GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-08-04 11:10:40 +08:00
Code Issues Packages Projects Releases Wiki Activity
8,780 Commits 14 Branches 89 Tags
Commit Graph

3309 Commits

Author SHA1 Message Date
Gael Guennebaud
ba05572dcb bump to 3.3-rc2 2016-11-04 09:09:06 +01:00
Benoit Steiner
ca0ba0d9a4 Improved AVX512 support 2016-11-03 04:00:49 -07:00
Benoit Steiner
c80587c92b Merged eigen/eigen into default 2016-11-03 03:55:11 -07:00
Gael Guennebaud
3f1d0cdc22 bug #1337: improve doc of homogeneous() and hnormalized() 2016-11-03 11:03:08 +01:00
Gael Guennebaud
78e93ac1ad bug #1330: Cholmod supports double precision only, so let's trigger a static assertion if the scalar type does not match this requirement. 2016-11-03 10:21:59 +01:00
Gael Guennebaud
a07bb428df bug #1004: improve accuracy of LinSpaced for abs(low) >> abs(high). 2016-11-02 11:34:38 +01:00
Gael Guennebaud
598de8b193 Add pinsertfirst function and implement pinsertlast for complex on SSE/AVX. 2016-11-02 10:38:13 +01:00
Benoit Steiner
7a0e96b80d Gate the code that refers to cuda fp16 primitives more thoroughly 2016-11-01 12:08:09 -07:00
Gael Guennebaud
3ecb343dc3 Fix regression in X = (X*X.transpose())/s with X rectangular by deferring resizing of the destination after the creation of the evaluator of the source expression. 2016-10-26 22:50:41 +02:00
Gael Guennebaud
97feea9d39 add a generic EIGEN_HAS_CXX11 2016-10-26 15:53:13 +02:00
Gael Guennebaud
ca6a2a5248 Fix warning with ICC 2016-10-26 14:13:05 +02:00
Gael Guennebaud
b15a5dc3f4 Fix ICC warnings 2016-10-25 22:20:24 +02:00
Gael Guennebaud
aad72f3c6d Add missing inline keywords 2016-10-25 20:20:09 +02:00
Benoit Steiner
3e194a6a73 Fixed a typo 2016-10-25 08:42:15 -07:00
Gael Guennebaud
58146be99b bug #1004: one more rewrite of LinSpaced for floating point numbers to guarantee both interpolation and monotonicity. 
This version simply does low+i*step plus a branch to return high if i==size-1.
Vectorization is accomplished with a branch and the help of pinsertlast.
Some quick benchmark revealed that the overhead is really marginal, even when filling small vectors.
 2016-10-25 16:53:09 +02:00
Gael Guennebaud
13fc18d3a2 Add a pinsertlast function replacing the last entry of a packet by a scalar. 
(useful to vectorize LinSpaced)
 2016-10-25 16:48:49 +02:00
Gael Guennebaud
2634f9386c bug #1333: fix bad usage of const_cast_derived. Better use .data() for that purpose. 2016-10-24 22:22:35 +02:00
Gael Guennebaud
9e8f07d7b5 Cleanup ArrayWrapper and MatrixWrapper by removing redundant accessors. 2016-10-24 22:16:48 +02:00
Gael Guennebaud
b027d7a8cf bug #1004: remove the inaccurate "sequential" path for LinSpaced, mark respective function as deprecated, and enforce strict interpolation of the higher range using a correction term. 
Now, even with floating point precision, both the 'low' and 'high' bounds are exactly reproduced at i=0 and i=size-1 respectively.
 2016-10-24 20:27:21 +02:00
Gael Guennebaud
53c77061f0 bug #698: rewrite LinSpaced for integer scalar types to avoid overflow and guarantee an even spacing when possible. 
Otherwise, the "high" bound is implicitly lowered to the largest value allowing for an even distribution.
