Fixed isfinite implementation and added test #348

thoasm · 2019-09-10T15:46:53Z

@yhmtsai discovered that both the implementation and the documentation is wrong for the self-written function isfinite (which is used in certain compiler combinations).
Now, both the implementation and the documentation of isfinite is (at least it should be) correct now.
Additionally, a test was added for the host functions to ensure the behavior is as expected.

In the process of fixing another compiler error, the whole previously created mess was cleaned up, leading to readable and maintainable code.

yhmtsai

It would be great if there is a cuda test for it.

core/test/base/math.cpp

codecov · 2019-09-11T18:13:41Z

Codecov Report

Merging #348 into develop will increase coverage by 0.01%.
The diff coverage is 100%.

@@             Coverage Diff             @@
##           develop     #348      +/-   ##
===========================================
+ Coverage    98.28%   98.29%   +0.01%     
===========================================
  Files          250      251       +1     
  Lines        18718    18752      +34     
===========================================
+ Hits         18397    18433      +36     
+ Misses         321      319       -2

Impacted Files	Coverage Δ
core/test/base/math.cpp	`100% <100%> (ø)`
include/ginkgo/core/base/math.hpp	`100% <100%> (+6.89%)`	⬆️

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update fbc409d...6f5b017. Read the comment docs.

yhmtsai · 2019-09-12T07:48:37Z

For Cuda test, I think comparing CUDA version and reference is okay for testing.
Google tests seems to work with *.cu (exception_helper.cu and executor.cu).
Add a __global__ function to call __device__ finite() in a cuda test file and compare its result and reference result.

thoasm · 2019-09-12T13:41:05Z

@yhmtsai I added a CUDA test now.
For some reason, I thought it would not be possible that way, but I should have checked.
Thanks for pointing that out.

yhmtsai · 2019-09-12T16:02:49Z

I got this error message with clean install.

ginkgo/cuda/test/base/math.cu(66): error: calling a constexpr __host__ function("isfinite") from a __global__ function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.
          detected during instantiation of "__nv_bool <unnamed>::IsFinite::test_real_isfinite_kernel<T>() [with T=float]" 
(124)

It seems to use the __host__ function not __device__.

cuda/test/base/math.cu

thoasm · 2019-09-12T16:10:22Z

@yhmtsai What compiler are you using? I had the same issue for some combination of compilers, which is why the the isfinite has quite a few #if defined(...) ....

yhmtsai · 2019-09-12T16:11:26Z

I use GCC 5.4 and NVCC 9.2

thoasm · 2019-09-12T16:32:51Z

@yhmtsai That is weird, I had pretty much the same thing happening with CLang, but never with GCC.
I will test that tomorrow thoroughly with the CI system (since it is too old to install it on my notebook).

pratikvn

Some questions but otherwise looks good.

core/test/base/math.cpp

cuda/test/base/math.cu

yhmtsai · 2019-09-13T09:31:49Z

In my environment (GCC 5.4 and NVCC 9.2), it uses std::isfinite function in the cuda code.
In (cuda-cmake)[/~https://github.com/ginkgo-project/ginkgo/blob/1b9b919d566cab75cee6291f3adf16cd922f06fb/cuda/CMakeLists.txt#L55-L62], we add --expt-relaxed-constexpr, so the compiler does not complain it in par-ilu.

By adding the similar settings in the create_test.cmake, it works in my environment

if(NOT CMAKE_CUDA_COMPILER_VERSION MATCHES "9.0")
    # remove false positive CUDA warnings when calling one<T>() and zero<T>()
    target_compile_options("${TEST_TARGET_NAME}"
        PRIVATE
            $<$<COMPILE_LANGUAGE:CUDA>:--expt-relaxed-constexpr>)
    endif()

However, I do not know why CI system does not face this problem.

yhmtsai · 2019-09-13T12:03:00Z

I would like to add the following to test the property of inf

// The followings are true
isfinite(inf - inf) // nan
isfinite(inf/inf) // nan
isfinite(inf * 2) // inf
isfinite(1.0/0.0) // inf
isfinite(0.0/0.0) // nan

thoasm · 2019-09-17T17:15:31Z

@yhmtsai could you try the current state again on your laptop at some point again? I changed a lot, and it would be nice if it would work in the current state.
However, if it does not, can you paste in the error that you get? I would assume it is a similar one I got with clang, but I would have to read it to be sure.

