fdurastante/psblas3
1
0
Fork 0
mirror of https://github.com/sfilippone/psblas3.git synced 2026-10-16 19:02:09 +00:00
Code Issues Packages Projects Releases Wiki Activity
3,286 Commits 105 Branches 48 Tags
cd0fa59647dfa07823ddfc8531c5fa060da3d927
Commit Graph
253 Commits
Author SHA1 Message Date
Stack-1 cd0fa59647 [UPDATE] Extended the 4-accumulator SpMV unrolling to s/c/z 
Generated from template-psblas, where the unrolled csmv_inner kernel was
added to the type-generic X_csr_impl.P90. The d version produced by the
template is byte-identical to the hand-written one in f7db723f, so this
commit only carries s/c/z.

Same dispatch as d: mean row length >= 16 nonzeros/row uses four
accumulators to break the FMA latency chain, shorter rows keep the
original loop. Applies to the beta==0/alpha==1 and beta==1/alpha==1
branches of psb_?_csr_csmv_inner.

Checked with mpif90 -O3 -fopenmp on all three types; the unrolled and
reference loops agree to rounding on rows of every length mod 4,
including empty rows.
 2026-08-07 12:40:46 +02:00
Stack-1 f7db723fdb [ADD] Added loop unrolling using 4 accumulators in csr SpMV 2026-07-12 17:07:54 +02:00
sfilippone d2fba9e952 Fix licensing text 2026-03-18 14:49:55 +01:00
Salvatore Filippone 25f73d15db Fix spspmm allocation to avoid double free error 2025-09-30 10:23:40 +02:00
sfilippone dc25c789ef Fixes for Intel compilation take 2. To be completed. 2025-09-18 20:27:26 +02:00
sfilippone da3f16ba56 Use less memory, check on symbmm 2025-07-09 11:30:41 +02:00
sfilippone 86a6ab413a Use less memory in spspmm 2025-06-27 14:03:46 +02:00
Salvatore Filippone 4a1610d705 Add check for allocation in certain OpenMP paths 2025-04-30 11:52:05 +02:00
sfilippone 922f7dd278 Tweak SIMD directives for CSR. 2025-04-08 13:40:33 +02:00
sfilippone 110a66e7d0 Switch off some debug write 2025-04-01 11:09:47 +02:00
sfilippone ea6c4181f7 Changed all defines with prefix PSB_ 2025-03-14 12:54:18 +01:00
sfilippone 701f297163 Silence warnings uncovered by INTEL 2025-02-27 16:55:06 +01:00
sfilippone 7700d4e0da Update copyright 2025-02-21 16:27:18 +01:00
sfilippone c9386c58d8 Remove default(none) from some OpenMP directives. 2025-02-10 15:49:49 +01:00
sfilippone 30810f804e Additional OpenMP tweaks, to be redesigned in the future. 2025-02-09 14:04:34 +01:00
sfilippone 772533b6f1 Avoid excessive OpenMP overhead in SPINS. Will have to be redesigned. 2025-01-23 15:38:05 +01:00
sfilippone 4e71fa971c Modify clean_zeros and have it only at base% 2025-01-08 17:18:54 +01:00
sfilippone 949499265e Simplify clean_zeros 2024-10-08 11:48:48 +02:00
sfilippone 1c235f9281 Improve clean_zeros 2024-10-04 10:28:52 +02:00
sfilippone 108d544fc1 Fix clean_zeros to always keep the diagonal 2024-10-04 08:37:45 +02:00
sfilippone 39cfcd3893 Fix allocation in coo_impl 2024-05-29 16:54:10 +02:00
sfilippone a38867be25 Fix allocation in coo_impl 2024-05-29 16:51:58 +02:00
sfilippone b9ad357648 Improve temp memory allocation in fix_coo 2024-05-29 16:25:49 +02:00
Salvatore Filippone 42293c62b6 Fix usage of sync() 2024-05-21 17:49:48 +02:00
sfilippone e19284eb6c Small omp addition 2024-05-16 10:10:08 +02:00
sfilippone 35096a2ef9 Cosmetic changes to coo_impl 2024-05-16 10:05:11 +02:00
sfilippone 70f51b9da8 Improve handling of fix_coo buffers with OpenMP 2024-05-15 12:05:15 +02:00
sfilippone ecccb13914 Fix COO fix_coo_inner_rowmajor not to overflow on integers. 2024-05-10 13:53:35 +02:00
sfilippone a613e963db First step in fix for coo_impl on OpenMP 2024-05-10 10:10:54 +02:00
sfilippone 4e0a9e5db8 Merge branch 'repackage' of github.com:sfilippone/psblas3 into repackage 2024-04-29 15:06:03 +02:00
sfilippone e72c0f0bf9 Fix OMP impl of sparse-sparse product 2024-04-29 15:04:23 +02:00
Salvatore Filippone d444a12879 Condition call to x%sync() in vect_mv 2024-04-26 11:35:44 +02:00
Salvatore Filippone 5e2e1e34fd Introduce set_host() in inner_vect_sv 2024-04-25 18:59:24 +02:00
sfilippone 025350a361 Make sure realloc is always called with size >0 2024-04-23 14:06:27 +02:00
sfilippone ba8c32c507 Define merge_nd method 2024-04-17 14:27:35 +02:00
sfilippone 3aa3c795e9 Refactor assembly and cnv 2023-12-23 13:15:01 +01:00
sfilippone 49e99a3e82 Fix conversion and product to enable overlap with GPU 2023-12-22 12:01:41 +01:00
sfilippone 74cf138a6c Merge branch 'repackage' into non-diag 2023-12-22 10:31:48 +01:00
sfilippone be7571f568 Fix missing directive 2023-12-20 13:45:43 +01:00
sfilippone 6c9ca58282 Silly bug in coo insert 2023-11-29 16:14:30 +01:00
sfilippone d3b2b7816d Fix coo insert OpenMP. Fix Make.inc.in 2023-11-29 15:35:55 +01:00
sfilippone ae7fad95d4 Merge branch 'development' into non-diag 2023-11-07 18:02:55 +01:00
sfilippone a6ec655a97 Prepare merge 2023-11-07 17:57:32 +01:00
sfilippone 41be1357c3 Set defaults for SPSPMM depending on OpenMP compilation. 2023-08-30 13:45:51 +02:00
Salvatore Filippone 0d8a5d3dc2 New SPSPMM implementation 2023-08-29 19:27:29 +02:00
Salvatore Filippone d0cacda995 Moved various modules related to RB around, into auxil, update Makefile. 2023-08-29 16:19:56 +02:00
wlthr 2322a9ce61 using end_idx to copy data from threads in gustavson and gustavson_1d 2023-08-28 11:06:30 +02:00
wlthr 0fe95c3c76 added use statement 2023-08-28 10:37:41 +02:00
wlthr 979a3da95f merged dev-openmp into omp-walther 2023-08-26 10:55:09 +02:00
wlthr 1af76c067c added parallel double precision spspmm implementations 2023-08-26 10:53:00 +02:00