ForkMaster/netgen
1
0
Fork 0
mirror of https://github.com/NGSolve/netgen.git synced 2024-11-12 00:59:16 +05:00
Code Issues Packages Projects Releases Wiki Activity
4,072 Commits 10 Branches 43 Tags 18 MiB
5851b558bb
Commit Graph

403 Commits

Author SHA1 Message Date
Joachim Schoeberl
2b9d4596ae AllReduce for Arrays 2022-03-05 11:42:00 +01:00
mhochsteger@cerbsim.com
4e8fe77098 fix some warnings 2022-03-02 14:14:38 +01:00
mhochsteger@cerbsim.com
fe838fbd75 fix ToLower() 2022-02-23 19:16:30 +01:00
mhochsteger@cerbsim.com
cf59f297a7 use std::filesystem::path 2022-02-23 11:56:00 +01:00
mhochsteger@cerbsim.com
eceb83171f allow implicit conversion from py::list to Array<> 2022-02-21 12:02:30 +01:00
mhochsteger@cerbsim.com
d3e0ae6fd7 use ngcore pybind list caster only for non-numpy types 
we should Array<T> for all numpy dtypes T
 2022-02-21 11:18:19 +01:00
mhochsteger@cerbsim.com
17e845df29 fix rpath of pyngcore 2022-02-16 09:44:00 +01:00
Matthias Hochsteger
d2378d9781 pyngcore as package (staying backward compatible) 2022-02-16 09:27:02 +01:00
Joachim Schoeberl
67c031cb77 fix unused-variable warning 2022-02-11 15:08:59 +01:00
Joachim Schoeberl
79ff65d224 added SIMD-wrapper for 'erf' 2022-02-11 13:56:28 +01:00
Joachim Schoeberl
1222356cf4 HSum(4) for M1, FlatArray!= 2022-02-04 17:01:31 +01:00
Stefan Brüns
5fce0b48b8 Fix signedness for ARM Neon mask type 
vbslq_f64 and vandq_u64 both require uint64x2_t types as mask arguments,
and the Neon intrinsics do not allow for implicit conversion.

Fixes build errors with current GCC 11.2.1:

