Fixed issue with the combine of CONCAT_VECTOR of 2 BUILD_VECTOR nodes - the optimisation wasn't ensuring that the scalar operands of both nodes were the same type/size for implicit truncation.
Test case spotted by Patrik Hagglund
llvm-svn: 235371
Summary:
This fixes http://llvm.org/bugs/show_bug.cgi?id=16439.
This is one possible way to approach this. The other would be to split InL>>(nbits-Amt) into (InL>>(nbits-1-Amt))>>1, which is also valid since since we only need to care about Amt up nbits-1. It's hard to tell which one is better since the shift might be expensive if this stage of expansion is not yet a legal machine integer, whereas comparisons with zero are relatively cheap at all sizes, but more expensive than a shift if the shift is on a legal machine type.
Patch by Keno Fischer!
Test Plan: regression test from http://reviews.llvm.org/D7752
Reviewers: chfast, resistor
Reviewed By: chfast, resistor
Subscribers: sanjoy, resistor, chfast, llvm-commits
Differential Revision: http://reviews.llvm.org/D4978
llvm-svn: 235370
X86ISD::ADDSUB, X86ISD::(F)HADD, X86ISD::(F)HSUB should not be selected
if the operand types do not match the result type because vector type
legalization cannot deal with this for custom nodes.
Testcase X86ISD::ADDSUB is attached. I could not create a testcase for
the FHADD/FHSUB cases because of: https://llvm.org/bugs/show_bug.cgi?id=23296
Differential Revision: http://reviews.llvm.org/D9120
llvm-svn: 235367
This fixes a regression introduced at revision 231243.
The target-independent selection algorithm in FastISel knows how to select
a SINT_TO_FP if the target is SSE but not AVX. That is because on X86, the
tablegen'd 'fastEmit' functions know how to select CVTSI2SSrr and CVTSI2SDrr.
Method X86FastISel::X86SelectSIToFP was therefore working under the
wrong assumption that the target was AVX. That assumption was incorrect since
we can have a target that is neither AVX nor SSE.
So, rather than asserting for the presence of AVX, we should have had an
early exit from 'X86SelectSIToFP' if the target was not AVX.
This patch fixes the issue replacing the invalid assertion with an early exit.
Thanks to Dimitry Andric for reporting this problem and for providing a small
reproducible testcase. Added test pr23273.ll.
llvm-svn: 235295
The fix ensures that scalar sources inserted into a vector are the correct bit size.
Integer scalar sources from BUILD_VECTOR and SCALAR_TO_VECTOR nodes may require truncation that this function doesn't currently support.
llvm-svn: 235281
This now emits simple, unoptimized xdata tables for __C_specific_handler
based on the handlers listed in @llvm.eh.actions calls produced by
WinEHPrepare.
This adds support for running __finally blocks when exceptions are
thrown, and removes the old landingpad fan-in codepath.
I ran some manual execution tests on small basic test cases with and
without optimization, as well as on Chrome base_unittests, which uses a
small amount of SEH. I'm sure there are bugs, and we may need to
revert.
llvm-svn: 235154
See r230786 and r230794 for similar changes to gep and load
respectively.
Call is a bit different because it often doesn't have a single explicit
type - usually the type is deduced from the arguments, and just the
return type is explicit. In those cases there's no need to change the
IR.
When that's not the case, the IR usually contains the pointer type of
the first operand - but since typed pointers are going away, that
representation is insufficient so I'm just stripping the "pointerness"
of the explicit type away.
This does make the IR a bit weird - it /sort of/ reads like the type of
the first operand: "call void () %x(" but %x is actually of type "void
()*" and will eventually be just of type "ptr". But this seems not too
bad and I don't think it would benefit from repeating the type
("void (), void () * %x(" and then eventually "void (), ptr %x(") as has
been done with gep and load.
This also has a side benefit: since the explicit type is no longer a
pointer, there's no ambiguity between an explicit type and a function
that returns a function pointer. Previously this case needed an explicit
type (eg: a function returning a void() function was written as
"call void () () * @x(" rather than "call void () * @x(" because of the
ambiguity between a function returning a pointer to a void() function
and a function returning void).
No ambiguity means even function pointer return types can just be
written alone, without writing the whole function's type.
This leaves /only/ the varargs case where the explicit type is required.
Given the special type syntax in call instructions, the regex-fu used
for migration was a bit more involved in its own unique way (as every
one of these is) so here it is. Use it in conjunction with the apply.sh
script and associated find/xargs commands I've provided in rr230786 to
migrate your out of tree tests. Do let me know if any of this doesn't
cover your cases & we can iterate on a more general script/regexes to
help others with out of tree tests.
About 9 test cases couldn't be automatically migrated - half of those
were functions returning function pointers, where I just had to manually
delete the function argument types now that we didn't need an explicit
function type there. The other half were typedefs of function types used
in calls - just had to manually drop the * from those.
import fileinput
import sys
import re
pat = re.compile(r'((?:=|:|^|\s)call\s(?:[^@]*?))(\s*$|\s*(?:(?:\[\[[a-zA-Z0-9_]+\]\]|[@%](?:(")?[\\\?@a-zA-Z0-9_.]*?(?(3)"|)|{{.*}}))(?:\(|$)|undef|inttoptr|bitcast|null|asm).*$)')
addrspace_end = re.compile(r"addrspace\(\d+\)\s*\*$")
func_end = re.compile("(?:void.*|\)\s*)\*$")
def conv(match, line):
if not match or re.search(addrspace_end, match.group(1)) or not re.search(func_end, match.group(1)):
return line
return line[:match.start()] + match.group(1)[:match.group(1).rfind('*')].rstrip() + match.group(2) + line[match.end():]
for line in sys.stdin:
sys.stdout.write(conv(re.search(pat, line), line))
llvm-svn: 235145
This is a major rewrite of the SelectionDAG switch lowering. The previous code
would lower switches as a binary tre, discovering clusters of cases
suitable for lowering by jump tables or bit tests as it went along. To increase
the likelihood of finding jump tables, the binary tree pivot was selected to
maximize case density on both sides of the pivot.
By not selecting the pivot in the middle, the binary trees would not always
be balanced, leading to performance problems in the generated code.
This patch rewrites the lowering to search for clusters of cases
suitable for jump tables or bit tests first, and then builds the binary
tree around those clusters. This way, the binary tree will always be balanced.
This has the added benefit of decoupling the different aspects of the lowering:
tree building and jump table or bit tests finding are now easier to tweak
separately.
For example, this will enable us to balance the tree based on profile info
in the future.
The algorithm for finding jump tables is O(n^2), whereas the previous algorithm
was O(n log n) for common cases, and quadratic only in the worst-case. This
doesn't seem to be major problem in practice, e.g. compiling a file consisting
of a 10k-case switch was only 30% slower, and such large switches should be rare
in practice. Compiling e.g. gcc.c showed no compile-time difference. If this
does turn out to be a problem, we could limit the search space of the algorithm.
This commit also disables all optimizations during switch lowering in -O0.
Differential Revision: http://reviews.llvm.org/D8649
llvm-svn: 235101
The only type that isn't an integer, isn't floating point, and isn't
a vector; ladies and gentlemen, the gift that keeps on giving: x86_mmx!
Fixes PR23246.
Original message (reverted in r235062):
[CodeGen] Combine concat_vectors of scalars into build_vector.
Combine something like:
(v8i8 concat_vectors (v2i8 bitcast (i16)) x4)
into:
(v8i8 (bitcast (v4i16 BUILD_VECTOR (i16) x4)))
If any of the scalars are floating point, use that throughout.
Differential Revision: http://reviews.llvm.org/D8948
llvm-svn: 235072
Remove 'inlinedAt:' from MDLocalVariable. Besides saving some memory
(variables with it seem to be single largest `Metadata` contributer to
memory usage right now in -g -flto builds), this stops optimization and
backend passes from having to change local variables.
The 'inlinedAt:' field was used by the backend in two ways:
1. To tell the backend whether and into what a variable was inlined.
2. To create a unique id for each inlined variable.
Instead, rely on the 'inlinedAt:' field of the intrinsic's `!dbg`
attachment, and change the DWARF backend to use a typedef called
`InlinedVariable` which is `std::pair<MDLocalVariable*, MDLocation*>`.
This `DebugLoc` is already passed reliably through the backend (as
verified by r234021).
This commit removes the check from r234021, but I added a new check
(that will survive) in r235048, and changed the `DIBuilder` API in
r235041 to require a `!dbg` attachment whose 'scope:` is in the same
`MDSubprogram` as the variable's.
If this breaks your out-of-tree testcases, perhaps the script I used
(mdlocalvariable-drop-inlinedat.sh) will help; I'll attach it to PR22778
in a moment.
llvm-svn: 235050
Add missing `!dbg` attachments to `@llvm.dbg.*` intrinsics. I updated
these using a script (add-dbg-to-intrinsics.sh) that I'll attach to
PR22778 for posterity.
llvm-svn: 235040
This is a 1-line patch (with a TODO for AVX because that will affect
even more regression tests) that lets us substitute the appropriate
64-bit store for the float/double/int domains.
It's not clear to me exactly what the difference is between the 0xD6 (MOVPQI2QImr) and
0x7E (MOVSDto64mr) opcodes, but this is apparently the right choice.
Differential Revision: http://reviews.llvm.org/D8691
llvm-svn: 235014
Set the transform bar at 2 divisions because the fastest current
x86 FP divider circuit is in SandyBridge / Haswell at 10 cycle
latency (best case) relative to a 5 cycle multiplier.
So that's the worst case for this transform (no latency win),
but multiplies are obviously pipelined while divisions are not,
so there's still a big throughput win which we would expect to
show up in typical FP code.
These are the sequences I'm comparing:
divss %xmm2, %xmm0
mulss %xmm1, %xmm0
divss %xmm2, %xmm0
Becomes:
movss LCPI0_0(%rip), %xmm3 ## xmm3 = mem[0],zero,zero,zero
divss %xmm2, %xmm3
mulss %xmm3, %xmm0
mulss %xmm1, %xmm0
mulss %xmm3, %xmm0
[Ignore for the moment that we don't optimize the chain of 3 multiplies
into 2 independent fmuls followed by 1 dependent fmul...this is the DAG
version of: https://llvm.org/bugs/show_bug.cgi?id=21768 ...if we fix that,
then the transform becomes even more profitable on all targets.]
