Joel E. Denny [Sat, 12 Oct 2019 16:00:35 +0000 (16:00 +0000)]
[lit] Try yet again to fix new tests that fail on Windows bots
I seem to have misread the bot logs on my last attempt. When lit's
internal diff runs on Windows under Python 2.7, it's text diffs not
binary diffs that need decoding to avoid this error when writing the
diff to stdout:
```
UnicodeEncodeError: 'ascii' codec can't encode characters in position 7-8: ordinal not in range(128)
```
There is no `decode` attribute in this case under Python 3.6.8 under
Ubuntu, so this patch checks for the `decode` attribute before using
it here. Hopefully nothing else is needed when `decode` isn't
available.
It might take a couple more attempts to figure out what error
handling, if any, is needed for this decoding.
Summary:
This is a recommit, this originally landed in rL370454 but was
subsequently reverted in rL370788 due to
https://bugs.llvm.org/show_bug.cgi?id=43206
The reduced testcase was added to bcmp-negative-tests.ll
as @pr43206_different_loops - we must ensure that the SCEV's
we got are both for the same loop we are currently investigating.
Original commit message:
@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}
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, 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?
* ???
Joel E. Denny [Sat, 12 Oct 2019 14:58:43 +0000 (14:58 +0000)]
[lit] Try again to fix new tests that fail on Windows bots
Based on the bot logs, when lit's internal diff runs on Windows, it
looks like binary diffs must be decoded also for Python 2.7.
Otherwise, writing the diff to stdout fails with:
```
UnicodeEncodeError: 'ascii' codec can't encode characters in position 7-8: ordinal not in range(128)
```
I did not need to decode using Python 2.7.15 under Ubuntu. When I do
it anyway in that case, `errors="backslashreplace"` fails for me:
```
TypeError: don't know how to handle UnicodeDecodeError in error callback
```
However, `errors="ignore"` works, so this patch uses that, hoping
it'll work on Windows as well.
This patch leaves `errors="backslashreplace"` for Python >= 3.5 as
there's no evidence yet that doesn't work and it produces more
informative binary diffs. This patch also adjusts some lit tests to
succeed for either error handler.
Joel E. Denny [Sat, 12 Oct 2019 11:58:30 +0000 (11:58 +0000)]
[lit] Fix internal diff's --strip-trailing-cr and use it
Using GNU diff, `--strip-trailing-cr` removes a `\r` appearing before
a `\n` at the end of a line. Without this patch, lit's internal diff
only removes `\r` if it appears as the last character. That seems
useless. This patch fixes that.
This patch also adds `--strip-trailing-cr` to some tests that fail on
Windows bots when D68664 is applied. Based on what I see in the bot
logs, I think the following is happening. In each test there, lit
diff is comparing a file with `\r\n` line endings to a file with `\n`
line endings. Without D68664, lit diff reads those files with
Python's universal newlines support activated, causing `\r` to be
dropped. However, with D68664, lit diff reads the files in binary
mode instead and thus reports that every line is different, just as
GNU diff does (at least under Ubuntu). Adding `--strip-trailing-cr`
to those tests restores the previous behavior while permitting the
behavior of lit diff to be more like GNU diff.
Joel E. Denny [Sat, 12 Oct 2019 11:58:03 +0000 (11:58 +0000)]
Reland r374392: [lit] Extend internal diff to support -U
To avoid breaking some tests, D66574, D68664, D67643, and D68668
landed together. However, D68664 introduced an issue now addressed by
D68839, with which these are now all relanding.
Joel E. Denny [Sat, 12 Oct 2019 11:57:41 +0000 (11:57 +0000)]
Reland r374390: [lit] Extend internal diff to support `-` argument
To avoid breaking some tests, D66574, D68664, D67643, and D68668
landed together. However, D68664 introduced an issue now addressed by
D68839, with which these are now all relanding.
Joel E. Denny [Sat, 12 Oct 2019 11:57:20 +0000 (11:57 +0000)]
Reland r374389: [lit] Clean up internal diff's encoding handling
To avoid breaking some tests, D66574, D68664, D67643, and D68668
landed together. However, D68664 introduced an issue now addressed by
D68839, with which these are now all relanding.