/home/abuild/rpmbuild/BUILD/netgen-6.2.2105/libsrc/core/simd_arm64.hpp:171:29: error: cannot convert '__Int64x2_t' to 'uint64x2_t'
 171 |     return vandq_u64 (a.Data(), b.Data());
 2022-01-27 14:16:51 +01:00
Joachim Schoeberl
44d626f727 BitArray indexing from the end 2021-12-29 21:06:52 +01:00
mhochsteger@cerbsim.com
f95332d0a1 archive optional<T> 2021-11-10 18:36:55 +01:00
mhochsteger@cerbsim.com
b258d7bd65 fix path to string conversion 2021-11-04 13:00:30 +01:00
mhochsteger@cerbsim.com
747367ab8a util function GetTempFilename() 2021-11-04 12:20:56 +01:00
mhochsteger@cerbsim.com
f0e7f843c3 util functions to get compile-time simd size and range check settings 2021-10-28 13:47:58 +02:00
mhochsteger@cerbsim.com
d9e8b815f5 pip installer 2021-10-27 16:51:41 +02:00
Christopher Lackner
a8da814c19 remove debug output 2021-09-28 22:51:06 +02:00
Christopher Lackner
bd29763b16 Add std::any to py::object caster for archive registered types 2021-09-28 22:34:11 +02:00
Joachim Schoeberl
968658a70e too much tracing for recursive tasks ? 2021-09-28 15:41:46 +02:00
Christopher Lackner
2b7347ce07 Export Flags.ToDict to python 2021-09-28 14:05:33 +02:00
mhochsteger@cerbsim.com
e1d71a78ab no need to remove archive type infos 2021-09-13 17:18:32 +02:00
Joachim Schoeberl
5a82e18ed1 default msg-level to 0 2021-09-06 17:36:53 +02:00
Joachim Schoeberl
77f07f8baf occ output control 2021-09-06 16:59:48 +02:00
Matthias Hochsteger
42faaacdb7 fix gcc build error 2021-09-03 13:12:23 +02:00
Matthias Hochsteger
6a26125889 Consistent Demangle on MSVC and gcc/clang 
Use regex list from here:
https://github.com/RobotLocomotion/drake/blob/master/common/nice_type_name.cc
 2021-09-01 16:56:01 +02:00
Matthias Hochsteger
48198d232d Revert "Revert "Merge branch 'parallel_meshing' into 'master'"" 
This reverts commit 65c5e2d244.
 2021-07-19 14:59:12 +02:00
Christopher Lackner
300835a446 fix table archive 2021-07-17 13:35:01 +02:00
Christopher Lackner
6f044faaba archive Table 2021-07-17 12:48:59 +02:00
Matthias Hochsteger
65c5e2d244 Revert "Merge branch 'parallel_meshing' into 'master'" 
This reverts commit d20a297cf1, reversing
changes made to fd50131a5b.
 2021-06-29 19:38:19 +02:00
Matthias Hochsteger
55ab122f35 Merge remote-tracking branch 'origin/master' into parallel_meshing 2021-06-28 12:10:08 +02:00
Joachim Schoeberl
fd50131a5b constexpr function 2021-06-28 01:35:23 +02:00
Joachim Schoeberl
91506aa71a static constexpr 2021-06-28 01:07:03 +02:00
Joachim Schoeberl
31d5ce8be9 packed archiving 2021-06-27 12:32:51 +02:00
Joachim Schoeberl
54db7941d0 tuning mesh(un)pickling 2021-06-26 12:14:17 +02:00
Joachim Schoeberl
e84d4e90c8 add header for std::byte 2021-06-25 18:58:25 +02:00
Matthias Hochsteger
34629749d6 Merge remote-tracking branch 'origin/master' into parallel_meshing 2021-06-25 17:32:16 +02:00
Joachim Schoeberl
72fb819def missing overloads for archive (byte) 2021-06-24 13:21:43 +02:00
Joachim Schoeberl
971d6bb465 little tuning of mesh pickling 2021-06-24 07:39:09 +02:00
Matthias Hochsteger
8be5b7d810 Merge remote-tracking branch 'origin/master' into parallel_meshing 2021-06-22 11:46:36 +02:00
Matthias Hochsteger
4b40a7eb31 backward-compatible Timer interface 2021-06-14 14:55:08 +02:00
Matthias Hochsteger
c835767115 Timer in ProcessTask() 2021-06-14 10:39:57 +02:00
Matthias Hochsteger
1de1a1800e Fix template argument deduction for Timer, remove ThreadRegionTimer 2021-06-11 15:19:30 +02:00
Matthias Hochsteger
6f7543c7dc Timer - convenience constructors to disable tracing and/or timing 
Examples:
Timer t0("name");
Timer t1("name", NoTracing);
Timer t2("name", NoTiming);
Timer t3("name", NoTracing, NoTiming);
Timer t4("name", NoTiming, NoTracing);
 2021-06-11 10:08:06 +02:00
Matthias Hochsteger
c5639a5706 Thread-safe Timer 
- use template arguments instead of run-time variable 'priority'
 - change in paje interface for tracing
 2021-06-11 09:52:58 +02:00
Matthias Hochsteger
a11294baf0 inline GetThreadI() (except on Windows, no dllexport for thread_local variables supported) 2021-06-11 09:51:23 +02:00
Matthias Hochsteger
a96a1e4624 separate memtracer.hpp 2021-06-10 09:38:00 +02:00
Joachim Schoeberl
abb2e43ccb optimize parallel load 2021-06-08 19:08:14 +02:00
Matthias Hochsteger
3ce5b1958e Initialize FlatArray members ( thx @mrambausek ) 2021-06-02 15:45:36 +02:00
Joachim Schoeberl
6dcc89ad04 some table py-features 2021-06-01 12:57:58 +02:00
Joachim Schoeberl
7c4f1cf53a minimal export of Table 2021-05-30 22:15:21 +02:00
Joachim Schoeberl
c3984fcc5b just use pointer for Array - iterators (on proposal of Matthias R) 2021-05-30 18:58:34 +02:00
Joachim Schoeberl
3258b27410 fix initialization order warning 2021-05-30 18:57:14 +02:00
Matthias Hochsteger
db494f4239 more Timers in Mesh 2021-05-12 10:56:21 +02:00
Matthias Hochsteger
4b53c63fba helper functions for table creation 2021-05-10 12:03:20 +02:00
Matthias Hochsteger
acf2b39680 Fix cross-platform archiving 
This is a non-backward compatible change for archives on Windows!
 2021-04-23 20:06:58 +02:00
Joachim Schoeberl
2d9e32ba70 ArrayMem from BaseArray ctor 2021-04-09 21:30:29 +02:00
Christopher Lackner
daa0985a41 trace memory free only when array owns memory 2021-04-07 09:58:53 +02:00
Christopher Lackner
4fad6e0631 fix pickling on arm, store long type platform independent 2021-04-01 10:48:13 +02:00
Joachim Schoeberl
1f45601387 Array<int> ia(n); ia.Range(2, END-1) 2021-03-29 22:39:57 +02:00
Christopher Lackner
001eaa32b6 DoArchive for LocalH 2021-03-29 13:55:23 +02:00
Stefan Brüns
85e8c09ff6 Fix GetTimeCounter for Aarch64 variants 
Neither GCC nor Clang define an __arm64__ preprocessor macro, but use
__aarch64__ (MSVC uses _MARM_64). Add a "64" suffix to the define, i.e.
NETGEN_ARCH_ARM64 to make it more obvious in only refers to aarch64, and
to be in line with NETGEN_ARCH_AMD64.

