]> granicus.if.org Git - llvm/log
llvm
5 years agoRemove an extra ";", NFC.
Haojian Wu [Fri, 30 Aug 2019 12:09:31 +0000 (12:09 +0000)]
Remove an extra ";", NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370465 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add tests for rotate matching. NFC
Amaury Sechet [Fri, 30 Aug 2019 11:35:28 +0000 (11:35 +0000)]
[X86] Add tests for rotate matching. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370464 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CodeGen] Introduce MachineBasicBlock::replacePhiUsesWith helper and use it. NFC
Bjorn Pettersson [Fri, 30 Aug 2019 11:23:10 +0000 (11:23 +0000)]
[CodeGen] Introduce MachineBasicBlock::replacePhiUsesWith helper and use it. NFC

Summary:
Found a couple of places in the code where all the PHI nodes
of a MBB is updated, replacing references to one MBB by
reference to another MBB instead.

This patch simply refactors the code to use a common helper
(MachineBasicBlock::replacePhiUsesWith) for such PHI node
updates.

Reviewers: t.p.northover, arsenm, uabelho

Subscribers: wdng, hiraditya, jsji, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66750

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370463 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombine] visitMULHS/visitMULHU - isBuildVectorAllZeros doesn't mean node is all...
Simon Pilgrim [Fri, 30 Aug 2019 10:42:14 +0000 (10:42 +0000)]
[DAGCombine] visitMULHS/visitMULHU - isBuildVectorAllZeros doesn't mean node is all zeros

Return a proper zero vector, just in case some elements are undef.

Noticed by inspection after dealing with a similar issue in PR43159.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370460 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix Wdocumentation warning. NFCI.
Simon Pilgrim [Fri, 30 Aug 2019 10:25:52 +0000 (10:25 +0000)]
Fix Wdocumentation warning. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370459 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-objcopy] Allow the visibility of symbols created by --binary and
Chris Jackson [Fri, 30 Aug 2019 10:17:16 +0000 (10:17 +0000)]
[llvm-objcopy] Allow the visibility of symbols created by --binary and
--add-symbol to be specified with --new-symbol-visibility

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370458 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Implement AANoAliasCallSiteArgument initialization
Hideto Ueno [Fri, 30 Aug 2019 10:00:32 +0000 (10:00 +0000)]
[Attributor] Implement AANoAliasCallSiteArgument initialization

Summary: This patch adds an appropriate `initialize` method for `AANoAliasCallSiteArgument`.

Reviewers: jdoerfert, sstefan1

Reviewed By: jdoerfert

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66927

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370456 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LoopIdiomRecognize] BCmp loop idiom recognition
Roman Lebedev [Fri, 30 Aug 2019 09:51:23 +0000 (09:51 +0000)]
[LoopIdiomRecognize] BCmp loop idiom recognition

Summary:
@mclow.lists brought up this issue up in IRC.
It is a reasonably common problem to compare some two values for equality.
Those may be just some integers, strings or arrays of integers.

In C, there is `memcmp()`, `bcmp()` functions.
In C++, there exists `std::equal()` algorithm.
One can also write that function manually.

libstdc++'s `std::equal()` is specialized to directly call `memcmp()` for
various types, but not `std::byte` from C++2a. https://godbolt.org/z/mx2ejJ

libc++ does not do anything like that, it simply relies on simple C++'s
`operator==()`. https://godbolt.org/z/er0Zwf (GOOD!)

