Commit Graph

43 Commits

Author SHA1 Message Date
Rafael Ravedutti
43259eb3cf Adjust neighbor lists layout to keep neighbor ids contiguous in memory
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-30 01:57:26 +02:00
Rafael Ravedutti
59145644e3 Last changes to 2xnn kernels
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 23:34:07 +02:00
Rafael Ravedutti
b15aa2f461 Optimize 4xn kernels
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 23:00:21 +02:00
Rafael Ravedutti
5c000444a4 Pre-compute masks for 4xn kernels
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 22:30:30 +02:00
Rafael Ravedutti
04ade6bcec Pre-compute masks for 2xnn kernel with full neighbor-lists
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 19:33:26 +02:00
Rafael Ravedutti
85f1484449 Specialize force kernel when there are no masks to be checked
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 18:04:18 +02:00
Rafael Ravedutti
965fda3879 Pre-compute masks in the same way as in the master branch
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 17:32:42 +02:00
Rafael Ravedutti
a86d214c73 Add working version with old masking
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-28 02:19:46 +02:00
Rafael Ravedutti
d138f975f6 Add diagonal checks
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-23 02:17:27 +01:00
Rafael Ravedutti
296a4c4e01 Set interaction masks as gromacs does
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-03-23 00:58:25 +01:00
Yannick Paschke
c61cf9a0ac Move likwid marker calls into OpenMP parallel region 2023-01-22 15:33:05 +01:00
Rafael Ravedutti
07f2f74561 Adjust force_iters stats for 4xN kernel
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-01-02 23:57:51 +01:00
Rafael Ravedutti
fe86c948a8 Adjust time and likwid measurements on 4xN kernels
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2023-01-02 14:19:59 +01:00
Rafael Ravedutti
fa4e38c6c4 Add IACA and stubbed measurements for GROMACS 4x8 FN kernel
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-11-18 01:00:20 +01:00
Jan Eitzinger
3d0f4b97ee Switch copyright header in source files. 2022-09-05 10:39:42 +02:00
Rafael Ravedutti
e48b3fb653 Add option to check if cj is local before applying reaction force
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-04-04 21:52:40 +02:00
Rafael Ravedutti
0e742766b7 Add working version of Simd4xn kernel with half neighbor lists
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-23 15:54:18 +01:00
Rafael Ravedutti
94521f03b3 Fix reference version with half neighbor lists
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-23 14:31:47 +01:00
Rafael Ravedutti
8709bc2a06 Add first version for half neighbor lists in GROMACS variant
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-22 23:47:05 +01:00
Rafael Ravedutti
d4b34e1fa4 Fix intrinsics for AVX2
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-17 00:35:21 +01:00
Rafael Ravedutti
4090f43095 Optimize partial forces reduction for compute_4xn kernel
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-16 17:54:52 +01:00
Rafael Ravedutti
d47173d7a2 Fix Simd2xNN kernel
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-15 19:59:10 +01:00
Rafael Ravedutti
d61576699d Add first compilable version of Gromacs with SP
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-15 02:40:56 +01:00
Rafael Ravedutti
8669f2f6d7 Fix LJ Simd4xN kernel
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-11 01:12:59 +01:00
Rafael Ravedutti
d79c3c2a1d Add first working version with 4x8 config (ref kernel)
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-10 22:33:41 +01:00
Rafael Ravedutti
22d0f0b958 Commit version that works for M=N
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-10 01:31:50 +01:00
Rafael Ravedutti
2b441e691e Make code compilable
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-09 17:23:49 +01:00
Rafael Ravedutti
c7360305c8 Add first draft version of GROMACS method separating i-clusters and j-clusters
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-03-09 02:25:39 +01:00
Rafael Ravedutti
c62e4ea4ad Add clusters efficiency on stats
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-28 16:10:09 +01:00
Rafael Ravedutti
e637a26844 Add percentage of atoms within cutoff radius when using GROMACS reference version
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-25 14:19:48 +01:00
Rafael Ravedutti
6a35a7a482 Update stats for cluster version
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-08 00:55:27 +01:00
Rafael Ravedutti
cdb1d5b9f1 Add version with AVX2 intrinsics for gromacs scheme
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-04 17:52:48 +01:00
Rafael Ravedutti
34ce407f18 Update stats for gromacs scheme
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-04 14:47:37 +01:00
Rafael Ravedutti
7b90800a2b Setting forces to zero before calculation is not required
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-04 14:05:04 +01:00
Rafael Ravedutti
9daf9e5f4d Fix exclusion masks and add SIMD debug tools
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-02 21:54:18 +01:00
Rafael Ravedutti
4c5f013bf4 Assign masked adds results to forces
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-02 18:07:56 +01:00
Rafael Ravedutti
6ad1e58a3e Add first kernel using SIMD instrinsics for 4xn cases
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-02 18:00:44 +01:00
Rafael Ravedutti
5fd2d422ee Adjust kernels to work with MxN loops
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-02 00:49:55 +01:00
Rafael Ravedutti
85e7954932 Check all clusters in cell when building neighbor lists because ghost clusters may not be sorted
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-02-01 20:16:04 +01:00
Rafael Ravedutti
e64c3345bc Fix a few more bugs on gromacs variant
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-01-31 23:46:20 +01:00
Rafael Ravedutti
e0e6b6a68c Perform a few fixes for gromacs variant
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-01-31 17:49:22 +01:00
Rafael Ravedutti
6691803910 Add first version of force calculation with cluster scheme
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-01-28 18:07:41 +01:00
Rafael Ravedutti
df09c2861e Add first version with more than one optimization scheme
Signed-off-by: Rafael Ravedutti <rafaelravedutti@gmail.com>
2022-01-17 14:15:02 +01:00