Differential Revision: http://reviews.llvm.org/D8941
llvm-svn: 235012
Summary:
This change moves creating calls to `llvm.uadd.with.overflow` from
InstCombine to CodeGenPrep. Combining overflow check patterns into
calls to the said intrinsic in InstCombine inhibits optimization because
it introduces an intrinsic call that not all other transforms and
analyses understand.
Depends on D8888.
Reviewers: majnemer, atrick
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D8889
llvm-svn: 234638
The bug manifests when there are two loads and two stores chained as follows in
a DAG,
(ld v3f32) -> (st f32) -> (ld v3f32) -> (st f32)
and the stores' values are extracted from the preceding vector loads.
MergeConsecutiveStores would replace the first store in the chain with the
merged vector store, which would create a cycle between the merged store node
and the last load node that appears in the chain.
This commits fixes the bug by replacing the last store in the chain instead.
rdar://problem/20275084
Differential Revision: http://reviews.llvm.org/D8849
llvm-svn: 234430
Fast isel used to zero extends immediates to 64 bits. This normally goes
unnoticed because the value is truncated to 32 bits for output.
Two cases were it is noticed:
* We fail to use smaller encodings.
* If the original constant was smaller than i32.
In the tests using i1 constants, codegen would change to use -1, which is fine
(and matches what regular isel does) since only the lowest bit is then used.
Instead, this patch then changes the ir to use i8 constants, which looks more
like what clang produces.
llvm-svn: 234249
This patch allows SSE4.1 targets to use (V)PINSRB to create 16i8 vectors by inserting i8 scalars directly into a XMM register instead of merging pairs of i8 scalars into a i16 and using the SSE2 PINSRW instruction.
This allows folding of byte loads and reduces scalar register usage as well.
Differential Revision: http://reviews.llvm.org/D8839
llvm-svn: 234193
This allows the compiler/assembly programmer to switch back to a
section. This in turn fixes the bootstrap failure on powerpc (tested
on gcc110) without changing the ppc codegen at all.
I will try to cleanup the various getELFSection overloads in a followup patch.
Just using a default argument now would lead to ambiguities.
llvm-svn: 234099
Previously the patterns didn't have high enough priority and we would only use the GR32 form if the only the upper 32 or 56 bits were zero.
Fixes PR23100.
llvm-svn: 234075
This patch attempts to fold the shuffling of 'scalar source' inputs - BUILD_VECTOR and SCALAR_TO_VECTOR nodes - if the shuffle node is the only user. This folds away a lot of unnecessary shuffle nodes, and allows quite a bit of constant folding that was being missed.
Differential Revision: http://reviews.llvm.org/D8516
llvm-svn: 234004
Without this patch, we split the 256-bit vector into halves and produced something like:
movzwl (%rdi), %eax
vmovd %eax, %xmm0
vxorps %xmm1, %xmm1, %xmm1
vblendps $15, %ymm0, %ymm1, %ymm0 ## ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
Now, we eliminate the xor and blend because those zeros are free with the vmovd:
movzwl (%rdi), %eax
vmovd %eax, %xmm0
This should be the final fix needed to resolve PR22685:
https://llvm.org/bugs/show_bug.cgi?id=22685
llvm-svn: 233941
For code like this:
define <8 x i32> @load_v8i32() {
ret <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
}
We produce this AVX code:
_load_v8i32: ## @load_v8i32
movl $7, %eax
vmovd %eax, %xmm0
vxorps %ymm1, %ymm1, %ymm1
vblendps $1, %ymm0, %ymm1, %ymm0 ## ymm0 = ymm0[0],ymm1[1,2,3,4,5,6,7]
retq
There are at least 2 bugs in play here:
We're generating a blend when a move scalar does the same job using 2 less instruction bytes (see FIXMEs).
We're not matching an existing pattern that would eliminate the xor and blend entirely. The zero bytes are free with vmovd.
The 2nd fix involves an adjustment of "AddedComplexity" [1] and mostly masks the 1st problem.
[1] AddedComplexity has close to no documentation in the source.
The best we have is this comment: "roughly corresponds to the number of nodes that are covered".
It appears that x86 has bastardized this definition by inflating its values for some other
undocumented reason. For example, we have a pattern with "AddedComplexity = 400" (!).
I searched my way to this page:
https://groups.google.com/forum/#!topic/llvm-dev/5UX-Og9M0xQ
Differential Revision: http://reviews.llvm.org/D8794
llvm-svn: 233931
I'm playing with supporting custom stack map formats with statepoints. While
doing so, I noticed that the existing implementation didn't indicate inherently
unsized frames. This change essentially just ports the functionality that already
exists for the default StackMaps section to custom stackmaps.
llvm-svn: 233891
addl has higher throughput and this was needlessly picking a suboptimal
encoding causing PR23098.
I wish there was a way of doing this without further duplicating tbl-
generated patterns, but so far I haven't found one.
llvm-svn: 233832
I suggested this change in D7898 (http://llvm.org/viewvc/llvm-project?view=revision&revision=231354)
It improves the v4i64 case although not optimally. This AVX codegen:
vmovq {{.*#+}} xmm0 = mem[0],zero
vxorpd %ymm1, %ymm1, %ymm1
vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3]
Becomes:
vmovsd {{.*#+}} xmm0 = mem[0],zero
Unfortunately, this doesn't completely solve PR22685. There are still at least 2 problems under here:
We're not handling v32i8 / v16i16.
We're not getting the FP / int domains right for instruction selection.
But since this patch alone appears to do no harm, reduces code duplication, and helps v4i64,
I'm submitting this patch ahead of fixing the above.
Differential Revision: http://reviews.llvm.org/D8341
llvm-svn: 233704
We used to miss non-Q YMM integer vectors, and, non-Q/D XMM integer
vectors.
While there, change the v4i32 patterns to prefer MOVNTDQ.
llvm-svn: 233668
Adds a test to verify the behavior that r233153 restored: 'optnone'
does not spuriously disable the DAG combiner, and in fact there are
cases where the DAG combiner must run (even at -O0 or 'optnone') in
order for codegen to succeed.
Differential Revision: http://reviews.llvm.org/D8614
llvm-svn: 233584
Fix testcases that don't pass the verifier after a WIP patch to check
`MDSubprogram` operands more effectively. I found the following issues:
- When `isDefinition: false`, the `variables:` field might point at
`!{i32 786468}`, or at a tuple that pointed at an empty tuple with
the comment "previously: invalid DW_TAG_base_type" (I vaguely recall
adding those comments during an upgrade script). In these cases, I
just dropped the array.
- The `variables:` field might point at something like `!{!{!8}}`,
where `!8` was an `MDLocation`. I removed the extra layer of
indirection.
- Invalid `type:` (not an `MDSubroutineType`).
llvm-svn: 233466
Fix debug info in these tests, which started failing with a WIP patch to
verify compile units and types. The problems look like they were all
caused by bitrot. They fell into these categories:
- Using `!{i32 0}` instead of `!{}`.
- Using `!{null}` instead of `!{}`.
- Using `!MDExpression()` instead of `!{}`.
- Using `!8` instead of `!{!8}`.
- `file:` references that pointed at `MDCompileUnit`s instead of the
same `MDFile` as the compile unit.
- `file:` references that were numerically off-by-one or (off-by-ten).
llvm-svn: 233415
nodes.
When a node is terminal it is pushed at the end of the list of the copies to
coalesce instead of being completely ignored. In effect, this reduces its
priority over non-terminal nodes.
Because of that, we do not miss the rematerialization opportunities, nor the
copies that can be merged with more complex, than the terminal rule,
interference checks.
Related to PR22768.
llvm-svn: 233395
Change `LLParser` to require a non-null `scope:` field for both
`MDLocation` and `MDLocalVariable`. There's no need to wait for the
verifier for this check. This also allows their `::getImpl()` methods
to assert that the incoming scope is non-null.
llvm-svn: 233394
This was discussed a while back and I left it optional for migration. Since it's been far more than the 'week or two' that was discussed, time to actually make this manditory.
llvm-svn: 233357
This patch adds support for explicitly provided spill slots in the GC arguments of a gc.statepoint. This is somewhat analogous to gcroot, but leverages the STATEPOINT MI node and StackMap infrastructure. The motivation for this is:
1) The stack spilling code for gc.statepoints hasn't advanced as fast as I'd like. One major option is to give up on doing spilling in the backend and do it at the IR level instead. We'd give up the ability to have gc values in registers, but that's a minor cost in practice. We are not neccessarily moving in that direction, but having the ability to prototype such a thing cheaply is interesting.
2) I want to port the gcroot lowering to use the statepoint infastructure. Given the metadata printers for gcroot expect a fixed set of stack roots, it's easiest to just reuse the explicit stack slots and pass them directly to the underlying statepoint.
I'm holding off on the documentation for the new feature until I'm reasonable sure this is going to stick around.
llvm-svn: 233356
This test returns nonnative integer types which aren't supported on all targets.
The real issue with the SelectionDAG scheduler is with x86 EFLAGS.
llvm-svn: 233355
Fix testcases whose variables are invalid. I'm working on a patch that
adds `Verifier` checks for `MDLocalVariable` (and `MDGlobalVariable`),
and these failed because:
- `scope:` fields need to point at `MDLocalScope` and can't be null.
- `file:` fields need to point at `MDFile`.
- `inlinedAt:` fields need to point at `MDLocation`.
llvm-svn: 233349
This patch teaches fast-isel how to select 128-bit vector load instructions.
Added test CodeGen/X86/fast-isel-vecload.ll
Differential Revision: http://reviews.llvm.org/D8605
llvm-svn: 233270
those are in the same basic block.
The previous approach was the topological order of the basic block.
By default this rule is disabled.
Related to PR22768.
llvm-svn: 233241
This patch adds supports for the vector constant folding of TRUNCATE and FP_EXTEND instructions and tidies up the SINT_TO_FP and UINT_TO_FP instructions to match.
It also moves the vector constant folding for the FNEG and FABS instructions to use the DAG.getNode() functionality like the other unary instructions.
