Joe Groff
|
1a3fd1dc8c
|
improve cuda library organization
|
2010-05-20 14:40:26 -07:00 |
|
Joe Groff
|
e286a8daef
|
cuda.devices: factor "context-device cuda-device-properties" into a "context-device-properties" word
|
2010-05-16 17:16:27 -07:00 |
|
Joe Groff
|
a0f28d4402
|
cuda.gl.ffi: add missing LIBRARY: cuda befor ffi function defs
|
2010-05-15 22:33:03 -07:00 |
|
Joe Groff
|
5f909cf443
|
cuda.gl: enum>number the flags argument to gl-buffer>resource so that single CUDA flag symbols can be used as arguments
|
2010-05-15 16:26:49 -07:00 |
|
Joe Groff
|
c1dbce85bb
|
cuda: move init-cuda from cuda.utils
|
2010-05-15 15:28:02 -07:00 |
|
Joe Groff
|
896cd58451
|
cuda.demos.prefix-sum: add init-cuda
|
2010-05-15 15:01:32 -07:00 |
|
Joe Groff
|
22e853ecb0
|
cuda: more API cleanups:
- remove useless with-cuda and with-cuda-program combinators
- eliminate redundant cuda-device, cuda-context variables
- rearrange arguments of with-*cuda-context to ( device flags quot -- )
- don't pass context to with-cuda-context quot
- add context-device word to ask for current device
|
2010-05-15 11:47:19 -07:00 |
|
Joe Groff
|
e32e0f1423
|
cuda.gl: add free-resource word to release interop handles
|
2010-05-14 16:17:03 -07:00 |
|
Joe Groff
|
f9a8e90c41
|
cuda: add "cuda.gl" vocab with words for cuda/opengl/gpu interop
|
2010-05-14 16:00:27 -07:00 |
|
Joe Groff
|
afdced0183
|
cuda: in with-cuda-context, sync context when cleaning up so that destroying context doesn't fail due to asynchronous errors
|
2010-05-14 16:00:27 -07:00 |
|
Joe Groff
|
4eb55a5c3b
|
Merge branch 'master' of factorcode.org:/git/factor
|
2010-05-12 13:49:02 -07:00 |
|
Joe Groff
|
860e871fe6
|
new cuda.types vocab containing CUDA vector types (int2, float4, etc.) with CUDA alignment
|
2010-05-12 13:48:44 -07:00 |
Slava Pestov
|
d1e2554ebf
|
cuda.utils: fix load error
|
2010-05-10 20:50:28 -04:00 |
|
Joe Groff
|
6d30ce485c
|
cuda.utils: add "sync-context" word
|
2010-05-10 15:06:15 -07:00 |
|
Joe Groff
|
077cd0397b
|
cuda.devices: fix distribute-jobs to look only at width of maxThreadsDim
|
2010-05-07 18:15:37 -07:00 |
|
Joe Groff
|
6d41ea32f5
|
cuda.devices: have distribute-jobs construct a launcher with the calculated grid-dim, block-dim, and shared-size
|
2010-05-07 18:09:21 -07:00 |
|
Joe Groff
|
5a980b58cb
|
remove cuda.constants; it's better to get those values from device-properties
|
2010-05-07 18:03:30 -07:00 |
|
Joe Groff
|
f26a5836e6
|
cuda.devices: "distribute-jobs" utility word that determines grid and block size for a job based on per-thread shared memory requirements and max block size
|
2010-05-07 18:02:42 -07:00 |
|
Joe Groff
|
097100cb12
|
cuda.utils: foo <c-object> ... foo memory>struct == foo <struct>
|
2010-05-07 17:18:08 -07:00 |
|
Joe Groff
|
a228e575a5
|
cdua.devices: remove redundant init-cuda calls
|
2010-05-07 17:16:28 -07:00 |
|
Joe Groff
|
fb619ec928
|
cuda: fix using single integer as grid size
|
2010-05-05 19:07:39 -07:00 |
|
Joe Groff
|
ae37cc730b
|
cuda.memory: dependencies
|
2010-05-05 17:38:59 -07:00 |
|
Joe Groff
|
1d916246b0
|
cuda: rearrange function-launcher grid and block dim to be in the right order
|
2010-05-05 17:38:37 -07:00 |
|
Joe Groff
|
7c89fb2438
|
add cuda.constants vocab
|
2010-05-04 19:49:44 -07:00 |
|
Joe Groff
|
76512260b0
|
cuda.memory, cuda.syntax: inline some more words, add cuda-malloc-type word that allocates n*heap-size bytes of memory
|
2010-05-04 19:46:53 -07:00 |
|
Joe Groff
|
152da8c93c
|
cuda: streamline memory api
|
2010-05-03 18:18:10 -07:00 |
Doug Coleman
|
cc546c3a94
|
init-cuda in a couple more places...
