Workaround for broken [[deprecated]] in PGI compiler #1581

pramodk · 2020-03-07T23:39:37Z

similar to Intel and NVCC, add workaround for PGI compiler
I tested this with:

kumbhar@bbpv1:~/tmp/fmt/build$ pgc++ --version

pgc++ 19.4-0 LLVM 64-bit target on x86-64 Linux -tp skylake
PGI Compilers and Tools
Copyright (c) 2019, NVIDIA CORPORATION.  All rights reserved.

kumbhar@bbpv1:~/tmp/fmt/build$ cmake .. -DCMAKE_BUILD_TYPE=Debug
-- CMake version: 3.15.3
-- Version: 6.1.3
-- Build type: Debug
-- CXX_STANDARD: 11
-- Required features: cxx_variadic_templates
CMake Warning (dev) at CMakeLists.txt:35 (set):
  implicitly converting 'STRINGS' to 'STRING' type.
Call Stack (most recent call first):
  CMakeLists.txt:253 (set_doc)
This warning is for project developers.  Use -Wno-dev to suppress it.

-- FMT_PEDANTIC: OFF
-- Configuring done
-- Generating done
-- Build files have been written to: /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/build

Currently two tests are failing:

$ make test

88% tests passed, 2 tests failed out of 17

Total Test time (real) =   0.29 sec

The following tests FAILED:
	  7 - format-test (SEGFAULT)
	  8 - format-impl-test (Failed)
Errors while running CTest


$ ctest -V

...
7: /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/format-test.cc:1111: Failure
7: Value of: format("{:.494}", 4.9406564584124654E-324)
7:   Actual: "0.0000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"
7: Expected: "4.9406564584124654417656879286822137236505980261432476442558568250067550" "727020875186529983636163599237979656469544571773092665671035593979639877" "479601078187812630071319031140452784581716784898210368871863605699873072" "305000638740915356498438731247339727316961514003171538539807412623856559" "117102665855668676818703956031062493194527159149245532930545654440112748" "012970999954193198940908041656332452475714786901472678015935523861155013" "480352649347201937902681071074917033322268447533357208324319361e-324"
7: Which is: "4.9406564584124654417656879286822137236505980261432476442558568250067550727020875186529983636163599237979656469544571773092665671035593979639877479601078187812630071319031140452784581716784898210368871863605699873072305000638740915356498438731247339727316961514003171538539807412623856559117102665855668676818703956031062493194527159149245532930545654440112748012970999954193198940908041656332452475714786901472678015935523861155013480352649347201937902681071074917033322268447533357208324319361e-324"
7: [  FAILED  ] FormatterTest.Precision (1 ms)

8: [ RUN      ] FormatTest.CountCodePoints
8: /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/format-impl-test.cc:434: Failure
8: Value of: fmt::internal::count_code_points( fmt::basic_string_view<char8_t>(reinterpret_cast<const char8_t*>("ёжик")))
8:   Actual: 1
8: Expected: 4
8: [  FAILED  ] FormatTest.CountCodePoints (0 ms)

I didn't look into details but if there are any suggestions, it will be helpful.

- similar to Intel and NVCC, add workaround for PGI compiler

- set CMAKE_CXX11_STANDARD_COMPILE_OPTION to remove -A : https://www.pgroup.com/userforum/viewtopic.php?f=4&t=7565&p=27308#p27308 - use forks of fmt and spdlog because of fmtlib/fmt#1581

vitaut

Thanks for the PR. Mostly looks good but please apply clang-format.

The FormatterTest.Precision failure is likely due to non-IEEE754 floating point. In this case you could try making the failing check conditional on is_iec559:

if (std::numeric_limits<double>::is_iec559) {
  EXPECT_EQ(
      "4.9406564584124654417656879286822137236505980261432476442558568250067550"
      "727020875186529983636163599237979656469544571773092665671035593979639877"
      "479601078187812630071319031140452784581716784898210368871863605699873072"
      "305000638740915356498438731247339727316961514003171538539807412623856559"
      "117102665855668676818703956031062493194527159149245532930545654440112748"
      "012970999954193198940908041656332452475714786901472678015935523861155013"
      "480352649347201937902681071074917033322268447533357208324319361e-324",
      format("{:.494}", 4.9406564584124654E-324));
}

It's hard to say why CountCodePoints is failing. You'll need to provide more information.

pramodk · 2020-03-08T21:39:54Z

The FormatterTest.Precision failure is likely due to non-IEEE754 floating point. In this case you could try making the failing check conditional on is_iec559

Thanks! In the compiler manual I saw flag -Kieee related to conformance of IEEE 754. This fixes FormatterTest.Precision test.