So likely, there exists a certain performance opportunities.
Let's compare performance of naive `std::equal()` (no `memcmp()`) with one that
is using `memcmp()` (in this case, compiled with modified compiler). {F8768213}

```
#include <algorithm>
#include <cmath>
#include <cstdint>
#include <iterator>
#include <limits>
#include <random>
#include <type_traits>
#include <utility>
#include <vector>

#include "benchmark/benchmark.h"

template <class T>
bool equal(T* a, T* a_end, T* b) noexcept {
  for (; a != a_end; ++a, ++b) {
    if (*a != *b) return false;
  }
  return true;
}

template <typename T>
std::vector<T> getVectorOfRandomNumbers(size_t count) {
  std::random_device rd;
  std::mt19937 gen(rd());
  std::uniform_int_distribution<T> dis(std::numeric_limits<T>::min(),
                                       std::numeric_limits<T>::max());
  std::vector<T> v;
  v.reserve(count);
  std::generate_n(std::back_inserter(v), count,
                  [&dis, &gen]() { return dis(gen); });
  assert(v.size() == count);
  return v;
}

struct Identical {
  template <typename T>
  static std::pair<std::vector<T>, std::vector<T>> Gen(size_t count) {
    auto Tmp = getVectorOfRandomNumbers<T>(count);
    return std::make_pair(Tmp, std::move(Tmp));
  }
};

struct InequalHalfway {
  template <typename T>
  static std::pair<std::vector<T>, std::vector<T>> Gen(size_t count) {
    auto V0 = getVectorOfRandomNumbers<T>(count);
    auto V1 = V0;
    V1[V1.size() / size_t(2)]++;  // just change the value.
    return std::make_pair(std::move(V0), std::move(V1));
  }
};

template <class T, class Gen>
void BM_bcmp(benchmark::State& state) {
  const size_t Length = state.range(0);

  const std::pair<std::vector<T>, std::vector<T>> Data =
      Gen::template Gen<T>(Length);
  const std::vector<T>& a = Data.first;
  const std::vector<T>& b = Data.second;
  assert(a.size() == Length && b.size() == a.size());

  benchmark::ClobberMemory();
  benchmark::DoNotOptimize(a);
  benchmark::DoNotOptimize(a.data());
  benchmark::DoNotOptimize(b);
  benchmark::DoNotOptimize(b.data());

  for (auto _ : state) {
    const bool is_equal = equal(a.data(), a.data() + a.size(), b.data());
    benchmark::DoNotOptimize(is_equal);
  }
  state.SetComplexityN(Length);
  state.counters["eltcnt"] =
      benchmark::Counter(Length, benchmark::Counter::kIsIterationInvariant);
  state.counters["eltcnt/sec"] =
      benchmark::Counter(Length, benchmark::Counter::kIsIterationInvariantRate);
  const size_t BytesRead = 2 * sizeof(T) * Length;
  state.counters["bytes_read/iteration"] =
      benchmark::Counter(BytesRead, benchmark::Counter::kDefaults,
                         benchmark::Counter::OneK::kIs1024);
  state.counters["bytes_read/sec"] = benchmark::Counter(
      BytesRead, benchmark::Counter::kIsIterationInvariantRate,
      benchmark::Counter::OneK::kIs1024);
}

template <typename T>
static void CustomArguments(benchmark::internal::Benchmark* b) {
  const size_t L2SizeBytes = []() {
    for (const benchmark::CPUInfo::CacheInfo& I :
         benchmark::CPUInfo::Get().caches) {
      if (I.level == 2) return I.size;
    }
    return 0;
  }();
  // What is the largest range we can check to always fit within given L2 cache?
  const size_t MaxLen = L2SizeBytes / /*total bufs*/ 2 /
                        /*maximal elt size*/ sizeof(T) / /*safety margin*/ 2;
  b->RangeMultiplier(2)->Range(1, MaxLen)->Complexity(benchmark::oN);
}

BENCHMARK_TEMPLATE(BM_bcmp, uint8_t, Identical)
    ->Apply(CustomArguments<uint8_t>);
BENCHMARK_TEMPLATE(BM_bcmp, uint16_t, Identical)
    ->Apply(CustomArguments<uint16_t>);
BENCHMARK_TEMPLATE(BM_bcmp, uint32_t, Identical)
    ->Apply(CustomArguments<uint32_t>);
BENCHMARK_TEMPLATE(BM_bcmp, uint64_t, Identical)
    ->Apply(CustomArguments<uint64_t>);

BENCHMARK_TEMPLATE(BM_bcmp, uint8_t, InequalHalfway)
    ->Apply(CustomArguments<uint8_t>);
BENCHMARK_TEMPLATE(BM_bcmp, uint16_t, InequalHalfway)
    ->Apply(CustomArguments<uint16_t>);
BENCHMARK_TEMPLATE(BM_bcmp, uint32_t, InequalHalfway)
    ->Apply(CustomArguments<uint32_t>);
BENCHMARK_TEMPLATE(BM_bcmp, uint64_t, InequalHalfway)
    ->Apply(CustomArguments<uint64_t>);
```
{F8768210}
```
$ ~/src/googlebenchmark/tools/compare.py --no-utest benchmarks build-{old,new}/test/llvm-bcmp-bench
RUNNING: build-old/test/llvm-bcmp-bench --benchmark_out=/tmp/tmpb6PEUx
2019-04-25 21:17:11
Running build-old/test/llvm-bcmp-bench
Run on (8 X 4000 MHz CPU s)
CPU Caches:
  L1 Data 16K (x8)
  L1 Instruction 64K (x4)
  L2 Unified 2048K (x4)
  L3 Unified 8192K (x1)
Load Average: 0.65, 3.90, 4.14
---------------------------------------------------------------------------------------------------
Benchmark                                         Time             CPU   Iterations UserCounters...
---------------------------------------------------------------------------------------------------
<...>
BM_bcmp<uint8_t, Identical>/512000           432131 ns       432101 ns         1613 bytes_read/iteration=1000k bytes_read/sec=2.20706G/s eltcnt=825.856M eltcnt/sec=1.18491G/s
BM_bcmp<uint8_t, Identical>_BigO               0.86 N          0.86 N
BM_bcmp<uint8_t, Identical>_RMS                   8 %             8 %
<...>
BM_bcmp<uint16_t, Identical>/256000          161408 ns       161409 ns         4027 bytes_read/iteration=1000k bytes_read/sec=5.90843G/s eltcnt=1030.91M eltcnt/sec=1.58603G/s
BM_bcmp<uint16_t, Identical>_BigO              0.67 N          0.67 N
BM_bcmp<uint16_t, Identical>_RMS                 25 %            25 %
<...>
BM_bcmp<uint32_t, Identical>/128000           81497 ns        81488 ns         8415 bytes_read/iteration=1000k bytes_read/sec=11.7032G/s eltcnt=1077.12M eltcnt/sec=1.57078G/s
BM_bcmp<uint32_t, Identical>_BigO              0.71 N          0.71 N
BM_bcmp<uint32_t, Identical>_RMS                 42 %            42 %
<...>
BM_bcmp<uint64_t, Identical>/64000            50138 ns        50138 ns        10909 bytes_read/iteration=1000k bytes_read/sec=19.0209G/s eltcnt=698.176M eltcnt/sec=1.27647G/s
BM_bcmp<uint64_t, Identical>_BigO              0.84 N          0.84 N
BM_bcmp<uint64_t, Identical>_RMS                 27 %            27 %
<...>
BM_bcmp<uint8_t, InequalHalfway>/512000      192405 ns       192392 ns         3638 bytes_read/iteration=1000k bytes_read/sec=4.95694G/s eltcnt=1.86266G eltcnt/sec=2.66124G/s
BM_bcmp<uint8_t, InequalHalfway>_BigO          0.38 N          0.38 N
BM_bcmp<uint8_t, InequalHalfway>_RMS              3 %             3 %
<...>
BM_bcmp<uint16_t, InequalHalfway>/256000     127858 ns       127860 ns         5477 bytes_read/iteration=1000k bytes_read/sec=7.45873G/s eltcnt=1.40211G eltcnt/sec=2.00219G/s
BM_bcmp<uint16_t, InequalHalfway>_BigO         0.50 N          0.50 N
BM_bcmp<uint16_t, InequalHalfway>_RMS             0 %             0 %
<...>
BM_bcmp<uint32_t, InequalHalfway>/128000      49140 ns        49140 ns        14281 bytes_read/iteration=1000k bytes_read/sec=19.4072G/s eltcnt=1.82797G eltcnt/sec=2.60478G/s
BM_bcmp<uint32_t, InequalHalfway>_BigO         0.40 N          0.40 N
BM_bcmp<uint32_t, InequalHalfway>_RMS            18 %            18 %
<...>
BM_bcmp<uint64_t, InequalHalfway>/64000       32101 ns        32099 ns        21786 bytes_read/iteration=1000k bytes_read/sec=29.7101G/s eltcnt=1.3943G eltcnt/sec=1.99381G/s
BM_bcmp<uint64_t, InequalHalfway>_BigO         0.50 N          0.50 N
BM_bcmp<uint64_t, InequalHalfway>_RMS             1 %             1 %
RUNNING: build-new/test/llvm-bcmp-bench --benchmark_out=/tmp/tmpQ46PP0
2019-04-25 21:19:29
Running build-new/test/llvm-bcmp-bench
Run on (8 X 4000 MHz CPU s)
CPU Caches:
  L1 Data 16K (x8)
  L1 Instruction 64K (x4)
  L2 Unified 2048K (x4)
  L3 Unified 8192K (x1)
Load Average: 1.01, 2.85, 3.71
---------------------------------------------------------------------------------------------------
Benchmark                                         Time             CPU   Iterations UserCounters...
---------------------------------------------------------------------------------------------------
<...>
BM_bcmp<uint8_t, Identical>/512000            18593 ns        18590 ns        37565 bytes_read/iteration=1000k bytes_read/sec=51.2991G/s eltcnt=19.2333G eltcnt/sec=27.541G/s
BM_bcmp<uint8_t, Identical>_BigO               0.04 N          0.04 N
BM_bcmp<uint8_t, Identical>_RMS                  37 %            37 %
<...>
BM_bcmp<uint16_t, Identical>/256000           18950 ns        18948 ns        37223 bytes_read/iteration=1000k bytes_read/sec=50.3324G/s eltcnt=9.52909G eltcnt/sec=13.511G/s
BM_bcmp<uint16_t, Identical>_BigO              0.08 N          0.08 N
BM_bcmp<uint16_t, Identical>_RMS                 34 %            34 %
<...>
BM_bcmp<uint32_t, Identical>/128000           18627 ns        18627 ns        37895 bytes_read/iteration=1000k bytes_read/sec=51.198G/s eltcnt=4.85056G eltcnt/sec=6.87168G/s
BM_bcmp<uint32_t, Identical>_BigO              0.16 N          0.16 N
BM_bcmp<uint32_t, Identical>_RMS                 35 %            35 %
<...>
BM_bcmp<uint64_t, Identical>/64000            18855 ns        18855 ns        37458 bytes_read/iteration=1000k bytes_read/sec=50.5791G/s eltcnt=2.39731G eltcnt/sec=3.3943G/s
BM_bcmp<uint64_t, Identical>_BigO              0.32 N          0.32 N
BM_bcmp<uint64_t, Identical>_RMS                 33 %            33 %
<...>
BM_bcmp<uint8_t, InequalHalfway>/512000        9570 ns         9569 ns        73500 bytes_read/iteration=1000k bytes_read/sec=99.6601G/s eltcnt=37.632G eltcnt/sec=53.5046G/s
BM_bcmp<uint8_t, InequalHalfway>_BigO          0.02 N          0.02 N
BM_bcmp<uint8_t, InequalHalfway>_RMS             29 %            29 %
<...>
BM_bcmp<uint16_t, InequalHalfway>/256000       9547 ns         9547 ns        74343 bytes_read/iteration=1000k bytes_read/sec=99.8971G/s eltcnt=19.0318G eltcnt/sec=26.8159G/s
BM_bcmp<uint16_t, InequalHalfway>_BigO         0.04 N          0.04 N
BM_bcmp<uint16_t, InequalHalfway>_RMS            29 %            29 %
<...>
BM_bcmp<uint32_t, InequalHalfway>/128000       9396 ns         9394 ns        73521 bytes_read/iteration=1000k bytes_read/sec=101.518G/s eltcnt=9.41069G eltcnt/sec=13.6255G/s
BM_bcmp<uint32_t, InequalHalfway>_BigO         0.08 N          0.08 N
BM_bcmp<uint32_t, InequalHalfway>_RMS            30 %            30 %
<...>
BM_bcmp<uint64_t, InequalHalfway>/64000        9499 ns         9498 ns        73802 bytes_read/iteration=1000k bytes_read/sec=100.405G/s eltcnt=4.72333G eltcnt/sec=6.73808G/s
BM_bcmp<uint64_t, InequalHalfway>_BigO         0.16 N          0.16 N
BM_bcmp<uint64_t, InequalHalfway>_RMS            28 %            28 %
Comparing build-old/test/llvm-bcmp-bench to build-new/test/llvm-bcmp-bench
Benchmark                                                  Time             CPU      Time Old      Time New       CPU Old       CPU New
---------------------------------------------------------------------------------------------------------------------------------------
<...>
BM_bcmp<uint8_t, Identical>/512000                      -0.9570         -0.9570        432131         18593        432101         18590
<...>
BM_bcmp<uint16_t, Identical>/256000                     -0.8826         -0.8826        161408         18950        161409         18948
<...>
BM_bcmp<uint32_t, Identical>/128000                     -0.7714         -0.7714         81497         18627         81488         18627
<...>
BM_bcmp<uint64_t, Identical>/64000                      -0.6239         -0.6239         50138         18855         50138         18855
<...>
BM_bcmp<uint8_t, InequalHalfway>/512000                 -0.9503         -0.9503        192405          9570        192392          9569
<...>
BM_bcmp<uint16_t, InequalHalfway>/256000                -0.9253         -0.9253        127858          9547        127860          9547
<...>
BM_bcmp<uint32_t, InequalHalfway>/128000                -0.8088         -0.8088         49140          9396         49140          9394
<...>
BM_bcmp<uint64_t, InequalHalfway>/64000                 -0.7041         -0.7041         32101          9499         32099          9498
```

What can we tell from the benchmark?
* Performance of naive equality check somewhat improves with element size,
  maxing out at eltcnt/sec=1.58603G/s for uint16_t, or bytes_read/sec=19.0209G/s
  for uint64_t. I think, that instability implies performance problems.
* Performance of `memcmp()`-aware benchmark always maxes out at around
  bytes_read/sec=51.2991G/s for every type. That is 2.6x the throughput of the
  naive variant!
* eltcnt/sec metric for the `memcmp()`-aware benchmark maxes out at
  eltcnt/sec=27.541G/s for uint8_t (was: eltcnt/sec=1.18491G/s, so 24x) and
  linearly decreases with element size.
  For uint64_t, it's ~4x+ the elements/second.
* The call obvious is more pricey than the loop, with small element count.
  As it can be seen from the full output {F8768210}, the `memcmp()` is almost
  universally worse, independent of the element size (and thus buffer size) when
  element count is less than 8.

So all in all, bcmp idiom does indeed pose untapped performance headroom.
This diff does implement said idiom recognition. I think a reasonable test
coverage is present, but do tell if there is anything obvious missing.

