QingShan Zhang [Thu, 3 Jan 2019 05:04:18 +0000 (05:04 +0000)]
[Power9] Enable the Out-of-Order scheduling model for P9 hw
When switched to the MI scheduler for P9, the hardware is modeled as out of order.
However, inside the MI Scheduler algorithm, we still use the in-order scheduling model
as the MicroOpBufferSize isn't set. The MI scheduler take it as the hw cannot buffer
the op. So, only when all the available instructions issued, the pending instruction
could be scheduled. That is not true for our P9 hw in fact.
This patch is trying to enable the Out-of-Order scheduling model. The buffer size 44 is
picked from the P9 hw spec, and the perf test indicate that, its value won't hurt the cpu2017.
With this patch, there are 3 specs improved over 3% and 1 spec deg over 3%. The detail is as follows:
Pete Cooper [Thu, 3 Jan 2019 01:38:08 +0000 (01:38 +0000)]
Teach ObjCARC optimizer about equivalent PHIs when eliminating autoreleaseRV/retainRV pairs
OptimizeAutoreleaseRVCall skips optimizing llvm.objc.autoreleaseReturnValue if it
sees a user which is llvm.objc.retainAutoreleasedReturnValue, and if they have
equivalent arguments (either identical or equivalent PHIs). It then assumes that
ObjCARCOpt::OptimizeRetainRVCall will optimize the pair instead.
Trouble is, ObjCARCOpt::OptimizeRetainRVCall doesn't know about equivalent PHIs
so optimizes in a different way and we are left with an unoptimized llvm.objc.autoreleaseReturnValue.
This teaches ObjCARCOpt::OptimizeRetainRVCall to also understand PHI equivalence.
Mike Spertus [Thu, 3 Jan 2019 00:52:54 +0000 (00:52 +0000)]
Fix MSVC visualizer for PointerUnion4
Calculate which item is being held and then display it with the appropriate type. We also
optimize the display of PointerUnion3 to take advantage of our knowing that the IntMask is
always 1 in PointerUnion types
Teresa Johnson [Wed, 2 Jan 2019 23:48:00 +0000 (23:48 +0000)]
[gold] emit assembly listing from gold plugin on LTO stage
Summary:
Sometimes it's useful to emit assembly after LTO stage to modify it manually. Emitting precodegen bitcode file (via save-temps plugin option) and then feeding it to llc doesn't always give the same binary as original.
This patch is simpler alternative to https://reviews.llvm.org/D24020.
Craig Topper [Wed, 2 Jan 2019 23:24:03 +0000 (23:24 +0000)]
[X86] Add test cases to show that we fail to fold loads into i8 smulo and i8/i16/i32/i64 umulo lowering without the assistance of the peephole pass. NFC
Craig Topper [Wed, 2 Jan 2019 19:01:05 +0000 (19:01 +0000)]
[X86] Remove X86ISD::INC/DEC. Just select them from X86ISD::ADD/SUB at isel time
INC/DEC are pretty much the same as ADD/SUB except that they don't update the C flag.
This patch removes the special nodes and just pattern matches from ADD/SUB during isel if the C flag isn't being used.
I had to avoid selecting DEC is the result isn't used. This will become a SUB immediate which will turned into a CMP later by optimizeCompareInstr. This lead to the one test change where we use a CMP instead of a DEC for an overflow intrinsic since we only checked the flag.
This also exposed a hole in our RMW flag matching use of hasNoCarryFlagUses. Our root node for the match is a store and there's no guarantee that all the flag users have been selected yet. So hasNoCarryFlagUses needs to check copyToReg and machine opcodes, but it also needs to check for the pre-match SETCC, SETCC_CARRY, BRCOND, and CMOV opcodes.
Craig Topper [Wed, 2 Jan 2019 18:19:07 +0000 (18:19 +0000)]
[DAGCombiner] After performing the division by constant optimization for a DIV or REM node, replace the users of the corresponding REM or DIV node if it exists.
Currently we expand the two nodes separately. This gives DAG combiner an opportunity to optimize the expanded sequence taking into account only one set of users. When we expand the other node we'll create the expansion again, but might not be able to optimize it the same way. So the nodes won't CSE and we'll have two similarish sequences in the same basic block. By expanding both nodes at the same time we'll avoid prematurely optimizing the expansion until both the division and remainder have been replaced.