|
2010-04-23 23:19:26 -05:00 |
|
Doug Coleman
|
8d76f2e612
|
Sprinkle init-cuda everywhere you could conceivably need it!
|
2010-04-23 23:17:47 -05:00 |
|
Doug Coleman
|
095763bcee
|
Move cuda.devices word out of cuda.utils
|
2010-04-23 18:27:13 -05:00 |
|
Doug Coleman
|
8323332838
|
Compile the .cu file with a relative pathname instead of absolute
|
2010-04-23 15:06:43 -05:00 |
|
Doug Coleman
|
b8c10dc2cd
|
Add using
|
2010-04-23 14:43:48 -05:00 |
|
Doug Coleman
|
b34c75a6a3
|
move cuda.utils library words to cuda.libraries
|
2010-04-23 14:43:13 -05:00 |
|
Doug Coleman
|
8ae600124c
|
remove init-cuda from a combinator because it's a startup-hook now
|
2010-04-23 14:29:24 -05:00 |
|
Doug Coleman
|
656f8987a1
|
Run hello-world on each CUDA device. fix a bug with returning the hello world string. add with-each-cuda-device combinator to run a program on each device. add an init-hook for cuda-init
|
2010-04-23 14:29:24 -05:00 |
|
Doug Coleman
|
d6036b6d5b
|
make host>device only take one parameter, add 2<<< for calling cuda functions where shared-memory is 0, simplify hello-world example
|
2010-04-23 14:29:23 -05:00 |
|
Joe Groff
|
a848b4b0c5
|
Merge branch 'symbolic-enums' of git://github.com/erikcharlebois/factor
|
2010-04-20 16:43:58 -07:00 |
|
Joe Groff
|
b71933f4d7
|
cuda.ptx: better representation of operands
|
2010-04-20 13:51:10 -07:00 |
|
Joe Groff
|
4bc915d526
|
C-ENUM: -> ENUM:
|
2010-04-19 20:07:16 -07:00 |
|
Joe Groff
|
1e7cc1613e
|
Merge branch 'master' of factorcode.org:/git/factor
|
2010-04-19 00:40:30 -07:00 |
|
Joe Groff
|
717dd1b10e
|
cuda.ptx: unit tests for instruction serialization
|
2010-04-19 00:40:10 -07:00 |
|
Doug Coleman
|
6b2024055f
|
Move some code to cuda.devices
|
2010-04-19 01:21:36 -05:00 |
|
Doug Coleman
|
1624903ae1
|
Split up cuda vocab some more, make CUDA-LIBRARY: work
|
2010-04-19 00:46:26 -05:00 |
|
Doug Coleman
|
eb3c6f414b
|
Fix the ptx path in cuda demo
|
2010-04-18 20:52:09 -05:00 |
|
Doug Coleman
|
5c17e6ee98
|
CUDA-FUNCTION: works, splitting up CUDA into more vocabs
|
2010-04-18 18:34:46 -05:00 |
|
Joe Groff
|
23cf6413dc
|
cuda.ptx: some unit tests
|
2010-04-18 11:59:37 -07:00 |
|
Joe Groff
|
0f2f54a195
|
cuda.ptx: rearrange maybe types so that tuple slots default to f properly, and adjust output of some insns
|
2010-04-17 16:10:53 -07:00 |
|
Joe Groff
|
f2de2222c7
|
cuda: AST representation for PTX code
|
2010-04-17 15:58:51 -07:00 |
|
Joe Groff
|
8501d5649e
|
cuda: add a single-block prefix sum sample kernel (so doug has a more substantive kernel to test with)
|
2010-04-16 16:42:46 -07:00 |
|
Joe Groff
|
8f56108702
|
cuda.ffi: add CUDA versions of double, longlong, ulonglong that always 8-byte align, and a >cuda-param-type function we can use to make structs that match kernel param space layout
|
2010-04-15 14:46:13 -07:00 |
|
Doug Coleman
|
c392ff2718
|
Working on a CUDA api. Add a hello world program.
|
2010-04-14 21:10:28 -05:00 |
|
Slava Pestov
|
e1ee2c82ea
|
cuda: 'not tested' tag
|
2010-04-13 18:44:10 -07:00 |
|
Doug Coleman
|
81c1e9fcb4
|
Playing around with a higher-level CUDA api. Also, changed some char* to c-string in cuda.ffi
|
2010-04-13 02:32:07 -07:00 |
|
Joe Groff
|
a46c624123
|
missed LIBRARY: cuda in cuda.ffi
|
2010-04-12 17:45:30 -07:00 |
|
Joe Groff
|
de50d0bfda
|
cuda.ffi binding
|
2010-04-12 17:37:06 -07:00 |