There are now two tests that are failing:

./bin/format-test segfaults for :

TEST(FormatTest, FormatUTF8Precision) {
  using str_type = std::basic_string<char8_t>;
  str_type format(reinterpret_cast<const char8_t*>(u8"{:.4}"));
  str_type str(reinterpret_cast<const char8_t*>(u8"caf\u00e9s"));  // cafés
  auto result = fmt::format(format, str);
  EXPECT_EQ(fmt::internal::count_code_points(result), 4);
  EXPECT_EQ(result.size(), 5);
  EXPECT_EQ(from_u8str(result), from_u8str(str.substr(0, 5)));
}

stack trance is:

(gdb) bt
#0  0x00000000005faa05 in fmt::v6::internal::parse_format_string (format_str=..., handler=0x7fffffff9330)
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/format.h:2562
#1  0x000000000060f1db in fmt::v6::vformat_to (out=..., format_str=..., args=..., loc=...)
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/format.h:3151
#2  0x00000000005fa904 in fmt::v6::internal::vformat_to (buf=0x7fffffff9480, format_str=..., args=...)
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/format.h:3306
#3  0x00000000005fa858 in fmt::v6::internal::vformat (format_str=..., args=...)
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/format.h:3424
#4  0x0000000000606340 in fmt::v6::format (format_str=0x7fffffff98d8, args=0x7fffffff98b0)
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/core.h:1546
#5  0x000000000047ad60 in FormatTest_FormatUTF8Precision_Test::TestBody ()
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/format-test.cc:2568
#6  0x000000000067cf8d in testing::internal::HandleSehExceptionsInMethodIfSupported (object=0x7fffe40008c0, method=...,
    location=0x6d1f2f <.S51215> "the test body") at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3447
#7  0x0000000000673653 in testing::internal::HandleExceptionsInMethodIfSupported (object=0x7fffe40008c0, method=...,
    location=0x6d1f2f <.S51215> "the test body") at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3483
#8  0x000000000064b152 in testing::Test::Run () at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3519
#9  0x000000000064c2c2 in testing::TestInfo () at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3690
#10 0x000000000064ccb3 in testing::TestCase () at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3807
#11 0x0000000000659147 in testing::internal::UnitTestImpl () at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:5648
#12 0x000000000067d10d in testing::internal::HandleSehExceptionsInMethodIfSupported (object=0x9bfd70, method=...,
    location=0x6d1e75 <.S52231> "auxiliary test code (environments or event listeners)")
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3447
#13 0x0000000000675103 in testing::internal::HandleExceptionsInMethodIfSupported (object=0x9bfd70, method=...,
    location=0x6d1e75 <.S52231> "auxiliary test code (environments or event listeners)")
    at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:3483
#14 0x000000000064d65d in testing::UnitTest () at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gmock-gtest-all.cc:5268
#15 0x000000000062b931 in RUN_ALL_TESTS () at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/gtest/gtest.h:20062
#16 0x000000000062b963 in main (argc=1, argv=0x7fffffff9eb8) at /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/test-main.cc:37

CountCodePoints test:

TEST(FormatTest, CountCodePoints) {
#ifndef __cpp_char8_t
  using fmt::char8_t;
#endif
  EXPECT_EQ(4, fmt::internal::count_code_points(
    fmt::basic_string_view<char8_t>(reinterpret_cast<const char8_t*>("ёжик"))));
}

produces:

8: [ RUN      ] FormatTest.CountCodePoints
8: /gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/test/format-impl-test.cc:434: Failure
8: Value of: fmt::internal::count_code_points( fmt::basic_string_view<char8_t>(reinterpret_cast<const char8_t*>("ёжик")))
8:   Actual: 1
8: Expected: 4

So seems something related to UTF-8 with PGI? I don't find any relevant info though.

pramodk · 2020-03-08T21:45:27Z

Mostly looks good but please apply clang-format.

Done

- Add newline at the end of files to avoid PGI warnings - Fix PGI internal assertion external/nmodl/src/codegen/codegen_c_visitor.cpp", line 3969: internal error: assertion failed at: "../src/lower_init.c", line 7946 auto statement = ShadowUseStatement{lhs, "+=", rhs}; - Fixes for PGI build: set CMAKE_CXX11_STANDARD_COMPILE_OPTION to remove -A : https://www.pgroup.com/userforum/viewtopic.php?f=4&t=7565&p=27308#p27308 - Use fork of spdlog because of fmtlib/fmt#1581 - Add necessary compiler flags by default

- Add newline at the end of files to avoid PGI warnings - Fix PGI internal assertion external/nmodl/src/codegen/codegen_c_visitor.cpp", line 3969: internal error: assertion failed at: "../src/lower_init.c", line 7946 auto statement = ShadowUseStatement{lhs, "+=", rhs}; - Fixes for PGI build: set CMAKE_CXX11_STANDARD_COMPILE_OPTION to remove https://www.pgroup.com/userforum/viewtopic.php?f=4&t=7565&p=27308#p27308 - Use fork of spdlog because of fmtlib/fmt#1581 - Add necessary compiler flags by default See #271, it was merged into wrong branch. Fixes #262

vitaut · 2020-03-14T16:32:43Z

So seems something related to UTF-8 with PGI?