This changeset also disable vectorization for this integer path.
 2016-10-24 15:50:27 +02:00
Gael Guennebaud
40f62974b7 bug #1328: workaround a compilation issue with gcc 4.2 2016-10-20 19:19:37 +02:00
Gael Guennebaud
692b30ca95 Fix previous merge. 2016-10-14 17:16:28 +02:00
Gael Guennebaud
050c681bdd Merged in rmlarsen/eigen2 (pull request PR-232) 
Improve performance of parallelized matrix multiply for rectangular matrices
 2016-10-14 14:51:09 +00:00
Benoit Steiner
737e4152c3 Merged in lukier/eigen (pull request PR-234) 
Enabling CUDA in Geometry
 2016-10-13 18:09:28 +00:00
Robert Lukierski
a94791b69a Fixes for min and abs after Benoit's comments, switched to numext. 2016-10-13 15:00:22 +01:00
Avi Ginsburg
ac63d6891c Patch to allow VS2015 & CUDA 8.0 to compile with Eigen included. I'm not sure 
whether to limit the check to this compiler combination
(` || (EIGEN_COMP_MSVC == 1900 &&  __CUDACC_VER__) `)
or to leave it as it is. I also don't know if this will have any affect on
including Eigen in device code (I'm not in my current project).
 2016-10-13 08:47:32 +00:00
Benoit Steiner
38b6048e14 Deleted redundant implementation of predux 2016-10-12 14:37:56 -07:00
Benoit Steiner
78d2926508 Merged eigen/eigen into default 2016-10-12 13:46:29 -07:00
Benoit Steiner
2e2f48e30e Take advantage of AVX512 instructions whenever possible to speedup the processing of 16 bit floats. 2016-10-12 13:45:39 -07:00
Robert Lukierski
471075f7ad Fixes min() warnings. 2016-10-12 18:59:05 +01:00
Gael Guennebaud
5c366fe1d7 Merged in rmlarsen/eigen (pull request PR-230) 
Fix a bug in psqrt for SSE and AVX when EIGEN_FAST_MATH=1
 2016-10-12 16:30:51 +00:00
Robert Lukierski
86711497c4 Adding EIGEN_DEVICE_FUNC in the Geometry module. 
Additional CUDA necessary fixes in the Core (mostly usage of
EIGEN_USING_STD_MATH).
 2016-10-12 16:35:17 +01:00
Rasmus Munk Larsen
47150af1c8 Fix copy-paste error: Must use _mm256_cmp_ps for AVX. 2016-10-12 08:34:39 -07:00
Gael Guennebaud
89e315152c bug #1325: fix compilation on NEON with clang 2016-10-12 16:55:47 +02:00
Benoit Steiner
5727e4d89c Reenabled the use of variadic templates on tegra x1 provides that the latest version (i.e. JetPack 2.3) is used. 2016-10-08 22:19:03 +00:00
Benoit Steiner
507b661106 Renamed predux_half into predux_downto4 2016-10-06 17:57:04 -07:00
Benoit Steiner
a498ff7df6 Fixed incorrect comment 2016-10-06 15:27:27 -07:00
Benoit Steiner
a7473d6d5a Fixed compilation error with gcc >= 5.3 2016-10-06 14:33:22 -07:00
Benoit Steiner
5e64cea896 Silenced a compilation warning 2016-10-06 14:24:17 -07:00
Benoit Steiner
d485d12c51 Added missing AVX intrinsics for fp16: in particular, implemented predux which is required by the matrix-vector code. 2016-10-06 10:41:03 -07:00
Rasmus Munk Larsen
48c635e223 Add a simple cost model to prevent Eigen's parallel GEMM from using too many threads when the inner dimension is small. 
Timing for square matrices is unchanged, but both CPU and Wall time are significantly improved for skinny matrices. The benchmarks below are for multiplying NxK * KxN matrices with test names of the form BM_OuterishProd/N/K.