For the --expt-relaxed-constexpr option: I am not a big fan of running host code on GPUs if we can prevent it. Host code might be optimized for CPU and might run bad on the GPU, in the worst case, it might result in Undefined Behavior or wrong results. Maybe, we can add it to the special case we have currently with CLang, although, I agree, that is not a very nice solution.

yhmtsai · 2019-09-17T19:15:01Z

@thoasm it does not work in my environment.

GCC configuration

Using built-in specs.
COLLECT_GCC=/home/u/mike/gcc-5.4/bin/gcc
COLLECT_LTO_WRAPPER=/home/u/mike/gcc-5.4/libexec/gcc/x86_64-unknown-linux-gnu/5.4.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-5.4.0/configure --prefix=/home/u/mike/gcc-5.4 --enable-languages=c,c++,fortran --disable-multilib
Thread model: posix
gcc version 5.4.0 (GCC)

The error message


[ 51%] Linking CXX executable math
[ 51%] Linking CXX executable types
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(64): error: calling a constexpr __host__ function("isfinite") from a __global__ function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.
          detected during instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]" 
(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(64): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(65): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(65): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(66): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(63): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=float]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(63): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=float]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(63): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=float]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(66): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=float]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
[ 51%] Linking CXX executable dim

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(67): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=float]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(67): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=float]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=float]"

(127): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(63): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=double]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=double]"

(130): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(63): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=double]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=double]"

(130): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(63): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=double]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=double]"

(130): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(66): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=double]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=double]"

(130): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(67): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=double]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=double]"

(130): here
/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(67): error: calling a constexpr host function("isfinite") from a global function("test_real_isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "void ::test_real_isfinite(__nv_bool *) [with T=double]"

(108): here

instantiation of "__nv_bool ::IsFinite::test_real_isfinite_kernel() [with T=double]"

(130): here
/home/u/mike/kit/review/ginkgo/include/ginkgo/core/base/math.hpp(546): warning: calling a constexpr host function("isfinite") from a host device function("isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "gko::xstd::enable_if_t<gko::detail::is_complex_impl::value, __nv_bool> gko::isfinite(const T &) [with T=gko::kernels::cuda::cuda_type<thrust::complex>]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(84): here

instantiation of "void ::test_complex_isfinite(__nv_bool *) [with ComplexType=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(117): here

instantiation of "__nv_bool ::IsFinite::test_complex_isfinite_kernel() [with T=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(135): here
/home/u/mike/kit/review/ginkgo/include/ginkgo/core/base/math.hpp(546): warning: calling a constexpr host function("isfinite") from a host device function("isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "gko::xstd::enable_if_t<gko::detail::is_complex_impl::value, __nv_bool> gko::isfinite(const T &) [with T=gko::kernels::cuda::cuda_type<thrust::complex>]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(84): here

instantiation of "void ::test_complex_isfinite(__nv_bool *) [with ComplexType=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(117): here

instantiation of "__nv_bool ::IsFinite::test_complex_isfinite_kernel() [with T=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(135): here
/home/u/mike/kit/review/ginkgo/include/ginkgo/core/base/math.hpp(546): warning: calling a constexpr host function("isfinite") from a host device function("isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "gko::xstd::enable_if_t<gko::detail::is_complex_impl::value, __nv_bool> gko::isfinite(const T &) [with T=gko::kernels::cuda::cuda_type<thrust::complex>]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(84): here

instantiation of "void ::test_complex_isfinite(__nv_bool *) [with ComplexType=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(117): here

instantiation of "__nv_bool ::IsFinite::test_complex_isfinite_kernel() [with T=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(141): here
/home/u/mike/kit/review/ginkgo/include/ginkgo/core/base/math.hpp(546): warning: calling a constexpr host function("isfinite") from a host device function("isfinite") is not allowed. The experimental flag '--expt-relaxed-constexpr' can be used to allow this.

detected during:

instantiation of "gko::xstd::enable_if_t<gko::detail::is_complex_impl::value, __nv_bool> gko::isfinite(const T &) [with T=gko::kernels::cuda::cuda_type<thrust::complex>]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(84): here

instantiation of "void ::test_complex_isfinite(__nv_bool *) [with ComplexType=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(117): here

instantiation of "__nv_bool ::IsFinite::test_complex_isfinite_kernel() [with T=thrust::complex]"

