We used to only combine intrinsics, and turn them into VLD1_UPD/VST1_UPD
when the base pointer is incremented after the load/store.
We can do the same thing for generic load/stores.
Note that we can only combine the first load/store+adds pair in
a sequence (as might be generated for a v16f32 load for instance),
because other combines turn the base pointer addition chain (each
computing the address of the next load, from the address of the last
load) into independent additions (common base pointer + this load's
offset).
Differential Revision: http://reviews.llvm.org/D6585
llvm-svn: 223862
It was missing from the VLD1/VST1 handling logic, even though the
corresponding instructions exist (same form as v2i64).
In preparation for a future patch.
llvm-svn: 223832
The load/store value type is currently not available when lowering the memcpy
intrinsic. Add the missing nullptr check to support this in 'computeAddress'.
Fixes rdar://problem/19178947.
llvm-svn: 223818
Lowering patterns were written through avx512_broadcast_pat multiclass as pattern generates VBROADCAST and COPY_TO_REGCLASS nodes.
Added lowering tests.
llvm-svn: 223804
With the foregoing three patches, VSX instructions can be used for
little endian. This patch removes the restriction that prevented
this, and re-enables the test cases from the first three patches.
llvm-svn: 223792
When performing instruction selection for ISD::VECTOR_SHUFFLE, there
is special code for handling v2f64 and v2i64 using VSX instructions.
This code must be adjusted for little-endian. Because the two inputs
are treated as a double-wide register, we must swap their order for
little endian. To get the appropriate mask elements to use with the
big-endian biased XXPERMDI instruction, we must reverse their order
and invert the bits.
A new test is added to test the 16 possible values of the shuffle
mask. It is initially disabled for reasons specified in the test. It
is re-enabled by patch 4/4.
llvm-svn: 223791
This optimization transforms code like:
bb1:
%0 = icmp ne i32 %a, 0
%1 = icmp ne i32 %b, 0
%or.cond = or i1 %0, %1
br i1 %or.cond, label %TrueBB, label %FalseBB
into a multiple branch instructions like:
bb1:
%0 = icmp ne i32 %a, 0
br i1 %0, label %TrueBB, label %bb2
bb2:
%1 = icmp ne i32 %b, 0
br i1 %1, label %TrueBB, label %FalseBB
This optimization is already performed by SelectionDAG, but not by FastISel.
FastISel cannot perform this optimization, because it cannot generate new
MachineBasicBlocks.
Performing this optimization at CodeGenPrepare time makes it available to both -
SelectionDAG and FastISel - and the implementation in SelectiuonDAG could be
removed. There are currenty a few differences in codegen for X86 and PPC, so
this commmit only enables it for FastISel.
Reviewed by Jim Grosbach
This fixes rdar://problem/19034919.
llvm-svn: 223786
This patch addresses the inherent big-endian bias in the lxvd2x,
lxvw4x, stxvd2x, and stxvw4x instructions. These instructions load
vector elements into registers left-to-right (with the first element
loaded into the high-order bits of the register), regardless of the
endian setting of the processor. However, these are the only
vector memory instructions that permit unaligned storage accesses, so
we want to use them for little-endian.
To make this work, a lxvd2x or lxvw4x is replaced with an lxvd2x
followed by an xxswapd, which swaps the doublewords. This works for
lxvw4x as well as lxvd2x, because for lxvw4x on an LE system the
vector elements are in LE order (right-to-left) within each
doubleword. (Thus after lxvw2x of a <4 x float> the elements will
appear as 1, 0, 3, 2. Following the swap, they will appear as 3, 2,
0, 1, as desired.) For stores, an stxvd2x or stxvw4x is replaced
with an stxvd2x preceded by an xxswapd.
Introduction of extra swap instructions provides correctness, but
obviously is not ideal from a performance perspective. Future patches
will address this with optimizations to remove most of the introduced
swaps, which have proven effective in other implementations.
The introduction of the swaps is performed during lowering of LOAD,
STORE, INTRINSIC_W_CHAIN, and INTRINSIC_VOID operations. The latter
are used to translate intrinsics that specify the VSX loads and stores
directly into equivalent sequences for little endian. Thus code that
uses vec_vsx_ld and vec_vsx_st does not have to be modified to be
ported from BE to LE.
We introduce new PPCISD opcodes for LXVD2X, STXVD2X, and XXSWAPD for
use during this lowering step. In PPCInstrVSX.td, we add new SDType
and SDNode definitions for these (PPClxvd2x, PPCstxvd2x, PPCxxswapd).
These are recognized during instruction selection and mapped to the
correct instructions.
Several tests that were written to use -mcpu=pwr7 or pwr8 are modified
to disable VSX on LE variants because code generation changes with
this and subsequent patches in this set. I chose to include all of
these in the first patch than try to rigorously sort out which tests
were broken by one or another of the patches. Sorry about that.
The new test vsx-ldst-builtin-le.ll, and the changes to vsx-ldst.ll,
are disabled until LE support is enabled because of breakages that
occur as noted in those tests. They are re-enabled in patch 4/4.
llvm-svn: 223783
missing barcelona CPU which that test uncovered, and remove the 32-bit
x86 CPUs which I really wasn't prepared to audit and test thoroughly.
If anyone wants to clean up the 32-bit only x86 CPUs, go for it.
Also, if anyone else wants to try to de-duplicate the AMD CPUs, that'd
be cool, but from the looks of it wouldn't save as much as it did for
the Intel CPUs.
llvm-svn: 223774
This handles the simplest case for mov -> push conversion:
1. x86-32 calling convention, everything is passed through the stack.
2. There is no reserved call frame.
3. Only registers or immediates are pushed, no attempt to combine a mem-reg-mem sequence into a single PUSHmm.
Differential Revision: http://reviews.llvm.org/D6503
llvm-svn: 223757
The aggressive anti-dep breaker, used by the PowerPC backend during post-RA
scheduling (but is available to all targets), did not handle early-clobber MI
operands (at all). When constructing the list of available registers for the
replacement of some def operand, check the using instructions, and remove
registers assigned to early-clobbered defs from the set.
Fixes PR21452.
llvm-svn: 223727
This fixes an issue with ScheduleDAGInstrs::buildSchedGraph
where stores without an underlying object would not be added
as a predecessor to the current BarrierChain.
llvm-svn: 223717
GCC accepts 'cc' as an alias for 'cr0', and we need to do the same when
processing inline asm constraints. This had previously been implemented using a
non-allocatable register, named 'cc', that was listed as an alias of 'cr0', but
the infrastructure does not seem to support this properly (neither the register
allocator nor the scheduler properly accounts for the alias). Instead, we can
just process this as a naming alias inside of the inline asm
constraint-processing code, so we'll do that instead.
There are two regression tests, one where the post-RA scheduler did the wrong
thing with the non-allocatable alias, and one where the register allocator did
the wrong thing. Fixes PR21742.
llvm-svn: 223708
Fix a compact unwind encoding logic bug which would try to encode
more callee saved registers than it should, leading to early bail out
in the encoding logic and abusive use of DWARF frame mode unnecessarily.
Also remove no-compact-unwind.ll which was testing the wrong thing
based on this bug and move it to valid 'compact unwind' tests. Added
other few more tests too.
llvm-svn: 223676
Teach ISel how to match a TZCNT/LZCNT from a conditional move if the
condition code is X86_COND_NE.
Existing tablegen patterns only allowed to match TZCNT/LZCNT from a
X86cond with condition code equal to X86_COND_E. To avoid introducing
extra rules, I added an 'ImmLeaf' definition that checks if the
condition code is COND_E or COND_NE.
llvm-svn: 223668
Before this patch, the backend sub-optimally expanded the non-constant shift
count of a v8i16 shift into a sequence of two 'movd' plus 'movzwl'.
