In commit r213915, Bill fixed little-endian usage of vmrgh* and vmrgl*
by swapping the input arguments. As it turns out, the exact same fix
is also required for the vpkuhum/vpkuwum patterns.
This fixes another regression in llvmpipe when vector support is
enabled.
Reviewed by Bill Schmidt.
llvm-svn: 214718
I ran into some test failures where common code changed vector division
by constant into a multiply-high operation (MULHU). But these are not
implemented by the back-end, so we failed to recognize the insn.
Fixed by marking MULHU/MULHS as Expand for vector types.
llvm-svn: 214716
This patch refactors code generation of vector comparisons.
This fixes a wrong code-gen bug for ISD::SETGE for floating-point types,
and improves generated code for vector comparisons in general.
Specifically, the patch moves all logic deciding how to implement vector
comparisons into getVCmpInst, which gets two extra boolean outputs
indicating to its caller whether its needs to swap the input operands
and/or negate the result of the comparison. Apart from implementing
these two modifications as directed by getVCmpInst, there is no need
to ever implement vector comparisons in any other manner; in particular,
there is never a need to perform two separate comparisons (e.g. one for
equal and one for greater-than, as code used to do before this patch).
Reviewed by Bill Schmidt.
llvm-svn: 214714
Summary:
This patch also fixes an issue with the way the Mips assembler enables/disables architecture
features. Before this patch, the assembler never disabled feature bits. For example,
.set mips64
.set mips32r2
would result in the 'OR' of mips64 with mips32r2 feature bits which isn't right.
Unfortunately this isn't trivial to fix because there's not an easy way to clear
feature bits as the algorithm in MCSubtargetInfo (ToggleFeature) only clears the bits
that imply the feature being cleared and not the implied bits by the feature (there's a
better explanation to the code I added).
Patch by Matheus Almeida and updated by Toma Tabacu
Reviewers: vmedic, matheusalmeida, dsanders
Reviewed By: dsanders
Subscribers: tomatabacu, llvm-commits
Differential Revision: http://reviews.llvm.org/D4123
llvm-svn: 214709
use of PACKUS. It's cleaner that way.
I looked at implementing clever combine-based folding of PACKUS chains
into PSHUFB but it is quite hard and doesn't seem likely to be worth it.
The most annoying part would be detecting that the correct masking had
been done to use PACKUS-style instructions as a blend operation rather
than there being any saturating as is indicated by its name. We generate
really nice code for what few test cases I've come up with that aren't
completely contrived for this by just directly prefering PSHUFB and so
let's go with that strategy for now. =]
llvm-svn: 214707
patterns of v16i8 shuffles.
This implements one of the more important FIXMEs for the SSE2 support in
the new shuffle lowering. We now generate the optimal shuffle sequence
for truncate-derived shuffles which show up essentially everywhere.
Unfortunately, this exposes a weakness in other parts of the shuffle
logic -- we can no longer form PSHUFB here. I'll add the necessary
support for that and other things in a subsequent commit.
llvm-svn: 214702
On Cygwin, getpagesize() returns 64k(AllocationGranularity).
In r214580, the size of X86GenInstrInfo.inc became 1499136.
FIXME: We should reorganize again getPageSize() on Win32.
MapFile allocates address along AllocationGranularity but view is mapped by physical page.
llvm-svn: 214681
I spent some time looking into a better or more principled way to handle
this. For example, by detecting arbitrary "unneeded" ORs... But really,
there wasn't any point. We just shouldn't build blatantly wrong code so
late in the pipeline rather than adding more stages and logic later on
to fix it. Avoiding this is just too simple.
llvm-svn: 214680
Fundamentally, there isn't a really portable way to test the constant
pool contents. Instead, pin this test to the bare-metal triple. This
also makes it a 64-bit triple which allows us to only match a single
constant pool rather than two. It can also just hard code the '.' prefix
as the format should be stable now that it has a fixed triple. Finally,
I've switched it to use CHECK-NEXT to be more precise in the instruction
sequence expected and to use variables rather than hard coding decisions
by the register allocator.
llvm-svn: 214679
combines) until they are legal.
