: Existing tests in MC/ARM/elf-reloc*.ll need to be converted over to .s
tests as well as expanded to cover the gamut.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131205 91177308-0d34-0410-b5e6-96231b3b80d8
about to be spilled.
This can only happen when two extra snippet registers are included in the spill,
and there is a copy between them. Hoisting the spill creates problems because
the hoist will mark the copy for later dead code elimination, and spilling the
second register will turn the copy into a spill.
<rdar://problem/9420853>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131192 91177308-0d34-0410-b5e6-96231b3b80d8
This change allows bugpoint to pinpoint the "opt" pass and bitcode
segment responsible for a crash caused by miscompilation. At least it
works well for me now, without having to create any custom execution
wrappers.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131186 91177308-0d34-0410-b5e6-96231b3b80d8
If there is a store after the load node, then there is a chain, which means
that there is another user. Thus, asking hasOneUser would fail. Instead we
ask hasNUsesOfValue on the 'data' value.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131183 91177308-0d34-0410-b5e6-96231b3b80d8
intrinsic call. This prevents it from being reordered so that it appears
*before* the setjmp intrinsic (thus making it completely useless).
<rdar://problem/9409683>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131174 91177308-0d34-0410-b5e6-96231b3b80d8
at the start of basic blocks to their common predecessor. It's actually quite
common (e.g. about 50 times in JM/lencod) and has shown to be a nice code size
benefit. e.g.
pushq %rax
testl %edi, %edi
jne LBB0_2
## BB#1:
xorb %al, %al
popq %rdx
ret
LBB0_2:
xorb %al, %al
callq _foo
popq %rdx
ret
=>
pushq %rax
xorb %al, %al
testl %edi, %edi
je LBB0_2
## BB#1:
callq _foo
LBB0_2:
popq %rdx
ret
rdar://9145558
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131172 91177308-0d34-0410-b5e6-96231b3b80d8
Still to do:
- Allow replacing / removing passes (infrastructure there, just needs an infrastructure exposed)
- Defining sets of passes to be added or removed as a group
- Extending the support to allow user-defined groups of optimisations
- Allow plugins to be specified for loading automatically (e.g. from plugins.conf or some similar mechanism)
Reviewed by Nick Lewycky.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131155 91177308-0d34-0410-b5e6-96231b3b80d8
test case; I've only seen this on a release branch, and I can't get it
to reproduce on trunk. rdar://problem/7662569
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131152 91177308-0d34-0410-b5e6-96231b3b80d8
this clang will use .debug_frame in, for example,
clang -g -c -m32 test.c
This matches gcc's behaviour. It looks like .debug_frame is a bit bigger
than .eh_frame, but has the big advantage of not being allocated.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131140 91177308-0d34-0410-b5e6-96231b3b80d8
DWARF stuff also gets fixed up by ELFARMAsmBackend::ApplyFixup(),
but the offset is not guaranteed to be mod 4 == 0 as in text/data.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131137 91177308-0d34-0410-b5e6-96231b3b80d8
The previous invalidation missed the alias interference caches.
Also add a stats counter for the number of repaired ranges.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131133 91177308-0d34-0410-b5e6-96231b3b80d8
It can happen that a live debug variable is the last use of a sub-register, and
the register allocator will pick a larger register class for the virtual
register. If the allocated register doesn't support the sub-register index,
just use %noreg for the debug variables instead of asserting.
In PR9872, a debug variable ends up in the sub_8bit_hi part of a GR32_ABCD
register. The register is split and one part is inflated to GR32 and assigned
%ESI because there are no more normal uses of sub_8bit_hi.
Since %ESI doesn't have that sub-register, substPhysReg asserted. Now it will
simply insert a %noreg instead, and the debug variable will be marked
unavailable in that range.
We don't currently have a way of saying: !"value" is in bits 8-15 of %ESI, I
don't know if DWARF even supports that.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131073 91177308-0d34-0410-b5e6-96231b3b80d8
the smaller encoding and this cuts 270336 bytes from a release version of
clang and 1246272 bytes from a debug build.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131067 91177308-0d34-0410-b5e6-96231b3b80d8
often expressed as "x >= y ? x : y", there is a good chance we can extract
the existing "x >= y" from it and use that as a replacement for "max(x,y)==x".
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131049 91177308-0d34-0410-b5e6-96231b3b80d8
As an ExecutionEngine class function, its definition
really belongs in ExecutionEngine.cpp, not JIT.cpp.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131027 91177308-0d34-0410-b5e6-96231b3b80d8
In particular, into EngineBuilder. This should only impact
the private API between the EE and EB classes, not external
clients, since JITCtor and MCJITCtor are both protected members.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131026 91177308-0d34-0410-b5e6-96231b3b80d8
This prepares for making JITCtor/MCJITCtor take a
TargetMachine* directly from clients like EngineBuilder.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131025 91177308-0d34-0410-b5e6-96231b3b80d8
This can't be just an assertion, users can always write impossible inline
assembly. Such an assembly statement should be included in the error message.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131024 91177308-0d34-0410-b5e6-96231b3b80d8
assert in the bitcode writer. No change needed because the ValueEnumerator holds
a whole-module numbering anyhow. Fixes PR9857!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131016 91177308-0d34-0410-b5e6-96231b3b80d8
After a virtual register is split, update any debug user variables that resided
in the old register. This ensures that the LiveDebugVariables are still correct
after register allocation.