With this patch the backend checks if the target features sse4.1. If so, then
it lets the shuffle legalizer deal with the expansion of the shift amount.
Example:
;;
define <8 x i16> @test(<8 x i16> %A, <8 x i16> %B) {
%shamt = shufflevector <8 x i16> %B, <8 x i16> undef, <8 x i32> zeroinitializer
%shl = shl <8 x i16> %A, %shamt
ret <8 x i16> %shl
}
;;
Before (with -mattr=+avx):
vmovd %xmm1, %eax
movzwl %ax, %eax
vmovd %eax, %xmm1
vpsllw %xmm1, %xmm0, %xmm0
retq
Now:
vpxor %xmm2, %xmm2, %xmm2
vpblendw $1, %xmm1, %xmm2, %xmm1
vpsllw %xmm1, %xmm0, %xmm0
retq
llvm-svn: 223660
matching offsets. I don't expect this to really matter, but its what the
latest incarnation of my script for maintaining these tests happens to
produce, and so its simpler for me if everything matches.
llvm-svn: 223613
This can significantly reduce the size of the switch, allowing for more
efficient lowering.
I also worked with the idea of exploiting unreachable defaults by
omitting the range check for jump tables, but always ended up with a
non-neglible binary size increase. It might be worth looking into some more.
SimplifyCFG currently does this transformation, but I'm working towards changing
that so we can optimize harder based on unreachable defaults.
Differential Revision: http://reviews.llvm.org/D6510
llvm-svn: 223566
Fix the poor codegen seen in PR21710 ( http://llvm.org/bugs/show_bug.cgi?id=21710 ).
Before we crack 32-byte build vectors into smaller chunks (and then subsequently
glue them back together), we should look for the easy case where we can just load
all elements in a single op.
An example of the codegen change is:
From:
vmovss 16(%rdi), %xmm1
vmovups (%rdi), %xmm0
vinsertps $16, 20(%rdi), %xmm1, %xmm1
vinsertps $32, 24(%rdi), %xmm1, %xmm1
vinsertps $48, 28(%rdi), %xmm1, %xmm1
vinsertf128 $1, %xmm1, %ymm0, %ymm0
retq
To:
vmovups (%rdi), %ymm0
retq
Differential Revision: http://reviews.llvm.org/D6536
llvm-svn: 223518
Summary:
Follow up to [x32] "Use ebp/esp as frame and stack pointer":
http://reviews.llvm.org/D4617
In that earlier patch, NaCl64 was made to always use rbp.
That's needed for most cases because rbp should hold a full
64-bit address within the NaCl sandbox so that load/stores
off of rbp don't require sandbox adjustment (zeroing the top
32-bits, then filling those by adding r15).
However, llvm.frameaddress returns a pointer and pointers
are 32-bit for NaCl64. In this case, use ebp instead, which
will make the register copy type check. A similar mechanism
may be needed for llvm.eh.return, but is not added in this change.
Test Plan: test/CodeGen/X86/frameaddr.ll
Reviewers: dschuff, nadav
Subscribers: jfb, llvm-commits
Differential Revision: http://reviews.llvm.org/D6514
llvm-svn: 223510
Update of some of the VSX test cases for Power to check fast-isel codegen as well as the regular codegen.
http://reviews.llvm.org/D6357
llvm-svn: 223509
SSE2/AVX non-constant packed shift instructions only use the lower 64-bit of
the shift count.
This patch teaches function 'getTargetVShiftNode' how to deal with shifts
where the shift count node is of type MVT::i64.
Before this patch, function 'getTargetVShiftNode' only knew how to deal with
shift count nodes of type MVT::i32. This forced the backend to wrongly
truncate the shift count to MVT::i32, and then zero-extend it back to MVT::i64.
llvm-svn: 223505
When lowering a vector shift node, the backend checks if the shift count is a
shuffle with a splat mask. If so, then it introduces an extra dag node to
extract the splat value from the shuffle. The splat value is then used
to generate a shift count of a target specific shift.
However, if we know that the shift count is a splat shuffle, we can use the
splat index 'I' to extract the I-th element from the first shuffle operand.
The advantage is that the splat shuffle may become dead since we no longer
use it.
Example:
;;
define <4 x i32> @example(<4 x i32> %a, <4 x i32> %b) {
%c = shufflevector <4 x i32> %b, <4 x i32> undef, <4 x i32> zeroinitializer
%shl = shl <4 x i32> %a, %c
ret <4 x i32> %shl
}
;;
Before this patch, llc generated the following code (-mattr=+avx):
vpshufd $0, %xmm1, %xmm1 # xmm1 = xmm1[0,0,0,0]
vpxor %xmm2, %xmm2
vpblendw $3, %xmm1, %xmm2, %xmm1 # xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
vpslld %xmm1, %xmm0, %xmm0
retq
With this patch, the redundant splat operation is removed from the code.
vpxor %xmm2, %xmm2
vpblendw $3, %xmm1, %xmm2, %xmm1 # xmm1 = xmm1[0,1],xmm2[2,3,4,5,6,7]
vpslld %xmm1, %xmm0, %xmm0
retq
llvm-svn: 223461
The test file test/CodeGen/ARM/build-attributes.ll was missing several
floating-point build attribute tests. The intention of this commit is that for
each CPU / architecture currently tested, there are now tests that make sure
the following attributes are sufficiently checked,
* Tag_ABI_FP_rounding
* Tag_ABI_FP_denormal
* Tag_ABI_FP_exceptions
* Tag_ABI_FP_user_exceptions
* Tag_ABI_FP_number_model
Also in this commit, the -unsafe-fp-math flag has been augmented with the full
suite of flags Clang sends to LLVM when you pass -ffast-math to Clang. That is,
`-unsafe-fp-math' has been changed to `-enable-unsafe-fp-math -disable-fp-elim
-enable-no-infs-fp-math -enable-no-nans-fp-math -fp-contract=fast'
Change-Id: I35d766076bcbbf09021021c0a534bf8bf9a32dfc
llvm-svn: 223454
Reverting this because, while it fixes the problem in the reduced test case, it
does not fix the problem in the full test case from the bug report.
llvm-svn: 223442
The scheduling dependency graph is built bottom-up within each scheduling
region, and ScheduleDAGInstrs::addPhysRegDeps is called to add output/anti
dependencies, based on physical registers, to the SUs for instructions
based on those that come before them.
In the test case, we start before post-RA scheduling with a block that looks
like this:
...
INLINEASM <...
andc $0,$0,$2
stdcx. $0,0,$3
bne- 1b
> [sideeffect] [mayload] [maystore] [attdialect], $0:[regdef-ec:G8RC], %X6<earlyclobber,def,dead>, $1:[mem], %X3<kill>, $2:[reguse:G8RC], %X5<kill>, $3:[reguse:G8RC], %X3, $4:[mem], %X3, $5:[clobber], %CC<earlyclobber,imp-def,dead>, <<badref>>
...
%X4<def,dead> = ANDIo8 %X4<kill>, 1, %CR0<imp-def,dead>, %CR0GT<imp-def>
...
%R29<def> = ISEL %R3<undef>, %R4<kill>, %CR0GT<kill>
where it is relevant that %CC is an alias to %CR0, and that %CR0GT is a
subregister of %CR0. However, for post-RA scheduling, no dependency was added
to prevent the INLINEASM from being scheduled in between the ANDIo8 and the
ISEL (which communicate via the %CR0GT register).
In ScheduleDAGInstrs::addPhysRegDeps, when called for the %CC operand, we'd
iterate over all of its aliases (which include %CC itself and also %CR0), and
look for previously-encountered defs of those registers. We'd find the ANDIo8,
but decide not to add a dependency between the INLINEASM and the ANDIo8 because
both the INLINEASM's def of %CC is dead, and also the ANDIo8 def of %CR0 is
dead. This ignores, however, that ANDIo8 has a non-dead def of %CR0GT, a
subregister of %CR0, and thus a dependency still must exist.