Joel E. Denny [Sat, 12 Oct 2019 11:56:57 +0000 (11:56 +0000)]
Reland r374388: [lit] Make internal diff work in pipelines
To avoid breaking some tests, D66574, D68664, D67643, and D68668
landed together. However, D68664 introduced an issue now addressed by
D68839, with which these are now all relanding.
Craig Topper [Sat, 12 Oct 2019 07:59:29 +0000 (07:59 +0000)]
[X86] Use pack instructions for packus/ssat truncate patterns when 256-bit is the largest legal vector and the result type is at least 256 bits.
Since the input type is larger than 256-bits we'll need to some
concatenating to reassemble the results. The pack instructions
ability to concatenate while packing make this a shorter/faster
sequence.
Craig Topper [Sat, 12 Oct 2019 07:59:24 +0000 (07:59 +0000)]
[X86] Test SKX cpu in the vector-trunc-packus/ssat/usat.ll tests instad of min-legal-vector-width.ll
This adds "min-legal-vector-width"="256" function attributes to
all the tests for a larger than 256-bit input. Also switch any
larger than 512-bit inputs to use a load. This makes the
arguments consistent with min-legal-vector-width attribute which
should usually be at least as large as the arguments.
The SKX configuration will avoid using zmm registers on the
modified test cases. For many of them we should use something
closer to the AVX2 codegen with pack instructions instead of
the avx512 saturating truncates.
Hubert Tong [Sat, 12 Oct 2019 04:08:31 +0000 (04:08 +0000)]
NFC: clang-format rL374420 and adjust comment wording
The commit of rL374420 had various formatting issues, including lines
that exceed 80 columns. This patch applies `git clang-format` on the
changes from commit 13bd3ef40d8b1586f26a022e01b21e56c91e05bd.
It further adjusts a comment to clarify the domain of inputs upon which
a newly added function is meant to operate. The adjustment to the
comment was suggested in a post-commit comment on D68721 and discussed
off-list with @sfertile.
Zi Xuan Wu [Sat, 12 Oct 2019 02:53:04 +0000 (02:53 +0000)]
recommit: [LoopVectorize][PowerPC] Estimate int and float register pressure separately in loop-vectorize
In loop-vectorize, interleave count and vector factor depend on target register number. Currently, it does not
estimate different register pressure for different register class separately(especially for scalar type,
float type should not be on the same position with int type), so it's not accurate. Specifically,
it causes too many times interleaving/unrolling, result in too many register spills in loop body and hurting performance.
So we need classify the register classes in IR level, and importantly these are abstract register classes,
and are not the target register class of backend provided in td file. It's used to establish the mapping between
the types of IR values and the number of simultaneous live ranges to which we'd like to limit for some set of those types.
For example, POWER target, register num is special when VSX is enabled. When VSX is enabled, the number of int scalar register is 32(GPR),
float is 64(VSR), but for int and float vector register both are 64(VSR). So there should be 2 kinds of register class when vsx is enabled,
and 3 kinds of register class when VSX is NOT enabled.
It runs on POWER target, it makes big(+~30%) performance improvement in one specific bmk(503.bwaves_r) of spec2017 and no other obvious degressions.
Vitaly Buka [Sat, 12 Oct 2019 02:29:26 +0000 (02:29 +0000)]
[sancov] Accommodate sancov and coverage report server for use under Windows
Summary:
This patch makes the following changes to SanCov and its complementary Python script in order to resolve issues pertaining to non-UNIX file paths in JSON symbolization information:
* Convert all paths to use forward slash.
* Update `coverage-report-server.py` to correctly handle paths to sources which contain spaces.
* Remove Linux platform restriction for all SanCov unit tests. All SanCov tests passed when ran on my local Windows machine.
Vitaly Buka [Sat, 12 Oct 2019 02:29:24 +0000 (02:29 +0000)]
[sancov] Use LLVM Support library JSON writer in favor of individual implementation
Summary:
In this diff, I've replaced the individual implementation of `JSONWriter` with `json::OStream` provided by `llvm/Support/JSON.h`.
Important Note: The output format of the JSON is considerably different compared to the original implementation. Important differences include:
* New line for each entry in an array (should make diffs cleaner)
* No space between keys and colon in attributed object entries.
