usage of the shuffle bitmask. Both work in 128-bit lanes without
crossing, but in the former the mask of the high part is the same
used by the low part while in the later both lanes have independent
masks. Handle this properly and and add support for vpermilpd.
llvm-svn: 136200
When splitting global live ranges, it is now possible to split for
multiple destination intervals at once. Previously, we only had the main
and stack intervals.
Each edge bundle is assigned to a split candidate, and splitAroundRegion
will insert copies between the candidate intervals and the stack
interval as needed.
The multi-way splitting is used to split around compact regions when
enabled with -compact-regions. The best candidate register still gets
all the bundles it wants, but everything outside the main interval is
first split around compact regions before we create single-block
intervals.
Compact region splitting still causes some regressions, so it is not
enabled by default.
llvm-svn: 136186
These copies would coalesce easily, but the resulting value would be
defined by a deleted instruction. Now we also remove the undefined value
number from the destination register.
This fixes PR10503.
llvm-svn: 136174
On x86 we can't encode an immediate LHS of a sub directly. If the RHS comes from a XOR with a constant we can
fold the negation into the xor and add one to the immediate of the sub. Then we can turn the sub into an add,
which can be commuted and encoded efficiently.
This code is generated for __builtin_clz and friends.
llvm-svn: 136167
Allow the rot_imm operand to be optional. This sets the stage for refactoring
away the "rr" versions from the multiclasses and replacing them with Pat<>s.
llvm-svn: 136154
Start of cleaning this up a bit. First step is to remove the encoder hook by
storing the operand as the bits it'll actually encode to so it can just be
directly used. Map it to the assembly source values 8/16/24 when we print it.
llvm-svn: 136152
When dead code elimination deletes a PHI value, the virtual register may
split into multiple connected components. In that case, revert each
component to the RS_Assign stage.
The new components are guaranteed to be smaller (the original value
numbers are distributed among the components), so this will always be
making progress. The components are now allowed to evict other live
ranges or be split again.
llvm-svn: 136034
The first problem to fix is to stop creating synthetic *Table_gen
targets next to all of the LLVM libraries. These had no real effect as
CMake specifies that add_custom_command(OUTPUT ...) directives (what the
'tablegen(...)' stuff expands to) are implicitly added as dependencies
to all the rules in that CMakeLists.txt.
These synthetic rules started to cause problems as we started more and
more heavily using tablegen files from *subdirectories* of the one where
they were generated. Within those directories, the set of tablegen
outputs was still available and so these synthetic rules added them as
dependencies of those subdirectories. However, they were no longer
properly associated with the custom command to generate them. Most of
the time this "just worked" because something would get to the parent
directory first, and run tablegen there. Once run, the files existed and
the build proceeded happily. However, as more and more subdirectories
have started using this, the probability of this failing to happen has
increased. Recently with the MC refactorings, it became quite common for
me when touching a large enough number of targets.
To add insult to injury, several of the backends *tried* to fix this by
adding explicit dependencies back to the parent directory's tablegen
rules, but those dependencies didn't work as expected -- they weren't
forming a linear chain, they were adding another thread in the race.
This patch removes these synthetic rules completely, and adds a much
simpler function to declare explicitly that a collection of tablegen'ed
files are referenced by other libraries. From that, we can add explicit
dependencies from the smaller libraries (such as every architectures
Desc library) on this and correctly form a linear sequence. All of the
backends are updated to use it, sometimes replacing the existing attempt
at adding a dependency, sometimes adding a previously missing dependency
edge.
Please let me know if this causes any problems, but it fixes a rather
persistent and problematic source of build flakiness on our end.
llvm-svn: 136023
shuffle before inserting on a 256-bit vector.
- Add AVX versions of movd/movq instructions
- Introduce a few COPY patterns to match insert_subvector instructions.
This turns a trivial insert_subvector instruction into a register copy,
coalescing the xmm into a ymm and avoid emiting on more instruction.
llvm-svn: 136002
Fix the Rn register encoding for both SSAT and USAT. Update the parsing of the
shift operand to correctly handle the allowed shift types and immediate ranges
and issue meaningful diagnostics when an illegal value or shift type is
specified. Add aliases to parse an ommitted shift operand (default value of
'lsl #0').
Add tests for diagnostics and proper encoding.
llvm-svn: 135990
refactorings. Several places that shouldn't have dependend on Target no
longer do. Also almost all of the CodeGen dependencies have gone away
for the MCDisassembler. Others add reasonable dependencies within the
target-specific layers.
llvm-svn: 135977
assert-path code, as previously we would have fallen off the end of the
function, but please review and let me know if this should go somewhere
else.