This may create DBG_VALUE instructions that place a user variable in a register
in parts of the function and in a stack slot in other parts. DwarfDebug
currently doesn't support that.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130998 91177308-0d34-0410-b5e6-96231b3b80d8
The post-ra scheduler was explicitly updating the depth of a node's
successors after scheduling it, regardless of whether the successor
was ready. This is quadratic for DAGs with transitively redundant
edges. I simply removed the useless update of depth, which is lazilly
computed later.
Fixes <rdar://problem/9044332> compiler takes way too long to build TextInput.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130992 91177308-0d34-0410-b5e6-96231b3b80d8
return the pointer being dereferenced, it returns the pointee, but a call
might return the pointer itself.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130979 91177308-0d34-0410-b5e6-96231b3b80d8
BuildSchedGraph was quadratic in the number of calls in the basic
block. After this fix, it keeps only a single call at the top of the
DefList so compile time doesn't blow up on large blocks. This reduces
postRA sched time on an external test case from 81s to 0.3s. Although
r130800 (reduced ARM register alias defs) also partially fixes the
issue by reducing the constant overhead of checking call interference
by an order of magnitude.
Fixes <rdar://problem/7662664> very poor compile time with post RA scheduling.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130943 91177308-0d34-0410-b5e6-96231b3b80d8
I tested both gdb on a bootstrapped clang and and the gdb testsuite on OS X (snow leopard)
and both are happy using __eh_frame.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130937 91177308-0d34-0410-b5e6-96231b3b80d8
who used this flag, and it now emits CFI and doesn't emit this anymore. All
other targets left this flag "false".
<rdar://problem/8486371>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130918 91177308-0d34-0410-b5e6-96231b3b80d8
Joining physregs is inherently dangerous because it uses a heuristic to avoid
creating invalid code. Linear scan had an emergency spilling mechanism to deal
with those rare cases. The new greedy allocator does not.
The greedy register allocator is much better at taking hints, so this has almost
no impact on code size and quality. The few cases where it matters show up as
unit tests that now have -join-physregs enabled explicitly.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130896 91177308-0d34-0410-b5e6-96231b3b80d8
It is OK for B to be any GR8_ABCD_H superclass, the returned register class
doesn't have to map surjectively onto B.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130892 91177308-0d34-0410-b5e6-96231b3b80d8
landing pad as its successor.
SjLj exception handling jumps to the correct landing pad via a switch statement
that's generated right before code-gen. Loosen the constraint in the machine
instruction verifier to allow for this. Note, this isn't the most rigorous check
since we cannot determine where that switch statement came from. But it's
marginally better than turning this check off when SjLj exceptions are used.
<rdar://problem/9187612>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130881 91177308-0d34-0410-b5e6-96231b3b80d8
Original message:
Teach MachineCSE how to do simple cross-block CSE involving physregs. This allows, for example, eliminating duplicate cmpl's on x86. Part of rdar://problem/8259436 .
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130877 91177308-0d34-0410-b5e6-96231b3b80d8
it is both inefficient and unexpected by dwarfdump. Change to
a DW_FORM_data4.
While in here, change the predicate name to reflect that the position
is not really absolute (it is an offset), just that the linker needs a
relocation.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130846 91177308-0d34-0410-b5e6-96231b3b80d8
but according to my super-optimizer there are only two missed simplifications
of -instsimplify kind when compiling bzip2, and this is one of them. It amuses
me to have bzip2 be perfectly optimized as far as instsimplify goes!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130840 91177308-0d34-0410-b5e6-96231b3b80d8
This adds functionality to remove size/zero extension during indvars
without generating a canonical IV and rewriting all IV users. It's
disabled by default so should have no effect on codegen. Work in progress.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130829 91177308-0d34-0410-b5e6-96231b3b80d8
LiveVariables doesn't understand that clobbering D0 and D1 completely overwrites
Q0, so if Q0 is live-in to a function, its live range will extend beyond a
function call that only clobbers D0 and D1. This shows up in the
ARM/2009-11-01-NeonMoves test case.
LiveVariables should probably implement the much stricter rules for physreg
liveness that RAFast imposes - a physreg is killed by the first use of any
alias.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130801 91177308-0d34-0410-b5e6-96231b3b80d8
Only create a canonical IV for backedge taken count if it will
actually be used by LinearFunctionTestReplace. And some related
cleanup, preparing to reduce dependence on canonical IVs.
No significant effect on x86 or arm in the test-suite.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130799 91177308-0d34-0410-b5e6-96231b3b80d8
Register coalescing can sometimes create live ranges that end in the middle of a
basic block without any killing instruction. When SplitKit detects this, it will
repair the live range by shrinking it to its uses.
Live range splitting also needs to know about this. When the range shrinks so
much that it becomes allocatable, live range splitting fails because it can't
find a good split point. It is paranoid about making progress, so an allocatable
range is considered an error.
The coalescer should really not be creating these bad live ranges. They appear
when coalescing dead copies.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130787 91177308-0d34-0410-b5e6-96231b3b80d8
max(a,b) >= a -> true. According to my super-optimizer, these are
by far the most common simplifications (of the -instsimplify kind)
that occur in the testsuite and aren't caught by -std-compile-opts.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130780 91177308-0d34-0410-b5e6-96231b3b80d8