Now, quality. This does succeed to build and pass the test-suite, at least
without any non-bundled elements. {F8768216} {F8768217}
This transform fires 91 times:
```
$ /build/test-suite/utils/compare.py -m loop-idiom.NumBCmp result-new.json
Tests: 1149
Metric: loop-idiom.NumBCmp

Program                                         result-new

MultiSourc...Benchmarks/7zip/7zip-benchmark    79.00
MultiSource/Applications/d/make_dparser         3.00
SingleSource/UnitTests/vla                      2.00
MultiSource/Applications/Burg/burg              1.00
MultiSourc.../Applications/JM/lencod/lencod     1.00
MultiSource/Applications/lemon/lemon            1.00
MultiSource/Benchmarks/Bullet/bullet            1.00
MultiSourc...e/Benchmarks/MallocBench/gs/gs     1.00
MultiSourc...gs-C/TimberWolfMC/timberwolfmc     1.00
MultiSourc...Prolangs-C/simulator/simulator     1.00
```
The size changes are:
I'm not sure what's going on with SingleSource/UnitTests/vla.test yet, did not look.
```
$ /build/test-suite/utils/compare.py -m size..text result-{old,new}.json --filter-hash
Tests: 1149
Same hash: 907 (filtered out)
Remaining: 242
Metric: size..text

Program                                        result-old result-new diff
test-suite...ingleSource/UnitTests/vla.test   753.00     833.00     10.6%
test-suite...marks/7zip/7zip-benchmark.test   1001697.00 966657.00  -3.5%
test-suite...ngs-C/simulator/simulator.test   32369.00   32321.00   -0.1%
test-suite...plications/d/make_dparser.test   89585.00   89505.00   -0.1%
test-suite...ce/Applications/Burg/burg.test   40817.00   40785.00   -0.1%
test-suite.../Applications/lemon/lemon.test   47281.00   47249.00   -0.1%
test-suite...TimberWolfMC/timberwolfmc.test   250065.00  250113.00   0.0%
test-suite...chmarks/MallocBench/gs/gs.test   149889.00  149873.00  -0.0%
test-suite...ications/JM/lencod/lencod.test   769585.00  769569.00  -0.0%
test-suite.../Benchmarks/Bullet/bullet.test   770049.00  770049.00   0.0%
test-suite...HMARK_ANISTROPIC_DIFFUSION/128    NaN        NaN        nan%
test-suite...HMARK_ANISTROPIC_DIFFUSION/256    NaN        NaN        nan%
test-suite...CHMARK_ANISTROPIC_DIFFUSION/64    NaN        NaN        nan%
test-suite...CHMARK_ANISTROPIC_DIFFUSION/32    NaN        NaN        nan%
test-suite...ENCHMARK_BILATERAL_FILTER/64/4    NaN        NaN        nan%
Geomean difference                                                   nan%
         result-old    result-new       diff
count  1.000000e+01  10.00000      10.000000
mean   3.152090e+05  311695.40000  0.006749
std    3.790398e+05  372091.42232  0.036605
min    7.530000e+02  833.00000    -0.034981
25%    4.243300e+04  42401.00000  -0.000866
50%    1.197370e+05  119689.00000 -0.000392
75%    6.397050e+05  639705.00000 -0.000005
max    1.001697e+06  966657.00000  0.106242
```

I don't have timings though.

And now to the code. The basic idea is to completely replace the whole loop.
If we can't fully kill it, don't transform.
I have left one or two comments in the code, so hopefully it can be understood.

Also, there is a few TODO's that i have left for follow-ups:
* widening of `memcmp()`/`bcmp()`
* step smaller than the comparison size
* Metadata propagation
* more than two blocks as long as there is still a single backedge?
* ???

Reviewers: reames, fhahn, mkazantsev, chandlerc, craig.topper, courbet

Reviewed By: courbet

Subscribers: hiraditya, xbolva00, nikic, jfb, gchatelet, courbet, llvm-commits, mclow.lists

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D61144

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370454 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC] SCEVExpander: add SetCurrentDebugLocation() / getCurrentDebugLocation() wrappers
Roman Lebedev [Fri, 30 Aug 2019 09:51:02 +0000 (09:51 +0000)]
[NFC] SCEVExpander: add SetCurrentDebugLocation() / getCurrentDebugLocation() wrappers

Summary:
The internal `Builder` is private, which means there is
currently no way to set the debuginfo locations for `SCEVExpander`.
This only adds the wrappers, but does not use them anywhere.

Reviewers: mkazantsev, sanjoy, gberry, jyknight, dneilson

Reviewed By: sanjoy

Subscribers: javed.absar, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D61007

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370453 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LiveDebugValues] Insert entry values after bundles
David Stenberg [Fri, 30 Aug 2019 09:06:50 +0000 (09:06 +0000)]
[LiveDebugValues] Insert entry values after bundles

Summary:
Change LiveDebugValues so that it inserts entry values after the bundle
which contains the clobbering instruction. Previously it would insert
the debug value after the bundle head using insertAfter(), breaking the
bundle.

Reviewers: djtodoro, NikolaPrica, aprantl, vsk

Reviewed By: vsk

Subscribers: hiraditya, llvm-commits

Tags: #debug-info, #llvm

Differential Revision: https://reviews.llvm.org/D66888

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370448 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agovim: add `immarg` keyword
Sven van Haastregt [Fri, 30 Aug 2019 08:52:55 +0000 (08:52 +0000)]
vim: add `immarg` keyword

The `immarg` attribute was added in r355981.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370443 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agogn build: Merge r370441
Nico Weber [Fri, 30 Aug 2019 08:26:37 +0000 (08:26 +0000)]
gn build: Merge r370441

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370442 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ADT] Removed VariadicFunction
Dmitri Gribenko [Fri, 30 Aug 2019 08:21:55 +0000 (08:21 +0000)]
[ADT] Removed VariadicFunction

Summary:
It is not used. It uses macro-based unrolling instead of variadic
templates, so it is not idiomatic anymore, and therefore it is a
questionable API to keep "just in case".

Subscribers: mgorny, dmgreen, dexonsmith, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66961

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370441 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LLD] [COFF] Support merging resource object files
Martin Storsjo [Fri, 30 Aug 2019 06:56:33 +0000 (06:56 +0000)]
[LLD] [COFF] Support merging resource object files

Extend WindowsResourceParser to support using a ResourceSectionRef for
loading resources from an object file.

Only allow merging resource object files in mingw mode; keep the
existing error on multiple resource objects in link mode.

If there only is one resource object file and no .res resources,
don't parse and recreate the .rsrc section, but just link it in without
inspecting it. This allows users to produce any .rsrc section (outside
of what the parser supports), just like before. (I don't have a specific
need for this, but it reduces the risk of this new feature.)

Separate out the .rsrc section chunks in InputFiles.cpp, and only include
them in the list of section chunks to link if we've determined that there
only was one single resource object. (We need to keep other chunks from
those object files, as they can legitimately contain other sections as
well, in addition to .rsrc section chunks.)

Differential Revision: https://reviews.llvm.org/D66824

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370436 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WindowsResource] Remove use of global variables in WindowsResourceParser
Martin Storsjo [Fri, 30 Aug 2019 06:56:02 +0000 (06:56 +0000)]
[WindowsResource] Remove use of global variables in WindowsResourceParser

Instead of updating a global variable counter for the next index of
strings and data blobs, pass along a reference to actual data/string
vectors and let the TreeNode insertion methods add their data/strings to
the vectors when a new entry is needed.

Additionally, if the resource tree had duplicates, that were ignored
with -force:multipleres in lld, we no longer store all versions of the
duplicated resource data, now we only keep the one that actually ends
up referenced.

Differential Revision: https://reviews.llvm.org/D66823

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370435 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WindowsResource] Avoid duplicating the input filenames for each resource. NFC.
Martin Storsjo [Fri, 30 Aug 2019 06:55:54 +0000 (06:55 +0000)]
[WindowsResource] Avoid duplicating the input filenames for each resource. NFC.

Differential Revision: https://reviews.llvm.org/D66821

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370434 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[COFF] Add a ResourceSectionRef method for getting resource contents
Martin Storsjo [Fri, 30 Aug 2019 06:55:49 +0000 (06:55 +0000)]
[COFF] Add a ResourceSectionRef method for getting resource contents

This allows llvm-readobj to print the contents of each resource
when printing resources from an object file or executable, like it
already does for plain .res files.

This requires providing the whole COFFObjectFile to ResourceSectionRef.

This supports both object files and executables. For executables,
the DataRVA field is used as is to look up the right section.

For object files, ideally we would need to complete linking of them
and fix up all relocations to know what the DataRVA field would end up
being. In practice, the only thing that makes sense for an RVA field
is an ADDR32NB relocation. Thus, find a relocation pointing at this
field, verify that it has the expected type, locate the symbol it
points at, look up the section the symbol points at, and read from the
right offset in that section.

This works both for GNU windres object files (which use one single
.rsrc section, with all relocations against the base of the .rsrc
section, with the original value of the DataRVA field being the
offset of the data from the beginning of the .rsrc section) and
cvtres object files (with two separate .rsrc$01 and .rsrc$02 sections,
and one symbol per data entry, with the original pre-relocated DataRVA
field being set to zero).

Differential Revision: https://reviews.llvm.org/D66820

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370433 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Lower uitofp
Petar Avramovic [Fri, 30 Aug 2019 05:51:12 +0000 (05:51 +0000)]
[MIPS GlobalISel] Lower uitofp

Add custom lowering for G_UITOFP for MIPS32.