To fix this problem, when calling registerDefIsDead on the SU with the def, we
also check all subregisters for possible non-dead defs, and add the dependency
if any are found.
Fixes PR21742.
llvm-svn: 223440
with fixes. Includes the move of tests for llvm-objdump for universal files to an X86
directory. And the fix where it was failing on linux Rafael tracked down with asan.
I had both Jim Grosbach and Adam Hemet look over the second fix since I could not
set up asan to reproduce with the old version but not with the fix.
llvm-svn: 223416
So there are a couple of issues with indirect calls on thumbv4t. First, the most
'obvious' instruction, 'blx' isn't available until v5t. And secondly, the
next-most-obvious sequence: 'mov lr, pc; bx rN' doesn't DTRT in thumb code
because the saved off pc has its thumb bit cleared, so when the callee returns
we end up in ARM mode.... yuck.
The solution is to 'bl' to a nearby landing pad with a 'bx rN' in it.
We could cut down on code size by sharing the landing pads between call sites
that are close enough, but for the moment let's do correctness first and look at
performance later.
Patch by: Iain Sandoe
http://reviews.llvm.org/D6519
llvm-svn: 223380
The current DAG combine turns a sequence of extracts from <4 x i32> followed by zexts into a store followed by scalar loads.
According to measurements by Martin Krastev (see PR 21269) for x86-64, a sequence of an extract, movs and shifts gives better performance. However, for 32-bit x86, the previous sequence still seems better.
Differential Revision: http://reviews.llvm.org/D6501
llvm-svn: 223360
According to a previous FIXME comment we now not only look at MBB
successors, but also handle code sinking past them:
x = computation
if () {} else {}
use x
The instruction could be sunk over the whole diamond for the
if/then/else (or loop, etc), allowing it to be sunk into other blocks
after that.
Modified test added in r204522, due to one spill less present.
Minor fixes in comments.
Patch provided by Jonas Paulsson. Reviewed by Hal Finkel.
llvm-svn: 223350
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
Commit on
- This patch fixes the bug described in
http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-May/062343.html
The fix allocates an extra slot just below the GPRs and stores the base pointer
there. This is done only for functions containing llvm.eh.sjlj.setjmp that also
need a base pointer. Because code containing llvm.eh.sjlj.setjmp saves all of
the callee-save GPRs in the prologue, the offset to the extra slot can be
computed before prologue generation runs.
Impact at run-time on affected functions is::
- One extra store in the prologue, The store saves the base pointer.
- One extra load after a llvm.eh.sjlj.setjmp. The load restores the base pointer.
Because the extra slot is just above a gap between frame-pointer-relative and
base-pointer-relative chunks of memory, there is no impact on other offset
calculations other than ensuring there is room for the extra slot.
http://reviews.llvm.org/D6388
Patch by Arch Robison <arch.robison@intel.com>
llvm-svn: 223329
We had mistakenly believed that GCC's 'cc' referred to the entire
condition-code register (cr0 through cr7) -- and implemented this in r205630 to
fix PR19326, but 'cc' is actually an alias only to 'cr0'. This is causing LLVM
to clobber too much with legacy code with inline asm using the 'cc' clobber.
Fixes PR21451.
llvm-svn: 223328
On PowerPC, inline asm memory operands might be expanded as 0($r), where $r is
a register containing the address. As a result, this register cannot be r0, and
we need to enforce this register subclass constraint to prevent miscompiling
the code (we'd get this constraint for free with the usual instruction
definitions, but that scheme has no knowledge of how we end up printing inline
asm memory operands, and so here we need to do it 'by hand'). We can accomplish
this within the current address-mode selection framework by introducing an
explicit COPY_TO_REGCLASS node.
Fixes PR21443.
llvm-svn: 223318
Prior to this commit, physical registers defined implicitly were considered free
right after their definition, i.e.. like dead definitions. Therefore, their uses
had to immediately follow their definitions, otherwise the related register may
be reused to allocate a virtual register.
This commit fixes this assumption by keeping implicit definitions alive until
they are actually used. The downside is that if the implicit definition was dead
(and not marked at such), we block an otherwise available register. This is
however conservatively correct and makes the fast register allocator much more
robust in particular regarding the scheduling of the instructions.
Fixes PR21700.
llvm-svn: 223317
Almost all immediates in PowerPC assembly (both 32-bit and 64-bit) are signed
numbers, and it is important that we print them as such. To make sure that
happens, we change PPCTargetLowering::LowerAsmOperandForConstraint so that it
does all intermediate checks on a signed-extended int64_t value, and then
creates the resulting target constant using MVT::i64. This will ensure that all
negative values are printed as negative values (mirroring what is done in other
backends to achieve the same sign-extension effect).
This came up in the context of inline assembly like this:
"add%I2 %0,%0,%2", ..., "Ir"(-1ll)
where we used to print:
addi 3,3,4294967295
and gcc would print:
addi 3,3,-1
and gas accepts both forms, but our builtin assembler (correctly) does not. Now
we print -1 like gcc does.
While here, I replaced a bunch of custom integer checks with isInt<16> and
friends from MathExtras.h.
Thanks to Paul Hargrove for the bug report.
llvm-svn: 223220
LLVM understands a -enable-sign-dependent-rounding-fp-math codegen option. When
the user has specified this option, the Tag_ABI_FP_rounding attribute should be
emitted with value 1. This option currently does not appear to disable
transformations and optimizations that assume default floating point rounding
behavior, AFAICT, but the intention should be recorded in the build attributes,
regardless of what the compiler actually does with the intention.
Change-Id: If838578df3dc652b6f2796b8d152545674bcb30e
llvm-svn: 223218
Select i1 logical ops directly to 64-bit SALU instructions.
Vector i1 values are always really in SGPRs, with each
bit for each item in the wave. This saves about 4 instructions
when and/or/xoring any condition, and also helps write conditions
that need to be passed in vcc.
This should work correctly now that the SGPR live range
fixing pass works. More work is needed to eliminate the VReg_1
pseudo regclass and possibly the entire SILowerI1Copies pass.
llvm-svn: 223206
We just needed to remove the assertion in
AMDGPURegisterInfo::getFrameRegister(), which is called when
initializing the parser for inline assembly.
llvm-svn: 223197
Patch by Ben Gamari!
This redefines the `prefix` attribute introduced previously and
introduces a `prologue` attribute. There are a two primary usecases
that these attributes aim to serve,
1. Function prologue sigils
2. Function hot-patching: Enable the user to insert `nop` operations
at the beginning of the function which can later be safely replaced
with a call to some instrumentation facility
3. Runtime metadata: Allow a compiler to insert data for use by the
runtime during execution. GHC is one example of a compiler that
needs this functionality for its tables-next-to-code functionality.
Previously `prefix` served cases (1) and (2) quite well by allowing the user
to introduce arbitrary data at the entrypoint but before the function
body. Case (3), however, was poorly handled by this approach as it
required that prefix data was valid executable code.
Here we redefine the notion of prefix data to instead be data which
occurs immediately before the function entrypoint (i.e. the symbol
address). Since prefix data now occurs before the function entrypoint,
there is no need for the data to be valid code.
The previous notion of prefix data now goes under the name "prologue
data" to emphasize its duality with the function epilogue.
The intention here is to handle cases (1) and (2) with prologue data and
case (3) with prefix data.
References
----------
This idea arose out of discussions[1] with Reid Kleckner in response to a
proposal to introduce the notion of symbol offsets to enable handling of
case (3).