Looks like it.

What does the following program print when compiled with PGI and run?

#include <fmt/format.h>

int main() {
  auto s = fmt::basic_string_view<char8_t>(
      reinterpret_cast<const char8_t*>("ёжик"));
  for (auto c: s) {
    fmt::print("\\x{:02x}", static_cast<unsigned>(c));
  }
}

pramodk · 2020-03-15T08:08:36Z

What does the following program print when compiled with PGI and run?

sure!

First I was getting this:

"/gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/format.h", line 3502:
          error: a literal operator template must have a template parameter
          list equivalent to "<char ...>"
  FMT_CONSTEXPR internal::udl_formatter<Char, CHARS...> operator""_format() {
                                                        ^
"../test.cpp", line 8: error: identifier "char8_t" is undefined
    auto s = fmt::basic_string_view<char8_t>(
                                    ^
"../test.cpp", line 9: error: identifier "char8_t" is undefined
        reinterpret_cast<const char8_t*>("ёжик"));

Then I added using fmt::char8_; and then got:

"/gpfs/bbp.cscs.ch/home/kumbhar/tmp/fmt/include/fmt/format.h", line 3502:
          error: a literal operator template must have a template parameter
          list equivalent to "<char ...>"
  FMT_CONSTEXPR internal::udl_formatter<Char, CHARS...> operator""_format() {

Then I set FMT_USE_UDL_TEMPLATE 0 and then managed to compiled the code. The output from your program is:

$ ./test
\xffffffd1\xffffffd0

By the way, PGI compile is free to download in case you want to test locally.

Let me know if I should test anything else.

vitaut · 2020-03-15T16:39:51Z

PGI butchered the string literal. The correct output should be: \xd1\x91\xd0\xb6\xd0\xb8\xd0\xba (https://godbolt.org/z/eKg2sM). Unfortunately I don't know how to work around this (other than just conditionally disable the tests). Suggestions/PRs are welcome.

* Workaround broken [[deprecated]] in PGI compiler - similar to Intel and NVCC, add workaround for PGI compiler

Workaround broken [[deprecated]] in PGI compiler

265c936

- similar to Intel and NVCC, add workaround for PGI compiler

pramodk mentioned this pull request Mar 7, 2020

Fixes for building with PGI compiler BlueBrain/nmodl#263

Closed

5 tasks

pramodk mentioned this pull request Mar 7, 2020

Fixes for building with PGI compiler BlueBrain/nmodl#271

Merged

4 tasks

vitaut requested changes Mar 8, 2020

View reviewed changes

run clang-format (v9.0.0)

0c04714

pramodk mentioned this pull request Mar 8, 2020

Build issues with PGI (v19.4) BlueBrain/nmodl#262

Closed

vitaut merged commit 5bb8856 into fmtlib:master Mar 9, 2020

pramodk mentioned this pull request Mar 9, 2020

Workaround for broken [[deprecated]] in PGI compiler gabime/spdlog#1463

Closed

pramodk mentioned this pull request Mar 11, 2020

Fixes for building with PGI compiler BlueBrain/nmodl#274

Merged

teajay-fr pushed a commit to teajay-fr/fmt that referenced this pull request Mar 18, 2020

Workaround for broken [[deprecated]] in PGI compiler (fmtlib#1581)

4db0248

* Workaround broken [[deprecated]] in PGI compiler - similar to Intel and NVCC, add workaround for PGI compiler

pramodk mentioned this pull request Aug 11, 2022

Reorganise "global" variables to not be global BlueBrain/nmodl#904

Merged

5 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Workaround for broken [[deprecated]] in PGI compiler #1581

Workaround for broken [[deprecated]] in PGI compiler #1581

pramodk commented Mar 7, 2020

vitaut left a comment

pramodk commented Mar 8, 2020

pramodk commented Mar 8, 2020

vitaut commented Mar 14, 2020

pramodk commented Mar 15, 2020

vitaut commented Mar 15, 2020 •

edited

Loading

Workaround for broken [[deprecated]] in PGI compiler #1581

Workaround for broken [[deprecated]] in PGI compiler #1581

Conversation

pramodk commented Mar 7, 2020

vitaut left a comment

Choose a reason for hiding this comment

pramodk commented Mar 8, 2020

pramodk commented Mar 8, 2020

vitaut commented Mar 14, 2020

pramodk commented Mar 15, 2020

vitaut commented Mar 15, 2020 • edited Loading

vitaut commented Mar 15, 2020 •

edited

Loading