Differential Revision: http://reviews.llvm.org/D8593
llvm-svn: 233224
This patch allows AVX blend instructions to handle insertion into the low
element of a 256-bit vector for the appropriate data types.
For f32, instead of:
vblendps $1, %xmm1, %xmm0, %xmm1 ## xmm1 = xmm1[0],xmm0[1,2,3]
vblendps $15, %ymm1, %ymm0, %ymm0 ## ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
we get:
vblendps $1, %ymm1, %ymm0, %ymm0 ## ymm0 = ymm1[0],ymm0[1,2,3,4,5,6,7]
For f64, instead of:
vmovsd %xmm1, %xmm0, %xmm1 ## xmm1 = xmm1[0],xmm0[1]
vblendpd $3, %ymm1, %ymm0, %ymm0 ## ymm0 = ymm1[0,1],ymm0[2,3]
we get:
vblendpd $1, %ymm1, %ymm0, %ymm0 ## ymm0 = ymm1[0],ymm0[1,2,3]
For the hardware-neglected integer data types, I left a TODO comment in the
code and added regression tests for a follow-on patch.
Differential Revision: http://reviews.llvm.org/D8609
llvm-svn: 233199
1. There were no CHECK-LABELs, so we could match instructions from the wrong function.
2. The use of zero operands meant multiple xor instructions could match some CHECKs.
3. The test was over-specified to need a Sandybridge CPU and Darwin triple.
llvm-svn: 233198
Reverts the code change from r221168 and the relevant test.
It was a mistake to disable the combiner, and based on the ultimate
definition of 'optnone' we shouldn't have considered the test case
as failing in the first place.
llvm-svn: 233153
We can't use TargetFrameLowering::getFrameIndexOffset directly, because
Win64 really wants the offset from the stack pointer at the end of the
prologue. Instead, use X86FrameLowering::getFrameIndexOffsetFromSP(),
which is a pretty close approximiation of that. It fails to handle cases
with interestingly large stack alignments, which is pretty uncommon on
Win64 and is TODO.
llvm-svn: 233137
vperm2x128 instructions have the special ability (aka free hardware capability)
to shuffle zero values into a vector.
This patch recognizes that type of shuffle and generates the appropriate
control byte.
https://llvm.org/bugs/show_bug.cgi?id=22984
Differential Revision: http://reviews.llvm.org/D8563
llvm-svn: 233100
While the uitofp scalar constant folding treats an integer as an unsigned value (from lang ref):
%X = sitofp i8 -1 to double ; yields double:-1.0
%Y = uitofp i8 -1 to double ; yields double:255.0
The vector constant folding was always using sitofp:
%X = sitofp <2 x i8> <i8 -1, i8 -1> to <2 x double> ; yields <double -1.0, double -1.0>
%Y = uitofp <2 x i8> <i8 -1, i8 -1> to <2 x double> ; yields <double -1.0, double -1.0>
This patch fixes this so that the correct opcode is used for sitofp and uitofp.
%X = sitofp <2 x i8> <i8 -1, i8 -1> to <2 x double> ; yields <double -1.0, double -1.0>
%Y = uitofp <2 x i8> <i8 -1, i8 -1> to <2 x double> ; yields <double 255.0, double 255.0>
Differential Revision: http://reviews.llvm.org/D8560
llvm-svn: 233033
Continue to simplify the `DIDescriptor` subclasses, so that they behave
more like raw pointers. Remove `getRaw()`, replace it with an
overloaded `get()`, and overload the arrow and cast operators. Two
testcases started to crash on the arrow operators with this change
because of `scope:` references that weren't real scopes. I fixed them.
Soon I'll add verifier checks for them too.
This also adds explicit dereference operators. Previously, the builtin
dereference against `operator MDNode *()` would have worked, but now the
builtins are ambiguous.
llvm-svn: 233030
As preparation for removing the getSubtargetImpl() call from
TargetMachine go ahead and flip the switch on caching the function
dependent subtarget and remove the bare getSubtargetImpl call
from the X86 port. As part of this add a few tests that show we
can generate code and assemble on X86 based on features/cpu on
the Function.
llvm-svn: 232879
With this patch, for this one exact case, we'll generate:
blendps %xmm0, %xmm1, $1
instead of:
insertps %xmm0, %xmm1, $0
If there's a memory operand available for load folding and we're
optimizing for size, we'll still generate the insertps.
The detailed performance data motivation for this may be found in D7866;
in summary, blendps has 2-3x throughput vs. insertps on widely used chips.
Differential Revision: http://reviews.llvm.org/D8332
llvm-svn: 232850
With the option -outline-optional-branches, LLVM will place optional
branches out of line (more details on r231230).
With this patch, this is not done for short optional branches. A short
optional branch is a branch containing a single block with an
instruction count below a certain threshold (defaulting to 3). Still
everything is guarded under -outline-optional-branches).
Outlining a short branch can't significantly improve code locality. It
can however decrease performance because of the additional jmp and in
cases where the optional branch is hot. This fixes a compile time
regression I have observed in a benchmark.
Review: http://reviews.llvm.org/D8108
llvm-svn: 232802
Another case of x86-specific shuffle strength reduction:
avoid generating insert*128 instructions with index 0 because
they are slower than their non-lane-changing blend equivalents.
Shuffle lowering already catches most of these cases, but
the zero vector case and some other paths such as in the
modified test in vector-shuffle-256-v32.ll were getting
through.
Differential Revision: http://reviews.llvm.org/D8366
llvm-svn: 232773
Currently v2i64 vectors shifts (non-equal shift amounts) are scalarized, costing 4 x extract, 2 x x86-shifts and 2 x insert instructions - and it gets even more awkward on 32-bit targets.
This patch separately shifts the vector by both shift amounts and then shuffles the partial results back together, costing 2 x shuffles and 2 x sse-shifts instructions (+ 2 movs on pre-AVX hardware).
Note - this patch only improves the SHL / LSHR logical shifts as only these are supported in SSE hardware.
Differential Revision: http://reviews.llvm.org/D8416
llvm-svn: 232660
When calculating the lanemask of a register class we have to include the
masks of subregisters supported by any of the class members, not just
the ones supported by all class members.
This fixes problems when coalescing towards a subclass with additional
subregisters available.
The attached testcase works fine as is, but does crash if you enable
subregister liveness on x86 without this change applied.
llvm-svn: 232652
The checks here were so vague that we could nuke intrinsics
from existence and still pass the test because we'd match
the function name.
llvm-svn: 232647
The 'vmovntdq' was only passing due to a fluke in
SandyBridge codegen that splits 32-byte stores in half,
but that meant that the test was not correctly checking
for the 32-byte store that we thought we were generating.
The lax checking in this file will be addressed in
another commit. There are bigger problems here.
llvm-svn: 232644
The two hot blocks are right next to each other and I verified that
there is no performance regression by compressing/uncompressing some
files with a minigzip built with the different options.
llvm-svn: 232629
These BEXTR cases are a check for the 64-bit load form and two negative cases where the bitrange is non-contiguous. From a private patch equivalent to r189742/PR17028.
llvm-svn: 232580
Targets which provide a rotate make it possible to replace a sequence of
(XOR (SHL 1, x), -1) with (ROTL ~1, x). This saves an instruction on
architectures like X86 and POWER(64).
Differential Revision: http://reviews.llvm.org/D8350
llvm-svn: 232572
COFF COMDATs (for selection kinds other than 'select any') require at
least one non-section symbol in the symbol table.
Satisfy this by morally enhancing the linkage from private to internal.
Differential Revision: http://reviews.llvm.org/D8394
llvm-svn: 232570
Summary:
COFF COMDATs (for selection kinds other than 'select any') require at
least one non-section symbol in the symbol table.
Satisfy this by morally enhancing the linkage from private to internal.
Reviewers: rafael
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D8374
llvm-svn: 232539
We removed @llvm.eh.typeid.for.i32 and replaced it with
@llvm.eh.typeid.for quite some time ago. Fix up some test cases which
never got updated.
llvm-svn: 232421
As part of PR22777, fix testcases that fail the debug info verifier.
The changes fall into the following categories:
- Empty `filename:` fields in `MDFile`s. Compile units and some types
require non-empty filenames. A number of testcases have empty
filenames, probably due to hand-reduction of testcases.
- Not-quite empty arrays: `!{i32 0}`. This used to be equivalent in
the debug info schema to `!{}`. They cause problems for
`!MDSubroutineType`'s `types:` array, since it requires all operands
to be valid types. (Note that `!{null}` is the correct type array
for functions that take no arguments and return `void`.)
- Significantly bitrotted testcases. Nodes got left behind a few
upgrades ago because of missing or invalid tags.
llvm-svn: 232415
Specifically, if there are copy-like instructions in the loop header
they are moved into the loop close to their uses. This reduces the live
intervals of the values and can avoid register spills.
This is working towards a fix for http://llvm.org/PR22230.
Review: http://reviews.llvm.org/D7259
Next steps:
- Find a better cost model (which non-copy instructions should be sunk?)
- Make this dependent on register pressure
llvm-svn: 232262
This fixes pr22854.
The core issue on the bug is that there are multiple instructions that
print the same in assembly. In fact, there doesn't seem to be any
syntax for specifying that a constant that fits in 8 bits should use a 32 bit
immediate.
The attached patch changes fast isel to consider i16immSExt8,
i32immSExt8, and i64immSExt8. They were disabled because fastisel didn’t know
to call the predicate back in the day.
llvm-svn: 232223
Similar to gep (r230786) and load (r230794) changes.
Similar migration script can be used to update test cases, which
successfully migrated all of LLVM and Polly, but about 4 test cases
needed manually changes in Clang.
(this script will read the contents of stdin and massage it into stdout
- wrap it in the 'apply.sh' script shown in previous commits + xargs to
apply it over a large set of test cases)
import fileinput
import sys
import re
rep = re.compile(r"(getelementptr(?:\s+inbounds)?\s*\()((<\d*\s+x\s+)?([^@]*?)(|\s*addrspace\(\d+\))\s*\*(?(3)>)\s*)(?=$|%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|zeroinitializer|<|\[\[[a-zA-Z]|\{\{)", re.MULTILINE | re.DOTALL)
def conv(match):
line = match.group(1)
line += match.group(4)
line += ", "
line += match.group(2)
return line
line = sys.stdin.read()
off = 0
for match in re.finditer(rep, line):
sys.stdout.write(line[off:match.start()])
sys.stdout.write(conv(match))
off = match.end()
sys.stdout.write(line[off:])
llvm-svn: 232184
This patch fixes a bug in the shuffle lowering logic implemented by function
'lowerV2X128VectorShuffle'.