[1] http://lists.cs.uiuc.edu/pipermail/llvmdev/2014-May/073235.html
Test Plan: testsuite
Differential Revision: http://reviews.llvm.org/D6454
llvm-svn: 223189
We need to use the custom expansion of readcyclecounter on all 32-bit targets
(even those with 64-bit registers). This should fix the ppc64 buildbot.
llvm-svn: 223182
A global variable without an explicit alignment specified should be assumed to
be ABI-aligned according to its type, like on other platforms. This allows us
to use better memory operations when accessing it.
rdar://18533701
llvm-svn: 223180
This frequently leads to cases like:
ldr xD, [xN, :lo12:var]
add xA, xN, :lo12:var
ldr xD, [xA, #8]
where the ADD would have been needed anyway, and the two distinct addressing
modes can prevent the formation of an ldp. Because of how we handle ADRP
(aggressively forming an ADRP/ADD pseudo-inst at ISel time), this pattern also
results in duplicated ADRP instructions (one on its own to cover the ldr, and
one combined with the add).
llvm-svn: 223172
4i32 shuffles for single insertions into zero vectors lowers to X86vzmovl which was using (v)blendps - causing domain switch stalls. This patch fixes this by using (v)pblendw instead.
The updated tests on test/CodeGen/X86/sse41.ll still contain a domain stall due to the use of insertps - I'm looking at fixing this in a future patch.
Differential Revision: http://reviews.llvm.org/D6458
llvm-svn: 223165
We've long supported readcyclecounter on PPC64, but it is easier there (the
read of the 64-bit time-base register can be accomplished via a single
instruction). This now provides an implementation for PPC32 as well. On PPC32,
the time-base register is still 64 bits, but can only be read 32 bits at a time
via two separate SPRs. The ISA manual explains how to do this properly (it
involves re-reading the upper bits and looping if the counter has wrapped while
being read).
This requires PPC to implement a custom integer splitting legalization for the
READCYCLECOUNTER node, turning it into a target-specific SDAG node, which then
gets turned into a pseudo-instruction, which is then expanded to the necessary
sequence (which has three SPR reads, the comparison and the branch).
Thanks to Paul Hargrove for pointing out to me that this was still unimplemented.
llvm-svn: 223161
Reduce the number of nops emitted for stackmap shadows on AArch64 by counting
non-stackmap instructions up to the next branch target towards the requested
shadow.
<rdar://problem/14959522>
llvm-svn: 223156
Summary:
Like N32/N64, they must be passed in the upper bits of the register.
The new code could be merged with the existing if-statements but I've
refrained from doing this since it will make porting the O32 implementation
to tablegen harder later.
Reviewers: vmedic
Reviewed By: vmedic
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6463
llvm-svn: 223148
This is the third patch in a small series. It contains the CodeGen support for lowering the gc.statepoint intrinsic sequences (223078) to the STATEPOINT pseudo machine instruction (223085). The change also includes the set of helper routines and classes for working with gc.statepoints, gc.relocates, and gc.results since the lowering code uses them.
With this change, gc.statepoints should be functionally complete. The documentation will follow in the fourth change, and there will likely be some cleanup changes, but interested parties can start experimenting now.
I'm not particularly happy with the amount of code or complexity involved with the lowering step, but at least it's fairly well isolated. The statepoint lowering code is split into it's own files and anyone not working on the statepoint support itself should be able to ignore it.
During the lowering process, we currently spill aggressively to stack. This is not entirely ideal (and we have plans to do better), but it's functional, relatively straight forward, and matches closely the implementations of the patchpoint intrinsics. Most of the complexity comes from trying to keep relocated copies of values in the same stack slots across statepoints. Doing so avoids the insertion of pointless load and store instructions to reshuffle the stack. The current implementation isn't as effective as I'd like, but it is functional and 'good enough' for many common use cases.
In the long term, I'd like to figure out how to integrate the statepoint lowering with the register allocator. In principal, we shouldn't need to eagerly spill at all. The register allocator should do any spilling required and the statepoint should simply record that fact. Depending on how challenging that turns out to be, we may invest in a smarter global stack slot assignment mechanism as a stop gap measure.
Reviewed by: atrick, ributzka
llvm-svn: 223137
Go through implicit defs of CSMI and MI, and clear the kill flags on
their uses in all the instructions between CSMI and MI.
We might have made some of the kill flags redundant, consider:
subs ... %NZCV<imp-def> <- CSMI
csinc ... %NZCV<imp-use,kill> <- this kill flag isn't valid anymore
subs ... %NZCV<imp-def> <- MI, to be eliminated
csinc ... %NZCV<imp-use,kill>
Since we eliminated MI, and reused a register imp-def'd by CSMI
(here %NZCV), that register, if it was killed before MI, should have
that kill flag removed, because it's lifetime was extended.
Also, add an exhaustive testcase for the motivating example.
Reviewed by: Juergen Ributzka <juergen@apple.com>
llvm-svn: 223133
The blocking code originated in ARM, which is more aggressive about casting
types to a canonical representative before doing anything else, so I missed out
most vector HFAs and broke the ABI. This should fix it.
llvm-svn: 223126
The default ARM floating-point mode does not support IEEE 754 mode exactly. Of
relevance to this patch is that input denormals are flushed to zero. The way in
which they're flushed to zero depends on the architecture,
* For VFPv2, it is implementation defined as to whether the sign of zero is
preserved.
* For VFPv3 and above, the sign of zero is always preserved when a denormal
is flushed to zero.
When FP support has been disabled, the strategy taken by this patch is to
assume the software support will mirror the behaviour of the hardware support
for the target *if it existed*. That is, for architectures which can only have
VFPv2, it is assumed the software will flush to positive zero. For later
architectures it is assumed the software will flush to zero preserving sign.
Change-Id: Icc5928633ba222a4ba3ca8c0df44a440445865fd
llvm-svn: 223110
Summary:
".weak" symbols cannot be consumed by ptxas (PR21685). This patch makes the
weak directive in MCAsmPrinter customizable, and disables emitting ".weak"
symbols for NVPTX.
Test Plan: weak-linkage.ll
Reviewers: jholewinski
Reviewed By: jholewinski
Subscribers: majnemer, jholewinski, llvm-commits
Differential Revision: http://reviews.llvm.org/D6455
llvm-svn: 223077
r208210 introduced an optimization that improves the vector select
codegen by doing the setcc on vectors directly.
This is a problem they the setcc operands are i1s, because the
optimization would create vectors of i1, which aren't legal.
Part of PR21549.
Differential Revision: http://reviews.llvm.org/D6308
llvm-svn: 223075
r213378 improved f16 bitcasts, so that they go directly through subregs,
instead of through the stack. That code now causes an assertion failure
for bitcasts from other 16-bits types (most importantly v2i8).
Correct that by doing the custom lowering for i16 bitcasts only when the
input is an f16.
Part of PR21549.
Differential Revision: http://reviews.llvm.org/D6307
llvm-svn: 223074
The MachineVerifier used to check that there was always exactly one
unconditional branch to a non-landingpad (normal) successor.
If that normal successor to an invoke BB is unreachable, it seems
reasonable to only have one successor, the landing pad.
On targets other than AArch64 (and on AArch64 with a different testcase),
the branch folder turns the branch to the landing pad into a fallthrough.
The MachineVerifier, which relies on AnalyzeBranch, is unable to check
the condition, and doesn't complain. However, it does in this specific
testcase, where the branch to the landing pad remained.
Make the MachineVerifier accept it.
llvm-svn: 223059
This can significantly reduce the size of the switch, allowing for more
efficient lowering.
I also worked with the idea of exploiting unreachable defaults by
omitting the range check for jump tables, but always ended up with a
non-neglible binary size increase. It might be worth looking into some more.
llvm-svn: 223049
Summary:
PowerPC DWARF unwind info defined CFA as SP + offset even in a function
where the stack had been dynamically realigned. This clearly doesn't
work because the offset from SP to CFA is not a constant. Fix it by
defining CFA as BP instead.
