I'm recommiting the codegen part of the patch.
The vectorizer part will be send to review again.
Masked Vector Load and Store Intrinsics.
Introduced new target-independent intrinsics in order to support masked vector loads and stores. The loop vectorizer optimizes loops containing conditional memory accesses by generating these intrinsics for existing targets AVX2 and AVX-512. The vectorizer asks the target about availability of masked vector loads and stores.
Added SDNodes for masked operations and lowering patterns for X86 code generator.
Examples:
<16 x i32> @llvm.masked.load.v16i32(i8* %addr, <16 x i32> %passthru, i32 4 /* align */, <16 x i1> %mask)
declare void @llvm.masked.store.v8f64(i8* %addr, <8 x double> %value, i32 4, <8 x i1> %mask)
Scalarizer for other targets (not AVX2/AVX-512) will be done in a separate patch.
http://reviews.llvm.org/D6191
llvm-svn: 223348
This complicates a few algorithms due to not having random access, but
not by a huge degree I don't think (open to debate/design
discussion/etc).
llvm-svn: 223261
This is the second patch in a small series. This patch contains the MachineInstruction and x86-64 backend pieces required to lower Statepoints. It does not include the code to actually generate the STATEPOINT machine instruction and as a result, the entire patch is currently dead code. I will be submitting the SelectionDAG parts within the next 24-48 hours. Since those pieces are by far the most complicated, I wanted to minimize the size of that patch. That patch will include the tests which exercise the functionality in this patch. The entire series can be seen as one combined whole in http://reviews.llvm.org/D5683.
The STATEPOINT psuedo node is generated after all gc values are explicitly spilled to stack slots. The purpose of this node is to wrap an actual call instruction while recording the spill locations of the meta arguments used for garbage collection and other purposes. The STATEPOINT is modeled as modifing all of those locations to prevent backend optimizations from forwarding the value from before the STATEPOINT to after the STATEPOINT. (Doing so would break relocation semantics for collectors which wish to relocate roots.)
The implementation of STATEPOINT is closely modeled on PATCHPOINT. Eventually, much of the code in this patch will be removed. The long term plan is to merge the functionality provided by statepoints and patchpoints. Merging their implementations in the backend is likely to be a good starting point.
Reviewed by: atrick, ributzka
llvm-svn: 223085
This reverts commit r222632 (and follow-up r222636), which caused a host
of LNT failures on an internal bot. I'll respond to the commit on the
list with a reproduction of one of the failures.
Conflicts:
lib/Target/X86/X86TargetTransformInfo.cpp
llvm-svn: 222936
Introduced new target-independent intrinsics in order to support masked vector loads and stores. The loop vectorizer optimizes loops containing conditional memory accesses by generating these intrinsics for existing targets AVX2 and AVX-512. The vectorizer asks the target about availability of masked vector loads and stores.
Added SDNodes for masked operations and lowering patterns for X86 code generator.
Examples:
<16 x i32> @llvm.masked.load.v16i32(i8* %addr, <16 x i32> %passthru, i32 4 /* align */, <16 x i1> %mask)
declare void @llvm.masked.store.v8f64(i8* %addr, <8 x double> %value, i32 4, <8 x i1> %mask)
Scalarizer for other targets (not AVX2/AVX-512) will be done in a separate patch.
http://reviews.llvm.org/D6191
llvm-svn: 222632
address of the stack guard was being spilled to the stack.
Previously the address of the stack guard would get spilled to the stack if it
was impossible to keep it in a register. This patch introduces a new target
independent node and pseudo instruction which gets expanded post-RA to a
sequence of instructions that load the stack guard value. Register allocator
can now just remat the value when it can't keep it in a register.
<rdar://problem/12475629>
llvm-svn: 213967
Add MSBuiltin which is similar in vein to GCCBuiltin. This allows for adding
intrinsics for Microsoft compatibility to individual instructions. This is
needed to permit the creation of ARM specific MSVC extensions.
This is not currently in use, and requires an associated change in clang to
enable use of the intrinsics defined by this new class. This merely sets the
LLVM portion of the infrastructure in place to permit the use of this
functionality. A separate set of changes will enable the new intrinsics.
llvm-svn: 212350
These are used in the ARM backends to aid type-checking on patterns involving
intrinsics. By making sure one argument is an extended/truncated version of
another.