Improvements in Wall time:

Run on [redacted] (12 X 3501 MHz CPUs); 2016-10-05T17:40:02.462497196-07:00
CPU: Intel Haswell with HyperThreading (6 cores) dL1:32KB dL2:256KB dL3:15MB
Benchmark                          Base (ns)  New (ns) Improvement
------------------------------------------------------------------
BM_OuterishProd/64/1                    3088      1610    +47.9%
BM_OuterishProd/64/4                    3562      2414    +32.2%
BM_OuterishProd/64/32                   8861      7815    +11.8%
BM_OuterishProd/128/1                  11363      6504    +42.8%
BM_OuterishProd/128/4                  11128      9794    +12.0%
BM_OuterishProd/128/64                 27691     27396     +1.1%
BM_OuterishProd/256/1                  33214     28123    +15.3%
BM_OuterishProd/256/4                  34312     36818     -7.3%
BM_OuterishProd/256/128               174866    176398     -0.9%
BM_OuterishProd/512/1                7963684    104224    +98.7%
BM_OuterishProd/512/4                7987913    112867    +98.6%
BM_OuterishProd/512/256              8198378   1306500    +84.1%
BM_OuterishProd/1k/1                 7356256    324432    +95.6%
BM_OuterishProd/1k/4                 8129616    331621    +95.9%
BM_OuterishProd/1k/512              27265418   7517538    +72.4%

Improvements in CPU time:

Run on [redacted] (12 X 3501 MHz CPUs); 2016-10-05T17:40:02.462497196-07:00
CPU: Intel Haswell with HyperThreading (6 cores) dL1:32KB dL2:256KB dL3:15MB
Benchmark                          Base (ns)  New (ns) Improvement
------------------------------------------------------------------
BM_OuterishProd/64/1                    6169      1608    +73.9%
BM_OuterishProd/64/4                    7117      2412    +66.1%
BM_OuterishProd/64/32                  17702     15616    +11.8%
BM_OuterishProd/128/1                  45415      6498    +85.7%
BM_OuterishProd/128/4                  44459      9786    +78.0%
BM_OuterishProd/128/64                110657    109489     +1.1%
BM_OuterishProd/256/1                 265158     28101    +89.4%
BM_OuterishProd/256/4                 274234    183885    +32.9%
BM_OuterishProd/256/128              1397160   1408776     -0.8%
BM_OuterishProd/512/1               78947048    520703    +99.3%
BM_OuterishProd/512/4               86955578   1349742    +98.4%
BM_OuterishProd/512/256             74701613  15584661    +79.1%
BM_OuterishProd/1k/1                78352601   3877911    +95.1%
BM_OuterishProd/1k/4                78521643   3966221    +94.9%
BM_OuterishProd/1k/512              258104736  89480530    +65.3%
 2016-10-06 10:33:10 -07:00
Benoit Steiner
4131074818 Deleted unecessary CMakeLists.txt file 2016-10-05 18:54:35 -07:00
Benoit Steiner
cb5cd69872 Silenced a compilation warning. 2016-10-05 18:50:53 -07:00
Benoit Steiner
78b569f685 Merged latest updates from trunk 2016-10-05 18:48:55 -07:00
Benoit Steiner
9c2b6c049b Silenced a few compilation warnings 2016-10-05 18:37:31 -07:00
Benoit Steiner
698ff69450 Properly characterize the CUDA packet primitives for fp16 as device only 2016-10-04 16:53:30 -07:00
Rasmus Munk Larsen
7f67e6dfdb Update comment for fast sqrt. 2016-10-04 15:09:11 -07:00
Rasmus Munk Larsen
765615609d Update comment for fast sqrt. 2016-10-04 15:08:41 -07:00
Rasmus Munk Larsen
3ed67cb0bb Fix a bug in the implementation of Carmack's fast sqrt algorithm in Eigen (enabled by EIGEN_FAST_MATH), which causes the vectorized parts of the computation to return -0.0 instead of NaN for negative arguments. 
Benchmark speed in Giga-sqrts/s
Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz
-----------------------------------------
                    SSE        AVX
Fast=1              2.529G     4.380G
Fast=0              1.944G     1.898G
Fast=1 fixed        2.214G     3.739G

This table illustrates the worst case in terms speed impact: It was measured by repeatedly computing the sqrt of an n=4096 float vector that fits in L1 cache. For large vectors the operation becomes memory bound and the differences between the different versions almost negligible.
 2016-10-04 14:22:56 -07:00
Benoit Steiner
881b90e984 Use explicit type casting to generate packets of zeros. 2016-10-04 08:23:38 -07:00