The are few cases where function 'lowerV2X128VectorShuffle' wrongly expands a
shuffle of two v4X64 vectors into a CONCAT_VECTORS of two EXTRACT_SUBVECTOR
nodes. The problematic expansion only occurs when the shuffle mask M has an
'undef' element at position 2, and M is equivalent to mask <0,1,4,5>.
In that case, the algorithm propagates the wrong vector to one of the two
new EXTRACT_SUBVECTOR nodes.
Example:
;;
define <4 x double> @test(<4 x double> %A, <4 x double> %B) {
entry:
%0 = shufflevector <4 x double> %A, <4 x double> %B, <4 x i32><i32 undef, i32 1, i32 undef, i32 5>
ret <4 x double> %0
}
;;
Before this patch, llc (-mattr=+avx) generated:
vinsertf128 $1, %xmm0, %ymm0, %ymm0
With this patch, llc correctly generates:
vinsertf128 $1, %xmm1, %ymm0, %ymm0
Added test lower-vec-shuffle-bug.ll
Differential Revision: http://reviews.llvm.org/D8259
llvm-svn: 232179
This should complete the job started in r231794 and continued in r232045:
We want to replace as much custom x86 shuffling via intrinsics
as possible because pushing the code down the generic shuffle
optimization path allows for better codegen and less complexity
in LLVM.
AVX2 introduced proper integer variants of the hacked integer insert/extract
C intrinsics that were created for this same functionality with AVX1.
This should complete the removal of insert/extract128 intrinsics.
The Clang precursor patch for this change was checked in at r232109.
llvm-svn: 232120
The permps and permd instructions have their operands swapped compared to the
intrinsic definition. Therefore, they do not fall into the INTR_TYPE_2OP
category.
I did not create a new category for those two, as they are the only one AFAICT
in that case.
<rdar://problem/20108262>
llvm-svn: 232085
Part of the folding logic implemented by function 'PerformISDSETCCCombine'
only worked under the assumption that the condition code in input could have
been either SETNE or SETEQ.
Unfortunately that assumption was incorrect, and in some cases the algorithm
ended up incorrectly folding SETCC nodes.
The incorrect folding only affected SETCC dag nodes where:
- one of the operands was a build_vector of all zeroes;
- the other operand was a SIGN_EXTEND from a vector of MVT:i1 elements;
- the condition code was neither SETNE nor SETEQ.
Example:
(setcc (v4i32 (sign_extend v4i1:%A)), (v4i32 VectorOfAllZeroes), setge)
Before this patch, the entire dag node sequence from the example was
incorrectly folded to node %A.
With this patch, the dag node sequence is folded to a
(xor %A, (v4i1 VectorOfAllOnes)).
Added test setcc-combine.ll.
Thanks to Greg Bedwell for spotting this issue.
llvm-svn: 232046
Now that we've replaced the vinsertf128 intrinsics,
do the same for their extract twins.
This is very much like D8086 (checked in at r231794):
We want to replace as much custom x86 shuffling via intrinsics
as possible because pushing the code down the generic shuffle
optimization path allows for better codegen and less complexity
in LLVM.
This is also the LLVM sibling to the cfe D8275 patch.
Differential Revision: http://reviews.llvm.org/D8276
llvm-svn: 232045
Instead, run both EH preparation passes, and have them both ignore
functions with unrecognized EH personalities. Pass delegation involved
some hacky code for creating an AnalysisResolver that we don't need now.
llvm-svn: 231995
CodeGen incorrectly ignores (assert from APInt) constant index bigger
than 2^64 in getelementptr instruction. This is a test and fix for that.
Patch by Paweł Bylica!
Reviewed By: rnk
Subscribers: majnemer, rnk, mcrosier, resistor, llvm-commits
Differential Revision: http://reviews.llvm.org/D8219
llvm-svn: 231984
If a function is going in an unique section (because of -ffunction-sections
for example), putting a jump table in .rodata will keep .rodata alive and
that will keep alive any other function that also has a jump table.
Instead, put the jump table in a unique section that is associated with the
function.
llvm-svn: 231961
Summary:
The generic ELF TargetObjectFile defaults to .ctors, but Linux's
defaults to .init_array by calling InitializeELF with the value of
UseInitArray from TargetMachine. Make NaCl's behavior match.
Reviewers: jvoung
Differential Revision: http://reviews.llvm.org/D8240
llvm-svn: 231934
- Use TargetLowering to check for the actual cost of each extension.
- Provide a factorized method to check for the cost of an extension:
TargetLowering::isExtFree.
- Provide a virtual method TargetLowering::isExtFreeImpl for targets to be able
to tune the cost of non-free extensions.
This refactoring offers a better granularity to model what really happens on
different targets.
No performance changes and very few code differences.
Part of <rdar://problem/19267165>
llvm-svn: 231855
Also it extracts getCopyFromRegs helper function in SelectionDAGBuilder as we need to be able to customize type of the register exported from basic block during lowering of the gc.result.
(Resubmitting this change after not being able to reproduce buildbot failure)
Differential Revision: http://reviews.llvm.org/D7760
llvm-svn: 231800
We want to replace as much custom x86 shuffling via intrinsics
as possible because pushing the code down the generic shuffle
optimization path allows for better codegen and less complexity
in LLVM.
This is the sibling patch for the Clang half of this change:
http://reviews.llvm.org/D8088
Differential Revision: http://reviews.llvm.org/D8086
llvm-svn: 231794
This fixes a subtle issue that was introduced in r205153.
When reusing a store for the extractelement expansion (to load directly
from it, inserting of going through the stack), later stores to the
same location might have overwritten the data we were expecting to
extract from.
To fix that, we need to explicitly replace the chain going out of the
reused store, so that later stores also have an explicit dependency on
the generated element-extracting loads, and can't clobber them.
rdar://20066785
Differential Revision: http://reviews.llvm.org/D8180
llvm-svn: 231721
Fix the double-deletion of AnalysisResolver when delegating through to
Dwarf EH preparation by creating one from scratch. Hopefully the new
pass manager simplifies this.
This reverts commit r229952.
llvm-svn: 231719
In the case where just tables are part of the function section, this produces
more readable assembly by avoiding switching to the eh section and back
to .text.
This would also break with non unique section names, as trying to switch to
a unique section actually creates a new one.
llvm-svn: 231677
There were cases where the backend computed a wrong permute mask for a VPERM2X128 node.
Example:
\code
define <8 x float> @foo(<8 x float> %a, <8 x float> %b) {
%shuffle = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 undef, i32 undef, i32 6, i32 7, i32 undef, i32 undef, i32 6, i32 7>
ret <8 x float> %shuffle
}
\code end
Before this patch, llc (with -mattr=+avx) emitted the following vperm2f128:
vperm2f128 $0, %ymm0, %ymm0, %ymm0 # ymm0 = ymm0[0,1,0,1]
With this patch, llc emits a vperm2f128 with a correct permute mask:
vperm2f128 $17, %ymm0, %ymm0, %ymm0 # ymm0 = ymm0[2,3,2,3]
Differential Revision: http://reviews.llvm.org/D8119
llvm-svn: 231601
This patch fixes the logic in the DAGCombiner that folds an AND node according
to rule: (and (X (load V)), C) -> (X (load V))
An AND between a vector load 'X' and a constant build_vector 'C' can be folded
into the load itself only if we can prove that the AND operation is redundant.
The algorithm implemented by 'visitAND' firstly computes the splat value 'S'
from C, and then checks if S has the lower 'B' bits set (where B is the size in
bits of the vector element type). The algorithm takes into account also the
'undef' bits in the splat mask.
Unfortunately, the algorithm only worked under the assumption that the size of S
is a multiple of the vector element type. With this patch, we conservatively
avoid folding the AND if the splat bits are not compatible with the vector
element type.
Added X86 test and-load-fold.ll
Differential Revision: http://reviews.llvm.org/D8085
llvm-svn: 231563
This patch attempts to convert a SCALAR_TO_VECTOR using an operand from an EXTRACT_VECTOR_ELT into a VECTOR_SHUFFLE.
This prevents many cases of spilling scalar data between the gpr + simd registers.
At present the optimization only accepts cases where there is no TRUNC of the scalar type (i.e. all types must match).
Differential Revision: http://reviews.llvm.org/D8132
llvm-svn: 231554
This is based on the following equivalences:
select(C0 & C1, X, Y) <=> select(C0, select(C1, X, Y), Y)
select(C0 | C1, X, Y) <=> select(C0, X, select(C1, X, Y))
Many target cannot perform and/or on the CPU flags and therefore the
right side should be choosen to avoid materializign the i1 flags in an
integer register. If the target can perform this operation efficiently
we normalize to the left form.
Differential Revision: http://reviews.llvm.org/D7622
llvm-svn: 231507
This patch reduces code size for all AVX targets and increases speed for some chips.
SSE 4.1 introduced the useless (see code comments) 2-register form of BLENDV and
only in the packed float/double flavors.
AVX subsequently made the instruction useful by adding a 4-register operand form.
So we just need to paper over the lack of scalar forms of this instruction, complicate
the code to choose float or double forms, and use blendv on scalars since all FP is in
xmm registers anyway.
This gives us an approximately 50% speed up for a blendv microbenchmark sequence
on SandyBridge and Haswell:
blendv : 29.73 cycles/iter
logic : 43.15 cycles/iter
No new test cases with this patch because:
1. fast-isel-select-sse.ll tests the positive side for regular X86 lowering and fast-isel
2. sse-minmax.ll and fp-select-cmp-and.ll confirm that we're not firing for scalar selects without AVX
3. fp-select-cmp-and.ll and logical-load-fold.ll confirm that we're not firing for scalar selects with constants.
http://llvm.org/bugs/show_bug.cgi?id=22483
Differential Revision: http://reviews.llvm.org/D8063
llvm-svn: 231408
Currently shuffles may only be combined if they are of the same type, despite the fact that bitcasts are often introduced in between shuffle nodes (e.g. x86 shuffle type widening).
