8720aecb80
Offload more loops
2020-06-10 12:57:55 -04:00
e97f3688db
Fix the HMC issue - kernel was launchnig asynchronously
2020-06-08 17:01:15 -04:00
89a1e78390
Merge branch 'sycl' of https://github.com/paboyle/Grid into sycl
2020-06-05 23:20:37 -04:00
5a73ef3647
Minor tweak to compile
2020-06-05 21:50:15 -04:00
87e5d2f4b7
Merge branch 'sycl' of https://www.github.com/paboyle/Grid into sycl
2020-06-05 17:32:21 -07:00
d720f10758
Liink error fix
2020-06-05 17:29:20 -07:00
14fcd0912a
Merge branch 'sycl' of https://github.com/paboyle/Grid into sycl
2020-06-05 19:14:17 -04:00
3111c0bd4f
Single precisiono hardwire
2020-06-05 19:13:27 -04:00
e03064490e
Merge branch 'sycl' of https://github.com/paboyle/Grid into sycl
2020-06-05 18:53:39 -04:00
1a4c8c3387
Global edit with change to View usage. autoView() creates a wrapper object that closes the view when scope closes.
2020-06-05 18:52:35 -04:00
2b1e259441
Decode of SYCL devices fix
2020-06-04 17:16:55 -07:00
f39c2a240b
Priintinig and device memory size detection
2020-06-04 14:58:03 -04:00
0d95805cde
Print improvement
2020-06-03 22:50:32 -04:00
f67830587f
Accelerator loop use
2020-06-03 22:50:09 -04:00
6bf7f839ff
Better printing and logging
2020-06-03 09:28:57 -04:00
e3147881a9
Cache scheme
2020-06-03 09:23:48 -04:00
fb559614ad
Initialise meemory manager
2020-06-03 09:12:47 -04:00
e93e12b6a4
More verbose SYCL setup
2020-06-03 09:12:11 -04:00
0c3112cd94
Use view mechanism
2020-06-03 09:11:51 -04:00
8cfd5d2639
Need lattice view
2020-06-03 09:11:28 -04:00
1c9f20b15e
Views must be closed
2020-06-03 09:10:29 -04:00
32237895bd
Reorg memory manager for O(1) hash table
2020-06-03 09:09:52 -04:00
1d252d0922
Accelerator inline
2020-05-28 11:45:25 -04:00
006cc8a8f1
Staggereed move to accelerator
2020-05-28 08:33:06 -04:00
cf2938688a
Sycl unhappy fix
2020-05-25 08:36:53 -07:00
ee63721bad
int unhappiness sycl fix
2020-05-25 08:36:24 -07:00
22c5168d70
Sycl happier
2020-05-25 08:35:56 -07:00
949ac3cd24
Must avoid non-trivial copy constructors
2020-05-25 08:35:28 -07:00
7bc0166c1c
SYCLL maknig happy - must avoid non ttrivial copy constructors
2020-05-25 08:34:19 -07:00
cb0d1b3399
hopefullly fix buildd fail
2020-05-24 21:27:00 -04:00
d1f1ccc705
HIP changes
2020-05-24 21:18:49 -04:00
c7519a237a
Assertions fail on HIP foor unknown reasons - dedbugging
2020-05-24 14:02:47 -04:00
32be2b13d3
Updates for HiP
2020-05-24 14:00:55 -04:00
92b342a477
Hip reduction too
2020-05-24 13:50:28 -04:00
556da86ac3
HIP fp16
2020-05-24 13:41:58 -04:00
8285e41574
View location / access mode
2020-05-21 16:14:41 -04:00
f999408e92
View locatoin and access mode
2020-05-21 16:14:20 -04:00
a7abda89e2
View location & access mode
2020-05-21 16:13:59 -04:00
7860a50f70
Make view specify where and drive data motion - first cut.
...
This is a compile tiime option --enable-unified=yes/no
2020-05-21 16:13:16 -04:00
ebb60330c9
Automatic data motion options beginning
2020-05-17 16:34:25 -04:00
a9847aa866
Dependence fix
2020-05-12 20:03:37 -04:00
d24d8e8398
Use X-direction as more bits meaningful on CUDA.
...
2^31-1 shoulddd always bee enough for SIMD and thread reduced local volume
e.g. 32*2^31 = 2^36 = (2^9)^4 or 512^4 ias big enough.
Where 32 is gpu_threads * Nsimd = 8*4
2020-05-12 10:35:49 -04:00
07c0c02f8c
Speed up Cshift
2020-05-11 17:02:01 -04:00
8c31c065b5
Keep the Vector fixed to protect it from realloc
2020-05-11 17:00:30 -04:00
bbbee5660d
First compiile on HiP
2020-05-10 05:28:09 -04:00
52081acfa5
NVCC compile fixes
2020-05-08 13:14:12 -04:00
f8b8e00090
Systematise the accelerator primitives and locate to Grid/threads/Accelerator.h / Accelerator.cc
...
Aim to reduce the amount of cuda and other code variations floating around all over the place.
Will move GpuInit iinto Accelerator.cc from Init.cc
Need to worry about SharedMemoryMPI.cc and the Peer2Peer windows
2020-05-08 06:23:55 -07:00
28a1fcaaff
First compile against SYCL
2020-05-05 11:13:27 -07:00
04927d2e40
SYCL prep - no sycl just make it compile through DPC++
2020-05-04 10:28:29 -07:00
7caed4edd9
dpc++ didn't like rdtsc()
2020-05-04 10:27:05 -07:00
Peter Boyle
u37294