Improves the test case from PR38217. There may be additional opportunities after this.
Nico Weber [Wed, 2 Jan 2019 18:13:14 +0000 (18:13 +0000)]
[gn build] Add fuzzers in llvm/tools that are needed for check-llvm
Also add a fuzzer() template for defining fuzzers that's similar to
add_llvm_fuzzer in the CMake build, and a build file for dependency
llvm/lib/FuzzMutate.
Also make `assert(defined(...` error strings a bit more self-consistent.
Craig Topper [Wed, 2 Jan 2019 18:09:41 +0000 (18:09 +0000)]
[X86] Adding full coverage of MC encoding for the XOP and LWP ISAs.
Adding MC regressions tests to cover the XOP isa set.
This patch is part of a larger task to cover MC encoding of all X86 isa sets started in revision: https://reviews.llvm.org/D39952
Craig Topper [Wed, 2 Jan 2019 17:58:30 +0000 (17:58 +0000)]
[LegalizeIntegerTypes] When promoting the result of an extract_vector_elt also promote the input type if necessary
By also promoting the input type we get a better idea for what scalar type to use. This can provide better results if the result of the extract is sign extended. What was previously happening is that the extract result would be legalized, sometime later the input of the sign extend would be legalized using the result of the extract. Then later the extract input would be legalized forcing a truncate into the input of the sign extend using a replace all uses. This requires DAG combine to combine out the sext/truncate pair. But sometimes we visited the truncate first and messed things up before the sext could be combined.
By creating the extract with the correct scalar type when we create legalize the result type, the truncate will be added right away. Then when the sign_extend input is legalized it will create an any_extend of the truncate which can be optimized by getNode to maybe remove the truncate. And then a sign_extend_inreg. Now DAG combine doesn't have to worry about getting rid of the extend.
Craig Topper [Wed, 2 Jan 2019 17:58:27 +0000 (17:58 +0000)]
[DAGCombiner][X86][PowerPC] Teach visitSIGN_EXTEND_INREG to fold (sext_in_reg (aext/sext x)) -> (sext x) when x has more than 1 sign bit and the sext_inreg is from one of them.
If x has multiple sign bits than it doesn't matter which one we extend from so we can sext from x's msb instead.
The X86 setcc-combine.ll changes are a little weird. It appears we ended up with a (sext_inreg (aext (trunc (extractelt)))) after type legalization. The sext_inreg+aext now gets optimized by this combine to leave (sext (trunc (extractelt))). Then we visit the trunc before we visit the sext. This ends up changing the truncate to an extractvectorelt from a bitcasted vector. I have a follow up patch to fix this.
Nico Weber [Wed, 2 Jan 2019 17:38:22 +0000 (17:38 +0000)]
[gn build] Add build files for bugpoint-passes and LLVMHello plugins
These two plugins are loaded into a host process that contains all LLVM
symbols, so they don't link against anything. This required minor readjustments
to the tablegen() setup of IR.
Wei Mi [Wed, 2 Jan 2019 17:07:23 +0000 (17:07 +0000)]
[PowerPC] Remove SeenUse check when optimizing conditional branch in
PPCPreEmitPeephole pass.
PPCPreEmitPeephole will convert a BC to B when the conditional branch is
based on a constant CR by CRSET or CRUNSET. This is added in
https://reviews.llvm.org/rL343100.
When the conditional branch is known to be always taken, all branches will
be removed and a new unconditional branch will be inserted. However, when
SeenUse is false the original patch will not remove the branches, but still
insert the new unconditional branch, update the successors and create
inconsistent IR. Compiling the synthetic testcase included can show the
problem we run into.
The patch simply removes the SeenUse condition when adding branches into
InstrsToErase set.
Hal Finkel [Wed, 2 Jan 2019 16:28:09 +0000 (16:28 +0000)]
[BasicAA] Support arbitrary pointer sizes (and fix an overflow bug)
Motivated by the discussion in D38499, this patch updates BasicAA to support
arbitrary pointer sizes by switching most remaining non-APInt calculations to
use APInt. The size of these APInts is set to the maximum pointer size (maximum
over all address spaces described by the data layout string).