This patch allows a single input shuffle to peek through bitcasts and if the input is another shuffle will merge them, shuffling using the smallest sized type, and re-applying the bitcasts at the inputs and output instead.
Dropped old ShuffleToZext test - this patch removes the use of the zext and vector-zext.ll covers these anyhow.
Differential Revision: http://reviews.llvm.org/D7939
llvm-svn: 231380
Added lowering for ISD::CONCAT_VECTORS and ISD::INSERT_SUBVECTOR for i1 vectors,
it is needed to pass all masked_memop.ll tests for SKX.
llvm-svn: 231371
Also it extracts getCopyFromRegs helper function in SelectionDAGBuilder as we need to be able to customize type of the register exported from basic block during lowering of the gc.result.
llvm-svn: 231366
just arbitrarily interleaving unrelated control flows once they get
moved "out-of-line" (both outside of natural CFG ordering and with
diamonds that cannot be fully laid out by chaining fallthrough edges).
This easy solution doesn't work in practice, and it isn't just a small
bug. It looks like a very different strategy will be required. I'm
working on that now, and it'll again go behind some flag so that
everyone can experiment and make sure it is working well for them.
llvm-svn: 231332
The target-independent selection algorithm in FastISel already knows how
to select a SINT_TO_FP if the target is SSE but not AVX.
On targets that have SSE but not AVX, the tablegen'd 'fastEmit' functions
for ISD::SINT_TO_FP know how to select instruction X86::CVTSI2SSrr
(for an i32 to f32 conversion) and X86::CVTSI2SDrr (for an i32 to f64
conversion).
This patch simplifies the logic in method X86SelectSIToFP knowing that
the code would not be reachable if the subtarget doesn't have AVX.
No functional change intended.
llvm-svn: 231243
a flag for now.
First off, thanks to Daniel Jasper for really pointing out the issue
here. It's been here forever (at least, I think it was there when
I first wrote this code) without getting really noticed or fixed.
The key problem is what happens when two reasonably common patterns
happen at the same time: we outline multiple cold regions of code, and
those regions in turn have diamonds or other CFGs for which we can't
just topologically lay them out. Consider some C code that looks like:
if (a1()) { if (b1()) c1(); else d1(); f1(); }
if (a2()) { if (b2()) c2(); else d2(); f2(); }
done();
Now consider the case where a1() and a2() are unlikely to be true. In
that case, we might lay out the first part of the function like:
a1, a2, done;
And then we will be out of successors in which to build the chain. We go
to find the best block to continue the chain with, which is perfectly
reasonable here, and find "b1" let's say. Laying out successors gets us
to:
a1, a2, done; b1, c1;
At this point, we will refuse to lay out the successor to c1 (f1)
because there are still un-placed predecessors of f1 and we want to try
to preserve the CFG structure. So we go get the next best block, d1.
... wait for it ...
Except that the next best block *isn't* d1. It is b2! d1 is waaay down
inside these conditionals. It is much less important than b2. Except
that this is exactly what we didn't want. If we keep going we get the
entire set of the rest of the CFG *interleaved*!!!
a1, a2, done; b1, c1; b2, c2; d1, f1; d2, f2;
So we clearly need a better strategy here. =] My current favorite
strategy is to actually try to place the block whose predecessor is
closest. This very simply ensures that we unwind these kinds of CFGs the
way that is natural and fitting, and should minimize the number of cache
lines instructions are spread across.
It also happens to be *dead simple*. It's like the datastructure was
specifically set up for this use case or something. We only push blocks
onto the work list when the last predecessor for them is placed into the
chain. So the back of the worklist *is* the nearest next block.
Unfortunately, a change like this is going to cause *soooo* many
benchmarks to swing wildly. So for now I'm adding this under a flag so
that we and others can validate that this is fixing the problems
described, that it seems possible to enable, and hopefully that it fixes
more of our problems long term.
llvm-svn: 231238
In a CFG with the edges A->B->C and A->C, B is an optional branch.
LLVM's default behavior is to lay the blocks out naturally, i.e. A, B,
C, in order to improve code locality and fallthroughs. However, if a
function contains many of those optional branches only a few of which
are taken, this leads to a lot of unnecessary icache misses. Moving B
out of line can work around this.
Review: http://reviews.llvm.org/D7719
llvm-svn: 231230
When trying to convert a BUILD_VECTOR into a shuffle, we try to split a single source vector that is twice as wide as the destination vector.
We can not do this when we also need the zero vector to create a blend.
This fixes PR22774.
Differential Revision: http://reviews.llvm.org/D8040
llvm-svn: 231219
The intrinsic is no longer generated by the front-end. Remove the intrinsic and
auto-upgrade it to a vector shuffle.
Reviewed by Nadav
This is related to rdar://problem/18742778.
llvm-svn: 231182
From:
int M, total;
void foo() {
int i;
for (i = 0; i < M; i++) {
total = total + i / 2;
}
}
This is the kernel loop:
.LBB0_2: # %for.body
=>This Inner Loop Header: Depth=1
movl %edx, %esi
movl %ecx, %edx
shrl $31, %edx
addl %ecx, %edx
sarl %edx
addl %esi, %edx
incl %ecx
cmpl %eax, %ecx
jl .LBB0_2
--------------------------
The first mov insn "movl %edx, %esi" could be removed if we change "addl %esi, %edx" to "addl %edx, %esi".
The IR before TwoAddressInstructionPass is:
BB#2: derived from LLVM BB %for.body
Predecessors according to CFG: BB#1 BB#2
%vreg3<def> = COPY %vreg12<kill>; GR32:%vreg3,%vreg12
%vreg2<def> = COPY %vreg11<kill>; GR32:%vreg2,%vreg11
%vreg7<def,tied1> = SHR32ri %vreg3<tied0>, 31, %EFLAGS<imp-def,dead>; GR32:%vreg7,%vreg3
%vreg8<def,tied1> = ADD32rr %vreg3<tied0>, %vreg7<kill>, %EFLAGS<imp-def,dead>; GR32:%vreg8,%vreg3,%vreg7
%vreg9<def,tied1> = SAR32r1 %vreg8<kill,tied0>, %EFLAGS<imp-def,dead>; GR32:%vreg9,%vreg8
%vreg4<def,tied1> = ADD32rr %vreg9<kill,tied0>, %vreg2<kill>, %EFLAGS<imp-def,dead>; GR32:%vreg4,%vreg9,%vreg2
%vreg5<def,tied1> = INC64_32r %vreg3<kill,tied0>, %EFLAGS<imp-def,dead>; GR32:%vreg5,%vreg3
CMP32rr %vreg5, %vreg0, %EFLAGS<imp-def>; GR32:%vreg5,%vreg0
%vreg11<def> = COPY %vreg4; GR32:%vreg11,%vreg4
%vreg12<def> = COPY %vreg5<kill>; GR32:%vreg12,%vreg5
JL_4 <BB#2>, %EFLAGS<imp-use,kill>
Now TwoAddressInstructionPass will choose vreg9 to be tied with vreg4. However, it doesn't see that there is copy from vreg4 to vreg11 and another copy from vreg11 to vreg2 inside the loop body. To remove those copies, it is necessary to choose vreg2 to be tied with vreg4 instead of vreg9. This code pattern commonly appears when there is reduction operation in a loop.
So check for a reversed copy chain and if we encounter one then we can commute the add instruction so we can avoid a copy.
Patch by Wei Mi.
http://reviews.llvm.org/D7806
llvm-svn: 231148
Ultimately, __CxxFrameHandler3 needs us to put a stack offset in a
table, and it will take responsibility for copying the exception object
into that slot. Modelling the exception object as an SSA value returned
by begincatch isn't going to work in general, so make it use an output
parameter.
Reviewers: andrew.w.kaylor
Differential Revision: http://reviews.llvm.org/D7920
llvm-svn: 231086
Move the specialized metadata nodes for the new debug info hierarchy
into place, finishing off PR22464. I've done bootstraps (and all that)
and I'm confident this commit is NFC as far as DWARF output is
concerned. Let me know if I'm wrong :).
The code changes are fairly mechanical:
- Bumped the "Debug Info Version".
- `DIBuilder` now creates the appropriate subclass of `MDNode`.
- Subclasses of DIDescriptor now expect to hold their "MD"
counterparts (e.g., `DIBasicType` expects `MDBasicType`).
- Deleted a ton of dead code in `AsmWriter.cpp` and `DebugInfo.cpp`
for printing comments.
- Big update to LangRef to describe the nodes in the new hierarchy.
Feel free to make it better.
Testcase changes are enormous. There's an accompanying clang commit on
its way.
If you have out-of-tree debug info testcases, I just broke your build.
- `upgrade-specialized-nodes.sh` is attached to PR22564. I used it to
update all the IR testcases.
- Unfortunately I failed to find way to script the updates to CHECK
lines, so I updated all of these by hand. This was fairly painful,
since the old CHECKs are difficult to reason about. That's one of
the benefits of the new hierarchy.
This work isn't quite finished, BTW. The `DIDescriptor` subclasses are
almost empty wrappers, but not quite: they still have loose casting
checks (see the `RETURN_FROM_RAW()` macro). Once they're completely
gutted, I'll rename the "MD" classes to "DI" and kill the wrappers. I
also expect to make a few schema changes now that it's easier to reason
about everything.
llvm-svn: 231082
This prevents the behavior observed in llvm.org/PR22369. I am not sure
whether I am reading the code correctly, but the early exit based on
isLiveOutPastPHIs() seems to make the wrong assumption that
RegisterCoalescer won't be able to coalesce those copies later.
This change hides the new behavior behind -no-phi-elim-live-out-early-exit
as it currently breaks four tests:
* Assertion in:
CodeGen/Hexagon/hwloop-cleanup.ll
* Worse code in:
CodeGen/X86/coalescer-commute4.ll
CodeGen/X86/phys_subreg_coalesce-2.ll
CodeGen/X86/zlib-longest-match.ll
The root cause here seems to be that the heuristic that determines
the visitation order in RegisterCoalescer gets less lucky.
llvm-svn: 231064
This lets us avoid a few copies that are otherwise hard to get rid of.
The way this is done is, the custom-inserter looks at the following
instruction for another CMOV, and replaces both at the same time.