Doing it the old way could, when the stars align *just* right, cause
a node to get into the combine set prior to being legalized. Then, when
the same node showed up as an operand to another node later on (but not
so much later on that it had been deleted as dead) we would fail to add
it back to the worklist thinking it had already been combined. This
would in turn cause it to not be legalized. Fortunately, we can also
walk the operands looking for uncombined (and thus potentially
un-legalized) nodes late. It will still ensure that we walk all operands
of all nodes and send all of them through both the legalizer without
changes and the combiner at least once. (Which was the original goal of
this).
I have a test case for this bug, but it is terribly brittle. For
example, it will stop finding the bug the moment I enable the new
shuffle lowering. I don't yet have any test case that reliably exercises
this bug, and it isn't clear that it will be possible to craft one. It
is entirely possible that with the new shuffle lowering the two forms of
doing this are precisely equivalent. That doesn't mean we shouldn't take
the more conservative approach of insisting on things in the combined
set having survived the legalizer.
llvm-svn: 214673
GCC 4.8.2 points out the ambiguity in evaluation of the assertion condition:
lib/Target/X86/X86FloatingPoint.cpp:949:49: warning: suggest parentheses around ‘&&’ within ‘||’ [-Wparentheses]
assert(STReturns == 0 || isMask_32(STReturns) && N <= 2);
llvm-svn: 214672
GCC 4.8.2 objects to the tautological condition in the assert as the unsigned
value is guaranteed to be >= 0. Simplify the assertion by dropping the
tautological condition.
llvm-svn: 214671
This is intended to be the minimal change needed to fix PR20354 ( http://llvm.org/bugs/show_bug.cgi?id=20354 ). The check for a vector operation was wrong; we need to check that the fabs itself is not a vector operation.
This patch will not generate the optimal code. A constant pool load and 'and' op will be generated instead of just returning a value that we can calculate in advance (as we do for the scalar case). I've put a 'TODO' comment for that here and expect to have that patch ready soon.
There is a very similar optimization that we can do in visitFNEG, so I've put another 'TODO' there and expect to have another patch for that too.
llvm-svn: 214670
sequence - AArch64 target support
This patch turns off madd/msub generation in the DAGCombiner and generates
them in the MachineCombiner instead. It replaces the original code sequence
with the combined sequence when it is beneficial to do so.
When there is no machine model support it always generates the madd/msub
instruction. This is true also when the objective is to optimize for code
size: when the combined sequence is shorter is always chosen and does not
get evaluated.
When there is a machine model the combined instruction sequence
is evaluated for critical path and resource length using machine
trace metrics and the original code sequence is replaced when it is
determined to be faster.
rdar://16319955
llvm-svn: 214669
sequence - target independent framework
When the DAGcombiner selects instruction sequences
it could increase the critical path or resource len.
For example, on arm64 there are multiply-accumulate instructions (madd,
msub). If e.g. the equivalent multiply-add sequence is not on the
crictial path it makes sense to select it instead of the combined,
single accumulate instruction (madd/msub). The reason is that the
conversion from add+mul to the madd could lengthen the critical path
by the latency of the multiply.
But the DAGCombiner would always combine and select the madd/msub
instruction.
This patch uses machine trace metrics to estimate critical path length
and resource length of an original instruction sequence vs a combined
instruction sequence and picks the faster code based on its estimates.
This patch only commits the target independent framework that evaluates
and selects code sequences. The machine instruction combiner is turned
off for all targets and expected to evolve over time by gradually
handling DAGCombiner pattern in the target specific code.
This framework lays the groundwork for fixing
rdar://16319955
llvm-svn: 214666
This makes EmitWindowsUnwindTables a virtual function and lowers the
implementation of the function to the X86WinCOFFStreamer. This method is a
target specific operation. This enables making the behaviour target dependent
by isolating it entirely to the target specific streamer.
llvm-svn: 214664
The frame information stored in this structure is driven by the requirements for
Windows NT unwinding rather than Windows 64 specifically. As a result, this
type can be shared across multiple architectures (ARM, AXP, MIPS, PPC, SH).