Most of this translation is straightforward, but this patch contains a fix for
a bug that revealed itself during this translation process. In order for
test/Analysis/BasicAA/gep-and-alias.ll to pass, which is run with 32-bit
pointers, the intermediate calculations must be performed using 64-bit
integers. This is because, as noted in the patch, when GetLinearExpression
decomposes an expression into C1*V+C2, and we then multiply this by Scale, and
distribute, to get (C1*Scale)*V + C2*Scale, it can be the case that, even
through C1*V+C2 does not overflow for relevant values of V, (C2*Scale) can
overflow. If this happens, later logic will draw invalid conclusions from the
(base) offset value. Thus, when initially applying the APInt conversion,
because the maximum pointer size in this test is 32 bits, it started failing.
Suspicious, I created a 64-bit version of this test (included here), and that
failed (miscompiled) on trunk for a similar reason (the multiplication can
overflow).
After fixing this overflow bug, the first test case (at least) in
Analysis/BasicAA/q.bad.ll started failing. This is also a 32-bit test, and was
relying on having 64-bit intermediate values to have BasicAA return an accurate
result. In order to fix this problem, and because I believe that it is not
uncommon to use i64 indexing expressions in 32-bit code (especially portable
code using int64_t), it seems reasonable to always use at least 64-bit
integers. In this way, we won't regress our analysis capabilities (and there's
a command-line option added, so experimenting with this should be easy).
As pointed out by Eli during the review, there are other potential overflow
conditions that this patch does not address. Fixing those is left to follow-up
work.
Patch by me with contributions from Michael Ferguson (mferguson@cray.com).
Philip Pfaffe [Wed, 2 Jan 2019 15:41:47 +0000 (15:41 +0000)]
Extend Module::getOrInsertGlobal to control the construction of the
GlobalVariable
Summary:
Extend Module::getOrInsertGlobal to accept a callback for creating a new
GlobalVariable if necessary instead of calling the GV constructor
directly using default arguments. Additionally overload
getOrInsertGlobal for the previous default behavior.
[MCA] Minor refactoring of method DefaultResourceStrategy::select. NFCI
Common code used by the default resource strategy to select pipeline resources
has been moved to an helper function.
The new selection logic has been slightly rewritten to get rid of a redundant
zero check on the `ReadyMask` value. Before this patch, method select internally
called function `PowerOf2Floor` to compute the next ready pipeline resource.
However, `PowerOf2Floor` forces an implicit (redundant) zero check on the input
value. By construction, `ReadyMask` can never be zero. This patch replaces the
call to `PowerOf2Floor` with an equivalent block of code which avoids the
redundant zero check. This gives a minor 3-3.5% speedup on a release build.
Craig Topper [Wed, 2 Jan 2019 06:40:11 +0000 (06:40 +0000)]
[X86] Remove the separate SMUL8/UMUL8 X86ISD opcodes by merging with SMUL/UMUL. Remove the second result from X86ISD::UMUL.
All of these use custom isel so we can pretty easily detect the differences in the custom code in X86ISelDAGToDAG. The ISD opcodes just need to express the desired semantics not the details of how they would be selected by isel. So unifying them lets us remove the special casing from lowering.
Craig Topper [Wed, 2 Jan 2019 05:46:03 +0000 (05:46 +0000)]
[X86] Allow LowerSELECT and LowerBRCOND to directly lower i8 UMULO/SMULO.
These require a different X86ISD node to be created than i16/i32/i64. I guess no one wanted to add the special code for that except in LowerXALUO. But now LowerXALUO, LowerSELECT, and LowerBRCOND all use a common helper function so they all share the special code.
Unfortunately, there are no test changes because we seem to correct the miss in a DAG combine later. I did verify it manually using test cases from xmulo.ll
Craig Topper [Wed, 2 Jan 2019 05:46:00 +0000 (05:46 +0000)]
[X86] Remove KNL specific check prefix from xmulo.ll test. NFC
This was added at a time when i1 was a legal type with avx512f and there was a bug. i1 is no longer considered a legal type with avx512f so there should be no codegen difference.