A previous version used a new CMOV2 opcode, but the custom inserter
is expected to be able to return a different basic block anyway, which
means it's OK - though far from ideal - to alter that block's contents.
Explicitly document that, in case it ever makes a difference.
Alternatives welcome!
Follow-up to r231045.
rdar://19767934
Closes http://reviews.llvm.org/D8019
llvm-svn: 231046
Fold and/or of setcc's to double CMOV:
(CMOV F, T, ((cc1 | cc2) != 0)) -> (CMOV (CMOV F, T, cc1), T, cc2)
(CMOV F, T, ((cc1 & cc2) != 0)) -> (CMOV (CMOV T, F, !cc1), F, !cc2)
When we can't use the CMOV instruction, it might increase branch
mispredicts. When we can, or when there is no mispredict, this
improves throughput and reduces register pressure.
These can't be catched by generic combines, because the pattern can
appear when legalizing some instructions (such as fcmp une).
rdar://19767934
http://reviews.llvm.org/D7634
llvm-svn: 231045
In the future, we should run the output of clang through instnamer to
make it easier to manually edit test cases.
No functionality change.
llvm-svn: 231037
We were missing a check for the following fold in DAGCombiner:
// fold (fmul (fmul x, c1), c2) -> (fmul x, (fmul c1, c2))
If 'x' is also a constant, then we shouldn't do anything. Otherwise, we could end up swapping the operands back and forth forever.
This should fix:
http://llvm.org/bugs/show_bug.cgi?id=22698
Differential Revision: http://reviews.llvm.org/D7917
llvm-svn: 230884
There are two types of files in the old (current) debug info schema.
!0 = !{!"some/filename", !"/path/to/dir"}
!1 = !{!"0x29", !0} ; [ DW_TAG_file_type ]
!1 has a wrapper class called `DIFile` which inherits from `DIScope` and
is referenced in 'scope' fields.
!0 is called a "file node", and debug info nodes with a 'file' field
point at one of these directly -- although they're built in `DIBuilder`
by sending in a `DIFile` and reaching into it.
In the new hierarchy, I unified these nodes as `MDFile` (which `DIFile`
is a lightweight wrapper for) in r230057. Moving the new hierarchy into
place (and upgrading testcases) caused CodeGen/X86/unknown-location.ll
to start failing -- apparently "0x29" was previously showing up in the
linetable as a filename, causing:
.loc 2 4 3
(where 2 points at filename "0x29") instead of:
.loc 1 4 3
(where 1 points at the actual filename).
Change the testcase to use the old schema correctly.
llvm-svn: 230880
Essentially the same as the GEP change in r230786.
A similar migration script can be used to update test cases, though a few more
test case improvements/changes were required this time around: (r229269-r229278)
import fileinput
import sys
import re
pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)")
for line in sys.stdin:
sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line))
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7649
llvm-svn: 230794
Summary:
Until now, we did this (among other things) based on whether or not the
target was Windows. This is clearly wrong, not just for Win64 ABI functions
on non-Windows, but for System V ABI functions on Windows, too. In this
change, we make this decision based on the ABI the calling convention
specifies instead.
Reviewers: rnk
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D7953
llvm-svn: 230793
One of several parallel first steps to remove the target type of pointers,
replacing them with a single opaque pointer type.
This adds an explicit type parameter to the gep instruction so that when the
first parameter becomes an opaque pointer type, the type to gep through is
still available to the instructions.
* This doesn't modify gep operators, only instructions (operators will be
handled separately)
* Textual IR changes only. Bitcode (including upgrade) and changing the
in-memory representation will be in separate changes.
* geps of vectors are transformed as:
getelementptr <4 x float*> %x, ...
->getelementptr float, <4 x float*> %x, ...
Then, once the opaque pointer type is introduced, this will ultimately look
like:
getelementptr float, <4 x ptr> %x
with the unambiguous interpretation that it is a vector of pointers to float.
* address spaces remain on the pointer, not the type:
getelementptr float addrspace(1)* %x
->getelementptr float, float addrspace(1)* %x
Then, eventually:
getelementptr float, ptr addrspace(1) %x
Importantly, the massive amount of test case churn has been automated by
same crappy python code. I had to manually update a few test cases that
wouldn't fit the script's model (r228970,r229196,r229197,r229198). The
python script just massages stdin and writes the result to stdout, I
then wrapped that in a shell script to handle replacing files, then
using the usual find+xargs to migrate all the files.
update.py:
import fileinput
import sys
import re
ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))")
def conv(match, line):
if not match:
return line
line = match.groups()[0]
if len(match.groups()[5]) == 0:
line += match.groups()[2]
line += match.groups()[3]
line += ", "
line += match.groups()[1]
line += "\n"
return line
for line in sys.stdin:
if line.find("getelementptr ") == line.find("getelementptr inbounds"):
if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("):
line = conv(re.match(ibrep, line), line)
elif line.find("getelementptr ") != line.find("getelementptr ("):
line = conv(re.match(normrep, line), line)
sys.stdout.write(line)
apply.sh:
for name in "$@"
do
python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name"
rm -f "$name.tmp"
done
The actual commands:
From llvm/src:
find test/ -name *.ll | xargs ./apply.sh
From llvm/src/tools/clang:
find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}"
From llvm/src/tools/polly:
find test/ -name *.ll | xargs ./apply.sh
After that, check-all (with llvm, clang, clang-tools-extra, lld,
compiler-rt, and polly all checked out).
The extra 'rm' in the apply.sh script is due to a few files in clang's test
suite using interesting unicode stuff that my python script was throwing
exceptions on. None of those files needed to be migrated, so it seemed
sufficient to ignore those cases.
Reviewers: rafael, dexonsmith, grosser
Differential Revision: http://reviews.llvm.org/D7636
llvm-svn: 230786
This work is currently being rethought along different lines and
if this work is needed it can be resurrected out of svn. Remove it
for now as no current work in ongoing on it and it's unused. Verified
with the authors before removal.
llvm-svn: 230780
Summary:
Currently fast-isel-abort will only abort for regular instructions,
and just warn for function calls, terminators, function arguments.
There is already fast-isel-abort-args but nothing for calls and
terminators.
This change turns the fast-isel-abort options into an integer option,
so that multiple levels of strictness can be defined.
This will help no being surprised when the "abort" option indeed does
not abort, and enables the possibility to write test that verifies
that no intrinsics are forgotten by fast-isel.
Reviewers: resistor, echristo
Subscribers: jfb, llvm-commits
Differential Revision: http://reviews.llvm.org/D7941
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 230775
This removes a bit of duplicated code and more importantly, remembers the
labels so that they don't need to be looked up by name.
This in turn allows for any name to be used and avoids a crash if the name
we wanted was already taken.
llvm-svn: 230772
vectors. This lets us fix the rest of the v16 lowering problems when
pshufb is clearly better.
We might still be able to improve some of the lowerings by enabling the
other combine-based rewriting to fire for non-128-bit vectors, but this
at least should remove any regressions from using the fancy v16i16
lowering strategy.
llvm-svn: 230753
repeated 128-bit lane shuffles of wider vector types and use it to lower
256-bit v16i16 vector shuffles where applicable.
This should let us perfectly lowering the pattern of pshuflw and pshufhw
even for AVX2 256-bit patterns.
I've not added AVX-512 support, but it should be trivial for someone
working on that to wire up.
Note that currently this generates bad, long shuffle chains because we
don't combine 256-bit target shuffles. The subsequent patches will fix
that.
llvm-svn: 230751
by mirroring v8i16 test cases across both 128-bit lanes. This should
highlight problems where we aren't correctly using 128-bit shuffles to
implement things.
llvm-svn: 230750
Summary:
This change causes us to actually save non-volatile registers in a Win64
ABI function that calls a System V ABI function, and vice-versa.
Reviewers: rnk
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D7919
llvm-svn: 230714
blend as legal.
We made the same mistake in two different places. Whenever we are custom
lowering a v32i8 blend we need to check whether we are custom lowering
it only for constant conditions that can be shuffled, or whether we
actually have AVX2 and full dynamic blending support on bytes. Both are
fixed, with comments added to make it clear what is going on and a new
test case.
llvm-svn: 230695
have the debugger step through each one individually. Turn off the
combine for adjacent stores at -O0 so we get this behavior.
Possibly, DAGCombine shouldn't run at all at -O0, but that's for
another day; see PR22346.
Differential Revision: http://reviews.llvm.org/D7181
llvm-svn: 230659
Turns out that after the past MMX commits, we don't need to rely on this
flag to get better codegen for MMX. Also update the tests to become
triple neutral.
llvm-svn: 230637
The Win64 epilogue structure is very restrictive, it permits a very
small number of opcodes and none of them are 'mov'.
This means that given:
mov %rbp, %rsp
pop %rbp
The mov isn't the epilogue, only the pop is. This is problematic unless
a frame pointer is present in which case we are free to do whatever we'd
like in the "body" of the function. If a frame pointer is present,
unwinding will undo the prologue operations in reverse order regardless
of the fact that we are at an instruction which is reseting the stack
pointer.
llvm-svn: 230543
(The change was landed in r230280 and caused the regression PR22674.
This version contains a fix and a test-case for PR22674).
When emitting the increment operation, SCEVExpander marks the
operation as nuw or nsw based on the flags on the preincrement SCEV.
This is incorrect because, for instance, it is possible that {-6,+,1}
is <nuw> while {-6,+,1}+1 = {-5,+,1} is not.
This change teaches SCEV to mark the increment as nuw/nsw only if it
can explicitly prove that the increment operation won't overflow.
Apart from the attached test case, another (more realistic)
manifestation of the bug can be seen in
Transforms/IndVarSimplify/pr20680.ll.
Differential Revision: http://reviews.llvm.org/D7778
llvm-svn: 230533
Reapply r230248.
Teach the peephole optimizer to work with MMX instructions by adding
entries into the foldable tables. This covers folding opportunities not
handled during isel.
llvm-svn: 230499
This patch unifies the comdat and non-comdat code paths. By doing this
it add missing features to the comdat side and removes the fixed
section assumptions from the non-comdat side.
In ELF there is no one true section for "4 byte mergeable" constants.