Differential Revision: https://reviews.llvm.org/D66930

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370432 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MIPS GlobalISel] Lower fptoui
Petar Avramovic [Fri, 30 Aug 2019 05:44:02 +0000 (05:44 +0000)]
[MIPS GlobalISel] Lower fptoui

Add lower for G_FPTOUI. Algorithm is similar to the SDAG version
in TargetLowering::expandFP_TO_UINT.
Lower G_FPTOUI for MIPS32.

Differential Revision: https://reviews.llvm.org/D66929

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370431 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CodeGen] Fix lowering for returning the result of an extractvalue
Dan Gohman [Fri, 30 Aug 2019 04:33:22 +0000 (04:33 +0000)]
[CodeGen] Fix lowering for returning the result of an extractvalue

When the number of return values exceeds the number of registers available,
SelectionDAGBuilder::visitRet transforms a function's return to use a
pointer to a buffer to hold return values. When the returned value is an
operator such as extractvalue, the value may have a non-zero result number.
Add that number to the indexing when obtaining the values to store.

This fixes https://bugs.llvm.org/show_bug.cgi?id=43132.

Differential Revision: https://reviews.llvm.org/D66978

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370430 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PowerPC][NFC] Use inline Subtarget->isPPC64()
Jinsong Ji [Fri, 30 Aug 2019 03:16:41 +0000 (03:16 +0000)]
[PowerPC][NFC] Use inline Subtarget->isPPC64()

To be consistent with all the other instances.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370428 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PowerPC][NFC] Use -mtriple in RUN line, remove target triple in tls.ll
Jinsong Ji [Fri, 30 Aug 2019 02:57:33 +0000 (02:57 +0000)]
[PowerPC][NFC] Use -mtriple in RUN line, remove target triple in tls.ll

To avoid confusion, especially when -mtriple are also added for PPC32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370427 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PPC32] Emit R_PPC_GOT_TPREL16 instead R_PPC_GOT_TPREL16_LO
Fangrui Song [Fri, 30 Aug 2019 02:20:49 +0000 (02:20 +0000)]
[PPC32] Emit R_PPC_GOT_TPREL16 instead R_PPC_GOT_TPREL16_LO

Unlike ppc64, which has ADDISgotTprelHA+LDgotTprelL pairs,
ppc32 just uses LDgotTprelL32, so it does not make lots of sense to use
_LO without a paired _HA.

Emit R_PPC_GOT_TPREL16 instead R_PPC_GOT_TPREL16_LO to match GCC, and
get better linker relocation check. Note, R_PPC_GOT_TPREL16_{HA,LO}
don't have good linker support:

(a) lld does not support R_PPC_GOT_TPREL16_{HA,LO}.
(b) Top of tree ld.bfd does not support R_PPC_GOT_REL16_HA Initial-Exec -> Local-Exec relaxation:

  // a.o
  addis 3, 3, tsd_tls@got@tprel@ha
  lwz 3, tsd_tls@got@tprel@l(3)
  add 3, 3, tsd_tls@tls
  // b.o
  .section .tdata,"awT"; .globl tsd_tls; tsd_tls:

  // ld/ld-new a.o b.o
  internal error, aborting at ../../bfd/elf32-ppc.c:7952 in ppc_elf_relocate_section

Reviewed By: adalava

Differential Revision: https://reviews.llvm.org/D66925

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370426 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Explicitly list all the always trivially rematerializable instructions.
Craig Topper [Fri, 30 Aug 2019 00:54:36 +0000 (00:54 +0000)]
[X86] Explicitly list all the always trivially rematerializable instructions.

Add a default with an llvm_unreachable for anything we don't expect.

This seems safer that just blindly returning true for anything
missing from the switch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370424 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoDebugInfo: add CodeView register mapping for ARM NT
Saleem Abdulrasool [Fri, 30 Aug 2019 00:16:02 +0000 (00:16 +0000)]
DebugInfo: add CodeView register mapping for ARM NT

Add the core registers and NEON registers mapping to the CodeView
register ID.  This is sufficient to compile a basic C program with debug
info using CodeView debug info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370423 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[WebAssembly] Make __attribute__((used)) not imply export.
Dan Gohman [Thu, 29 Aug 2019 22:40:00 +0000 (22:40 +0000)]
[WebAssembly] Make __attribute__((used)) not imply export.

Add an WASM_SYMBOL_NO_STRIP flag, so that __attribute__((used)) doesn't
need to imply exporting. When targeting Emscripten, have
WASM_SYMBOL_NO_STRIP imply exporting.

Differential Revision: https://reviews.llvm.org/D62542

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370415 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Tests] Precommit a few cases where we're missing oppurtunities for block local simpl...
Philip Reames [Thu, 29 Aug 2019 22:08:17 +0000 (22:08 +0000)]
[Tests] Precommit a few cases where we're missing oppurtunities for block local simplications off assumes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370414 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PowerPC] Support extended mnemonics mffprwz etc.
Jinsong Ji [Thu, 29 Aug 2019 21:53:59 +0000 (21:53 +0000)]
[PowerPC] Support extended mnemonics mffprwz etc.

Summary:
Reported in https://github.com/opencv/opencv/issues/15413.

We have serveral extended mnemonics for Move To/From Vector-Scalar Register Instructions
eg: mffprd,mtfprd etc.

We only support one of them, this patch add the others.

Reviewers: nemanjai, steven.zhang, hfinkel, #powerpc

Reviewed By: hfinkel

Subscribers: wuzish, qcolombet, hiraditya, kbarton, MaskRay, shchenz, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66963

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370411 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][GlobalISel] Select arithmetic extended register patterns
Jessica Paquette [Thu, 29 Aug 2019 21:53:58 +0000 (21:53 +0000)]
[AArch64][GlobalISel] Select arithmetic extended register patterns

This teaches GISel to select patterns which fold an extend plus optional shift
into the addressing mode. In particular, adds and subs.

Factor out the arith extended register ComplexPatterns in AArch64InstrFormats.td
and create GISel equivalents.

Add some equivalent functions to the ones in AArch64ISelDAGToDAG:

- `selectArithExtendedRegister`
- `narrowExtendRegIfNeeded`
- `getExtendTypeForInst`

`getExtendTypeForInst` includes the checks for loads and stores. This will be
used for WRO addressing modes in loads + stores.

Teach selectCopy to properly handle subregister copies on the same bank in
order to support `narrowExtendRegIfNeeded`. The extended register must be a
GPR32, so we need to support same-bank subregister copies.

Fix a bug in getSubRegForClass which would cause registers on things like
GPR32common to end up getting ssub. Just change the check to look for FPR32
rather than GPR32.

For tests:

- Add select-arith-extended-reg.mir
- Update addsub_ext.ll to include GlobalISel checks

Differential Revision: https://reviews.llvm.org/D66835

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370410 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Don't emit unreachable stack adjustments
Reid Kleckner [Thu, 29 Aug 2019 21:24:41 +0000 (21:24 +0000)]
[X86] Don't emit unreachable stack adjustments

Summary:
This is a minor improvement on our past attempts to do this. Fixes
PR43155.

Reviewers: hans

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66905

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370409 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAllow '@' to appear in x86 mingw symbols
Reid Kleckner [Thu, 29 Aug 2019 21:15:02 +0000 (21:15 +0000)]
Allow '@' to appear in x86 mingw symbols

Summary:
There is no reason to differ in assembler behavior here between -msvc
and -gnu targets. Without this setting, the text after the '@' is
interpreted as a symbol variable, like foo@IMGREL.

Reviewers: mstorsjo

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66974

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370408 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] add possible bswap as widening shuffle test; NFC
Sanjay Patel [Thu, 29 Aug 2019 20:57:50 +0000 (20:57 +0000)]
[InstCombine] add possible bswap as widening shuffle test; NFC

Goes with the proposal in D66965.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370407 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix the build for MSVC builds using M_PI
Reid Kleckner [Thu, 29 Aug 2019 20:32:53 +0000 (20:32 +0000)]
Fix the build for MSVC builds using M_PI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370405 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][SSE] combinePMULDQ - pmuldq(x, 0) -> zero vector (PR43159)
Simon Pilgrim [Thu, 29 Aug 2019 20:22:08 +0000 (20:22 +0000)]
[X86][SSE] combinePMULDQ - pmuldq(x, 0) -> zero vector (PR43159)

ISD::isBuildVectorAllZeros permits undef elements to be present, which means we can't return it as a zero vector. PMULDQ/PMULUDQ is an extending multiply so a multiply by zero of the lower 32-bits should result in a zero 64-bit element.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370404 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ASan] Version mismatch check follow-up
Julian Lettner [Thu, 29 Aug 2019 20:20:05 +0000 (20:20 +0000)]
[ASan] Version mismatch check follow-up

Follow-up for:
[ASan] Make insertion of version mismatch guard configurable
3ae9b9d5e40d1d9bdea1fd8e6fca322df920754a

This tiny change makes sure that this test passes on our internal bots
as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370403 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAMDGPU/GlobalISel: Legalize sin/cos
Matt Arsenault [Thu, 29 Aug 2019 20:06:48 +0000 (20:06 +0000)]
AMDGPU/GlobalISel: Legalize sin/cos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370402 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] reduce duplicated code; NFC
Sanjay Patel [Thu, 29 Aug 2019 19:36:18 +0000 (19:36 +0000)]
[InstCombine] reduce duplicated code; NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370399 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert [MBP] Disable aggressive loop rotate in plain mode
Jordan Rupprecht [Thu, 29 Aug 2019 19:03:58 +0000 (19:03 +0000)]
Revert [MBP] Disable aggressive loop rotate in plain mode

This reverts r369664 (git commit 51f48295cbe8fa3a44db263b528dd9f7bae7bf9a)

It causes many benchmark regressions, internally and in llvm's benchmark suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370398 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoRevert enabling MemorySSA.
Alina Sbirlea [Thu, 29 Aug 2019 19:01:23 +0000 (19:01 +0000)]
Revert enabling MemorySSA.

