C++17: Work-Around NVCC gatherParticles #2596
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The `noexcept` lambda does not compile in C++17 mode due to an NVCC
compiler bug, at least in NVCC 11.3.109.
Compiles in C++14 mode with the same compiler.
Error:
```
nvcc_internal_extended_lambda_implementation:732:77: error: ‘operator()’ is not a member of ‘int (ParticleBoundaryBuffer::gatherParticles(MultiParticleContainer&, const amrex::Vector<const amrex::MultiFab*>&)::<lambda(const SrcData&, int)>::*)(const amrex::ConstParticleTileData<0, 0, 4, 0>&, int) const noexcept’
Source/Particles/ParticleBoundaryBuffer.cpp: In member function ‘void ParticleBoundaryBuffer::gatherParticles(MultiParticleContainer&, const amrex::Vector<const amrex::MultiFab*>&)’:
Source/Particles/ParticleBoundaryBuffer.cpp:245:34: error: no matching function for call to ‘__nv_hdl_create_wrapper_t<false, false, __nv_dl_tag<void (ParticleBoundaryBuffer::*)(MultiParticleContainer&, const amrex::Vector<const amrex::MultiFab*>&), &ParticleBoundaryBuffer::gatherParticles, 1>, const GetParticlePosition, amrex::Array4<const float>, amrex::GpuArray<float, 3>, amrex::GpuArray<float, 3> >::__nv_hdl_create_wrapper(ParticleBoundaryBuffer::gatherParticles(MultiParticleContainer&, const amrex::Vector<const amrex::MultiFab*>&)::<lambda(const SrcData&, int)>, const GetParticlePosition&, amrex::Array4<const float>&, amrex::GpuArray<float, 3>&, amrex::GpuArray<float, 3>&)’
245 | },
| ^
```
atmyers
approved these changes
Nov 23, 2021
ax3l
commented
Nov 24, 2021
| @@ -232,7 +232,8 @@ void ParticleBoundaryBuffer::gatherParticles (MultiParticleContainer& mypc, | |||
| int timestep = warpx_instance.getistep(0); | |||
| using SrcData = WarpXParticleContainer::ParticleTileType::ConstParticleTileDataType; | |||
| auto count = amrex::filterAndTransformParticles(ptile_buffer, ptile, | |||
| [=] AMREX_GPU_HOST_DEVICE (const SrcData& /*src*/, const int ip) noexcept | |||
| [=] AMREX_GPU_HOST_DEVICE (const SrcData& /*src*/, const int ip) | |||
| /* NVCC 11.3.109 chokes in C++17 on this: noexcept */ | |||
There was a problem hiding this comment.
Submitted as Nvidia developer bug no. 3447924
dpgrote
pushed a commit
to dpgrote/WarpX
that referenced
this pull request
Nov 29, 2021
The `noexcept` lambda does not compile in C++17 mode due to an NVCC compiler bug, at least in NVCC 11.3.109. Compiles in C++14 mode with the same compiler.
roelof-groenewald
added a commit
to ModernElectron/WarpX
that referenced
this pull request
Dec 3, 2021
* Fix Init of Vector Members (ECP-WarpX#2595) Fix default init of `Vector` member variables. The old construct is not valid C++. https://stackoverflow.com/a/11491003/2719194 * C++17: Work-Around NVCC gatherParticles (ECP-WarpX#2596) The `noexcept` lambda does not compile in C++17 mode due to an NVCC compiler bug, at least in NVCC 11.3.109. Compiles in C++14 mode with the same compiler. * requirements.txt - PICMI development version (ECP-WarpX#2588) Document in `requirements.txt` on how to install a pre-release version of PICMI. * CONTRIBUTING: Update/Modernize (ECP-WarpX#2600) * CONTRIBUTING: Update/Modernize - Add GitHub account setup - Add local git setup - Modernize for CMake workflows * Apply suggestions by Edoardo Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com> * Replaced duplicated current deposition documentation (ECP-WarpX#2604) * Throwing a warning if particle_shape>1 with EB (ECP-WarpX#2592) * Aborting if particle_shape!