Replace the (Clang specific) __builtin_readcyclecounter with inline
asm:
- The function return cycles (i.e. varies with CPU frequency), not time
- It may return 0, depending on the PMU settings
- It may cause an illegal instruction, in case it is not trapped by the
  kernel, e.g. on FreeBSD.

Reading the generic timer/counter CNTVCT_EL0 instead of PMCCNTR_EL0 avoids
these pitfalls. The inline asm works on GCC and Clang, instead of
Clang only for the builtin.
 2021-03-03 17:30:33 +01:00
Joachim Schoeberl
979a695f62 fixing warnings 2021-02-18 10:30:01 +01:00
Joachim Schoeberl
87e472b6fc start face-hierarchy in Netgen 2021-02-17 14:54:14 +01:00
Christopher Lackner
0c2430f3dc add std::any symboltable to Flags to store arbitrary objects 2021-02-08 15:44:15 +01:00
Joachim Schoeberl
25011c8407 arm-simd: HSum, tuple support 2021-02-05 11:59:03 +01:00
Joachim Schöberl
9a9828d3af some more arm-simds 2021-01-31 16:31:47 +01:00
Joachim Schöberl
18f5a933a9 arm-simd working 2021-01-30 21:02:49 +01:00
Joachim Schöberl
f53c069308 prepare SIMD for arm64 2021-01-30 20:05:28 +01:00
Joachim Schöberl
ea7f6c1e94 fnma intrinsic for avx512 2020-12-22 13:06:08 +01:00
Joachim Schöberl
c1c10174be FNMA asm-instruction 2020-12-22 09:37:21 +01:00
Matthias Hochsteger
94ecf8de92 Fix private linking of Python 
The CMake export of Interface libraries also exports PRIVATE build
settings, which leads to build errors with non-existing include paths and .lib files for binary distributions.