We are better off computing the required properties of the section
and asking the context for it.
llvm-svn: 230411
Author: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Mon Feb 23 23:04:28 2015 +0000
Fix based on post-commit comment on D7816 & rL230177 - BUILD_VECTOR operand truncation was using the the BV's output scalar type instead of the input type.
and
Author: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sun Feb 22 18:17:28 2015 +0000
[DagCombiner] Generalized BuildVector Vector Concatenation
The CONCAT_VECTORS combiner pass can transform the concat of two BUILD_VECTOR nodes into a single BUILD_VECTOR node.
This patch generalises this to support any number of BUILD_VECTOR nodes, and also permits UNDEF nodes to be included as well.
This was noticed as AVX vec128 -> vec256 canonicalization sometimes creates a CONCAT_VECTOR with a real vec128 lower and an vec128 UNDEF upper.
Differential Revision: http://reviews.llvm.org/D7816
as the root cause of PR22678 which is causing an assertion inside the DAG combiner.
I'll follow up to the main thread as well.
llvm-svn: 230358
When emitting the increment operation, SCEVExpander marks the
operation as nuw or nsw based on the flags on the preincrement SCEV.
This is incorrect because, for instance, it is possible that {-6,+,1}
is <nuw> while {-6,+,1}+1 = {-5,+,1} is not.
This change teaches SCEV to mark the increment as nuw/nsw only if it
can explicitly prove that the increment operation won't overflow.
Apart from the attached test case, another (more realistic) manifestation
of the bug can be seen in Transforms/IndVarSimplify/pr20680.ll.
NOTE: this change was landed with an incorrect commit message in
rL230275 and was reverted for that reason in rL230279. This commit
message is the correct one.
Differential Revision: http://reviews.llvm.org/D7778
llvm-svn: 230280
230275 got committed with an incorrect commit message due to a mixup
on my side. Will re-land in a few moments with the correct commit
message.
llvm-svn: 230279
This patch teaches the backend how to expand a double-half conversion into
a double-float conversion immediately followed by a float-half conversion.
We do this only under fast-math, and if float-half conversions are legal
for the target.
Added test CodeGen/X86/fastmath-float-half-conversion.ll
Differential Revision: http://reviews.llvm.org/D7832
llvm-svn: 230276
The bug was a result of getPreStartForExtend interpreting nsw/nuw
flags on an add recurrence more strongly than is legal. {S,+,X}<nsw>
implies S+X is nsw only if the backedge of the loop is taken at least
once.
Differential Revision: http://reviews.llvm.org/D7808
llvm-svn: 230275
Prologue emission, in some cases, requires calls to a stack probe helper
function. The amount of stack to probe is passed as a register
argument in the Win64 ABI but the instruction sequence used is
pessimistic: it assumes that the number of bytes to probe is greater
than 4 GB.
Instead, select a more appropriate opcode depending on the number of
bytes we are going to probe.
llvm-svn: 230270
Teach the peephole optimizer to work with MMX instructions by adding
entries into the foldable tables. This covers folding opportunities not
handled during isel.
llvm-svn: 230226
Add tests to cover the RR form of the pslli, psrli and psrai intrinsics.
In the next commit, the loads are going to be folded and the
instructions use the RM form.
llvm-svn: 230224
The CONCAT_VECTORS combiner pass can transform the concat of two BUILD_VECTOR nodes into a single BUILD_VECTOR node.
This patch generalises this to support any number of BUILD_VECTOR nodes, and also permits UNDEF nodes to be included as well.
This was noticed as AVX vec128 -> vec256 canonicalization sometimes creates a CONCAT_VECTOR with a real vec128 lower and an vec128 UNDEF upper.
Differential Revision: http://reviews.llvm.org/D7816
llvm-svn: 230177
Stack realignment occurs after the prolog, not during, for Win64.
Because of this, don't factor in the maximum stack alignment when
establishing a frame pointer.
This fixes PR22572.
llvm-svn: 230113
This patch teaches X86FastISel how to select intrinsic 'convert_from_fp16' and
intrinsic 'convert_to_fp16'.
If the target has F16C, we can select VCVTPS2PHrr for a float-half conversion,
and VCVTPH2PSrr for a half-float conversion.
Differential Revision: http://reviews.llvm.org/D7673
llvm-svn: 230043
The new shuffle lowering has been the default for some time. I've
enabled the new legality testing by default with no really blocking
regressions. I've fuzz tested this very heavily (many millions of fuzz
test cases have passed at this point). And this cleans up a ton of code.
=]
Thanks again to the many folks that helped with this transition. There
was a lot of work by others that went into the new shuffle lowering to
make it really excellent.
In case you aren't using a diff algorithm that can handle this:
X86ISelLowering.cpp: 22 insertions(+), 2940 deletions(-)
llvm-svn: 229964
is going well, remove the flag and the code for the old legality tests.
This is the first step toward removing the entire old vector shuffle
lowering. *Much* more code to delete coming up next.
llvm-svn: 229963
reflects the fact that the x86 backend can in fact lower any shuffle you
want it to with reasonably high code quality.
My recent work on the new vector shuffle has made this regress *very*
little. The diff in the test cases makes me very, very happy.
llvm-svn: 229958
one test case that is only partially tested in 32-bits into two test
cases so that the script doesn't generate massive spews of tests for the
cases we don't care about.
llvm-svn: 229955
Today a simple function that only catches exceptions and doesn't run
destructor cleanups ends up containing a dead call to _Unwind_Resume
(PR20300). We can't remove these dead resume instructions during normal
optimization because inlining might introduce additional landingpads
that do have cleanups to run. Instead we can do this during EH
preparation, which is guaranteed to run after inlining.
Fixes PR20300.
Reviewers: majnemer
Differential Revision: http://reviews.llvm.org/D7744
llvm-svn: 229944
X86 load folding is fragile; eg, the tests here
don't work without AVX even though they should. This
is because we have a mix of tablegen patterns that have
been added over time, and we have a load folding table
used by the peephole optimizer that has to be kept in
sync with the ever-changing ISA and tablegen defs.
llvm-svn: 229870
systematic lowering of v8i16.
This required a slight strategy shift to prefer unpack lowerings in more
places. While this isn't a cut-and-dry win in every case, it is in the
overwhelming majority. There are only a few places where the old
lowering would probably be a touch faster, and then only by a small
margin.
In some cases, this is yet another significant improvement.
llvm-svn: 229859
addition to lowering to trees rooted in an unpack.
This saves shuffles and or registers in many various ways, lets us
handle another class of v4i32 shuffles pre SSE4.1 without domain
crosses, etc.
llvm-svn: 229856
terribly complex partial blend logic.
This code path was one of the more complex and bug prone when it first
went in and it hasn't faired much better. Ultimately, with the simpler
basis for unpack lowering and support bit-math blending, this is
completely obsolete. In the worst case without this we generate
different but equivalent instructions. However, in many cases we
generate much better code. This is especially true when blends or pshufb
is available.
This does expose one (minor) weakness of the unpack lowering that I'll
try to address.
In case you were wondering, this is actually a big part of what I've
been trying to pull off in the recent string of commits.
llvm-svn: 229853
needed, and significantly improve the SSSE3 path.
This makes the new strategy much more clear. If we can blend, we just go
with that. If we can't blend, we try to permute into an unpack so
that we handle cases where the unpack doing the blend also simplifies
the shuffle. If that fails and we've got SSSE3, we now call into
factored-out pshufb lowering code so that we leverage the fact that
pshufb can set up a blend for us while shuffling. This generates great
code, especially because we *know* we don't have a fast blend at this
point. Finally, we fall back on decomposing into permutes and blends
because we do at least have a bit-math-based blend if we need to use
that.
This pretty significantly improves some of the v8i16 code paths. We
never need to form pshufb for the single-input shuffles because we have
effective target-specific combines to form it there, but we were missing
its effectiveness in the blends.
llvm-svn: 229851
them into permutes and a blend with the generic decomposition logic.
This works really well in almost every case and lets the code only
manage the expansion of a single input into two v8i16 vectors to perform
the actual shuffle. The blend-based merging is often much nicer than the
pack based merging that this replaces. The only place where it isn't we
end up blending between two packs when we could do a single pack. To
handle that case, just teach the v2i64 lowering to handle these blends
by digging out the operands.
With this we're down to only really random permutations that cause an
explosion of instructions.
llvm-svn: 229849
v16i8 shuffles, and replace it with new facilities.
This uses precise patterns to match exact unpacks, and the new
generalized unpack lowering only when we detect a case where we will
have to shuffle both inputs anyways and they terminate in exactly
a blend.
This fixes all of the blend horrors that I uncovered by always lowering
blends through the vector shuffle lowering. It also removes *sooooo*
much of the crazy instruction sequences required for v16i8 lowering
previously. Much cleaner now.
The only "meh" aspect is that we sometimes use pshufb+pshufb+unpck when
it would be marginally nicer to use pshufb+pshufb+por. However, the
difference there is *tiny*. In many cases its a win because we re-use
the pshufb mask. In others, we get to avoid the pshufb entirely. I've
left a FIXME, but I'm dubious we can really do better than this. I'm
actually pretty happy with this lowering now.
For SSE2 this exposes some horrors that were really already there. Those
will have to fixed by changing a different path through the v16i8
lowering.
llvm-svn: 229846
lowering paths. I'm going to be leveraging this to simplify a lot of the
overly complex lowering of v8 and v16 shuffles in pre-SSSE3 modes.
Sadly, this isn't profitable on v4i32 and v2i64. There, the float and
double blending instructions for pre-SSE4.1 are actually pretty good,
and we can't beat them with bit math. And once SSE4.1 comes around we
have direct blending support and this ceases to be relevant.
Also, some of the test cases look odd because the domain fixer
canonicalizes these to floating point domain. That's OK, it'll use the
integer domain when it matters and some day I may be able to update
enough of LLVM to canonicalize the other way.
This restores almost all of the regressions from teaching x86's vselect
lowering to always use vector shuffle lowering for blends. The remaining
problems are because the v16 lowering path is still doing crazy things.
I'll be re-arranging that strategy in more detail in subsequent commits
to finish recovering the performance here.
llvm-svn: 229836
First, don't combine bit masking into vector shuffles (even ones the
target can handle) once operation legalization has taken place. Custom
legalization of vector shuffles may exist for these patterns (making the
predicate return true) but that custom legalization may in some cases
produce the exact bit math this matches. We only really want to handle
this prior to operation legalization.