However, there's no reason to limit them to just vectors types. For example
AArch64 has the instruction "uqshrn sD, dN, #imm" which would naturally use an
intrinsic taking an i64 and returning an i32.
llvm-svn: 205003
The "noduplicate" function attribute exists to prevent certain optimizations
from duplicating calls to the function. This is important on platforms where
certain function call duplications are unsafe (for example execution barriers
for CUDA and OpenCL).
This patch makes it possible to specify intrinsics as "noduplicate" and
translates that to the appropriate function attribute.
llvm-svn: 204200
The old system was fairly convoluted:
* A temporary label was created.
* A single PROLOG_LABEL was created with it.
* A few MCCFIInstructions were created with the same label.
The semantics were that the cfi instructions were mapped to the PROLOG_LABEL
via the temporary label. The output position was that of the PROLOG_LABEL.
The temporary label itself was used only for doing the mapping.
The new CFI_INSTRUCTION has a 1:1 mapping to MCCFIInstructions and points to
one by holding an index into the CFI instructions of this function.
I did consider removing MMI.getFrameInstructions completelly and having
CFI_INSTRUCTION own a MCCFIInstruction, but MCCFIInstructions have non
trivial constructors and destructors and are somewhat big, so the this setup
is probably better.
The net result is that we don't create temporary labels that are never used.
llvm-svn: 203204
The convention used to specify the PowerPC ISA is that bits are numbered in
reverse order (0 is the index of the high bit). To support this "little endian"
encoding convention, CodeEmitterGen will reverse the bit numberings prior to
generating the encoding tables. In order to generate a disassembler,
FixedLenDecoderEmitter needs to do the same.
This moves the bit reversal logic out of CodeEmitterGen and into CodeGenTarget
(where it can be used by both CodeEmitterGen and FixedLenDecoderEmitter). This
is prep work for disassembly support in the PPC backend (which is the only
in-tree user of this little-endian encoding support).
llvm-svn: 197532
Patch by Ana Pazos.
1.Added support for v1ix and v1fx types.
2.Added Scalar Pairwise Reduce instructions.
3.Added initial implementation of Scalar Arithmetic instructions.
llvm-svn: 191263
Most places can use PrintFatalError as the unwinding mechanism was not
used for anything other than printing the error. The single exception
was CodeGenDAGPatterns.cpp, where intermediate errors during type
resolution were ignored to simplify incremental platform development.
This use is replaced by an error flag in TreePattern and bailout earlier
in various places if it is set.
llvm-svn: 166712
Currently, TableGen just guesses instruction properties when it can't
infer them form patterns.
This adds a guessInstructionProperties flag to the instruction set
definition that will be used to disable guessing. The flag is intended
as a migration aid. It will be removed again when no more targets need
their properties guessed.
llvm-svn: 162460
subtarget CPU descriptions and support new features of
MachineScheduler.
MachineModel has three categories of data:
1) Basic properties for coarse grained instruction cost model.
2) Scheduler Read/Write resources for simple per-opcode and operand cost model (TBD).
3) Instruction itineraties for detailed per-cycle reservation tables.
These will all live side-by-side. Any subtarget can use any
combination of them. Instruction itineraries will not change in the
near term. In the long run, I expect them to only be relevant for
in-order VLIW machines that have complex contraints and require a
precise scheduling/bundling model. Once itineraries are only actively
used by VLIW-ish targets, they could be replaced by something more
appropriate for those targets.
This tablegen backend rewrite sets things up for introducing
MachineModel type #2: per opcode/operand cost model.
llvm-svn: 159891
This reverts commit 151760.
We want to move getSubReg() from TargetRegisterInfo into MCRegisterInfo,
but to do that, the type of the lookup table needs to be the same for
all targets.
llvm-svn: 151814
1. Added opcode BUNDLE
2. Taught MachineInstr class to deal with bundled MIs
3. Changed MachineBasicBlock iterator to skip over bundled MIs; added an iterator to walk all the MIs
4. Taught MachineBasicBlock methods about bundled MIs
llvm-svn: 145975
Manage Inits in a FoldingSet. This provides several benefits:
- Memory for Inits is properly managed
- Duplicate Inits are folded into Flyweights, saving memory
- It enforces const-correctness, protecting against certain classes
of bugs
The above benefits allow Inits to be used in more contexts, which in
turn provides more dynamism to TableGen. This enhanced capability
will be used by the AVX code generator to a fold common patterns
together.
llvm-svn: 134907
Make the Elements vector private and expose an ArrayRef through
getOrder() instead. getOrder will eventually provide multiple
user-specified allocation orders.
