This patch re-introduces the MCContext member that was removed from
MCDisassembler in r206063, and requires that an MCContext be passed in at
MCDisassembler construction time. (Previously the MCContext member had been
initialized in an ad-hoc fashion after construction). The MCCContext member
can be used by MCDisassembler sub-classes to construct constant or
target-specific MCExprs.
This patch updates disassemblers for in-tree targets, and provides the
MCRegisterInfo instance that some disassemblers were using through the
MCContext (previously those backends were constructing their own
MCRegisterInfo instances).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206241 91177308-0d34-0410-b5e6-96231b3b80d8
MCDisassembler has an MCSymbolizer member that is meant to take care of
symbolizing during disassembly, but it also has several methods that enable the
disassembler to do symbolization internally (i.e. without an attached symbolizer
object). There is no need for this duplication, but ARM64 had been making use of
it. This patch moves the ARM64 symbolization logic out of ARM64Disassembler and
into an ARM64ExternalSymbolizer class, and removes the duplicated MCSymbolizer
functionality from the MCDisassembler interface. Symbolization will now be
done exclusively through MCSymbolizers.
There should be no impact on disassembly for any platform, but this allows us to
tidy up the MCDisassembler interface and simplify the process of (and invariants
related to) disassembler setup.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@206063 91177308-0d34-0410-b5e6-96231b3b80d8
When rematerializing through truncates, the coalescer may produce instructions
with dead defs, but live implicit-defs of subregs:
E.g.
%X1<def,dead> = MOVi64imm 2, %W1<imp-def>; %X1:GPR64, %W1:GPR32
These instructions are live, and their definitions should not be rewritten.
Fixes <rdar://problem/16492408>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205565 91177308-0d34-0410-b5e6-96231b3b80d8
I should have read that comment a little more carefully. ;)
Regression test in the works, committing in the mean time to un-break people.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205511 91177308-0d34-0410-b5e6-96231b3b80d8
TargetInstrInfo::findCommutedOpIndices to enable VFMA*231 commutation, rather
than abusing commuteInstruction.
Thanks very much for the suggestion guys!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205489 91177308-0d34-0410-b5e6-96231b3b80d8
opportunities in the current basic block, rather than just the last one seen.
<rdar://problem/16478629>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205481 91177308-0d34-0410-b5e6-96231b3b80d8
on FMA3 memory operands. FMA3 instructions are VEX encoded, so they can load
from unaligned memory.
Testcase to follow, along with related patch.
<rdar://problem/16478629>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@205472 91177308-0d34-0410-b5e6-96231b3b80d8
The edge data structure (EdgeEntry) now holds the indices of its entries in the
adjacency lists of the nodes it connects. This trades a little ugliness for
faster insertion/removal, which is now O(1) with a cheap constant factor. All
of this is implementation detail within the PBQP graph, the external API remains
unchanged.
Individual register allocations are likely to change, since the adjacency lists
will now be ordered differently (or rather, will now be unordered). This
shouldn't affect the average quality of allocations however.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204841 91177308-0d34-0410-b5e6-96231b3b80d8
RTDyldMemoryManager, regardless of whether it thinks they're "required for
execution".
Currently, RuntimeDyld only passes sections that are "required for execution"
to the RTDyldMemoryManager, and takes "required for execution" to mean exactly
"contains symbols or relocations". There are two problems with this:
(1) It can drop sections with anonymous data that is referenced by code.
(2) It leaves the JIT client no way to inspect interesting sections that aren't
actually required to run the program (e.g dwarf sections).
A test case is still in the works.
Future work: We may want to replace this with a generic section filtering
mechanism, but that will require more consideration. For now, this flag at least
allows clients to volunteer to do the filtering themselves.
Fixes <rdar://problem/15177691>.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204398 91177308-0d34-0410-b5e6-96231b3b80d8
- Adds support for inserting vzerouppers before tail-calls.
This is enabled implicitly by having MachineInstr::copyImplicitOps preserve
regmask operands, which allows VZeroUpperInserter to see where tail-calls use
vector registers.
- Fixes a bug that caused the previous version of this optimization to miss some
vzeroupper insertion points in loops. (Loops-with-vector-code that followed
loops-without-vector-code were mistakenly overlooked by the previous version).
- New algorithm never revisits instructions.
Fixes <rdar://problem/16228798>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204021 91177308-0d34-0410-b5e6-96231b3b80d8
relevant subclasses of RuntimeDyldImpl. This allows construction of
RuntimeDyldImpl instances to be deferred until after the target architecture is
known.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@203352 91177308-0d34-0410-b5e6-96231b3b80d8
The previous PBQP solver was very robust but consumed a lot of memory,
performed a lot of redundant computation, and contained some unnecessarily tight
coupling that prevented experimentation with novel solution techniques. This new
solver is an attempt to address these shortcomings.