Breaks sanitizers bots.

Differential Revision: https://reviews.llvm.org/D58311

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370397 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove what little support we had for MPX
Craig Topper [Thu, 29 Aug 2019 18:09:02 +0000 (18:09 +0000)]
[X86] Remove what little support we had for MPX

-Deprecate -mmpx and -mno-mpx command line options
-Remove CPUID detection of mpx for -march=native
-Remove MPX from all CPUs
-Remove MPX preprocessor define

I've left the "mpx" string in the backend so we don't fail on old IR, but its not connected to anything.

gcc has also deprecated these command line options. https://www.phoronix.com/scan.php?page=news_item&px=GCC-Patch-To-Drop-MPX

Differential Revision: https://reviews.llvm.org/D66669

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370393 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoGlobalISel: Don't compute known bits for non-integral GEP
Matt Arsenault [Thu, 29 Aug 2019 17:55:05 +0000 (17:55 +0000)]
GlobalISel: Don't compute known bits for non-integral GEP

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370392 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LoopUnrollAndJam] Use Lazy strategy for DTU.
Florian Hahn [Thu, 29 Aug 2019 17:47:58 +0000 (17:47 +0000)]
[LoopUnrollAndJam] Use Lazy strategy for DTU.

We can also apply the earlier updates to the lazy DTU, instead of
applying them directly.

Reviewers: kuhar, brzycki, asbirlea, SjoerdMeijer

Reviewed By: brzycki, asbirlea, SjoerdMeijer

Differential Revision: https://reviews.llvm.org/D66918

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370391 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoGlobalISel: Add maskedValueIsZero and signBitIsZero to known bits
Matt Arsenault [Thu, 29 Aug 2019 17:24:36 +0000 (17:24 +0000)]
GlobalISel: Add maskedValueIsZero and signBitIsZero to known bits

I dropped the DemandedElts since it seems to be missing from some of
the new interfaces, but not others.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370389 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoGlobalISel: Add known bits to InstructionSelector
Matt Arsenault [Thu, 29 Aug 2019 17:24:32 +0000 (17:24 +0000)]
GlobalISel: Add known bits to InstructionSelector

AMDGPU uses this for some addressing mode selection patterns. The
analysis run itself doesn't do anything so it seems easier to just
always require this than adding a way to opt in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370388 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[MemorySSA & LoopPassManager] Enable MemorySSA as loop dependency. Update tests.
Alina Sbirlea [Thu, 29 Aug 2019 17:08:13 +0000 (17:08 +0000)]
[MemorySSA & LoopPassManager] Enable MemorySSA as loop dependency. Update tests.

Summary:
I'm not planning to check this in at the moment, but feedback is very welcome, in particular how this affects performance.
The feedback obtains here will guide the next steps towards enabling this.

This patch enables the use of MemorySSA in the loop pass manager.

Passes that currently use MemorySSA:
 - EarlyCSE
Passes that use MemorySSA after this patch:
 - EarlyCSE
 - LICM
 - SimpleLoopUnswitch
Loop passes that update MemorySSA (and do not use it yet, but could use it after this patch):
 - LoopInstSimplify
 - LoopSimplifyCFG
 - LoopUnswitch
 - LoopRotate
 - LoopSimplify
 - LCSSA
Loop passes that do *not* update MemorySSA:
 - IndVarSimplify
 - LoopDelete
 - LoopIdiom
 - LoopSink
 - LoopUnroll
 - LoopInterchange
 - LoopUnrollAndJam
 - LoopVectorize
 - LoopReroll
 - IRCE

Reviewers: chandlerc, george.burgess.iv, davide, sanjoy, gberry

Subscribers: jlebar, Prazek, dmgreen, jdoerfert, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D58311

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370384 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel][AArch64] Select llvm.aarch64.stxr* intrinsics.
Jessica Paquette [Thu, 29 Aug 2019 16:55:55 +0000 (16:55 +0000)]
[GlobalISel][AArch64] Select llvm.aarch64.stxr* intrinsics.

Add a GISelPredicateCode to the stxr_* PatFrags in AArch64InstrAtomics.td.

This allows us to select these intrinsics.

Differential Revision: https://reviews.llvm.org/D65779

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370382 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] add tests for bswap disguised as shuffle; NFC
Sanjay Patel [Thu, 29 Aug 2019 16:48:00 +0000 (16:48 +0000)]
[InstCombine] add tests for bswap disguised as shuffle; NFC

Somewhat motivating case In PR43146:
https://bugs.llvm.org/show_bug.cgi?id=43146

But that's a lot more complicated.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370381 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[GlobalISel][AArch64] Use a GISelPredicateCode to select llvm.aarch64.stlxr.*
Jessica Paquette [Thu, 29 Aug 2019 16:45:19 +0000 (16:45 +0000)]
[GlobalISel][AArch64] Use a GISelPredicateCode to select llvm.aarch64.stlxr.*

Remove manual selection code for this intrinsic and use a GISelPredicateCode
instead.

This allows us to fully select this intrinsic without any tricky custom C++
matching.

Differential Revision: https://reviews.llvm.org/D65780

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370380 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][GlobalISel] Select @llvm.aarch64.ldxr.* intrinsics
Jessica Paquette [Thu, 29 Aug 2019 16:33:01 +0000 (16:33 +0000)]
[AArch64][GlobalISel] Select @llvm.aarch64.ldxr.* intrinsics

Same thing as D66897, but for ldxr.* instead. Add a GISelPredicateCode to the
ldxr_* definitions, which allows us to import them.

Add select-ldxr-intrin.mir, and update arm64-ldxr-stxr.ll.

Differential Revision: https://reviews.llvm.org/D66898

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370378 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[AArch64][GlobalISel] Select @llvm.aarch64.ldaxr.* intrinsics
Jessica Paquette [Thu, 29 Aug 2019 16:16:38 +0000 (16:16 +0000)]
[AArch64][GlobalISel] Select @llvm.aarch64.ldaxr.* intrinsics

Add a GISelPredicateCode to ldaxr_*. This allows us to import the patterns for
@llvm.aarch64.ldaxr.*, and thus select them.

Add `isLoadStoreOfNumBytes` for the GISelPredicateCode, since each of these
intrinsics involves the same check.

Add select-ldaxr-intrin.mir, and update arm64-ldxr-stxr.ll.

Differential Revision: https://reviews.llvm.org/D66897

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370377 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SimplifyCFG] Skip sinking common lifetime markers of `alloca`.
Michael Liao [Thu, 29 Aug 2019 16:12:05 +0000 (16:12 +0000)]
[SimplifyCFG] Skip sinking common lifetime markers of `alloca`.

Summary:
- Similar to the workaround in fix of PR30188, skip sinking common
  lifetime markers of `alloca`. They are mostly left there after
  inlining functions in branches.

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66950

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370376 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[PowerPC][NFC] Update fp-int-conversions-direct-moves.ll using script
Jinsong Ji [Thu, 29 Aug 2019 15:38:02 +0000 (15:38 +0000)]
[PowerPC][NFC] Update fp-int-conversions-direct-moves.ll using script

Also add -ppc-asm-full-reg-names,-ppc-vsr-nums-as-vr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370375 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[NFC][SimplifyCFG] 'Safely extract low bits' pattern will also benefit from -phi...
Roman Lebedev [Thu, 29 Aug 2019 14:46:49 +0000 (14:46 +0000)]
[NFC][SimplifyCFG] 'Safely extract low bits' pattern will also benefit from -phi-node-folding-threshold=3

This is the naive implementation of x86 BZHI/BEXTR instruction:
it takes input and bit count, and extracts low nbits up to bit width.
I.e. unlike shift it does not have any UB when nbits >= bitwidth.
Which means we don't need a while PHI here, simple select will do.
And if it's a select, it should then be trivial to fix codegen
to select it to BEXTR/BZHI.

See https://bugs.llvm.org/show_bug.cgi?id=34704

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370369 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombine] Fix shadow variable warnings. NFCI.
Simon Pilgrim [Thu, 29 Aug 2019 14:34:07 +0000 (14:34 +0000)]
[DAGCombine] Fix shadow variable warnings. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370365 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoDWARFDebugLoc: Make parsing and error reporting more robust
Pavel Labath [Thu, 29 Aug 2019 14:26:05 +0000 (14:26 +0000)]
DWARFDebugLoc: Make parsing and error reporting more robust

Summary:
While examining this class for possible use in lldb, I noticed two
things:
- it spits out parsing errors directly to stderr
- the loclists parser can incorrectly return valid location lists when
  parsing malformed (truncated) data

I improve the stderr situation by making the parseOneLocationList
functions return Expected<T>s. The errors are still dumped to stderr by
their callers, so this is only a partial fix, but it is enough for my
use case, as I intend to parse the locations lists one by one.

I fix the behavior in the truncated scenario by using the newly
introduced DataExtractor Cursor API.