This fixes a Clang warning:
lib/MC/MCMachOStreamer.cpp:201:11: error: enumeration value 'MCSA_IndirectSymbol' not handled in switch [-Werror,-Wswitch-enum]
switch (Attribute) {
^
1 error generated.
llvm-svn: 135976
The shift immediate encoding, printing, etc. is handled directly by the
enclosing operand definition, so it should be a vanilla immediate, not a
nested complex operand (shift_imm).
llvm-svn: 135968
Remove some inititalizers that are the same as the default, move defs next to
their (singular) uses and generally simplify some formatting of asm operand
definitions.
llvm-svn: 135946
unwind encoding for that function. This simply crawls through the prolog looking
for machine instrs marked as "frame setup". It can calculate from these what the
compact unwind should look like.
This is currently disabled because of needed linker support. But initial tests
look good.
llvm-svn: 135922
The .local, .hidden, .internal, and .protected are not legal for all supported
file formats (in particular, they're invalid for MachO). Move the parsing for
them into the ELF assembly parser since that's the format they're for.
Similarly, .weak is used by COFF and ELF, but not MachO, so move the parsing
to the COFF and ELF asm parsers. Previously, using any of these directives
on Darwin would result in an assertion failure in the parser; now we get
a diagnostic as we should.
rdar://9827089
llvm-svn: 135921
This mechanism already exists, but the RS_Split2 stage makes it clearer.
When live range splitting creates ranges that may not be making
progress, they are marked RS_Split2 instead of RS_New. These ranges may
be split again, but only in a way that can be proven to make progress.
For local ranges, that means they must be split into ranges used by
strictly fewer instructions.
For global ranges, region splitting is bypassed and the RS_Split2
ranges go straight to per-block splitting.
llvm-svn: 135912
The stage is used to control where a live range is going, not where it
is coming from. Live ranges created by splitting will usually be marked
RS_New, but some are marked RS_Spill to avoid wasting time trying to
split them again.
The old RS_Global and RS_Local stages are merged - they are really the
same thing for local and global live ranges.
llvm-svn: 135911
This fixes PR10463. A two-address instruction with an <undef> use
operand was incorrectly rewritten so the def and use no longer used the
same register, violating the tie constraint.
Fix this by always rewriting <undef> operands with the register a def
operand would use.
llvm-svn: 135885
This method computes the edge bundles that should be live when splitting
around a compact region. This is independent of interference.
The function returns false if the live range was already a compact
region, or the compact region doesn't have any live bundles - it would
be the same as splitting around basic blocks.
Compact regions are computed using the normal spill placement code. We
pretend there is interference in all live-through blocks that don't use
the live range. This removes all edges from the Hopfield network used
for spill placement, so it converges instantly.
llvm-svn: 135847
If there is no interference and no last split point, we cannot
enterIntvBefore(Stop) - that function needs a real instruction.
Use enterIntvAtEnd instead for that very easy case.
This code doesn't currently run, it is needed by multi-way splitting.
llvm-svn: 135846
A split candidate can have a null PhysReg which means that it doesn't
map to a real interference pattern. Instead, pretend that all through
blocks have interference.
This makes it possible to generate compact regions where the live range
doesn't go through blocks that don't use it. The live range will still
be live between directly connected blocks with uses.
Splitting around a compact region tends to produce a live range with a
high spill weight, so it may evict a less dense live range.
llvm-svn: 135845
This method matches addLinks - All the listed blocks are considered to
have interference, so they add a negative bias to their bundles.
This could also be done by addConstraints, but that requires building a
separate BlockConstraint array.
llvm-svn: 135844
This makes TargetRegisterClass slightly slower. Next step will be making contains faster.
Eventually TargetRegisterClass will be killed entirely.
llvm-svn: 135835
The immediate is in the range 1-32, but is encoded as 0-31 in a 5-bit bitfield.
Update the representation such that we store the operand as 0-31, allowing us
to remove the encoder method and the special case handling in the disassembler.
Update the assembly parser and the instruction printer accordingly.
llvm-svn: 135823
The header file was already properly located. The previous need for it
in Support had to do with the version string printing which was fixed in
r135757.
Also update build dependencies where libraries that needed the
functionality of the Target library (in the form of the TargetRegistry)
were picking it up via Support. This is pretty pervasive, essentially
every TargetInfo library (ARMInfo, etc) uses TargetRegistry, making it
depend on Target. All of these were previously just sneaking by.
llvm-svn: 135760
Evan's recent refactorings (I believe). Specifically, MCDisassembler no
longer depends on Target, and ARMDisassembler no longer depends on
CodeGen. The added dependencies from ARMAsmParser to ARMDesc looks
correct based on header file inclusion.
llvm-svn: 135759
the Support library. Now its part of the TargetRegistry, and the three
commands that care about this explicitly register this extra bit of
version information.