* Attributes with empty strings will now print the attribute name and a quote pair rather than excluding the attribute altogether
Examples of these differences can be seen in the changes to the sancov tests which compare the JSON output.
Craig Topper [Sat, 12 Oct 2019 00:01:08 +0000 (00:01 +0000)]
[X86] Fold a VTRUNCS/VTRUNCUS+store into a saturating truncating store.
We already did this for VTRUNCUS with a specific combination of
types. This extends this to VTRUNCS and handles any types where
a truncating store is legal.
David Blaikie [Fri, 11 Oct 2019 21:52:41 +0000 (21:52 +0000)]
DebugInfo: Use base address selection entries for debug_loc
Unify the range and loc emission (for both DWARFv4 and DWARFv5 style lists) and take advantage of that unification to use strategic base addresses for loclists.
Simon Atanasyan [Fri, 11 Oct 2019 21:51:33 +0000 (21:51 +0000)]
[mips] Store 64-bit `li.d' operand as a single 8-byte value
Now assembler generates two consecutive `.4byte` directives to store
64-bit `li.d' operand. The first directive stores high 4-byte of the
value. The second directive stores low 4-byte of the value. But on
64-bit system we load this value at once and get wrong result if the
system is little-endian.
This patch fixes the bug. It stores the `li.d' operand as a single
8-byte value.
Simon Atanasyan [Fri, 11 Oct 2019 21:51:23 +0000 (21:51 +0000)]
[mips] Use less instruction to load zero into FPR by li.s / li.d pseudos
If `li.s` or `li.d` loads zero into a FPR, it's not necessary to load
zero into `at` GPR register and then move its value into a floating
point register. We can use as a source register the `zero / $0` one.
Quentin Colombet [Fri, 11 Oct 2019 20:58:26 +0000 (20:58 +0000)]
[GISel][UnitTest] Fix a bunch of tests that were not doing anything
After r368065, all the tests using GISelMITest must call setUp() before
doing anything, otherwise the TargetMachine is not going to be set up.
A few tests added after that commit were not doing that and ended up
testing effectively nothing.
Fix the setup of all the tests and fix the failing tests.
Quentin Colombet [Fri, 11 Oct 2019 20:22:47 +0000 (20:22 +0000)]
[MachineIRBuilder] Fix an assertion failure with buildMerge
Teach buildMerge how to deal with scalar to vector kind of requests.
Prior to this patch, buildMerge would issue either a G_MERGE_VALUES
when all the vregs are scalars or a G_CONCAT_VECTORS when the destination
vreg is a vector.
G_CONCAT_VECTORS was actually not the proper instruction when the source
vregs were scalars and the compiler would assert that the sources must
be vectors. Instead we want is to issue a G_BUILD_VECTOR when we are
in this situation.
Sanjay Patel [Fri, 11 Oct 2019 14:17:56 +0000 (14:17 +0000)]
[DAGCombiner] fold vselect-of-constants to shift
The diffs suggest that we are missing some more basic
analysis/transforms, but this keeps the vector path in
sync with the scalar (rL374397). This is again a
preliminary step for introducing the reverse transform
in IR as proposed in D63382.
Kai Nacke [Fri, 11 Oct 2019 12:50:57 +0000 (12:50 +0000)]
[Tests] Output of od can be lower or upper case (llvm-objcopy/yaml2obj).
The command `od -t x` is used to dump data in hex format.
The LIT tests assumes that the hex characters are in lowercase.
However, there are also platforms which use uppercase letter.
To solve this issue the tests are updated to use the new
`--ignore-case` option of FileCheck.
Simon Atanasyan [Fri, 11 Oct 2019 12:33:12 +0000 (12:33 +0000)]
[mips] Fix loading "double" immediate into a GPR and FPR
If a "double" (64-bit) value has zero low 32-bits, it's possible to load
such value into a GP/FP registers as an instruction immediate. But now
assembler loads only high 32-bits of the value.
For example, if a target register is GPR the `li.d $4, 1.0` instruction
converts into the `lui $4, 16368` one. As a result, we get `0x3FF00000`
in the register. While a correct representation of the `1.0` value is
`0x3FF0000000000000`. The patch fixes that.