I also add tests for handling the error cases, as they currently have no
coverage.

Reviewers: dblaikie, JDevlieghere, probinson

Subscribers: lldb-commits, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D63591

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370363 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RISCV] Fix callee-saved-gprs.ll test ABIs
Luis Marques [Thu, 29 Aug 2019 14:05:59 +0000 (14:05 +0000)]
[RISCV] Fix callee-saved-gprs.ll test ABIs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370359 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAllow replaceAndRecursivelySimplify to list unsimplified visitees.
Joerg Sonnenberger [Thu, 29 Aug 2019 13:22:30 +0000 (13:22 +0000)]
Allow replaceAndRecursivelySimplify to list unsimplified visitees.

This is part of D65280 and split it to avoid ABI changes on the 9.0
release branch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370355 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[mips] Inline emitStoreWithSymOffset and emitLoadWithSymOffset methods. NFC
Simon Atanasyan [Thu, 29 Aug 2019 13:19:50 +0000 (13:19 +0000)]
[mips] Inline emitStoreWithSymOffset and emitLoadWithSymOffset methods. NFC

Both methods `MipsTargetStreamer::emitStoreWithSymOffset` and
`MipsTargetStreamer::emitLoadWithSymOffset` are almost the same and
differ argument names only. These methods are used in the single place
so it's better to inline their code and remove original methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370354 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[mips] Fix expanding `lw/sw $reg1, symbol($reg2)` instruction
Simon Atanasyan [Thu, 29 Aug 2019 13:19:38 +0000 (13:19 +0000)]
[mips] Fix expanding `lw/sw $reg1, symbol($reg2)` instruction

When a "base" in the `lw/sw $reg1, symbol($reg2)` instruction is
a register and generated code is position independent, backend
does not add the "base" value to the symbol address.
```
lw     $reg1, %got(symbol)($gp)
lw/sw  $reg1, 0($reg1)
```

This patch fixes the bug and adds the missed `addu` instruction by
passing `BaseReg` into the `loadAndAddSymbolAddress` routine and handles
the case when the `BaseReg` is the zero register to escape redundant
`move reg, reg` instruction:
```
lw     $reg1, %got(symbol)($gp)
addu   $reg1, $reg1, $reg2
lw/sw  $reg1, 0($reg1)
```

Differential Revision: https://reviews.llvm.org/D66894

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370353 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstSimplify] Drop leftover "division-by-zero guard" around `@llvm.umul.with.overflo...
Roman Lebedev [Thu, 29 Aug 2019 12:48:04 +0000 (12:48 +0000)]
[InstSimplify] Drop leftover "division-by-zero guard" around `@llvm.umul.with.overflow` inverted overflow bit

Summary:
Now that with D65143/D65144 we've produce `@llvm.umul.with.overflow`,
and with D65147 we've flattened the CFG, we now can see that
the guard may have been there to prevent division by zero is redundant.
We can simply drop it:
```
----------------------------------------
Name: no overflow or zero
  %iszero = icmp eq i4 %y, 0
  %umul = smul_overflow i4 %x, %y
  %umul.ov = extractvalue {i4, i1} %umul, 1
  %umul.ov.not = xor %umul.ov, -1
  %retval.0 = or i1 %iszero, %umul.ov.not
  ret i1 %retval.0
=>
  %iszero = icmp eq i4 %y, 0
  %umul = smul_overflow i4 %x, %y
  %umul.ov = extractvalue {i4, i1} %umul, 1
  %umul.ov.not = xor %umul.ov, -1
  %retval.0 = or i1 %iszero, %umul.ov.not
  ret i1 %umul.ov.not

Done: 1
Optimization is correct!
```
Note that this is inverted from what we have in a previous patch,
here we are looking for the inverted overflow bit.
And that inversion is kinda problematic - given this particular
pattern we neither hoist that `not` closer to `ret` (then the pattern
would have been identical to the one without inversion,
and would have been handled by the previous patch), neither
do the opposite transform. But regardless, we should handle this too.
I've filled [[ https://bugs.llvm.org/show_bug.cgi?id=42720 | PR42720 ]].

Reviewers: nikic, spatel, xbolva00, RKSimon

Reviewed By: spatel

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65151

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370351 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstSimplify] Drop leftover "division-by-zero guard" around `@llvm.umul.with.overflo...
Roman Lebedev [Thu, 29 Aug 2019 12:47:50 +0000 (12:47 +0000)]
[InstSimplify] Drop leftover "division-by-zero guard" around `@llvm.umul.with.overflow` overflow bit

Summary:
Now that with D65143/D65144 we've produce `@llvm.umul.with.overflow`,
and with D65147 we've flattened the CFG, we now can see that
the guard may have been there to prevent division by zero is redundant.
We can simply drop it:
```
----------------------------------------
Name: no overflow and not zero
  %iszero = icmp ne i4 %y, 0
  %umul = umul_overflow i4 %x, %y
  %umul.ov = extractvalue {i4, i1} %umul, 1
  %retval.0 = and i1 %iszero, %umul.ov
  ret i1 %retval.0
=>
  %iszero = icmp ne i4 %y, 0
  %umul = umul_overflow i4 %x, %y
  %umul.ov = extractvalue {i4, i1} %umul, 1
  %retval.0 = and i1 %iszero, %umul.ov
  ret %umul.ov

Done: 1
Optimization is correct!
```

Reviewers: nikic, spatel, xbolva00

Reviewed By: spatel

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65150

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370350 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[SimplifyCFG] FoldTwoEntryPHINode(): don't bailout on i1 PHI's if we can hoist a...
Roman Lebedev [Thu, 29 Aug 2019 12:47:34 +0000 (12:47 +0000)]
[SimplifyCFG] FoldTwoEntryPHINode(): don't bailout on i1 PHI's if we can hoist a 'not' from incoming values

Summary:
As it can be seen in the tests in D65143/D65144, even though we have formed an '@llvm.umul.with.overflow'
and got rid of potential for division-by-zero, the control flow remains, we still have that branch.

We have this condition:
```
  // Don't fold i1 branches on PHIs which contain binary operators
  // These can often be turned into switches and other things.
  if (PN->getType()->isIntegerTy(1) &&
      (isa<BinaryOperator>(PN->getIncomingValue(0)) ||
       isa<BinaryOperator>(PN->getIncomingValue(1)) ||
       isa<BinaryOperator>(IfCond)))
    return false;
```
which was added back in rL121764 to help with `select` formation i think?

That check prevents us to flatten the CFG here, even though we know
we no longer need that guard and will be able to drop everything
but the '@llvm.umul.with.overflow' + `not`.

As it can be seen from tests, we end here because the `not` is being
sinked into the PHI's incoming values by InstCombine,
so we can't workaround this by hoisting it to after PHI.

Thus i suggest that we relax that check to not bailout if we'd get to hoist the `not`.

Reviewers: craig.topper, spatel, fhahn, nikic

Reviewed By: spatel

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65147

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370349 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Fold '((%x * %y) u/ %x) != %y' to '@llvm.umul.with.overflow' + overflow...
Roman Lebedev [Thu, 29 Aug 2019 12:47:20 +0000 (12:47 +0000)]
[InstCombine] Fold '((%x * %y) u/ %x) != %y' to '@llvm.umul.with.overflow' + overflow bit extraction

Summary:
`((%x * %y) u/ %x) != %y` is one of (3?) common ways to check that
some unsigned multiplication (will not) overflow.
Currently, we don't catch it. We could:
```
$ /repositories/alive2/build-Clang-unknown/alive -root-only ~/llvm-patch1.ll
Processing /home/lebedevri/llvm-patch1.ll..

----------------------------------------
Name: no overflow
  %o0 = mul i4 %y, %x
  %o1 = udiv i4 %o0, %x
  %r = icmp ne i4 %o1, %y
  ret i1 %r
=>
  %n0 = umul_overflow i4 %x, %y
  %o0 = extractvalue {i4, i1} %n0, 0
  %o1 = udiv %o0, %x
  %r = extractvalue {i4, i1} %n0, 1
  ret %r

Done: 1
Optimization is correct!

----------------------------------------
Name: no overflow
  %o0 = mul i4 %y, %x
  %o1 = udiv i4 %o0, %x
  %r = icmp eq i4 %o1, %y
  ret i1 %r
=>
  %n0 = umul_overflow i4 %x, %y
  %o0 = extractvalue {i4, i1} %n0, 0
  %o1 = udiv %o0, %x
  %n1 = extractvalue {i4, i1} %n0, 1
  %r = xor %n1, -1
  ret i1 %r

Done: 1
Optimization is correct!