=1 with EB * Throw warning instead of aborting * Checking at runtime if EB is initialized * Added missing preprocessor directive * Ignoring an unused variable * Fix typo * Improve style Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> * Fix segfault when importing _libwarpx without initializing WarpX (ECP-WarpX#2580) * Added check for if warpx was initialized when calling finalize * Renamed to be warpx_initialized * Fixed reference to global variable Co-authored-by: Peter Scherpelz <31747262+peterscherpelz@users.noreply.github.com> * Changed global variable to member of libwarpx * Fixed syntax errors * Remove custom arg from argv to avoid parmparse error Co-authored-by: Peter Scherpelz <31747262+peterscherpelz@users.noreply.github.com> * Added parallel pragma to ApplyBoundaryConditions (ECP-WarpX#2612) * Note that CCache 4.2 introduced large CUDA improvements (ECP-WarpX#2606) * Dimensionality Docs: Default (ECP-WarpX#2609) Just adds the note that 3D is the default geometry. * AMReX: Weekly Update (ECP-WarpX#2613) * MergeBuffersForPlotfile: Barrier (ECP-WarpX#2608) Make sure that all MPI ranks are in sync, i.e., have closed the files that they wrote, before trying to merge them. * Fix installation location for libraries (ECP-WarpX#2583) During configuration the installation location for libraries is given by dumping the cmake variable `CMAKE_INSTALL_LIBDIR`. This commit adjusts the installation of WarpX libraries (WarpX_LIB=ON) to respect this setting. Co-authored-by: Rolf Pfeffertal <tropf@users.noreply.github.com> * Release 21.12 (ECP-WarpX#2614) * AMReX: 21.12 * PICSAR: 21.12 * WarpX: 21.12 * div(E,B) Cleaning Options for PSATD (ECP-WarpX#2403) * Implement div(E)/div(B) Cleaning with Standard PSATD * Cleaning * Update Benchmark * Add Nodal Synchronization of F,G * OneStep_multiJ: Nodal Syncs, Damp PML * OneStep_multiJ: Push PSATD Fields in PML * div Cleaning Defaults (Domain v. PML) * Include Fix of ECP-WarpX#2429 until Merged * Reset Benchmark of Langmuir_multi_psatd_div_cleaning * Multi-J: Remove PML Support * Include Fix of ECP-WarpX#2474 Until Merged * Exchange All Guard Cells for F,G * Fix Defaults * Update Test, Reset Benchmark * Fix Defaults * Cleaning * Default update_with_rho=1 if do_dive_cleaning=1 * Update CI Test pml_psatd_dive_divb_cleaning * Replace Warning with Abort * Add 2D Langmuir Test w/ MR & PSATD (ECP-WarpX#2605) * Add 2D Langmuir Test w/ MR & PSATD * Add Missing Compile String * Fix out-of-bound in Inverse FFT of F,G (ECP-WarpX#2619) * Mention that the potentail should be constant inside EB (ECP-WarpX#2618) * Mention that the potentail should be constant inside EB * Update text * Replace AMREX_SPACEDIM: Boundary & Parallelization (ECP-WarpX#2620) * AMREX_SPACEDIM : Boundary Conditions * AMREX_SPACEDIM : Parallelization * Fix compilation * Update Source/Parallelization/WarpXComm_K.H * Fix out-of-bound in the initialization of EB (ECP-WarpX#2607) * Call FillBoundary when initializing EB * Avoid out-of-bound * Bug fix * Apply suggestions from code review * update version number Co-authored-by: Axel Huebl <axel.huebl@plasma.ninja> Co-authored-by: Edoardo Zoni <59625522+EZoni@users.noreply.github.com> Co-authored-by: Remi Lehe <remi.lehe@normalesup.org> Co-authored-by: Lorenzo Giacomel <47607756+lgiacome@users.noreply.github.com> Co-authored-by: Kevin Z. Zhu <86268612+KZhu-ME@users.noreply.github.com> Co-authored-by: Peter Scherpelz <31747262+peterscherpelz@users.noreply.github.com> Co-authored-by: David Grote <grote1@llnl.gov> Co-authored-by: Phil Miller <phil@intensecomputing.com> Co-authored-by: s9105947 <80697868+s9105947@users.noreply.github.com> Co-authored-by: Rolf Pfeffertal <tropf@users.noreply.github.com> Co-authored-by: Prabhat Kumar <89051199+prkkumar@users.noreply.github.com>
|
Good news! Nvidia reported that they fixed the issue (3447924) in a "future release" as of May 16, 2022. So should be gone with CUDA 12+ 🎉 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The
noexceptlambda does not compile in C++17 mode (#2300) due to an NVCC compiler bug, at least in NVCC 11.3.109.Compiles in C++14 mode with the same compiler.
Error:
Thx to @atmyers for the hint :)
To Do
cmake -S . -B build -DWarpX_COMPUTE=CUDA -DWarpX_OPENPMD=ON -DAMReX_CUDA_ARCH=6.0 -DWarpX_EB=ON -DWarpX_PRECISION=SINGLE -DWarpX_PSATD=ON -DAMReX_CUDA_ERROR_CROSS_EXECUTION_SPACE_CALL=ON -DAMReX_CUDA_ERROR_CAPTURE_THIS=ON -DCMAKE_VERBOSE_MAKEFILE=ON cmake --build build