The set of commands which care was computed by intersecting those which
use the Support library's version string printing and those that
initialize all the registered targets in a way that produces
a meaningful list. The only odd ball out is that 'clang -cc1as -version'
no longer prints the registered targets. I don't think anyone is really
interested in that (especially as the fact that llvm-mc does so is under
a FIXME), but if someone really does want this back I'll happily apply
the same patch there.
llvm-svn: 135757
function on the TargetRegistry. Also clean it up and use the modern LLVM
utility libraries available instead of rolling a few things manually.
llvm-svn: 135756
register extra version information to be printed. This is designed to
allow those tools which link in various targets to also print those
registered targets under --version.
Currently this printing logic is embedded into the Support library
directly; a huge layering violation. This is the first step to hoisting
it out into the tools without adding lots of duplicated code.
llvm-svn: 135755
the way to go. Doing this here will prevent several node matches later,
and would have to force looking all the way through several
VINSERTF128/VEXTRACTF128 chains to optimize simple things.
llvm-svn: 135730
and was actually very wrong, fix it and make it simpler. Also remove the
ConcatVectors function, which is unused now.
- Fix a introduction of useless nodes in r126664 and r126264. The
VUNPCKL* should never be introduced cause we don't want duplicate
nodes for 128 AVX and non-AVX modes, the actual instruction
difference only exists during isel, but not for target specific DAG
nodes. We only introduce V* target nodes when there is no 128-bit
version already there.
- Fix a fragile test and make it more useful.
llvm-svn: 135729
size but different element types, so that it filters out the cases
that CreateShuffleVectorCast doesn't handle. This fixes rdar://9786827.
llvm-svn: 135721
Aliases for LDM/STM. The single-register versions should encode to LDR/STR
with writeback, but we don't (yet) get that correct. Neither does Darwin's
system assembler, though, so that's not a deal-breaker of a limitation.
llvm-svn: 135702
Stefanovic. I removed the part that actually emits the instructions cause
I want that to get in better shape first and in incremental steps. This
also makes it easier to review the upcoming parts.
llvm-svn: 135678
instruction introduced in AVX, which can operate on 128 and 256-bit vectors.
It considers a 256-bit vector as two independent 128-bit lanes. It can permute
any 32 or 64 elements inside a lane, and restricts the second lane to
have the same permutation of the first one. With the improved splat support
introduced early today, adding codegen for this instruction enable more
efficient 256-bit code:
Instead of:
vextractf128 $0, %ymm0, %xmm0
punpcklbw %xmm0, %xmm0
punpckhbw %xmm0, %xmm0
vinsertf128 $0, %xmm0, %ymm0, %ymm1
vinsertf128 $1, %xmm0, %ymm1, %ymm0
vextractf128 $1, %ymm0, %xmm1
shufps $1, %xmm1, %xmm1
movss %xmm1, 28(%rsp)
movss %xmm1, 24(%rsp)
movss %xmm1, 20(%rsp)
movss %xmm1, 16(%rsp)
vextractf128 $0, %ymm0, %xmm0
shufps $1, %xmm0, %xmm0
movss %xmm0, 12(%rsp)
movss %xmm0, 8(%rsp)
movss %xmm0, 4(%rsp)
movss %xmm0, (%rsp)
vmovaps (%rsp), %ymm0
We get:
vextractf128 $0, %ymm0, %xmm0
punpcklbw %xmm0, %xmm0
punpckhbw %xmm0, %xmm0
vinsertf128 $0, %xmm0, %ymm0, %ymm1
vinsertf128 $1, %xmm0, %ymm1, %ymm0
vpermilps $85, %ymm0, %ymm0
llvm-svn: 135662
refactor the code and add a bunch of comments. The final shuffle
emitted by handling 256-bit types is suitable for the VPERM shuffle
instruction which is going to be introduced in a next commit (with
a testcase which cover this commit)
llvm-svn: 135661
Move the shift operator and special value (32 encoded as 0 for PKHTB) handling
into the instruction printer. This cleans up a bit of the disassembler
special casing for these instructions, more easily handles not printing the
operand at all for "lsl #0" and prepares for correct asm parsing of these
operands.
llvm-svn: 135626
Move common definitions for ARM and Thumb2 into ARMInstrFormats.td and rename
them to be a bit more descriptive that they're for the PKH instructions.
llvm-svn: 135617
The shift type is implied by the instruction (PKHBT vs. PKHTB) and so shouldn't
be also encoded as part of the shift value immediate. Otherwise we're able to
represent invalid instructions, plus it needlessly complicates the
representation. Preparatory work for asm parsing of these instructions.
llvm-svn: 135616
- Not great yet, but it's a start.
- Requires an object file with a symbol table. (I really want to fix this, but it'll need a whole new algorithm)
- ELF and COFF won't work at the moment due to libObject shortcomings.
To try it out run
$ llvm-objdump -d --cfg foo.o
This will create a graphviz file for every symbol in the object file's text section containing a CFG.
llvm-svn: 135608
- Introduce JITDefault code model. This tells targets to set different default
code model for JIT. This eliminates the ugly hack in TargetMachine where
code model is changed after construction.
llvm-svn: 135580
Eli Friedman