Sanjay Patel [Tue, 1 Jan 2019 21:51:39 +0000 (21:51 +0000)]
[InstCombine] canonicalize raw IR rotate patterns to funnel shift
The final piece of IR-level analysis to allow this was committed with:
rL350188
Using the intrinsics should improve transforms based on cost models
like vectorization and inlining.
The backend should be prepared too, so we can now canonicalize more
sequences of shift/logic to the intrinsics and know that the end
result should be equal or better to the original code even if the
target does not have an actual rotate instruction.
Craig Topper [Tue, 1 Jan 2019 19:34:11 +0000 (19:34 +0000)]
[X86] Factor the core code out of LowerXALUO into a helper function. Use it in LowerBRCOND and LowerSELECT to avoid some duplicated code.
This makes it easier to keep the LowerBRCOND and LowerSELECT code in sync with LowerXALUO so they always pick the same operation for overflowing instructions.
This is inspired by the helper functions used by ARM and AArch64 for the same purpose.
The test change is because LowerSELECT was not in sync with LowerXALUO with regard to INC/DEC for SADDO/SSUBO.
Craig Topper [Tue, 1 Jan 2019 18:44:44 +0000 (18:44 +0000)]
[X86] Remove KNL specific check prefix from xaluo.ll test. NFC
This was added at a time when i1 was a legal type with avx512f and there was a bug. i1 is no longer considered a legal type with avx512f so there should be no codegen difference.
Craig Topper [Tue, 1 Jan 2019 18:44:42 +0000 (18:44 +0000)]
[X86] Add test cases to show where LowerSELECT doesn't select SADDO/SSUBO to INC/DEC, but LowerXALUOOp does. Leading to duplicate code.
When SADDO/SSUBO is used as a part of a condition, the X86 backend has to lower the instruction twice. One for the flags use and then once for the data use. These two selections should be kept in sync so they end up with one node providing the data and the flags. This doesn't seem to be happening for INC/DEC.
Nikita Popov [Tue, 1 Jan 2019 10:17:35 +0000 (10:17 +0000)]
[BDCE] Remove -instsimplify from BDCE test; NFC
To make it more obvious which part of the transformation is carried
out by BDCE. Also drop the CHECK-IO lines which only run -instsimplify
as they don't really seem meaningful if the main check doesn't run
-instsimplify either.
Nikita Popov [Tue, 1 Jan 2019 10:05:26 +0000 (10:05 +0000)]
Reapply "[BDCE][DemandedBits] Detect dead uses of undead instructions"
This (mostly) fixes https://bugs.llvm.org/show_bug.cgi?id=39771.
BDCE currently detects instructions that don't have any demanded bits
and replaces their uses with zero. However, if an instruction has
multiple uses, then some of the uses may be dead (have no demanded bits)
even though the instruction itself is still live. This patch extends
DemandedBits/BDCE to detect such uses and replace them with zero.
While this will not immediately render any instructions dead, it may
lead to simplifications (in the motivating case, by converting a rotate
into a simple shift), break dependencies, etc.
The implementation tries to strike a balance between analysis power and
complexity/memory usage. Originally I wanted to track demanded bits on
a per-use level, but ultimately we're only really interested in whether
a use is entirely dead or not. I'm using an extra set to track which uses
are dead. However, as initially all uses are dead, I'm not storing uses
those user is also dead. This case is checked separately instead.
The previous attempt to land this lead to miscompiles, because cases
where uses were initially dead but were later found to be live during
further analysis were not always correctly removed from the DeadUses
set. This is fixed now and the added test case demanstrates such an
instance.
Ayonam Ray [Tue, 1 Jan 2019 06:37:50 +0000 (06:37 +0000)]
Omit range checks from jump tables when lowering switches with unreachable
default
During the lowering of a switch that would result in the generation of a jump
table, a range check is performed before indexing into the jump table, for the
switch value being outside the jump table range and a conditional branch is
inserted to jump to the default block. In case the default block is
unreachable, this conditional jump can be omitted. This patch implements
omitting this conditional branch for unreachable defaults.
Nico Weber [Mon, 31 Dec 2018 23:48:22 +0000 (23:48 +0000)]
[gn build] Add some llvm/tools: llvm-exegesis, llvm-extract, llvm-link
Also add build file for dependency llvm/lib/ExecutionEngine/MCJIT.
