cuda.devices: "distribute-jobs" utility word that determines grid and block size...

diff --git a/extra/cuda/cuda.factor b/extra/cuda/cuda.factor
index 1b144632fbbee9da6521edf5b516c1233814dbd3..dd3f5b8f9e0470e6d9e3875804551aabaafa8640 100644 (file)
--- a/extra/cuda/cuda.factor
+++ b/extra/cuda/cuda.factor
@@ -7,7 +7,7 @@ destructors fry init io io.backend io.encodings.string
 io.encodings.utf8 kernel lexer locals macros math math.parser
 namespaces nested-comments opengl.gl.extensions parser
 prettyprint quotations sequences words cuda.libraries ;
-QUALIFIED-WITH: alien.c-types a
+QUALIFIED-WITH: alien.c-types c
 IN: cuda
 
 TUPLE: launcher
@@ -41,11 +41,11 @@ dim-grid dim-block shared-size stream ;
 
 : c-type>cuda-setter ( c-type -- n cuda-type )
     {
-        { [ dup a:int = ] [ drop 4 [ cuda-int* ] ] }
-        { [ dup a:uint = ] [ drop 4 [ cuda-int* ] ] }
-        { [ dup a:float = ] [ drop 4 [ cuda-float* ] ] }
-        { [ dup a:pointer? ] [ drop 4 [ cuda-int* ] ] }
-        { [ dup a:void* = ] [ drop 4 [ cuda-int* ] ] }
+        { [ dup c:int = ] [ drop 4 [ cuda-int* ] ] }
+        { [ dup c:uint = ] [ drop 4 [ cuda-int* ] ] }
+        { [ dup c:float = ] [ drop 4 [ cuda-float* ] ] }
+        { [ dup c:pointer? ] [ drop 4 [ cuda-int* ] ] }
+        { [ dup c:void* = ] [ drop 4 [ cuda-int* ] ] }
     } cond ;
 
 <PRIVATE
@@ -85,5 +85,5 @@ MACRO: cuda-arguments ( c-types -- quot: ( args... function -- ) )
             [ run-function-launcher ] 2bi
         ]
     ]
-    [ 2nip \ function-launcher suffix a:void function-effect ]
+    [ 2nip \ function-launcher suffix c:void function-effect ]
     3bi define-declared ;

diff --git a/extra/cuda/devices/devices-tests.factor b/extra/cuda/devices/devices-tests.factor
new file mode 100644 (file)
index 0000000..fc648d4
--- /dev/null
+++ b/extra/cuda/devices/devices-tests.factor
@@ -0,0 +1,13 @@
+! (c)2010 Joe Groff bsd license
+USING: cuda.devices tools.test ;
+IN: cuda.devices.tests
+
+[ 1  5 100 ] [  5 20 100 10 (distribute-jobs) ] unit-test 
+[ 2  5 100 ] [ 10 20 100 10 (distribute-jobs) ] unit-test 
+[ 2  5 100 ] [ 10 20 200  5 (distribute-jobs) ] unit-test 
+[ 2  5 100 ] [ 10 20 300  6 (distribute-jobs) ] unit-test 
+[ 2  6 120 ] [ 11 20 300  6 (distribute-jobs) ] unit-test 
+[ 1 10 200 ] [ 10 20 200 10 (distribute-jobs) ] unit-test 
+[ 1 10   0 ] [ 10  0 200 10 (distribute-jobs) ] unit-test 
+[ 2  5   0 ] [ 10  0 200  9 (distribute-jobs) ] unit-test 
+

diff --git a/extra/cuda/devices/devices.factor b/extra/cuda/devices/devices.factor
index e5f72f50d61a7ea6788b775a1da8c8f0d4d5d515..17e098c8df483c00887dcd0250d647d66f528118 100644 (file)
--- a/extra/cuda/devices/devices.factor
+++ b/extra/cuda/devices/devices.factor
@@ -1,9 +1,9 @@
 ! Copyright (C) 2010 Doug Coleman.
 ! See http://factorcode.org/license.txt for BSD license.
-USING: alien.c-types alien.data alien.strings arrays assocs
-byte-arrays classes.struct combinators cuda cuda.ffi cuda.utils
-fry io io.encodings.utf8 kernel math.parser prettyprint
-sequences ;
+USING: accessors alien.c-types alien.data alien.strings arrays
+assocs byte-arrays classes.struct combinators cuda cuda.ffi
+cuda.utils fry io io.encodings.utf8 kernel locals math
+math.order math.parser namespaces prettyprint sequences ;
 IN: cuda.devices
 
 : #cuda-devices ( -- n )
@@ -19,9 +19,8 @@ IN: cuda.devices
     [ enumerate-cuda-devices ] dip '[ <launcher> _ with-cuda ] each ; inline
 
 : cuda-device-properties ( n -- properties )
-    [ CUdevprop <c-object> ] dip
-    [ cuDeviceGetProperties cuda-error ] 2keep drop
-    CUdevprop memory>struct ;
+    [ CUdevprop <struct> ] dip
+    [ cuDeviceGetProperties cuda-error ] 2keep drop ;
 
 : cuda-devices ( -- assoc )
     enumerate-cuda-devices [ dup cuda-device-properties ] { } map>assoc ;
@@ -68,3 +67,20 @@ IN: cuda.devices
     "CUDA Version: " write cuda-version number>string print nl
     #cuda-devices iota [ nl ] [ cuda-device. ] interleave ;
 
+: up/i ( x y -- z )
+    [ 1 - + ] keep /i ; inline
+
+:: (distribute-jobs) ( job-count per-job-shared max-shared-size max-block-size
+                       -- grid-size block-size per-block-shared )
+    per-job-shared [ max-block-size ] [ max-shared-size swap /i max-block-size min ] if-zero
+        job-count min :> job-max-block-size
+    job-count job-max-block-size up/i :> grid-size
+    job-count grid-size up/i          :> block-size
+    block-size per-job-shared *       :> per-block-shared
+
+    grid-size block-size per-block-shared ; inline
+
+: distribute-jobs ( job-count per-job-shared -- grid-size block-size per-block-shared )
+    cuda-device get cuda-device-properties 
+    [ sharedMemPerBlock>> ] [ maxThreadsDim>> ] bi
+    (distribute-jobs) ; inline

diff --git a/extra/cuda/utils/utils.factor b/extra/cuda/utils/utils.factor
index a85a0b35e972d4e9725d2f438331f3b0b6912d81..269ebbe401659cf94b1171f23e533a756feb529c 100644 (file)
--- a/extra/cuda/utils/utils.factor
+++ b/extra/cuda/utils/utils.factor
@@ -1,8 +1,8 @@
 ! Copyright (C) 2010 Doug Coleman.
 ! See http://factorcode.org/license.txt for BSD license.
 USING: accessors alien.c-types alien.data alien.strings arrays
-assocs byte-arrays classes.struct combinators cuda.ffi io
-io.backend io.encodings.utf8 kernel math.parser namespaces
+assocs byte-arrays classes.struct combinators cuda.devices cuda.ffi
+io io.backend io.encodings.utf8 kernel math.parser namespaces
 prettyprint sequences ;
 IN: cuda.utils
 
@@ -92,5 +92,3 @@ ERROR: throw-cuda-error n ;
 : function-shared-size ( n -- )
     [ cuda-function get ] dip
     cuFuncSetSharedSize cuda-error ;
-
-: distribute-jobs ( job-count per-job-shared -- grid-size block-size per-block-shared )