This was causing the AddressSanitizer null_deref test to fail 50% of
the time, depending on whether SP happened to be 32-byte aligned on
entry to a particular function or not.
Reviewers: willschm, uweigand, hfinkel
Reviewed By: hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6410
llvm-svn: 222996
This commit fixes a bug in stack protector pass where edge weights were not set
when new basic blocks were added to lists of successor basic blocks.
Differential Revision: http://reviews.llvm.org/D5766
llvm-svn: 222987
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
Allow unaligned 16-byte memop codegen for btver2. No functional changes for any other subtargets.
Replace the existing supposed small memcpy test with an actual test of a small memcpy.
The previous test wasn't using FileCheck either.
This patch should allow us to close PR21541 ( http://llvm.org/bugs/show_bug.cgi?id=21541 ).
Differential Revision: http://reviews.llvm.org/D6360
llvm-svn: 222925
The AAPCS treats small structs and homogeneous floating (or vector) aggregates
specially, and guarantees they either get passed as a contiguous block of
registers, or prevent any future use of those registers and get passed on the
stack.
This concept can fit quite neatly into LLVM's own type system, mapping an HFA
to [N x float] and so on, and small structs to [N x i64]. Doing so allows
front-ends to emit AAPCS compliant code without having to duplicate the
register counting logic.
llvm-svn: 222903
The string data for string-valued build attributes were being unconditionally
uppercased. There is no mention in the ARM ABI addenda about case conventions,
so it's technically implementation defined as to whether the data are
capitialised in some way or not. However, there are good reasons not to
captialise the data.
* It's less work.
* Some vendors may legitimately have case-sensitive checks for these
attributes which would fail on LLVM generated object files.
* There could be locale issues with uppercasing.
The original reasons for uppercasing appear to have stemmed from an
old codesourcery toolchain behaviour, see
http://comments.gmane.org/gmane.comp.compilers.llvm.cvs/87133
This patch makes the object file emitted no longer captialise string
data, it encodes as seen in the assembly source.
Change-Id: Ibe20dd6e60d2773d57ff72a78470839033aa5538
llvm-svn: 222882
This mostly entails adding relocations, however there are a couple of
changes to existing relocations:
1. R_AARCH64_NONE is defined to be zero rather than 256
R_AARCH64_NONE has been defined to be zero for a long time elsewhere
e.g. binutils and glibc since the submission of the AArch64 port in
2012 so this is required for compatibility.
2. R_AARCH64_TLSDESC_ADR_PAGE renamed to R_AARCH64_TLSDESC_ADR_PAGE21
I don't think there is any way for relocation names to leak out of LLVM
so this should not break anything.
Tested with check-all with no regressions.
llvm-svn: 222821
including SAE mode and memory operand.
Added AVX512_maskable_scalar template, that should cover all scalar instructions in the future.
The main difference between AVX512_maskable_scalar<> and AVX512_maskable<> is using X86select instead of vselect.
I need it, because I can't create vselect node for MVT::i1 mask for scalar instruction.
http://reviews.llvm.org/D6378
llvm-svn: 222820
Since (v)pslldq / (v)psrldq instructions resolve to a single input argument it is useful to match it much earlier than we currently do - this prevents more complicated shuffles (notably insertion into a zero vector) matching before it.
Differential Revision: http://reviews.llvm.org/D6409
llvm-svn: 222796
The pattern matching failed to recognize all instances of "-1", because when
comparing against "-1" we didn't use an APInt of the same bitwidth.
This commit fixes this and also adds inverse versions of the conditon to catch
more cases.
llvm-svn: 222722
This does not matter on newer cores (where we can use reciprocal estimates in
fast-math mode anyway), but for older cores this allows us to generate better
fast-math code where we have multiple FDIVs with a common divisor.
llvm-svn: 222710
This patch teaches function 'transformVSELECTtoBlendVECTOR_SHUFFLE' how to
convert VSELECT dag nodes to shuffles on targets that do not have SSE4.1.
On pre-SSE4.1 targets, we can still perform blend operations using movss/movsd.
Also, removed a target specific combine that performed a premature lowering of
VSELECT nodes to target specific MOVSS/MOVSD nodes.
llvm-svn: 222647
r222375 made some improvements to build_vector lowering of v4x32 and v4xf32 into an insertps, but it missed a case where:
1. A single extracted element is used twice.
2. The lower of the two non-zero indexes should be preserved, and the higher should be used for the dest mask.
This caused a crash, since the source value for the insertps ends-up uninitialized.
Differential Revision: http://reviews.llvm.org/D6377
llvm-svn: 222635
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
has a remarkably unique and efficient lowering.
While we get this some of the time already, we miss a few cases and
there wasn't a principled reason we got it. We should at least test
this. v8 already has tests for this pattern.
llvm-svn: 222607
This s_mov_b32 will write to a virtual register from the M0Reg
class and all the ds instructions now take an extra M0Reg explicit
argument.
This change is necessary to prevent issues with the scheduler
mixing together instructions that expect different values in the m0
registers.
llvm-svn: 222583
filler such as if delay slot filler have to put NOP instruction into the
delay slot of microMIPS BEQ or BNE instruction which uses the register $0,
then instead of emitting NOP this instruction is replaced by the corresponding
microMIPS compact branch instruction, i.e. BEQZC or BNEZC.
Differential Revision: http://reviews.llvm.org/D3566
llvm-svn: 222580
This patch adds a feature flag to avoid unaligned 32-byte load/store AVX codegen
for Sandy Bridge and Ivy Bridge. There is no functionality change intended for
those chips. Previously, the absence of AVX2 was being used as a proxy to detect
this feature. But that hindered codegen for AVX-enabled AMD chips such as btver2
that do not have the 32-byte unaligned access slowdown.
Performance measurements are included in PR21541 ( http://llvm.org/bugs/show_bug.cgi?id=21541 ).
Differential Revision: http://reviews.llvm.org/D6355
llvm-svn: 222544
shuffle lowering to allow much better blend matching.
Specifically, with the new structure the code seems clearer to me and we
correctly can hit the cases where merging two 128-bit lanes is a clear
win and can be shuffled cheaply afterward.
llvm-svn: 222539
a bunch more improvements.
Non-lane-crossing is fine, the key is that lane merging only makes sense
for single-input shuffles. Not sure why I got so turned around here. The
code all works, I was just using the wrong model for it.
This only updates v4 and v8 lowering. The v16 and v32 lowering requires
restructuring the entire check sequence.
llvm-svn: 222537
Before this patch, the DAGCombiner only tried to convert build_vector dag nodes
into shuffles if all operands were either extract_vector_elt or undef.
This patch improves that logic and teaches the DAGCombiner how to deal with
build_vector dag nodes where one or more operands are zero. A build_vector
dag node with some zero operands is turned into a shuffle only if the resulting
shuffle mask is legal for the target.
llvm-svn: 222536
lanes.
By special casing these we can often either reduce the total number of
shuffles significantly or reduce the number of (high latency on Haswell)
AVX2 shuffles that potentially cross 128-bit lanes. Even when these
don't actually cross lanes, they have much higher latency to support
that. Doing two of them and a blend is worse than doing a single insert
across the 128-bit lanes to blend and then doing a single interleaved
shuffle.
While this seems like a narrow case, it kept cropping up on me and the
difference is *huge* as you can see in many of the test cases. I first
hit this trying to perfectly fix the interleaving shuffle patterns used
by Halide for AVX2.
llvm-svn: 222533
merging 128-bit subvectors and also shuffling all the elements of those
subvectors. Currently we generate pretty bad code for many of these, but
I'm testing a patch that should dramatically improve this in addition to
making the shuffle lowering robust to other changes.
llvm-svn: 222525
E.g., ( a / D; b / D ) -> ( recip = 1.0 / D; a * recip; b * recip)
A hook is added to allow the target to control whether it needs to do such combine.