The exegesis stuff is pretty hairy and knows a lot about Target internals (in
general, not specifically in the GN build). I put the llvm-tblgen -gen-exegesis
call in llvm/tools/llvm-exegesis/lib/X86, instead of in llvm/lib/Target/X86
where it is in CMake land, and asked on D52932 why it's in that place in the
CMake build.
Michal Gorny [Mon, 31 Dec 2018 13:48:12 +0000 (13:48 +0000)]
[test] Fix propagating HOME envvar to unittests
Propagate HOME environment variable to unittests. This is necessary
to fix test failures resulting from pw_home pointing to a non-existing
directory while being overriden with HOME. Apparently Gentoo users
hit this sometimes when they override build directory for Portage.
Original bug report: https://bugs.gentoo.org/674088
MSan used to report false positives in the case the argument of
llvm.is.constant intrinsic was uninitialized.
In fact checking this argument is unnecessary, as the intrinsic is only
used at compile time, and its value doesn't depend on the value of the
argument.
Nico Weber [Mon, 31 Dec 2018 00:10:47 +0000 (00:10 +0000)]
[gn build] Make `ninja check-clang` also run Clang's unit tests
Also add a build file for clang/lib/ASTMatchers/Dynamic, which is only needed
by tests (and clang/tools/extra).
Also make llvm/utils/gn/build/sync_source_lists_from_cmake.py check that every
CMakeLists.txt file below {lld,clang}/unittests has a corresponding BUILD.gn
file, so we notice if new test binaries get added (since the failure mode for
missing GN build files for tests is just the tests silently not running in the
GN build).
Also add a unittest() macro for defining unit test targets, and add a lengthy
comment there about where the unit test binaries go and why.
With this, the build files for //clang are complete.
Kang Zhang [Sun, 30 Dec 2018 15:13:51 +0000 (15:13 +0000)]
[PowerPC] Fix machine verify pass error for PATCHPOINT pseudo instruction that bad machine code
Summary:
For SDAG, we pretend patchpoints aren't special at all until we emit the code for the pseudo.
Then the verifier runs and it seems like we have a use of an undefined register (the register will
be reserved later, but the verifier doesn't know that).
So this patch call setUsesTOCBasePtr before emit the code for the pseudo, so verifier can know
X2 is a reserved register.
Craig Topper [Sun, 30 Dec 2018 03:05:07 +0000 (03:05 +0000)]
[X86] Don't mark SEXTLOAD from v4i8/v4i16/v8i8 as Custom on pre-sse4.1.
This seems to be getting in the way more than its helping. This does mean we stop scalarizing some cases, but I'm not convinced the scalarization was really better.
Some of the changes to vsel-cmp-load.ll are a regression but D56156 should fix it.
Craig Topper [Sun, 30 Dec 2018 02:30:34 +0000 (02:30 +0000)]
[X86] Add custom type legalization for SIGN_EXTEND_VECTOR_INREG from 16i16/v32i8 to v4i64 when v4i64 needs splitting.
This allows us to sign extend to v4i32 first. And then share that extension to implement the final steps to v4i64 using a pcmpgt and punpckl and punpckh.
We already do something similar for SIGN_EXTEND with -x86-experimental-vector-widening-legalization.
Nemanja Ivanovic [Sat, 29 Dec 2018 16:13:11 +0000 (16:13 +0000)]
[PowerPC][NFC] Macro for register set defs for the Asm Parser
We have some unfortunate code in the back end that defines a bunch of register
sets for the Asm Parser. Every time another class is needed in the parser, we
have to add another one of those definitions with explicit lists of registers.
This NFC patch simply provides macros to use to condense that code a little bit.
Nemanja Ivanovic [Sat, 29 Dec 2018 13:40:48 +0000 (13:40 +0000)]
[PowerPC] Complete the custom legalization of vector int to fp conversion
A recent patch has added custom legalization of vector conversions of
v2i16 -> v2f64. This just rounds it out for other types where the input vector
has an illegal (narrower) type than the result vector. Specifically, this will
handle the following conversions:
Nemanja Ivanovic [Sat, 29 Dec 2018 11:43:54 +0000 (11:43 +0000)]
[PowerPC] Fix CR Bit spill pseudo expansion
The current CRBIT spill pseudo-op expansion creates a KILL instruction
that kills the CRBIT and defines the enclosing CR field. However, this
paints a false picture to the register allocator that all bits in the CR
field are killed so copies of other bits out of the field become dead and
removable.