```

Reviewers: nikic, spatel, efriedma, xbolva00, RKSimon

Reviewed By: nikic

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65144

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370348 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Fold '(-1 u/ %x) u< %y' to '@llvm.umul.with.overflow' + overflow bit...
Roman Lebedev [Thu, 29 Aug 2019 12:47:08 +0000 (12:47 +0000)]
[InstCombine] Fold '(-1 u/ %x) u< %y' to '@llvm.umul.with.overflow' + overflow bit extraction

Summary:
`(-1 u/ %x) u< %y` is one of (3?) common ways to check that
some unsigned multiplication (will not) overflow.
Currently, we don't catch it. We could:
```
----------------------------------------
Name: no overflow
  %o0 = udiv i4 -1, %x
  %r = icmp ult i4 %o0, %y
=>
  %o0 = udiv i4 -1, %x
  %n0 = umul_overflow i4 %x, %y
  %r = extractvalue {i4, i1} %n0, 1

Done: 1
Optimization is correct!

----------------------------------------
Name: no overflow, swapped
  %o0 = udiv i4 -1, %x
  %r = icmp ugt i4 %y, %o0
=>
  %o0 = udiv i4 -1, %x
  %n0 = umul_overflow i4 %x, %y
  %r = extractvalue {i4, i1} %n0, 1

Done: 1
Optimization is correct!

----------------------------------------
Name: overflow
  %o0 = udiv i4 -1, %x
  %r = icmp uge i4 %o0, %y
=>
  %o0 = udiv i4 -1, %x
  %n0 = umul_overflow i4 %x, %y
  %n1 = extractvalue {i4, i1} %n0, 1
  %r = xor %n1, -1

Done: 1
Optimization is correct!

----------------------------------------
Name: overflow
  %o0 = udiv i4 -1, %x
  %r = icmp ule i4 %y, %o0
=>
  %o0 = udiv i4 -1, %x
  %n0 = umul_overflow i4 %x, %y
  %n1 = extractvalue {i4, i1} %n0, 1
  %r = xor %n1, -1

Done: 1
Optimization is correct!
```

As it can be observed from tests, while simply forming the `@llvm.umul.with.overflow`
is easy, if we were looking for the inverted answer, then more work needs to be done
to cleanup the now-pointless control-flow that was guarding against division-by-zero.
This is being addressed in follow-up patches.

Reviewers: nikic, spatel, efriedma, xbolva00, RKSimon

Reviewed By: nikic, xbolva00

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D65143

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370347 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[CostModel] Model all `extractvalue`s as free.
Roman Lebedev [Thu, 29 Aug 2019 11:50:30 +0000 (11:50 +0000)]
[CostModel] Model all `extractvalue`s as free.

Summary:
As disscussed in https://reviews.llvm.org/D65148#1606412,
`extractvalue` don't actually generate any code,
so we should treat them as free.

Reviewers: craig.topper, RKSimon, jnspaulsson, greened, asb, t.p.northover, jmolloy, dmgreen

Reviewed By: jmolloy

Subscribers: javed.absar, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66098

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370339 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] LiveDebugValues: correctly discriminate kinds of variable locations
Jeremy Morse [Thu, 29 Aug 2019 11:20:54 +0000 (11:20 +0000)]
[DebugInfo] LiveDebugValues: correctly discriminate kinds of variable locations

The missing line added by this patch ensures that only spilt variable
locations are candidates for being restored from the stack. Otherwise,
register or constant-value information can be interpreted as a spill
location, through a union.

The added regression test replicates a scenario where this occurs: the
stack load from [rsp] causes the register-location DBG_VALUE to be
"restored" to rsi, when it should be left alone. See PR43058 for details.

Un x-fail a test that was suffering from this from a previous patch.

Differential Revision: https://reviews.llvm.org/D66895

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370334 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix signed/unsigned comparison warning. NFCI.
Simon Pilgrim [Thu, 29 Aug 2019 11:18:53 +0000 (11:18 +0000)]
Fix signed/unsigned comparison warning. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370333 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix shadow variable warning. NFCI.
Simon Pilgrim [Thu, 29 Aug 2019 11:16:32 +0000 (11:16 +0000)]
Fix shadow variable warning. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370332 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[yaml2obj] - Allow placing local symbols after globals.
George Rimar [Thu, 29 Aug 2019 10:58:47 +0000 (10:58 +0000)]
[yaml2obj] - Allow placing local symbols after globals.

This allows us to produce broken binaries with local
symbols placed after global in '.dynsym'/'.symtab'

Also, simplifies the code.

Differential revision: https://reviews.llvm.org/D66799

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370331 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj/llvm-readelf] - Report a proper warning when dumping a broken dynamic...
George Rimar [Thu, 29 Aug 2019 10:55:57 +0000 (10:55 +0000)]
[llvm-readobj/llvm-readelf] - Report a proper warning when dumping a broken dynamic relocation.

When we have a dynamic relocation with a broken symbol's st_name,
tools report a useless error: "Invalid data was encountered while parsing the file".

After this change we report a warning + "<corrupt>" as a symbol name.

Differential revision: https://reviews.llvm.org/D66734

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370330 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] MVE Masked loads and stores
David Green [Thu, 29 Aug 2019 10:54:35 +0000 (10:54 +0000)]
[ARM] MVE Masked loads and stores

Masked loads and store fit naturally with MVE, the instructions being easily
predicated. This adds lowering for the simple cases of masked loads and stores.
It does not yet deal with widening/narrowing or pre/post inc.

The llvm masked load intrinsic will accept a "passthru" value, dictating the
values used for the zero masked lanes. In MVE the instructions write 0 to the
zero predicated lanes, so we need to match a passthru that isn't 0 (or undef)
with a select instruction to pull in the correct data after the load.

We also need to do something with unaligned loads/stores. Currently this uses a
similar method used in big endian, using an VLDRB.8 (and potentially a VREV in
BE). This does mean that the predicate mask is converted from, for example, a
v4i1 to a v16i1. The VLDR instructions are defined as using the first bit of
the relevant mask lane, so this could potentially load different results if the
predicate is little odd. As the input is a v4i1 however, I believe this is OK
and all the bits required should be set in the predicate, making the VLDRB.8
load the same data.

Differential Revision: https://reviews.llvm.org/D66534

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370329 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DebugInfo] LiveDebugValues should always revisit backedges if it skips them
Jeremy Morse [Thu, 29 Aug 2019 10:53:29 +0000 (10:53 +0000)]
[DebugInfo] LiveDebugValues should always revisit backedges if it skips them

The "join" method in LiveDebugValues does not attempt to join unseen
predecessor blocks if their out-locations aren't yet initialized, instead
the block should be re-visited later to see if any locations have changed
validity. However, because the set of blocks were all being "process"'d
once before "join" saw them, that logic in "join" was actually ignoring
legitimate out-locations on the first pass through. This meant that some
invalidated locations were not removed from the head of loops, allowing
illegal locations to persist.

Fix this by removing the run of "process" before the main join/process loop
in ExtendRanges. Now the unseen predecessors that "join" skips truly are
uninitialized, and we come back to the block at a later time to re-run
"join", see the @baz function added.

This also fixes another fault where stack/register transfers in the entry
block (or any other before-any-loop-block) had their tranfers initially
ignored, and were then never revisited. The MIR test added tests for this
behaviour.

XFail a test that exposes another bug; a fix for this is coming in D66895.

Differential Revision: https://reviews.llvm.org/D66663

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370328 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86][CodeGen][NFC] Delay `combineIncDecVector()` from DAGCombine to X86DAGToDAGISel
Roman Lebedev [Thu, 29 Aug 2019 10:50:09 +0000 (10:50 +0000)]
[X86][CodeGen][NFC] Delay `combineIncDecVector()` from DAGCombine to X86DAGToDAGISel

Summary:
We were previously doing it in DAGCombine.
But we also want to do `sub %x, C` -> `add %x, (sub 0, C)` for vectors in DAGCombine.
So if we had `sub %x, -1`, we'll transform it to `add %x, 1`,
which `combineIncDecVector()` will immediately transform back into `sub %x, -1`,
and here we go again...

I've marked this as NFC since not a single test changes,
but since that 'changes' DAGCombine, probably this isn't fully NFC.

Reviewers: RKSimon, craig.topper, spatel

Reviewed By: craig.topper

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D62327

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370327 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[DAGCombiner] (insert_vector_elt (vector_shuffle X, Y), (extract_vector_elt X, N...
Amaury Sechet [Thu, 29 Aug 2019 10:35:51 +0000 (10:35 +0000)]
[DAGCombiner] (insert_vector_elt (vector_shuffle X, Y), (extract_vector_elt X, N), IdxC) -> (vector_shuffle X, Y)

Summary: This is beneficial when the shuffle is only used once and end up being generated in a few places when some node is combined into a shuffle.

Reviewers: craig.topper, efriedma, RKSimon, lebedev.ri

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66718

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370326 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ARM] Masked load and store and predicate tests. NFC
David Green [Thu, 29 Aug 2019 10:32:12 +0000 (10:32 +0000)]
[ARM] Masked load and store and predicate tests. NFC

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370325 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[InstCombine] Shift amount reassociation in bittest: trunc-of-lshr (PR42399)
Roman Lebedev [Thu, 29 Aug 2019 10:26:23 +0000 (10:26 +0000)]
[InstCombine] Shift amount reassociation in bittest: trunc-of-lshr (PR42399)

Summary:
Finally, the fold i was looking forward to :)

The legality check is muddy, i doubt  i've groked the full generalization,
but it handles all the cases i care about, and can come up with:
https://rise4fun.com/Alive/26j

I.e. we can perform the fold if **any** of the following is true:
* The shift amount is either zero or one less than widest bitwidth
* Either of the values being shifted has at most lowest bit set
* The value that is being shifted by `shl` (which is not truncated) should have no less leading zeros than the total shift amount;
* The value that is being shifted by `lshr` (which **is** truncated) should have no less leading zeros than the widest bit width minus total shift amount minus one

I strongly suspect there is some better generalization, but i'm not aware of it as of right now.
For now i also avoided using actual `computeKnownBits()`, but restricted it to constants.