/home/u/mike/kit/review/ginkgo/cuda/test/base/math.cu(141): here
17 errors detected in the compilation of "/tmp/tmpxft_000047a0_00000000-6_math.cpp1.ii".

cuda/test/base/CMakeFiles/cuda_test_base_math.dir/build.make:62: recipe for target 'cuda/test/base/CMakeFiles/cuda_test_base_math.dir/math.cu.o' failed

make[2]: *** [cuda/test/base/CMakeFiles/cuda_test_base_math.dir/math.cu.o] Error 1

CMakeFiles/Makefile2:1330: recipe for target 'cuda/test/base/CMakeFiles/cuda_test_base_math.dir/all' failed

make[1]: *** [cuda/test/base/CMakeFiles/cuda_test_base_math.dir/all] Error 2

make[1]: *** Waiting for unfinished jobs....

[ 51%] Built target core_test_base_math

thoasm · 2019-09-18T17:01:51Z

@yhmtsai I was able to simplify the whole mess with the different isfinite implementations to a point where I am happy with the result. I think the current state should be a lot more robust to different CUDA and compiler versions because it does not depend on them anymore.

I would appreciate it if you could test it again on your system (hopefully the last time).

thoasm · 2019-09-19T23:31:39Z

Finally, everything seems to compile and work.
@yhmtsai I know I annoy you with this, but could you please try it on your system again, please? In theory, nothing should change for you since you are not using CLang, but at this point, I do not trust my theory.

yhmtsai · 2019-09-20T08:07:58Z

@thoasm Sure. It works well in my environment

cuda/base/math.hpp

tcojean

LGTM. Some minor comments.

cuda/base/math.hpp

yhmtsai

LGTM.

cuda/base/math.hpp

Both the implementation and the documentation of the self-written `isfinite` was wrong and is now fixed. Also added a test to check the implementation of `isfinite`.

Also: test kernels are now called with complex types instead of the real-type.

Reimplementation was necessary since wrong results got produced with some compilers in release mode. Also, moved all the CUDA kernels in the cuda/base/math.hpp file instead of leaving them in the public header.

Simplified the dependency of `isfinite` to the CUDA version in both core and cuda and removed all self-implemented versions of it. This should both improve readability as well as maintanability.

It was necessary to readd the custom `isfinite` kernel since clang always replaced the CUDA math function with their builtin version.

Another try to make the clang compiler accept the custom `isfinite` function by also providing a complex version (calling the regular one). This only happens for clang and toolkit < 9.2.