Important/interesting changes:
1) The domain-independent PBQP solver class, HeuristicSolverImpl, is gone.
It is replaced by a register allocation specific solver, PBQP::RegAlloc::Solver
(see RegAllocSolver.h).
The optimal reduction rules and the backpropagation algorithm have been extracted
into stand-alone functions (see ReductionRules.h), which can be used to build
domain specific PBQP solvers. This provides many more opportunities for
domain-specific knowledge to inform the PBQP solvers' decisions. In theory this
should allow us to generate better solutions. In practice, we can at least test
out ideas now.
As a side benefit, I believe the new solver is more readable than the old one.
2) The solver type is now a template parameter of the PBQP graph.
This allows the graph to notify the solver of any modifications made (e.g. by
domain independent rules) without the overhead of a virtual call. It also allows
the solver to supply policy information to the graph (see below).
3) Significantly reduced memory overhead.
Memory management policy is now an explicit property of the PBQP graph (via
the CostAllocator typedef on the graph's solver template argument). Because PBQP
graphs for register allocation tend to contain many redundant instances of
single values (E.g. the value representing an interference constraint between
GPRs), the new RASolver class uses a uniquing scheme. This massively reduces
memory consumption for large register allocation problems. For example, looking
at the largest interference graph in each of the SPEC2006 benchmarks (the
largest graph will always set the memory consumption high-water mark for PBQP),
the average memory reduction for the PBQP costs was 400x. That's times, not
percent. The highest was 1400x. Yikes. So - this is fixed.
"PBQP: No longer feasting upon every last byte of your RAM".
Minor details:
- Fully C++11'd. Never copy-construct another vector/matrix!
- Cute tricks with cost metadata: Metadata that is derived solely from cost
matrices/vectors is attached directly to the cost instances themselves. That way
if you unique the costs you never have to recompute the metadata. 400x less
memory means 400x less cost metadata (re)computation.
Special thanks to Arnaud de Grandmaison, who has been the source of much
encouragement, and of many very useful test cases.
This new solver forms the basis for future work, of which there's plenty to do.
I will be adding TODO notes shortly.
- Lang.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@202551 91177308-0d34-0410-b5e6-96231b3b80d8
findOrEmitSection).
Vaidas Gasiunas's patch, r201259, fixed one instance where we were always
allocating sections as text. This patch fixes the remaining buggy call sites.
No test case: This isn't breaking anything that I know of, it's just
inconsistent.
<rdar://problem/15943542>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201605 91177308-0d34-0410-b5e6-96231b3b80d8
required for all sections in a module. This can be useful when targets or
code-models place strict requirements on how sections must be laid out
in memory.
If RTDyldMemoryManger::needsToReserveAllocationSpace() is overridden to return
true then the JIT will call the following method on the memory manager, which
can be used to preallocate the necessary memory.
void RTDyldMemoryManager::reserveAllocationSpace(uintptr_t CodeSize,
uintptr_t DataSizeRO,
uintptr_t DataSizeRW)
Patch by Vaidas Gasiunas. Thanks very much Viadas!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201259 91177308-0d34-0410-b5e6-96231b3b80d8
uintptr_t. An unsigned could overflow for large sections.
No test case - anything big enough to overflow an unsigned is going to take an
appreciable time to zero when the test passes.
The choice of uintptr_t was made to match the RTDyldMemoryManager APIs, but
these should probably be hardcoded to uint64_ts: It is legitimate to JIT for
64-bit targets from a 32-bit host/compiler.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@201127 91177308-0d34-0410-b5e6-96231b3b80d8
Commuting the 231 and 132 variants would swap addends and
multiplicands/multipliers, which isn't valid.
I'm still trying to reduce a decent test case for this.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200792 91177308-0d34-0410-b5e6-96231b3b80d8
It looks like these pseudos were only used for pattern matching. Def pats are
the appropriate way to do that. As a bonus, these intrinsics will now have
memory operands folded properly, and better FMA3 variants selected where
appropriate (see r199933).
<rdar://problem/15611947>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200577 91177308-0d34-0410-b5e6-96231b3b80d8
Also replaces testcase for r180790 (support for absolute non-externs relocs)
with a more robust version.
<rdar://problem/15864721>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@200404 91177308-0d34-0410-b5e6-96231b3b80d8
promotion code, Tablegen will now select FPExt for floating point promotions
(previously it had returned AExt, which is not valid for floating point types).