Rename this class in preparation for adding support for supporting unwinding
information for Windows on ARM.
Take the opportunity to constify the members as everything except the
ChainedParent is read-only. This required some adjustment to the label
handling.
llvm-svn: 214663
This slipped in in r214467, so something like
V_MOV_B32_e32 v0, ... is now printed with 2 spaces
between the instruction name and first operand.
llvm-svn: 214660
when let can do the same thing. Keep the 64bit variants as codegen-only.
While they have a different register class, the encoding is the same for
32bit and 64bit mode. Having both present would otherwise confuse the
disassembler.
llvm-svn: 214636
Darwin x86 asm comment prefix designed to work around GAS on that
platform. That makes the comment-matching of the test much more stable.
llvm-svn: 214629
lowering with a small addition to it and adding PSHUFB combining.
There is one obvious place in the new vector shuffle lowering where we
should form PSHUFBs directly: when without them we will unpack a vector
of i8s across two different registers and do a potentially 4-way blend
as i16s only to re-pack them into i8s afterward. This is the crazy
expensive fallback path for i8 shuffles and we can just directly use
pshufb here as it will always be cheaper (the unpack and pack are
two instructions so even a single shuffle between them hits our
three instruction limit for forming PSHUFB).
However, this doesn't generate very good code in many cases, and it
leaves a bunch of common patterns not using PSHUFB. So this patch also
adds support for extracting a shuffle mask from PSHUFB in the X86
lowering code, and uses it to handle PSHUFBs in the recursive shuffle
combining. This allows us to combine through them, combine multiple ones
together, and generally produce sufficiently high quality code.
Extracting the PSHUFB mask is annoyingly complex because it could be
either pre-legalization or post-legalization. At least this doesn't have
to deal with re-materialized constants. =] I've added decode routines to
handle the different patterns that show up at this level and we dispatch
through them as appropriate.
The two primary test cases are updated. For the v16 test case there is
still a lot of room for improvement. Since I was going through it
systematically I left behind a bunch of FIXME lines that I'm hoping to
turn into ALL lines by the end of this.
llvm-svn: 214628
of normally binary shuffle instructions like PUNPCKL and MOVLHPS.
This detects cases where a single register is used for both operands
making the shuffle behave in a unary way. We detect this and adjust the
mask to use the unary form which allows the existing DAG combine for
shuffle instructions to actually work at all.
As a consequence, this uncovered a number of obvious bugs in the
existing DAG combine which are fixed. It also now canonicalizes several
shuffles even with the existing lowering. These typically are trying to
match the shuffle to the domain of the input where before we only really
modeled them with the floating point variants. All of the cases which
change to an integer shuffle here have something in the integer domain, so
there are no more or fewer domain crosses here AFAICT. Technically, it
might be better to go from a GPR directly to the floating point domain,
but detecting floating point *outputs* despite integer inputs is a lot
more code and seems unlikely to be worthwhile in practice. If folks are
seeing domain-crossing regressions here though, let me know and I can
hack something up to fix it.
Also as a consequence, a bunch of missed opportunities to form pshufb
now can be formed. Notably, splats of i8s now form pshufb.
Interestingly, this improves the existing splat lowering too. We go from
3 instructions to 1. Yes, we may tie up a register, but it seems very
likely to be worth it, especially if splatting the 0th byte (the
common case) as then we can use a zeroed register as the mask.
llvm-svn: 214625
so using a single helper which adds operands back onto the worklist.
Several places didn't rigorously do this but a couple already did.
Factoring them together and doing it rigorously is important to delete
things recursively early on in the combiner and get a chance to see
accurate hasOneUse values. While no existing test cases change, an
upcoming patch to add DAG combining logic for PSHUFB requires this to
work correctly.
llvm-svn: 214623
during DAGCombine in certain circumstances. Unfortunately, the circumstances required
to trigger the issue seem to require a pretty specific interaction of DAGCombines,
and I haven't been able to find a testcase that reproduces on X86, ARM, or AArch64.
The functionality added here is replicated in essentially every other DAG combine,
so it seems pretty obviously correct.
llvm-svn: 214622