The Ginkgo team is proud to announce the new minor release of Ginkgo version 1.1.0. This release brings several performance improvements, adds Windows support, adds support for factorizations inside Ginkgo and a new ILU preconditioner based on ParILU algorithm, among other things. For detailed information, check the respective issue. Supported systems and requirements: + For all platforms, cmake 3.9+ + Linux and MacOS + gcc: 5.3+, 6.3+, 7.3+, 8.1+ + clang: 3.9+ + Intel compiler: 2017+ + Apple LLVM: 8.0+ + CUDA module: CUDA 9.0+ + Windows + MinGW and CygWin: gcc 5.3+, 6.3+, 7.3+, 8.1+ + Microsoft Visual Studio: VS 2017 15.7+ + CUDA module: CUDA 9.0+, Microsoft Visual Studio + OpenMP module: MinGW or CygWin. The current known issues can be found in the [known issues page](/~https://github.com/ginkgo-project/ginkgo/wiki/Known-Issues). Additions: + Upper and lower triangular solvers ([#327](#327), [#336](#336), [#341](#341), [#342](#342)) + New factorization support in Ginkgo, and addition of the ParILU algorithm ([#305](#305), [#315](#315), [#319](#319), [#324](#324)) + New ILU preconditioner ([#348](#348), [#353](#353)) + Windows MinGW and Cygwin support ([#347](#347)) + Windows Visual studio support ([#351](#351)) + New example showing how to use ParILU as a preconditioner ([#358](#358)) + New example on using loggers for debugging ([#360](#360)) + Add two new 9pt and 27pt stencil examples ([#300](#300), [#306](#306)) + Allow benchmarking CuSPARSE spmv formats through Ginkgo's benchmarks ([#303](#303)) + New benchmark for sparse matrix format conversions ([#312](/~https://github.com/ginkgo-project/ginkgo/issues/312)[#317](/~https://github.com/ginkgo-project/ginkgo/issues/317)) + Add conversions between CSR and Hybrid formats ([#302](#302), [#310](#310)) + Support for sorting rows in the CSR format by column idices ([#322](#322)) + Addition of a CUDA COO SpMM kernel for improved performance ([#345](#345)) + Addition of a LinOp to handle perturbations of the form (identity + scalar * basis * projector) ([#334](#334)) + New sparsity matrix representation format with Reference and OpenMP kernels ([#349](#349), [#350](#350)) Fixes: + Accelerate GMRES solver for CUDA executor ([#363](#363)) + Fix BiCGSTAB solver convergence ([#359](#359)) + Fix CGS logging by reporting the residual for every sub iteration ([#328](#328)) + Fix CSR,Dense->Sellp conversion's memory access violation ([#295](#295)) + Accelerate CSR->Ell,Hybrid conversions on CUDA ([#313](#313), [#318](#318)) + Fixed slowdown of COO SpMV on OpenMP ([#340](#340)) + Fix gcc 6.4.0 internal compiler error ([#316](#316)) + Fix compilation issue on Apple clang++ 10 ([#322](#322)) + Make Ginkgo able to compile on Intel 2017 and above ([#337](#337)) + Make the benchmarks spmv/solver use the same matrix formats ([#366](#366)) + Fix self-written isfinite function ([#348](#348)) + Fix Jacobi issues shown by cuda-memcheck Tools and ecosystem: + Multiple improvements to the CI system and tools ([#296](#296), [#311](#311), [#365](#365)) + Multiple improvements to the Ginkgo containers ([#328](#328), [#361](#361)) + Add sonarqube analysis to Ginkgo ([#304](#304), [#308](#308), [#309](#309)) + Add clang-tidy and iwyu support to Ginkgo ([#298](#298)) + Improve Ginkgo's support of xSDK M12 policy by adding the `TPL_` arguments to CMake ([#300](#300)) + Add support for the xSDK R7 policy ([#325](#325)) + Fix examples in html documentation ([#367](#367))

The Ginkgo team is proud to announce the new minor release of Ginkgo version 1.1.0. This release brings several performance improvements, adds Windows support, adds support for factorizations inside Ginkgo and a new ILU preconditioner based on ParILU algorithm, among other things. For detailed information, check the respective issue. Supported systems and requirements: + For all platforms, cmake 3.9+ + Linux and MacOS + gcc: 5.3+, 6.3+, 7.3+, 8.1+ + clang: 3.9+ + Intel compiler: 2017+ + Apple LLVM: 8.0+ + CUDA module: CUDA 9.0+ + Windows + MinGW and Cygwin: gcc 5.3+, 6.3+, 7.3+, 8.1+ + Microsoft Visual Studio: VS 2017 15.7+ + CUDA module: CUDA 9.0+, Microsoft Visual Studio + OpenMP module: MinGW or Cygwin. The current known issues can be found in the [known issues page](/~https://github.com/ginkgo-project/ginkgo/wiki/Known-Issues). ### Additions + Upper and lower triangular solvers ([#327](#327), [#336](#336), [#341](#341), [#342](#342)) + New factorization support in Ginkgo, and addition of the ParILU algorithm ([#305](#305), [#315](#315), [#319](#319), [#324](#324)) + New ILU preconditioner ([#348](#348), [#353](#353)) + Windows MinGW and Cygwin support ([#347](#347)) + Windows Visual Studio support ([#351](#351)) + New example showing how to use ParILU as a preconditioner ([#358](#358)) + New example on using loggers for debugging ([#360](#360)) + Add two new 9pt and 27pt stencil examples ([#300](#300), [#306](#306)) + Allow benchmarking CuSPARSE spmv formats through Ginkgo's benchmarks ([#303](#303)) + New benchmark for sparse matrix format conversions ([#312](/~https://github.com/ginkgo-project/ginkgo/issues/312)[#317](/~https://github.com/ginkgo-project/ginkgo/issues/317)) + Add conversions between CSR and Hybrid formats ([#302](#302), [#310](#310)) + Support for sorting rows in the CSR format by column idices ([#322](#322)) + Addition of a CUDA COO SpMM kernel for improved performance ([#345](#345)) + Addition of a LinOp to handle perturbations of the form (identity + scalar * basis * projector) ([#334](#334)) + New sparsity matrix representation format with Reference and OpenMP kernels ([#349](#349), [#350](#350)) ### Fixes + Accelerate GMRES solver for CUDA executor ([#363](#363)) + Fix BiCGSTAB solver convergence ([#359](#359)) + Fix CGS logging by reporting the residual for every sub iteration ([#328](#328)) + Fix CSR,Dense->Sellp conversion's memory access violation ([#295](#295)) + Accelerate CSR->Ell,Hybrid conversions on CUDA ([#313](#313), [#318](#318)) + Fixed slowdown of COO SpMV on OpenMP ([#340](#340)) + Fix gcc 6.4.0 internal compiler error ([#316](#316)) + Fix compilation issue on Apple clang++ 10 ([#322](#322)) + Make Ginkgo able to compile on Intel 2017 and above ([#337](#337)) + Make the benchmarks spmv/solver use the same matrix formats ([#366](#366)) + Fix self-written isfinite function ([#348](#348)) + Fix Jacobi issues shown by cuda-memcheck ### Tools and ecosystem improvements + Multiple improvements to the CI system and tools ([#296](#296), [#311](#311), [#365](#365)) + Multiple improvements to the Ginkgo containers ([#328](#328), [#361](#361)) + Add sonarqube analysis to Ginkgo ([#304](#304), [#308](#308), [#309](#309)) + Add clang-tidy and iwyu support to Ginkgo ([#298](#298)) + Improve Ginkgo's support of xSDK M12 policy by adding the `TPL_` arguments to CMake ([#300](#300)) + Add support for the xSDK R7 policy ([#325](#325)) + Fix examples in html documentation ([#367](#367)) Related PR: #370