However, the x86 backend, in a fit of awesome, relied on this. What it
would do is mark VSELECTs as expand, which would turn them into
arithmetic, which this would then match back into vector shuffles, which
we would then lower properly. Amazing.
Instead, the second change is to teach the x86 backend to directly form
vector shuffles from VSELECT nodes with constant conditions, and to mark
all of the vector types we support lowering blends as shuffles as custom
VSELECT lowering. We still mark the forms which actually support
variable blends as *legal* so that the custom lowering is bypassed, and
the legal lowering can even be used by the vector shuffle legalization
(yes, i know, this is confusing. but that's how the patterns are
written).
This makes the VSELECT lowering much more sensible, and in fact should
fix a bunch of bugs with it. However, as you'll see in the test cases,
right now what it does is point out the *hilarious* deficiency of the
new vector shuffle lowering when it comes to blends. Fortunately, my
very next patch fixes that. I can't submit it yet, because that patch,
somewhat obviously, forms the exact and/or pattern that the DAG combine
is matching here! Without this patch, teaching the vector shuffle
lowering to produce the right code infloops in the DAG combiner. With
this patch alone, we produce terrible code but at least lower through
the right paths. With both patches, all the regressions here should be
fixed, and a bunch of the improvements (like using 2 shufps with no
memory loads instead of 2 andps with memory loads and an orps) will
stay. Win!
There is one other change worth noting here. We had hilariously wrong
vectorization cost estimates for vselect because we fell through to the
code path that assumed all "expand" vector operations are scalarized.
However, the "expand" lowering of VSELECT is vector bit math, most
definitely not scalarized. So now we go back to the correct if horribly
naive cost of "1" for "not scalarized". If anyone wants to add actual
modeling of shuffle costs, that would be cool, but this seems an
improvement on its own. Note the removal of 16 and 32 "costs" for doing
a blend. Even in SSE2 we can blend in fewer than 16 instructions. ;] Of
course, we don't right now because of OMG bad code, but I'm going to fix
that. Next patch. I promise.
llvm-svn: 229835
1) We should not try to simplify if the sext has multiple uses
2) There is no need to simplify is the source value is already sign-extended.
Patch by Gil Rapaport <gil.rapaport@intel.com>
Differential Revision: http://reviews.llvm.org/D6949
llvm-svn: 229659
code.
While this didn't have the miscompile (it used MatchLeft consistently)
it missed some cases where it could use right shifts. I've added a test
case Craig Topper came up with to exercise the right shift matching.
This code is really identical between the two. I'm going to merge them
next so that we don't keep two copies of all of this logic.
llvm-svn: 229655
track state.
I didn't like this in the code review because the pattern tends to be
error prone, but I didn't see a clear way to rewrite it. Turns out that
there were bugs here, I found them when fuzz testing our shuffle
lowering for correctness on x86.
The core of the problem is that we need to consistently test all our
preconditions for the same directionality of shift and the same input
vector. Instead, formulate this as two predicates (one doesn't depend on
the input in any way), pass things like the directionality and input
vector as inputs, and loop over the alternatives.
This fixes a pattern of very rare miscompiles coming out of this code.
Turned up roughly 4 out of every 1 million v8 shuffles in my fuzz
testing. The new code is over half a million test runs with no failures
yet. I've also fuzzed every other function in the lowering code with
over 3.5 million test cases and not discovered any other miscompiles.
llvm-svn: 229642
This patch teaches fast-isel how to select a (V)CVTSI2SSrr for an integer to
float conversion, and how to select a (V)CVTSI2SDrr for an integer to double
conversion.
Added test 'fast-isel-int-float-conversion.ll'.
Differential Revision: http://reviews.llvm.org/D7698
llvm-svn: 229589
The problem in the original patch was not switching back to .text after printing
an eh table.
Original message:
On ELF, put PIC jump tables in a non executable section.
Fixes PR22558.
llvm-svn: 229586
If an EH table is printed in between the function and the jump table we would
fail to switch back to the text section to print the jump table.
llvm-svn: 229580
Change the memory operands in sse12_fp_packed_scalar_logical_alias from scalars to vectors.
That's what the hardware packed logical FP instructions define: 128-bit memory operands.
There are no scalar versions of these instructions...because this is x86.
Generating the wrong code (folding a scalar load into a 128-bit load) is still possible
using the peephole optimization pass and the load folding tables. We won't completely
solve this bug until we either fix the lowering in fabs/fneg/fcopysign and any other
places where scalar FP logic is created or fix the load folding in foldMemoryOperandImpl()
to make sure it isn't changing the size of the load.
Differential Revision: http://reviews.llvm.org/D7474
llvm-svn: 229531
This is a follow-on patch to:
http://reviews.llvm.org/D7093
That patch canonicalized constant splats as build_vectors,
and this patch removes the constant check so we can canonicalize
all splats as build_vectors.
This fixes the 2nd test case in PR22283:
http://llvm.org/bugs/show_bug.cgi?id=22283
The unfortunate code duplication between SelectionDAG and DAGCombiner
is discussed in the earlier patch review. At least this patch is just
removing code...
This improves an existing x86 AVX test and changes codegen in an ARM test.
Differential Revision: http://reviews.llvm.org/D7389
llvm-svn: 229511
Flag -fast-isel-abort is required in order to verify that X86FastISel
never fails to select FPExt (float-to-double) and FPTrunc (double-to-float).
No Functional change intended.
llvm-svn: 229489
- added mask types v8i1 and v16i1 to possible function parameters
- enabled passing 512-bit vectors in standard CC
- added a test for KNL intel_ocl_bi conventions
llvm-svn: 229482
Vector zext tends to get legalized into a vector anyext, represented as a vector shuffle with an undef vector + a bitcast, that gets ANDed with a mask that zeroes the undef elements.
Combine this into an explicit shuffle with a zero vector instead. This allows shuffle lowering to match it as a zext, instead of matching it as an anyext and emitting an explicit AND.
This combine only covers a subset of the cases, but it's a start.
Differential Revision: http://reviews.llvm.org/D7666
llvm-svn: 229480
This allows it to match still more places where previously we would have
to fall back on floating point shuffles or other more complex lowering
strategies.
I'm hoping to replace some of the hand-rolled unpack matching with this
routine is it gets more and more clever.
llvm-svn: 229463
This test was failing on non-x86 hosts because it specified a cpu of x86_64,
but not an architecture. x86_64 is obviously not a valid cpu on all
architectures.
llvm-svn: 229460
Patch to explicitly add the SSE MOVQ (rr,mr,rm) instructions to SSEPackedInt domain - prevents a number of costly domain switches.
Differential Revision: http://reviews.llvm.org/D7600
llvm-svn: 229439
to generically lower blends and is particularly nice because it is
available frome SSE2 onward. This removes a lot of the remaining domain
crossing blends in SSE2 code.
I'm hoping to replace some of the "interleaved" lowering hacks with
something closer to this which should be more principled. First, this
needs to learn how to detect and use other interleavings besides that of
the natural type provided. That will be a follow-up patch though.
llvm-svn: 229378
This blend instruction is ... really lame. The register usage is insane.
As a consequence this is probably only *barely* better than 2 pshufbs
followed by a por, and that mostly because it only has to read from
a single memory location.
However, this doesn't fix as much as I kind of expected, so more to go.
Pretty sure that the ordering and delegation of v16i8 is just really,
really bad.
llvm-svn: 229373
advantage of the existence of a reasonable blend instruction.
The 256-bit vector shuffle lowering has leveraged the general technique
of decomposed shuffles and blends for quite some time, but this never
made it back into the 128-bit code, and there are a large number of
patterns where this is substantially better. For example, this removes
almost all domain crossing in vector shuffles that involve some blend
and some permutation with SSE4.1 and later. See the massive reduction
in 'shufps' for integer test cases in this commit.
This isn't perfect yet for a few reasons:
1) The v8i16 shuffle lowering continues to plague me. We don't always
form an unpack-based blend when that would be better. But the wins
pretty drastically outstrip the losses here.
2) The v16i8 shuffle lowering is just a disaster here. I never went and
implemented blend support here for some terrible reason. I'll do
that next probably. I've not updated it for now.
More variations on this technique are coming as well -- we don't
shuffle-into-unpack or shuffle-into-palignr, both of which would also be
profitable.
Note that some test cases grow significantly in the number of
instructions, but I expect to actually be faster. We use
pshufd+pshufd+blendw instead of a single shufps, but the pshufd's are
very likely to pipeline well (two ports on most modern intel chips) and
the blend is a *very* fast instruction. The domain switch penalty will
essentially always be more than a blend instruction, which is the only
increase in tree height.
llvm-svn: 229350
This patch refactors the existing lowerVectorShuffleAsByteShift function to add support for 256-bit vectors on AVX2 targets.
It also fixes a tablegen issue that prevented the lowering of vpslldq/vpsrldq vec256 instructions.
Differential Revision: http://reviews.llvm.org/D7596
llvm-svn: 229311
when that will allow it to lower with a single permute instead of
multiple permutes.
It tries to detect when it will only have to do a single permute in
either case to maximize folding of loads and such.
This cuts a *lot* of the avx2 shuffle permute counts in half. =]
llvm-svn: 229309
directly into blends of the splats.
These patterns show up even very late in the vector shuffle lowering
where we don't have any chance for DAG combining to kick in, and
blending is a tremendously simpler operation to model. By coercing the
shuffle into a blend we can much more easily match and lower shuffles of
splats.
Immediately with this change there are significantly more blends being
matched in the x86 vector shuffle lowering.
llvm-svn: 229308
I was somewhat surprised this pattern really came up, but it does. It
seems better to just directly handle it than try to special case every
place where we end up forming a shuffle that devolves to a shuffle of
a zero vector.
llvm-svn: 229301
subvectors from buildvectors. That doesn't really make any sense and it
breaks all of the down-stream matching of buildvectors to cleverly lower
shuffles.
With this, we now get the shift-based lowering of 256-bit vector
shuffles with AVX1 when we split them into 128-bit vectors. We also do
much better on the zero-extension patterns, although there remains quite
a bit of room for improvement here.
llvm-svn: 229299