Any out-of-tree targets that were relying on AExt being returned for FP
promotions will need to update their code check for FPExt instead.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@199252 91177308-0d34-0410-b5e6-96231b3b80d8
root path to which object files managed by the LLIObjectCache instance should be
written. This option defaults to "", in which case objects are cached in the
same directory as the bitcode they are derived from.
The load-object-a.ll test has been rewritten to use this option to support
testing in environments where the test directory is not writable.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@198852 91177308-0d34-0410-b5e6-96231b3b80d8
I believe the bot failures on linux systems were due to overestimating the
alignment of object-files within archives, which are only guaranteed to be
two-byte aligned. I have reduced the alignment in
RuntimeDyldELF::createObjectImageFromFile accordingly.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@198737 91177308-0d34-0410-b5e6-96231b3b80d8
target independent.
Most of the x86 specific stackmap/patchpoint handling was necessitated by the
use of the native address-mode format for frame index operands. PEI has now
been modified to treat stackmap/patchpoint similarly to DEBUG_INFO, allowing
us to use a simple, platform independent register/offset pair for frame
indexes on stackmap/patchpoints.
Notes:
- Folding is now platform independent and automatically supported.
- Emiting patchpoints with direct memory references now just involves calling
the TargetLoweringBase::emitPatchPoint utility method from the target's
XXXTargetLowering::EmitInstrWithCustomInserter method. (See
X86TargetLowering for an example).
- No more ugly platform-specific operand parsers.
This patch shouldn't change the generated output for X86.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195944 91177308-0d34-0410-b5e6-96231b3b80d8
<def,dead> ones.
Add an assertion to make sure we catch this in the future.
Fixes <rdar://problem/15464559>.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@195401 91177308-0d34-0410-b5e6-96231b3b80d8
X86AsmPrinter::EmitInstruction, rather than X86MCInstLower::Lower.
The aim is to improve the reusability of the X86MCInstLower class by making it
more function-like. The X86::MORESTACK_RET_RESTORE_R10 pseudo broke the
function model by emitting an extra instruction to the MCStreamer attached to
the AsmPrinter.
The patch should have no impact on generated code.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194431 91177308-0d34-0410-b5e6-96231b3b80d8
The new graph structure replaces the node and edge linked lists with vectors.
Free lists (well, free vectors) are used for fast insertion/deletion.
The ultimate aim is to make PBQP graphs cheap to clone. The motivation is that
the PBQP solver destructively consumes input graphs while computing a solution,
forcing the graph to be fully reconstructed for each round of PBQP. This
imposes a high cost on large functions, which often require several rounds of
solving/spilling to find a final register allocation. If we can cheaply clone
the PBQP graph and incrementally update it between rounds then hopefully we can
reduce this cost. Further, once we begin pooling matrix/vector values (future
work), we can cache some PBQP solver metadata and share it between cloned
graphs, allowing the PBQP solver to re-use some of the computation done in
earlier rounds.
For now this is just a data structure update. The allocator and solver still
use the graph the same way as before, fully reconstructing it between each
round. I expect no material change from this update, although it may change
the iteration order of the nodes, causing ties in the solver to break in
different directions, and this could perturb the generated allocations
(hopefully in a completely benign way).
Thanks very much to Arnaud Allard de Grandmaison for encouraging me to get back
to work on this, and for a lot of discussion and many useful PBQP test cases.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@194300 91177308-0d34-0410-b5e6-96231b3b80d8
an MCExpr, in order to avoid writing an encoded zero value in the immediate
field.
When getUnconditionalBranchTargetOpValue is called with an MCExpr target, we
don't know what the final immediate field value should be. We shouldn't
explicitly set the immediate field to an encoded zero value as zero is encoded
with a non-zero bit pattern. This leads to bits being set that pollute the
final immediate value. The nature of the encoding is such that the polluted
bits only affect very large immediate values, explaining why this hasn't
caused problems earlier.
Fixes <rdar://problem/15155975>.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193535 91177308-0d34-0410-b5e6-96231b3b80d8
the instruction defenitions and ISEL reflect this.
Prior to this patch these instructions took an i32i8imm, and the high bits were
dropped during encoding. This led to incorrect behavior for shifts by
immediates higher than 255. This patch fixes that issue by detecting large
immediate shifts and returning constant zero (for logical shifts) or capping
the shift amount at an encodable value (for arithmetic shifts).
Fixes <rdar://problem/14968098>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193096 91177308-0d34-0410-b5e6-96231b3b80d8
same way as X86_64_GOT relocations. The 'Load' part of GOTLoad is just an
optimization hint for the linker anyway, and can be safely ignored.