Use the work-around mentioned here to circumvent this behavior:
https://gitlab.kitware.com/cmake/cmake/-/issues/15415#note_849405
 2020-12-18 11:05:10 +01:00
Joachim Schöberl
d30accdc1a Merge branch 'apple_silicon' into 'master' 
Support for Apple M1

See merge request jschoeberl/netgen!359
 2020-12-16 20:47:21 +00:00
Matthias Hochsteger
eb6ac164e7 int64_t for masks 2020-12-16 21:00:12 +01:00
Matthias Hochsteger
d97a9a6594 Alignment for generic SIMD classes 2020-12-16 17:20:18 +01:00
Matthias Hochsteger
e68d8cea9b workaround for missing intrinsic on GCC 7 2020-12-16 10:58:01 +01:00
Matthias Hochsteger
9c0dbec8c9 Fix SIMD<mask64> ctor and Unpack 2020-12-15 15:31:17 +01:00
Matthias Hochsteger
dbe894fea3 Support for Apple M1 2020-12-15 10:43:11 +01:00
Matthias Hochsteger
1b55c51da5 remove AlignedAlloc, use alignas 2020-12-15 09:40:43 +01:00
Matthias Hochsteger
1f3aebcec0 Fix AVX-Operators for int64_t simd (use generic ones) 2020-12-15 09:40:43 +01:00
mhochsteger
f213a7a5b1 fix fabs for AVX on Windows 2020-12-14 15:50:27 +01:00
Matthias Hochsteger
248145bbf0 fix wrong simd operators 2020-12-14 12:47:53 +01:00
Matthias Hochsteger
fc44eb95df simd - array and variadic ctor 2020-12-11 23:12:34 +01:00
Matthias Hochsteger
2d667a08dc move (refactored) SIMD headers from ngsolve into ngcore 2020-12-11 20:54:41 +01:00
Christopher Lackner
cb0d8295bf fix hashing of bitarray (uninitialized value in HashArchive & random 
values at end)
 2020-11-25 22:07:07 +01:00
Matthias Hochsteger
91f127ef71 memory tracer - fix memory accumulation of children 2020-11-25 14:34:29 +01:00
Matthias Hochsteger
b55264e0ee memory tracing - handle multiple consecutive tracers correctly 2020-11-24 19:20:21 +01:00
Matthias Hochsteger
efdc57885a memory tracing - store parents array instead of children table 2020-11-24 17:12:39 +01:00
Christopher Lackner
922ad16213 if more memory is deallocated than allocated set memtracer to 0 not 
negative values
 2020-11-21 22:32:41 +01:00
Christopher Lackner
a69cdc9000 mem tracing compile time option, simplify by MemoryTracer as member 2020-11-21 15:49:07 +01:00
Matthias Hochsteger
87623981a6 export PajeTrace.WriteMemoryChart() to python 2020-11-19 19:29:04 +01:00
Matthias Hochsteger
f0152baacf mem tracing - TraceMemorySwap helper function 2020-11-19 17:35:29 +01:00
Matthias Hochsteger
6f98123e98 mem tracing - use topological sorting, some fixes 2020-11-19 16:16:39 +01:00
Matthias Hochsteger
a17066a387 html chart for peak memory consumption, some Array tracing fixes 2020-11-19 14:57:45 +01:00
Matthias Hochsteger
f143995f27 clean up memory tracing 2020-11-18 21:45:00 +01:00
Matthias Hochsteger
1a93fb3fa5 first attempt on memory tracing 2020-11-18 20:20:35 +01:00
Joachim Schöberl
cddfb4a0b5 fixing delaunay2d point search, non-parallel for small meshes 2020-10-26 11:20:12 +01:00
Joachim Schöberl
bfbef51996 remove bitarray in delaunay2d, just one hashtable position 2020-10-23 19:40:47 +02:00
Matthias Hochsteger
832485e41a pybind11 compatibility 2020-10-22 12:11:19 +02:00