Use the sorted member set for member and subclass tests. Clean up a lot
of ad hoc searches.
llvm-svn: 133040
I'll be moving some more code there to gather all of the
register-specific stuff in one place. Currently it is shared between
CodeGenTarget and RegisterInfoEmitter.
The plan is that CodeGenRegisters can compute the full register bank
structure while RegisterInfoEmitter only will handle the printing part.
llvm-svn: 132788
Some register classes are only used for instruction operand constraints.
They should never be used for virtual registers. Previously, those
register classes were given an empty allocation order, but now you can
say 'let isAllocatable=0' in the register class definition.
TableGen calculates if a register is part of any allocatable register
class, and makes that information available in TargetRegisterDesc::inAllocatableClass.
The goal here is to eliminate use cases for overriding allocation_order_*
methods.
llvm-svn: 132508
switch. With this newfound organization, teach tblgen how not to give
all intrinsics the 'nounwind' attribute. Introduce a new intrinsic,
llvm.eh.resume, which does not have this attribute. Documentation and uses
to follow.
llvm-svn: 132252
On the x86-64 and thumb2 targets, some registers are more expensive to encode
than others in the same register class.
Add a CostPerUse field to the TableGen register description, and make it
available from TRI->getCostPerUse. This represents the cost of a REX prefix or a
32-bit instruction encoding required by choosing a high register.
Teach the greedy register allocator to prefer cheap registers for busy live
ranges (as indicated by spill weight).
llvm-svn: 129864
passed the root of the match, even though only a few patterns
actually needed this (one in X86, several in ARM [which should
be refactored anyway], and some in CellSPU that I don't feel
like detangling). Instead of requiring all ComplexPatterns to
take the dead root, have targets opt into getting the root by
putting SDNPWantRoot on the ComplexPattern.
llvm-svn: 114471
The COPY instruction is intended to replace the target specific copy
instructions for virtual registers as well as the EXTRACT_SUBREG and
INSERT_SUBREG instructions in MachineFunctions. It won't we used in a selection
DAG.
COPY is lowered to native register copies by LowerSubregs.
llvm-svn: 107529
list of predefined instructions appear. Add some consistency checks.
Ideally, TargetOpcodes.h should be produced by TableGen from Target.td, but it
is hardly worth the effort.
llvm-svn: 107520
structure that represents a mapping without any dependencies on SubRegIndex
numbering.
This brings us closer to being able to remove the explicit SubRegIndex
numbering, and it is now possible to specify any mapping without inventing
*_INVALID register classes.
llvm-svn: 104563
sub-register indices and outputs a single super register which is formed from
a consecutive sequence of registers.
This is used as register allocation / coalescing aid and it is useful to
represent instructions that output register pairs / quads. For example,
v1024, v1025 = vload <address>
where v1024 and v1025 forms a register pair.
This really should be modelled as
v1024<3>, v1025<4> = vload <address>
but it would violate SSA property before register allocation is done.
Currently we use insert_subreg to form the super register:
v1026 = implicit_def
v1027 - insert_subreg v1026, v1024, 3
v1028 = insert_subreg v1027, v1025, 4
...
= use v1024
= use v1028
But this adds pseudo live interval overlap between v1024 and v1025.
We can now modeled it as
v1024, v1025 = vload <address>
v1026 = REG_SEQUENCE v1024, 3, v1025, 4
...
= use v1024
= use v1026
After coalescing, it will be
v1026<3>, v1025<4> = vload <address>
...
= use v1026<3>
= use v1026
llvm-svn: 102815
record* -> instrinfo instead of std::string -> instrinfo.
This speeds up tblgen on cellcpu from 7.28 -> 5.98s with a debug
build (20%).
llvm-svn: 98916
changing the primary datastructure from being a
"std::vector<unsigned char>" to being a new TypeSet class
that actually has (gasp) invariants!
This changes more things than I remember, but one major
innovation here is that it enforces that named input
values agree in type with their output values.
This also eliminates code that transparently assumes (in
some cases) that SDNodeXForm input/output types are the
same, because this is wrong in many case.
This also eliminates a bug which caused a lot of ambiguous
patterns to go undetected, where a register class would
sometimes pick the first possible type, causing an
ambiguous pattern to get arbitrary results.
With all the recent target changes, this causes no
functionality change!
llvm-svn: 98534
Craig Topper