This patch also fixes some minor issues with the relocations introduced while
processing an X86_64_GOT[Load]: the addend for the GOT entry should always be
zero, and the addend for the replacement relocation at the original offset
should be the same as the addend of the relocation being replaced.
I haven't come up with a good way of testing this yet, but I'm working on it.
This fixes <rdar://problem/14651564>.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188499 91177308-0d34-0410-b5e6-96231b3b80d8
indirect branches correctly. Under some circumstances, this led to the deletion
of basic blocks that were the destination of indirect branches. In that case it
left indirect branches to nowhere in the code.
This patch replaces, and is more general than either of the previous fixes for
indirect-branch-analysis issues, r181161 and r186461.
For other branches (not indirect) this refactor should have *almost* identical
behavior to the previous version. There are some corner cases where this
refactor is able to analyze blocks that the previous version could not (e.g.
this necessitated the update to thumb2-ifcvt2.ll).
<rdar://problem/14464830>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186735 91177308-0d34-0410-b5e6-96231b3b80d8
block. Blocks that have an indirect branch terminator, even if it's not the
last terminator, should still be treated as unanalyzable.
<rdar://problem/14437274>
Reducing a useful regression test case is proving difficult - I hope to have
one soon.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186461 91177308-0d34-0410-b5e6-96231b3b80d8
should expand ATOMIC_CMP_SWAP nodes the same way that it does for ATOMIC_SWAP.
Since ATOMIC_LOADs on some targets (e.g. older ARM variants) get legalized to
ATOMIC_CMP_SWAPs, the missing case had been causing i64 atomic loads to crash
during isel.
<rdar://problem/14074644>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185186 91177308-0d34-0410-b5e6-96231b3b80d8
return values are bitcasts.
The chain had previously been being clobbered with the entry node to
the dag, which sometimes caused other code in the function to be
erroneously deleted when tailcall optimization kicked in.
<rdar://problem/13827621>
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@181696 91177308-0d34-0410-b5e6-96231b3b80d8
When the SlotIndexes pass was introduced it was intended to support insertion
of code during register allocation. Removal of code was a minor consideration
(and raised the question of what to do about dangling SlotIndex objects pointing
to the erased index), so I opted to keep all indexes around indefinitely and
simply null out those that weren't being used.
Nowadays people are moving more code around (e.g. via HandleMove), which means
more zombie indexes. I want to start killing off indexes when we're done with
them to reclaim the resources they use up.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@179834 91177308-0d34-0410-b5e6-96231b3b80d8
codegen passes. This brings it in to line with clang and llc's codegen setup,
and tidies up the code.
If I understand correctly, adding ModulePasses to a FunctionPassManager is
bogus. It only seems to explode if an added ModulePass depends on a
FunctionPass though, which might be why this code has survived so long.
Fixes <rdar://problem/13386816>.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176977 91177308-0d34-0410-b5e6-96231b3b80d8
GlobalValue linkage up to ExternalLinkage in the ExtractGV pass. This
prevents linkonce and linkonce_odr symbols from being DCE'd.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176459 91177308-0d34-0410-b5e6-96231b3b80d8
InitSections is called before the MCContext is initialized it could cause
duplicate temporary symbols to be emitted later (after context initialization
resets the temporary label counter).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@169785 91177308-0d34-0410-b5e6-96231b3b80d8
checkRegMaskInterference only initializes the bitmask on the first interference.
This fixes PR14027 and (re)fixes PR13945.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@165608 91177308-0d34-0410-b5e6-96231b3b80d8
findLastUseBefore was previous considering virtreg liveness only, leading to
incorrect live intervals for reg units when instrs with physreg operands were
moved up.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163685 91177308-0d34-0410-b5e6-96231b3b80d8
boolean flag to an enum: { Fast, Standard, Strict } (default = Standard).
This option controls the creation by optimizations of fused FP ops that store
intermediate results in higher precision than IEEE allows (E.g. FMAs). The
behavior of this option is intended to match the behaviour specified by a
soon-to-be-introduced frontend flag: '-ffuse-fp-ops'.
Fast mode - allows formation of fused FP ops whenever they're profitable.
Standard mode - allow fusion only for 'blessed' FP ops. At present the only
blessed op is the fmuladd intrinsic. In the future more blessed ops may be
added.
Strict mode - allow fusion only if/when it can be proven that the excess
precision won't effect the result.
Note: This option only controls formation of fused ops by the optimizers. Fused
operations that are explicitly requested (e.g. FMA via the llvm.fma.* intrinsic)
will always be honored, regardless of the value of this option.
Internally TargetOptions::AllowExcessFPPrecision has been replaced by
TargetOptions::AllowFPOpFusion.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@158956 91177308-0d34-0410-b5e6-96231b3b80d8
Lang Hames