George Rimar [Fri, 11 Oct 2019 12:27:11 +0000 (12:27 +0000)]
[llvm-readobj] - Remove excessive fields when dumping "Version symbols".
This removes a few fields that are not useful:
"Section Name", "Address", "Offset" and "Link"
(they duplicated the information available under
the "Sections [" tag).
Oliver Stannard [Fri, 11 Oct 2019 11:59:55 +0000 (11:59 +0000)]
Dead Virtual Function Elimination
Currently, it is hard for the compiler to remove unused C++ virtual
functions, because they are all referenced from vtables, which are referenced
by constructors. This means that if the constructor is called from any live
code, then we keep every virtual function in the final link, even if there
are no call sites which can use it.
This patch allows unused virtual functions to be removed during LTO (and
regular compilation in limited circumstances) by using type metadata to match
virtual function call sites to the vtable slots they might load from. This
information can then be used in the global dead code elimination pass instead
of the references from vtables to virtual functions, to more accurately
determine which functions are reachable.
To make this transformation safe, I have changed clang's code-generation to
always load virtual function pointers using the llvm.type.checked.load
intrinsic, instead of regular load instructions. I originally tried writing
this using clang's existing code-generation, which uses the llvm.type.test
and llvm.assume intrinsics after doing a normal load. However, it is possible
for optimisations to obscure the relationship between the GEP, load and
llvm.type.test, causing GlobalDCE to fail to find virtual function call
sites.
The existing linkage and visibility types don't accurately describe the scope
in which a virtual call could be made which uses a given vtable. This is
wider than the visibility of the type itself, because a virtual function call
could be made using a more-visible base class. I've added a new
!vcall_visibility metadata type to represent this, described in
TypeMetadata.rst. The internalization pass and libLTO have been updated to
change this metadata when linking is performed.
This doesn't currently work with ThinLTO, because it needs to see every call
to llvm.type.checked.load in the linkage unit. It might be possible to
extend this optimisation to be able to use the ThinLTO summary, as was done
for devirtualization, but until then that combination is rejected in the
clang driver.
To test this, I've written a fuzzer which generates random C++ programs with
complex class inheritance graphs, and virtual functions called through object
and function pointers of different types. The programs are spread across
multiple translation units and DSOs to test the different visibility
restrictions.
I've also tried doing bootstrap builds of LLVM to test this. This isn't
ideal, because only classes in anonymous namespaces can be optimised with
-fvisibility=default, and some parts of LLVM (plugins and bugpoint) do not
work correctly with -fvisibility=hidden. However, there are only 12 test
failures when building with -fvisibility=hidden (and an unmodified compiler),
and this change does not cause any new failures for either value of
-fvisibility.
On the 7 C++ sub-benchmarks of SPEC2006, this gives a geomean code-size
reduction of ~6%, over a baseline compiled with "-O2 -flto
-fvisibility=hidden -fwhole-program-vtables". The best cases are reductions
of ~14% in 450.soplex and 483.xalancbmk, and there are no code size
increases.
I've also run this on a set of 8 mbed-os examples compiled for Armv7M, which
show a geomean size reduction of ~3%, again with no size increases.
I had hoped that this would have no effect on performance, which would allow
it to awlays be enabled (when using -fwhole-program-vtables). However, the
changes in clang to use the llvm.type.checked.load intrinsic are causing ~1%
performance regression in the C++ parts of SPEC2006. It should be possible to
recover some of this perf loss by teaching optimisations about the
llvm.type.checked.load intrinsic, which would make it worth turning this on
by default (though it's still dependent on -fwhole-program-vtables).
Kai Nacke [Fri, 11 Oct 2019 11:59:14 +0000 (11:59 +0000)]
[FileCheck] Implement --ignore-case option.
The FileCheck utility is enhanced to support a `--ignore-case`
option. This is useful in cases where the output of Unix tools
differs in case (e.g. case not specified by Posix).
Aleksandr Urakov [Fri, 11 Oct 2019 09:03:29 +0000 (09:03 +0000)]
[Windows] Use information from the PE32 exceptions directory to construct unwind plans
This patch adds an implementation of unwinding using PE EH info. It allows to
get almost ideal call stacks on 64-bit Windows systems (except some epilogue
cases, but I believe that they can be fixed with unwind plan disassembly
augmentation in the future).