Reviewed in http://reviews.llvm.org/D6334
llvm-svn: 222510
This mirrors r222331, which enabled SeparateConstOffsetFromGEP on AArch64, in
the PowerPC backend. Yields, on a POWER7 machine, a 30% speedup on
SingleSource/Benchmarks/Shootout/nestedloop (this might just be from LICM,
there is a store moved out of the inner loop) and a potential speedup on
MultiSource/Benchmarks/mediabench/mpeg2/mpeg2dec/mpeg2decode. Regardless, it
makes some code look cleaner, and synchronizing the backends in this regard
seems like a generally good thing.
llvm-svn: 222504
Windows itanium targets the MSVCRT, and the stack probe symbol is provided by
MSVCRT. This corrects the emission of stack probes on i686-windows-itanium.
llvm-svn: 222439
This patch improves the lowering of v4f32 and v4i32 build_vector dag nodes
that are known to have at least two non-zero elements.
With this patch, a build_vector that performs a blend with zero is
converted into a shuffle. This is done to let the shuffle legalizer expand
the dag node in a optimal way. For example, if we know that a build_vector
performs a blend with zero, we can try to lower it as a movq/blend instead of
always selecting an insertps.
This patch also improves the logic that lowers a build_vector into a insertps
with zero masking. See for example the extra test cases added to test sse41.ll.
Differential Revision: http://reviews.llvm.org/D6311
llvm-svn: 222375
A register operand that has a common sub-class with its instruction's
defined register class is not always legal. For example,
SReg_32 and M0Reg both have a common sub-class, but we can't
use an SReg_32 in instructions that expect a M0Reg.
This prevents the llvm.SI.sendmsg.ll test from failing when the fold
operand pass is added.
llvm-svn: 222368
This patch builds on http://reviews.llvm.org/D5598 to perform byte rotation shuffles (lowerVectorShuffleAsByteRotate) on pre-SSSE3 (palignr) targets - pre-SSSE3 is only enabled on i8 and i16 vector targets where it is a more definite performance gain.
I've also added a separate byte shift shuffle (lowerVectorShuffleAsByteShift) that makes use of the ability of the SLLDQ/SRLDQ instructions to implicitly shift in zero bytes to avoid the need to create a zero register if we had used palignr.
Differential Revision: http://reviews.llvm.org/D5699
llvm-svn: 222340
SeparateConstOffsetFromGEP can gives more optimizaiton opportunities related to GEPs, which benefits EarlyCSE
and LICM. By enabling these passes we can have better address calculations and generate a better addressing
mode. Some SPEC 2006 benchmarks (astar, gobmk, namd) have obvious improvements on Cortex-A57.
Reviewed in http://reviews.llvm.org/D5864.
llvm-svn: 222331
This partially makes up for not having address spaces
used for alias analysis in some simple cases.
This is not yet enabled by default so shouldn't change anything yet.
llvm-svn: 222286
Under many circumstances the stack is not 32-byte aligned, resulting in the use of the vmovups/vmovupd/vmovdqu instructions when inserting ymm reloads/spills.
This minor patch adds these instructions to the isFrameLoadOpcode/isFrameStoreOpcode helpers so that they can be correctly identified and not be treated as folded reloads/spills.
This has also been noticed by http://llvm.org/bugs/show_bug.cgi?id=18846 where it was causing redundant spills - I've added a reduced test case at test/CodeGen/X86/pr18846.ll
Differential Revision: http://reviews.llvm.org/D6252
llvm-svn: 222281
shift-right for booleans (i1).
Arithmetic shift-right immediate with sign-/zero-extensions also works for
boolean values. Update the assert and the test cases to reflect that fact.
llvm-svn: 222272
shift-right for booleans (i1).
Logical shift-right immediate with sign-/zero-extensions also works for boolean
values. Update the assert and the test cases to reflect that fact.
llvm-svn: 222270
Shifts also perform sign-/zero-extends to larger types, which requires us to emit
an integer extend instead of a simple COPY.
Related to PR21594.
llvm-svn: 222257
This should expose more of the actually used VALU
instructions to the machine optimization passes.
This also should help getting i1 handling into a better state.
For not entirly understood reasons, this fixes the split-scalar-i64-add.ll
test where a 64-bit add would only partially be moved to the VALU
resulting in use of undefined VCC.
llvm-svn: 222256
This change emits a COPY for a shift-immediate with a "zero" shift value.
This fixes PR21594 where we emitted a shift instruction with an incorrect
immediate operand.
llvm-svn: 222247
The triple parser should only accept existing architecture names
when the triple starts with armv, armebv, thumbv or thumbebv.
Patch by Gabor Ballabas.
llvm-svn: 222129
This was motivated by a bug which caused code like this to be
miscompiled:
declare void @take_ptr(i8*)
define void @test() {
%addr1.32 = alloca i8
%addr2.32 = alloca i32, i32 1028
call void @take_ptr(i8* %addr1)
ret void
}
This was emitting the following assembly to get the value of %addr1:
add r0, sp, #1020
add r0, r0, #8
However, "add r0, r0, #8" is not a valid Thumb1 instruction, and this
could not be assembled. The generated object file contained this,
resulting in r0 holding SP+8 rather tha SP+1028:
add r0, sp, #1020
add r0, sp, #8
This function looked like it could have caused miscompilations for
other combinations of registers and offsets (though I don't think it is
currently called with these), and the heuristic it used did not match
the emitted code in all cases.
llvm-svn: 222125
Some optimisations in DAGCombiner cause miscompilations for targets that use
TargetLowering::UndefinedBooleanContent, because they assume that the results
of a SELECT_CC node are boolean values, and can be safely ANDed, ORed and
XORed. These optimisations are only valid for targets that use
ZeroOrOneBooleanContent or ZeroOrNegativeOneBooleanContent.
This is a follow-up to D6210/r221693.
llvm-svn: 222123
This patch teaches the DAGCombiner how to combine shuffles according to rules:
shuffle(shuffle(A, Undef, M0), B, M1) -> shuffle(B, A, M2)
shuffle(shuffle(A, B, M0), B, M1) -> shuffle(B, A, M2)
shuffle(shuffle(A, B, M0), A, M1) -> shuffle(B, A, M2)
llvm-svn: 222090
Updated X86TargetLowering::isShuffleMaskLegal to match SHUFP masks with commuted inputs and PSHUFD masks that reference the second input.
As part of this I've refactored isPSHUFDMask to work in a more general manner and allow it to match against either the first or second input vector.
Differential Revision: http://reviews.llvm.org/D6287
llvm-svn: 222087
This gets the correct NaN behavior based on the compare type
the hardware uses. This now passes the new piglit test I have
for this on SI.
Add stricter tests for the operand order.
llvm-svn: 222079
We use to track quite a few "adjusted" offsets through the FrameLowering code
to account for changes in the prologue instructions as we went and allow the
emission of correct CFA annotations. However, we were missing a couple of cases
and the code was almost impenetrable.
It's easier to just add any stack-adjusting instruction to a list and emit them
together.
llvm-svn: 222057
When we folded the DPR alignment gap into a push, we weren't noting the extra
distance from the beginning of the push to the FP, and so FP ended up pointing
at an incorrect offset.
The .cfi_def_cfa_offset directives are still wrong in this case, but I think
that can be improved by refactoring.
llvm-svn: 222056
The test's DWARF stubs were there just to trigger the emission of .cfi
directives. Fortunately, the NetBSD ABI already demands proper DWARF unwind
info, so it's easier to just use that triple.
llvm-svn: 222055
If we have spilled the value of the m0 register, then we need to restore
it with v_readlane_b32 to a regular sgpr, because v_readlane_b32 can't
write to m0.
v_readlane_b32 can't write to m0, so
llvm-svn: 222036
This patch adds builtin support for xvdivdp and xvdivsp, along with a
test case. Straightforward stuff.