Reviewers: spatel, nikic, xbolva00

Reviewed By: spatel

Subscribers: hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66383

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370324 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoLegalizeSetCCCondCode - Reduce scope of NeedSwap to fix cppcheck warning. NFCI.
Simon Pilgrim [Thu, 29 Aug 2019 10:11:34 +0000 (10:11 +0000)]
LegalizeSetCCCondCode - Reduce scope of NeedSwap to fix cppcheck warning. NFCI.

No need for this to be defined outside the only switch case its used in.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370320 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix variable set but no used warnings on NDEBUG builds. NFCI.
Simon Pilgrim [Thu, 29 Aug 2019 10:08:45 +0000 (10:08 +0000)]
Fix variable set but no used warnings on NDEBUG builds. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370319 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoFix variable set but no used warning on NDEBUG builds. NFCI.
Simon Pilgrim [Thu, 29 Aug 2019 09:58:47 +0000 (09:58 +0000)]
Fix variable set but no used warning on NDEBUG builds. NFCI.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370317 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[COFF] Add a ResourceSectionRef method for getting the data entry, print it in llvm...
Martin Storsjo [Thu, 29 Aug 2019 09:00:14 +0000 (09:00 +0000)]
[COFF] Add a ResourceSectionRef method for getting the data entry, print it in llvm-readobj

Differential Revision: https://reviews.llvm.org/D66819

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370311 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[COFF] Add a bounds checking helper for iterating a coff_resource_dir_table
Martin Storsjo [Thu, 29 Aug 2019 08:59:56 +0000 (08:59 +0000)]
[COFF] Add a bounds checking helper for iterating a coff_resource_dir_table

Instead of blindly incrementing pointers in llvm-readobj, use this
helper, which does bounds checking against the available section
data.

Differential Revision: https://reviews.llvm.org/D66818

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370310 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[COFF] Fix error handling in ResourceSectionRef
Martin Storsjo [Thu, 29 Aug 2019 08:59:41 +0000 (08:59 +0000)]
[COFF] Fix error handling in ResourceSectionRef

Previously, the expression (Reader.readFoo()) was expanded twice,
triggering asserts as one of the Error types ends up not checked
(and as it was expanded twice, the method would end up called twice
if it failed first).

Differential Revision: https://reviews.llvm.org/D66817

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370309 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj] Print the resource type textually for .res files
Martin Storsjo [Thu, 29 Aug 2019 08:59:31 +0000 (08:59 +0000)]
[llvm-readobj] Print the resource type textually for .res files

This already is done when dumping resources from coff objects.

Differential Revision: https://reviews.llvm.org/D66816

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370308 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[llvm-readobj] Remove a leftover string trim operation. NFC.
Martin Storsjo [Thu, 29 Aug 2019 08:59:05 +0000 (08:59 +0000)]
[llvm-readobj] Remove a leftover string trim operation. NFC.

This became unnecessary in SVN r359153.

Differential Revision: https://reviews.llvm.org/D66815

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370307 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove isel patterns with X86VBroadcast+scalar_to_vector+load.
Craig Topper [Thu, 29 Aug 2019 06:36:16 +0000 (06:36 +0000)]
[X86] Remove isel patterns with X86VBroadcast+scalar_to_vector+load.

The DAG should have these as X86VBroadcast+load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370299 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Remove some unneeded X86VBroadcast isel patterns that have larger than 128...
Craig Topper [Thu, 29 Aug 2019 06:02:11 +0000 (06:02 +0000)]
[X86] Remove some unneeded X86VBroadcast isel patterns that have larger than 128 bit input types.

We should always be shrinking the input to 128 bits or smaller
when the node is created.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370296 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Deduce "noalias" attribute
Hideto Ueno [Thu, 29 Aug 2019 05:52:00 +0000 (05:52 +0000)]
[Attributor] Deduce "noalias" attribute

Summary:
This patch adds very basic deduction for noalias.

Reviewers: jdoerfert, sstefan1

Reviewed By: jdoerfert

Tags: LLVM

Differential Revision: https://reviews.llvm.org/D66207

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370295 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Add a DAG combine to combine INSERTPS and VBROADCAST of a scalar load. Remove...
Craig Topper [Thu, 29 Aug 2019 05:48:48 +0000 (05:48 +0000)]
[X86] Add a DAG combine to combine INSERTPS and VBROADCAST of a scalar load. Remove corresponding isel patterns.

We had an isel pattern to perform this, but its better to
do it in DAG combine as a simplification. This also fixes the lack
of patterns for AVX512 targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370294 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Make inline assembly 'x' and 'v' constraints work for f128.
Craig Topper [Thu, 29 Aug 2019 05:13:56 +0000 (05:13 +0000)]
[X86] Make inline assembly 'x' and 'v' constraints work for f128.

Including a type legalizer fix to make bitcast operand promotion
work correctly when getSoftenedFloat returns f128 instead of i128.

Fixes PR43157

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370293 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[LoopUnroll] Use Lazy strategy for DTU used for MergeBlockIntoPredecessor.
Florian Hahn [Thu, 29 Aug 2019 04:26:29 +0000 (04:26 +0000)]
[LoopUnroll] Use Lazy strategy for DTU used for MergeBlockIntoPredecessor.

We do not access the DT in the loop, so we do not have to apply updates
eagerly. We can apply them lazyly and flush them after we are done
merging blocks.

As follow-up work, we might be able to use the DTU above as well,
instead of manually updating the DT.

This brings the example from PR43134 from ~100s to ~4s for a relase +
assertions build on my machine.

Reviewers: efriedma, kuhar, asbirlea, brzycki

Reviewed By: kuhar, brzycki

Differential Revision: https://reviews.llvm.org/D66911

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370292 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[ObjectYAML] Fix lifetime issue in dumpDebugLines
Vitaly Buka [Thu, 29 Aug 2019 02:36:48 +0000 (02:36 +0000)]
[ObjectYAML] Fix lifetime issue in dumpDebugLines

Subscribers: llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D66901

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370289 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Improve messages in iteration verify mode
Johannes Doerfert [Thu, 29 Aug 2019 01:29:44 +0000 (01:29 +0000)]
[Attributor] Improve messages in iteration verify mode

When we now verify the iteration count we will see the actual count
and the expected count before the assertion is triggered.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370285 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor][NFC] Add const to map key
Johannes Doerfert [Thu, 29 Aug 2019 01:28:30 +0000 (01:28 +0000)]
[Attributor][NFC] Add const to map key

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370284 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor][Fix] Indicate change correctly
Johannes Doerfert [Thu, 29 Aug 2019 01:26:58 +0000 (01:26 +0000)]
[Attributor][Fix] Indicate change correctly

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370283 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[Attributor] Fix typo
Johannes Doerfert [Thu, 29 Aug 2019 01:26:09 +0000 (01:26 +0000)]
[Attributor] Fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370282 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAMDGPU: Don't use frame virtual registers
Matt Arsenault [Thu, 29 Aug 2019 01:13:47 +0000 (01:13 +0000)]
AMDGPU: Don't use frame virtual registers

SGPR spills aren't really handled after SILowerSGPRSpills. In order to
directly control what happens if the scavenger needs to spill, the
scavenger needs to be used directly. There is an alternative to
spilling in these contexts anyway since the frame register can be
increment and restored.

This does present another possible issue if spilling is needed for the
unused carry out if an add is needed. I think this can be avoided by
using a scalar add (although that clobbers SCC, which happens anyway).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370281 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoGlobalISel/TableGen: Handle setcc patterns
Matt Arsenault [Thu, 29 Aug 2019 01:13:41 +0000 (01:13 +0000)]
GlobalISel/TableGen: Handle setcc patterns

This is a special case because one node maps to two different G_
instructions, and the operand order is changed.

This mostly enables G_FCMP for AMDPGPU. G_ICMP is still manually
selected for now since it has the SALU and VALU complication to deal
with.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370280 91177308-0d34-0410-b5e6-96231b3b80d8

5 years agoAdd requirement to test.
Richard Trieu [Thu, 29 Aug 2019 00:46:57 +0000 (00:46 +0000)]
Add requirement to test.

-debug-only option for llc is only available in debug builds so
"REQUIRES: asserts" is needed in the tes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370279 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[X86] Fix a couple isel patterns to not shrink a volatile load.
Craig Topper [Wed, 28 Aug 2019 23:45:10 +0000 (23:45 +0000)]
[X86] Fix a couple isel patterns to not shrink a volatile load.

Also add a FIXME because I'm not sure why these patterns exist. Looks like a missing combine.

And another FIXME because the AVX512 equivalent one of the patterns is missing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370276 91177308-0d34-0410-b5e6-96231b3b80d8

5 years ago[RISCV] Avoid generating AssertZext for LP64 ABI when lowering floating LibCall
Shiva Chen [Wed, 28 Aug 2019 23:40:37 +0000 (23:40 +0000)]
[RISCV] Avoid generating AssertZext for LP64 ABI when lowering floating LibCall

The patch fixed the issue that RV64 didn't clear the upper bits
when return complex floating value with lp64 ABI.

float _Complex
complex_add(float _Complex a, float _Complex b)
{
   return a + b;
}

RealResult = zero_extend(RealA + RealB)
ImageResult = ImageA + ImageB
Return (RealResult | (ImageResult << 32))

The patch introduces shouldExtendTypeInLibCall target hook to suppress
the AssertZext generation when lowering floating LibCall.

Thanks to Eli's comments from the Bugzilla
https://bugs.llvm.org/show_bug.cgi?id=42820

Differential Revision: https://reviews.llvm.org/D65497

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@370275 91177308-0d34-0410-b5e6-96231b3b80d8