To achieve the goal the CallFrameInfo abstraction was made. It is based on the
DWARFCallFrameInfo class interface with a few changes to make it less
DWARF-specific.
To implement the new interface for PECOFF object files the class PECallFrameInfo
was written. It uses the next helper classes:
- UnwindCodesIterator helps to iterate through UnwindCode structures (and
processes chained infos transparently);
- EHProgramBuilder with the use of UnwindCodesIterator constructs EHProgram;
- EHProgram is, by fact, a vector of EHInstructions. It creates an abstraction
over the low-level unwind codes and simplifies work with them. It contains
only the information that is relevant to unwinding in the unified form. Also
the required unwind codes are read from the object file only once with it;
- EHProgramRange allows to take a range of EHProgram and to build an unwind row
for it.
So, PECallFrameInfo builds the EHProgram with EHProgramBuilder, takes the ranges
corresponding to every offset in prologue and builds the rows of the resulted
unwind plan. The resulted plan covers the whole range of the function except the
epilogue.
QingShan Zhang [Fri, 11 Oct 2019 08:36:54 +0000 (08:36 +0000)]
[TableGen] Fix a bug that MCSchedClassDesc is interfered between different SchedModel
Assume that, ModelA has scheduling resource for InstA and ModelB has scheduling resource for InstB. This is what the llvm::MCSchedClassDesc looks like:
llvm::MCSchedClassDesc ModelBSchedClasses[] = {
...
InstA, -1,...
InstB, 0,...
};
The -1 means invalid num of macro ops, while it is valid if it is >=0. This is what we look like now:
llvm::MCSchedClassDesc ModelBSchedClasses[] = {
...
InstA, 0,...
InstB, 0,...
};
And compiler hit the assertion here because the SCDesc is valid now for both InstA and InstB.
Craig Topper [Fri, 11 Oct 2019 07:24:36 +0000 (07:24 +0000)]
[X86] Add v8i64->v8i8 ssat/usat/packus truncate tests to min-legal-vector-width.ll
I wonder if we should split the v8i8 stores in order to form
two v4i8 saturating truncating stores. This would remove the
unpckl needed to concatenated the v4i8 results to make a
single store.
Pavel Labath [Fri, 11 Oct 2019 07:16:19 +0000 (07:16 +0000)]
Fix modules build for r374337
A modules build failed with the following error:
call to function 'operator&' that is neither visible in the template definition nor found by argument-dependent lookup
Fix that by declaring the appropriate operators in the llvm::minidump
namespace.
Philip Reames [Fri, 11 Oct 2019 03:48:56 +0000 (03:48 +0000)]
[CVP] Remove a masking operation if range information implies it's a noop
This is really a known bits style transformation, but known bits isn't context sensitive. The particular case which comes up happens to involve a range which allows range based reasoning to eliminate the mask pattern, so handle that case specifically in CVP.
InstCombine likes to generate the mask-by-low-bits pattern when widening an arithmetic expression which includes a zext in the middle.
Lang Hames [Fri, 11 Oct 2019 01:58:12 +0000 (01:58 +0000)]
[JITLink] Disable the MachO/AArch64 testcase while investigating bot failures.
The windows bots are failing due to a memory layout error. Temporarily disabling
while I investigate whether this can be worked around, or whether the test
should be disabled on Windows.
Volodymyr Sapsai [Fri, 11 Oct 2019 00:57:41 +0000 (00:57 +0000)]
[Stats] Add ALWAYS_ENABLED_STATISTIC enabled regardless of LLVM_ENABLE_STATS.
The intended usage is to measure relatively expensive operations. So the
cost of the statistic is negligible compared to the cost of a measured
operation and can be enabled all the time without impairing the
compilation time.
Craig Topper [Fri, 11 Oct 2019 00:38:41 +0000 (00:38 +0000)]
[X86] Update trunc_packus_v32i32_v32i8 test in min-legal-vector-width.ll to use a load for the large type and add the min-legal-vector-width attribute.
The attribute is needed to avoid zmm registers. Using memory
avoids argument splitting for large vectors.