There's a companion patch for Clang.
llvm-svn: 221983
getTargetConstant should only be used when you can guarantee the instruction
selected will be able to cope with the raw value. BUILD_VECTOR is rather too
generic for this so we should use getConstant instead. In that case, an
instruction can still consume the constant, but if it doesn't it'll be
materialised through its own round of ISel.
Should fix PR21352.
llvm-svn: 221961
Summary:
This has most of what is needed for mips fast-isel call lowering for O32.
What is missing I will add on the next patch because this patch is already too large.
It should not be doing anything wrong but it will punt on some cases that it is basically
capable of doing.
The mechanism is there for parameters to be passed on the stack but I have not enabled it because it serves as a way for now to prevent some of the strange cases of O32 register passing that I have not fully checked yet and have some issues.
The Mips O32 abi rules are very complicated as far how data is passed in floating and integer registers.
However there is a way to think about this all very simply and this implementation reflects that.
Basically, the ABI rules are written as if everything is passed on the stack and aligned as such.
Once that is conceptually done, it is nearly trivial to reassign those locations to registers and
then all the complexity disappears.
So I have told tablegen that all the data is passed on the stack and during the lowering I fix
this by assigning to registers as per the ABI doc.
This has been my approach and you can line up what I did with the ABI document and see 1 to 1 what
is going on.
Test Plan: callabi.ll
Reviewers: dsanders
Reviewed By: dsanders
Subscribers: jholewinski, echristo, ahatanak, llvm-commits, rfuhler
Differential Revision: http://reviews.llvm.org/D5714
llvm-svn: 221948
in-lane shuffles that aren't always handled well by the current vector
shuffle lowering.
No functionality change yet, that will follow in a subsequent commit.
llvm-svn: 221938
The generic FastISel code would bail, because it can't emit a sign-extend for
AArch64. This copies the code over and uses AArch64 specific emit functions.
This is not ideal and 'computeAddress' should handles this, so it can fold the
address computation into the memory operation.
I plan to clean up 'computeAddress' anyways, so I will add that in a future
commit.
Related to rdar://problem/18962471.
llvm-svn: 221923
If a function is just an unreachable, this would hit a
"this is not a MachO target" assertion because of setting
HasSubsectionViaSymbols.
llvm-svn: 221920
e.g. v_mad_f32 a, b, c -> v_mad_f32 b, a, c
This simplifies matching v_madmk_f32.
This looks somewhat surprising, but it appears to be
OK to do this. We can commute src0 and src1 in all
of these instructions, and that's all that appears
to matter.
llvm-svn: 221910
Normally entries can only move to a lower address, but when that wasn't viable,
the user's block was considered anyway. Unfortunately, it went via
createNewWater which wasn't designed to handle the case where there's already
an island after the block.
Unfortunately, the test we have is slow and fragile, and I couldn't reduce it
to anything sane even with the @llvm.arm.space intrinsic. The test change here
is recreating the previous one after the change.
rdar://problem/18545506
llvm-svn: 221905
We were using a naive heuristic to determine whether a basic block already had
an unconditional branch at the end. This mostly corresponded to reality
(assuming branches got optimised) because there's not much point in a branch to
the next block, but could go wrong.
llvm-svn: 221904
Creating tests for the ConstantIslands pass is very difficult, since it depends
on precise layout details. Having the ability to precisely inject a number of
bytes into the stream helps greatly.
llvm-svn: 221903
between splitting a vector into 128-bit lanes and recombining them vs.
decomposing things into single-input shuffles and a final blend.
This handles a large number of cases in AVX1 where the cross-lane
shuffles would be much more expensive to represent even though we end up
with a fast blend at the root. Instead, we can do a better job of
shuffling in a single lane and then inserting it into the other lanes.
This fixes the remaining bits of Halide's regression captured in PR21281
for AVX1. However, the bug persists in AVX2 because I've made this
change reasonably conservative. The cases where it makes sense in AVX2
to split into 128-bit lanes are much more rare because we can often do
full permutations across all elements of the 256-bit vector. However,
the particular test case in PR21281 is an example of one of the rare
cases where it is *always* better to work in a single 128-bit lane. I'm
going to try to teach the logic to detect and form the good code even in
AVX2 next, but it will need to use a separate heuristic.
Finally, there is one pesky regression here where we previously would
craftily use vpermilps in AVX1 to shuffle both high and low halves at
the same time. We no longer pull that off, and not for any really good
reason. Ultimately, I think this is just another missing nuance to the
selection heuristic that I'll try to add in afterward, but this change
already seems strictly worth doing considering the magnitude of the
improvements in common matrix math shuffle patterns.
As always, please let me know if this causes a surprising regression for
you.
llvm-svn: 221861
re-combining shuffles because nothing was available in the wider vector
type.
The key observation (which I've put in the comments for future
maintainers) is that at this point, no further combining is really
possible. And so even though these shuffles trivially could be combined,
we need to actually do that as we produce them when producing them this
late in the lowering.
This fixes another (huge) part of the Halide vector shuffle regressions.
As it happens, this was already well covered by the tests, but I hadn't
noticed how bad some of these got. The specific patterns that turn
directly into unpckl/h patterns were occurring *many* times in common
vector processing code.
There are still more problems here sadly, but trying to incrementally
tease them apart and it looks like this is the core of the problem in
the splitting logic.
There is some chance of regression here, you can see it in the test
changes. Specifically, where we stop forming pshufb in some cases, it is
possible that pshufb was in fact faster. Intel "says" that pshufb is
slower than the instruction sequences replacing it.
llvm-svn: 221852
Prior to this patch the TypePromotionHelper was promoting only sign extensions.
Supporting zero extensions changes:
- How constants are extended.
- How sign extensions, zero extensions, and truncate are composed together.
- How the type of the extended operation is recorded. Now we need to know the
kind of the extension as well as its type.
Each change is fairly small, unlike the diff.
Most of the diff are comments/variable renaming to say "extension" instead of
"sign extension".
The performance improvements on the test suite are within the noise.
Related to <rdar://problem/18310086>.
llvm-svn: 221851
This folds the compare emission into the select emission when possible, so we
can directly use the flags and don't have to emit a separate compare.
Related to rdar://problem/18960150.
llvm-svn: 221847
This is a follow-on to r221706 and r221731 and discussed in more detail in PR21385.
This patch also loosens the testcase checking for btver2. We know that the "1.0" will be loaded, but
we can't tell exactly when, so replace the CHECK-NEXT specifiers with plain CHECKs. The CHECK-NEXT
sequence relied on a quirk of post-RA-scheduling that may change independently of anything in these tests.
llvm-svn: 221819
Summary:
Large-model was added first. With the addition of support for multiple PIC
models in LLVM, now add small-model PIC for 32-bit PowerPC, SysV4 ABI. This
generates more optimal code, for shared libraries with less than about 16380
data objects.
Test Plan: Test cases added or updated
Reviewers: joerg, hfinkel
Reviewed By: hfinkel
Subscribers: jholewinski, mcrosier, emaste, llvm-commits
Differential Revision: http://reviews.llvm.org/D5399
llvm-svn: 221791
cases from Halide folks. This initial step was extracted from
a prototype change by Clay Wood to try and address regressions found
with Halide and the new vector shuffle lowering.
llvm-svn: 221779
This patch enables the vec_vsx_ld and vec_vsx_st intrinsics for
PowerPC, which provide programmer access to the lxvd2x, lxvw4x,
stxvd2x, and stxvw4x instructions.
New LLVM intrinsics are provided to represent these four instructions
in IntrinsicsPowerPC.td. These are patterned after the similar
intrinsics for lvx and stvx (Altivec). In PPCInstrVSX.td, these
intrinsics are tied to the code gen patterns, with additional patterns
to allow plain vanilla loads and stores to still generate these
instructions.