This changes the expansion to preserve the KILL flag on the CRBIT as an
implicit use and to treat the CR field as an undef input.
Thanks to Hal Finkel for the review and Uli Weigand for implementation input.
Simon Atanasyan [Sat, 29 Dec 2018 10:10:02 +0000 (10:10 +0000)]
[mips] Show an error on attempt to use 64-bit PC-relative relocation
The following code requests 64-bit PC-relative relocations unsupported
by MIPS ABI. Now it triggers an assertion. It's better to show an error
message.
```
foo:
.quad bar - foo
```
Craig Topper [Sat, 29 Dec 2018 01:17:11 +0000 (01:17 +0000)]
[X86] Don't mark SEXTLOAD v4i8->v4i64 and v8i8->v8i64 as custom under vector widening legalization.
This was tricking us into making these operations and then letting them get scalarized later. But I can't prove that the scalarized version is actually better.
Craig Topper [Fri, 28 Dec 2018 19:19:39 +0000 (19:19 +0000)]
[X86] Directly emit X86ISD::PMULUDQ from the ReplaceNodeResults handling of v2i8/v2i16/v2i32 multiply.
Previously we emitted a multiply and some masking that was supposed to matched to PMULUDQ, but the masking could sometimes be removed before we got a chance to match it. So instead just emit the PMULUDQ directly.
Remove the DAG combine that was added when the ReplaceNodeResults code was originally added. Add a new DAG combine to avoid regressions in shrink_vmul.ll
Some of the shrink_vmul.ll test cases now pick PMULUDQ instead of PMADDWD/PMULLD, but I think this should be an improvement on most CPUs.
I think all of this can go away if/when we switch to -x86-experimental-vector-widening-legalization
Diogo N. Sampaio [Fri, 28 Dec 2018 17:14:58 +0000 (17:14 +0000)]
[AArch64] Add command-line option for SB
SB (Speculative Barrier) is only mandatory from 8.5
onwards but is optional from Armv8.0-A. This patch adds a command
line option to enable SB, as it was previously only possible to
enable by selecting -march=armv8.5-a.
This patch also moves to FeatureSB the old FeatureSpecRestrict.
Hiroshi Inoue [Fri, 28 Dec 2018 08:00:39 +0000 (08:00 +0000)]
[PowerPC] handle ISD:TRUNCATE in BitPermutationSelector
This is the last one in a series of patches to support better code generation for bitfield insert.
BitPermutationSelector already support ISD::ZERO_EXTEND but not TRUNCATE.
This patch adds support for ISD:TRUNCATE in BitPermutationSelector.
For example of this test case,
struct s64b {
int a:4;
int b:16;
int c:24;
};
void bitfieldinsert64b(struct s64b *p, unsigned char v) {
p->b = v;
}
By handling truncate in the BitPermutationSelector, we can use information from AssertZext when selecting t19 and skip the mask operation corresponding to t18.
So the generated sequences with and without this patch are
without this patch
rlwinm 5, 5, 0, 28, 11 # corresponding to t18
rlwimi 5, 4, 4, 20, 27
with this patch
rlwimi 5, 4, 4, 12, 27
Max Kazantsev [Fri, 28 Dec 2018 06:08:51 +0000 (06:08 +0000)]
[LoopSimplifyCFG] Delete dead blocks in RPO
Deletion of dead blocks in arbitrary order may lead to failure
of assertion in `DeleteDeadBlock` that requires that we have
deleted all predecessors before we can delete the current block.
We should instead delete them in RPO order.
QingShan Zhang [Fri, 28 Dec 2018 03:38:09 +0000 (03:38 +0000)]
[PowerPC] Remove the implicit use of the register if it is replaced by Imm
If we are changing the MI operand from Reg to Imm, we need also handle its implicit use if have.
projects / llvm / log