thoasm added is:bug Something looks wrong. mod:core This is related to the core module. 1:ST:ready-for-review This PR is ready for review labels Sep 10, 2019

thoasm requested review from pratikvn, yhmtsai, hartwiganzt and tcojean September 10, 2019 15:46

thoasm self-assigned this Sep 10, 2019

thoasm force-pushed the fix_isfinite branch from 1d2de8d to dfdfa75 Compare September 10, 2019 15:48

yhmtsai requested changes Sep 11, 2019

View reviewed changes

core/test/base/math.cpp Outdated Show resolved Hide resolved

core/test/base/math.cpp Show resolved Hide resolved

thoasm force-pushed the fix_isfinite branch from be85815 to f970ad6 Compare September 12, 2019 16:08

yhmtsai requested changes Sep 12, 2019

View reviewed changes

cuda/test/base/math.cu Outdated Show resolved Hide resolved

cuda/test/base/math.cu Outdated Show resolved Hide resolved

cuda/test/base/math.cu Outdated Show resolved Hide resolved

thoasm added 1:ST:WIP This PR is a work in progress. Not ready for review. and removed 1:ST:ready-for-review This PR is ready for review labels Sep 12, 2019

pratikvn approved these changes Sep 12, 2019

View reviewed changes

core/test/base/math.cpp Show resolved Hide resolved

cuda/test/base/math.cu Show resolved Hide resolved

cuda/test/base/math.cu Outdated Show resolved Hide resolved

thoasm added 1:ST:do-not-merge Please do not merge PR this yet. 1:ST:ready-for-review This PR is ready for review and removed 1:ST:WIP This PR is a work in progress. Not ready for review. labels Sep 18, 2019

thoasm added 1:ST:WIP This PR is a work in progress. Not ready for review. 1:ST:ready-for-review This PR is ready for review and removed 1:ST:ready-for-review This PR is ready for review 1:ST:WIP This PR is a work in progress. Not ready for review. labels Sep 18, 2019

yhmtsai requested changes Sep 20, 2019

View reviewed changes

cuda/base/math.hpp Show resolved Hide resolved

tcojean approved these changes Sep 20, 2019

View reviewed changes

cuda/base/math.hpp Show resolved Hide resolved

cuda/base/math.hpp Outdated Show resolved Hide resolved

thoasm force-pushed the fix_isfinite branch from 84df9e8 to 41d1576 Compare September 20, 2019 12:59

yhmtsai approved these changes Sep 20, 2019

View reviewed changes

cuda/base/math.hpp Outdated Show resolved Hide resolved

cuda/base/math.hpp Outdated Show resolved Hide resolved

thoasm force-pushed the fix_isfinite branch from 41d1576 to 06224cd Compare September 20, 2019 14:38