At -O1 and higher the intrinsics are immediately converted to loads
and stores in InstCombineCalls.cpp. This will open up more
optimization opportunities while still allowing the correct
instructions to be generated. (Similar code exists for aligned
Altivec loads and stores.)
The new intrinsics are added to the code that checks for consecutive
loads and stores in PPCISelLowering.cpp, as well as to
PPCTargetLowering::getTgtMemIntrinsic().
There's a new test to verify the correct instructions are generated.
The loads and stores tend to be reordered, so the test just counts
their number. It runs at -O2, as it's not very effective to test this
at -O0, when many unnecessary loads and stores are generated.
I ended up having to modify vsx-fma-m.ll. It turns out this test case
is slightly unreliable, but I don't know a good way to prevent
problems with it. The xvmaddmdp instructions read and write the same
register, which is one of the multiplicands. Commutativity allows
either to be chosen. If the FMAs are reordered differently than
expected by the test, the register assignment can be different as a
result. Hopefully this doesn't change often.
There is a companion patch for Clang.
llvm-svn: 221767
This commit adds a new pass that can inject checks before indirect calls to
make sure that these calls target known locations. It supports three types of
checks and, at compile time, it can take the name of a custom function to call
when an indirect call check fails. The default failure function ignores the
error and continues.
This pass incidentally moves the function JumpInstrTables::transformType from
private to public and makes it static (with a new argument that specifies the
table type to use); this is so that the CFI code can transform function types
at call sites to determine which jump-instruction table to use for the check at
that site.
Also, this removes support for jumptables in ARM, pending further performance
analysis and discussion.
Review: http://reviews.llvm.org/D4167
llvm-svn: 221708
This is a first step for generating SSE rcp instructions for reciprocal
calcs when fast-math allows it. This is very similar to the rsqrt optimization
enabled in D5658 ( http://reviews.llvm.org/rL220570 ).
For now, be conservative and only enable this for AMD btver2 where performance
improves significantly both in terms of latency and throughput.
We may never enable this codegen for Intel Core* chips because the divider circuits
are just too fast. On SandyBridge, divss can be as fast as 10 cycles versus the 21
cycle critical path for the rcp + mul + sub + mul + add estimate.
Follow-on patches may allow configuration of the number of Newton-Raphson refinement
steps, add AVX512 support, and enable the optimization for more chips.
More background here: http://llvm.org/bugs/show_bug.cgi?id=21385
Differential Revision: http://reviews.llvm.org/D6175
llvm-svn: 221706
My original support for the general dynamic and local dynamic TLS
models contained some fairly obtuse hacks to generate calls to
__tls_get_addr when lowering a TargetGlobalAddress. Rather than
generating real calls, special GET_TLS_ADDR nodes were used to wrap
the calls and only reveal them at assembly time. I attempted to
provide correct parameter and return values by chaining CopyToReg and
CopyFromReg nodes onto the GET_TLS_ADDR nodes, but this was also not
fully correct. Problems were seen with two back-to-back stores to TLS
variables, where the call sequences ended up overlapping with unhappy
results. Additionally, since these weren't real calls, the proper
register side effects of a call were not recorded, so clobbered values
were kept live across the calls.
The proper thing to do is to lower these into calls in the first
place. This is relatively straightforward; see the changes to
PPCTargetLowering::LowerGlobalTLSAddress() in PPCISelLowering.cpp.
The changes here are standard call lowering, except that we need to
track the fact that these calls will require a relocation. This is
done by adding a machine operand flag of MO_TLSLD or MO_TLSGD to the
TargetGlobalAddress operand that appears earlier in the sequence.
The calls to LowerCallTo() eventually find their way to
LowerCall_64SVR4() or LowerCall_32SVR4(), which call FinishCall(),
which calls PrepareCall(). In PrepareCall(), we detect the calls to
__tls_get_addr and immediately snag the TargetGlobalTLSAddress with
the annotated relocation information. This becomes an extra operand
on the call following the callee, which is expected for nodes of type
tlscall. We change the call opcode to CALL_TLS for this case. Back
in FinishCall(), we change it again to CALL_NOP_TLS for 64-bit only,
since we require a TOC-restore nop following the call for the 64-bit
ABIs.
During selection, patterns in PPCInstrInfo.td and PPCInstr64Bit.td
convert the CALL_TLS nodes into BL_TLS nodes, and convert the
CALL_NOP_TLS nodes into BL8_NOP_TLS nodes. This replaces the code
removed from PPCAsmPrinter.cpp, as the BL_TLS or BL8_NOP_TLS
nodes can now be emitted normally using their patterns and the
associated printTLSCall print method.
Finally, as a result of these changes, all references to get-tls-addr
in its various guises are no longer used, so they have been removed.
There are existing TLS tests to verify the changes haven't messed
anything up). I've added one new test that verifies that the problem
with the original code has been fixed.
llvm-svn: 221703
The ISel lowering for global TLS access in PIC mode was creating a pseudo
instruction that is later expanded to a call, but the code was not
setting the hasCalls flag in the MachineFrameInfo alongside the adjustsStack
flag. This caused some functions to be mistakenly recognized as leaf functions,
and this in turn affected the decision to eliminate the frame pointer.
With the fix, hasCalls is properly set and the leaf frame pointer is correctly
preserved.
llvm-svn: 221695
LLVM replaces the SelectionDAG pattern (xor (set_cc cc x y) 1) with
(set_cc !cc x y), which is only correct when the xor has type i1.
Instead, we should check that the constant operand to the xor is all
ones.
llvm-svn: 221693
Summary:
This patch enables code generation for the MIPS II target. Pre-Mips32
targets don't have the MUL instruction, so we add the correspondent
pattern that uses the MULT/MFLO combination in order to retrieve the
product.
This is WIP as we don't support code generation for select nodes due to
the lack of conditional-move instructions.
Reviewers: dsanders
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D6150
llvm-svn: 221686
This fixes an issue with matching trunc -> assertsext -> zext on x86-64, which would not zero the high 32-bits. See PR20494 for details.
Recommitting - This time, with a hopefully working test.
Differential Revision: http://reviews.llvm.org/D6128
llvm-svn: 221672
AVX2 is available.
According to IACA, the new lowering has a throughput of 8 cycles instead of 13
with the previous one.
Althought this lowering kicks in some SPECs benchmarks, the performance
improvement was within the noise.
Correctness testing has been done for the whole range of uint32_t with the
following program:
uint4 v = (uint4) {0,1,2,3};
uint32_t i;
//Check correctness over entire range for uint4 -> float4 conversion
for( i = 0; i < 1U << (32-2); i++ )
{
float4 t = test(v);
float4 c = correct(v);
if( 0xf != _mm_movemask_ps( t == c ))
{
printf( "Error @ %vx: %vf vs. %vf\n", v, c, t);
return -1;
}
v += 4;
}
Where "correct" is the old lowering and "test" the new one.
The patch adds a test case for the two custom lowering instruction.
It also modifies the vector cost model, which is why cast.ll and uitofp.ll are
modified.
2009-02-26-MachineLICMBug.ll is also modified because we now hoist 7
instructions instead of 4 (3 more constant loads).
rdar://problem/18153096>
llvm-svn: 221657
In the case we optimize an integer extend away and replace it directly with the
source register, we also have to clear all kill flags at all its uses.
This is necessary, because the orignal IR instruction might be trivially dead,
but we replaced it with a nop at MI level.
llvm-svn: 221628
This fixes an issue with matching trunc -> assertsext -> zext on x86-64, which would not zero the high 32-bits.
See PR20494 for details.
Differential Revision: http://reviews.llvm.org/D6128
llvm-svn: 221626
Ahmed Bougacha