Thomas Grützmacher added 11 commits September 23, 2019 12:18

Fixed isfinite implementation and added test

1d1feb2

Both the implementation and the documentation of the self-written `isfinite` was wrong and is now fixed. Also added a test to check the implementation of `isfinite`.

Review update

d84d44c

Added CUDA test for isfinite

f513292

Removed copy paste mistake.

20212fc

Added more tests

bd883f8

Also: test kernels are now called with complex types instead of the real-type.

Reimplemented custom isfinite

8232d9c

Reimplementation was necessary since wrong results got produced with some compilers in release mode. Also, moved all the CUDA kernels in the cuda/base/math.hpp file instead of leaving them in the public header.

Removed debug and unnecesary code

cda5516

Simplified the isfinite handling

dab33a4

Simplified the dependency of `isfinite` to the CUDA version in both core and cuda and removed all self-implemented versions of it. This should both improve readability as well as maintanability.

Readded custom isfinite for clang

1c8cd64

It was necessary to readd the custom `isfinite` kernel since clang always replaced the CUDA math function with their builtin version.

Added complex isfinite for clang compiler

f0e1122

Another try to make the clang compiler accept the custom `isfinite` function by also providing a complex version (calling the regular one). This only happens for clang and toolkit < 9.2.

Changed doc. according to reviewer comments

6f5b017

thoasm force-pushed the fix_isfinite branch from 06224cd to 6f5b017 Compare September 23, 2019 10:18

thoasm merged commit 26b6c44 into develop Sep 23, 2019

thoasm added 1:ST:ready-to-merge This PR is ready to merge. and removed 1:ST:do-not-merge Please do not merge PR this yet. 1:ST:ready-for-review This PR is ready for review labels Sep 23, 2019

thoasm deleted the fix_isfinite branch September 23, 2019 14:48

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fixed isfinite implementation and added test #348

Fixed isfinite implementation and added test #348

thoasm commented Sep 10, 2019 •

edited

Loading

yhmtsai left a comment

codecov bot commented Sep 11, 2019 •

edited

Loading

yhmtsai commented Sep 12, 2019

thoasm commented Sep 12, 2019

yhmtsai commented Sep 12, 2019

thoasm commented Sep 12, 2019

yhmtsai commented Sep 12, 2019

thoasm commented Sep 12, 2019

pratikvn left a comment

yhmtsai commented Sep 13, 2019

yhmtsai commented Sep 13, 2019

thoasm commented Sep 17, 2019

yhmtsai commented Sep 17, 2019

thoasm commented Sep 18, 2019

thoasm commented Sep 19, 2019

yhmtsai commented Sep 20, 2019

tcojean left a comment

yhmtsai left a comment

Fixed isfinite implementation and added test #348

Fixed isfinite implementation and added test #348

Conversation

thoasm commented Sep 10, 2019 • edited Loading

yhmtsai left a comment

Choose a reason for hiding this comment

codecov bot commented Sep 11, 2019 • edited Loading

Codecov Report

yhmtsai commented Sep 12, 2019

thoasm commented Sep 12, 2019

yhmtsai commented Sep 12, 2019

thoasm commented Sep 12, 2019

yhmtsai commented Sep 12, 2019

thoasm commented Sep 12, 2019

pratikvn left a comment

Choose a reason for hiding this comment

yhmtsai commented Sep 13, 2019

yhmtsai commented Sep 13, 2019

thoasm commented Sep 17, 2019

yhmtsai commented Sep 17, 2019

thoasm commented Sep 18, 2019

thoasm commented Sep 19, 2019

yhmtsai commented Sep 20, 2019

tcojean left a comment

Choose a reason for hiding this comment

yhmtsai left a comment

Choose a reason for hiding this comment

thoasm commented Sep 10, 2019 •

edited

Loading

codecov bot commented Sep 11, 2019 •

edited

Loading