… | |
… | |
304 | |
304 | |
305 | glXSwapBuffers; |
305 | glXSwapBuffers; |
306 | |
306 | |
307 | select undef, undef, undef, 1/60; |
307 | select undef, undef, undef, 1/60; |
308 | } |
308 | } |
|
|
309 | |
|
|
310 | =head2 How to modify the previous example to not rely on GL sharing. |
|
|
311 | |
|
|
312 | For those poor souls with only a sucky CPU OpenCL implementation, you |
|
|
313 | currently have to read the image into some perl scalar, and then modify a |
|
|
314 | texture or use glDrawPixels or so). |
|
|
315 | |
|
|
316 | First, when you don't need gl sharing, you can create the context much simpler: |
|
|
317 | |
|
|
318 | $ctx = $platform->context (undef, [$dev]) |
|
|
319 | |
|
|
320 | To use a texture, you would modify the above example by creating an |
|
|
321 | OpenCL::Image manually instead of deriving it from a texture: |
|
|
322 | |
|
|
323 | my $tex = $ctx->image2d (OpenCL::MEM_WRITE_ONLY, OpenCL::RGBA, OpenCL::UNORM_INT8, $S, $S); |
|
|
324 | |
|
|
325 | And in the darw loop, intead of acquire_gl_objects/release_gl_objects, you |
|
|
326 | would read the image2d after the kernel has written it: |
|
|
327 | |
|
|
328 | $queue->read_image ($tex, 0, 0, 0, 0, $S, $S, 1, 0, 0, my $data); |
|
|
329 | |
|
|
330 | And then you would upload the pixel data to the texture (or use glDrawPixels): |
|
|
331 | |
|
|
332 | glTexSubImage2D_s GL_TEXTURE_2D, 0, 0, 0, $S, $S, GL_RGBA, GL_UNSIGNED_BYTE, $data; |
|
|
333 | |
|
|
334 | The fully modified example can be found at |
|
|
335 | L<http://cvs.schmorp.de/OpenCL/examples/juliaflight-nosharing>. |
309 | |
336 | |
310 | =head1 DOCUMENTATION |
337 | =head1 DOCUMENTATION |
311 | |
338 | |
312 | =head2 BASIC CONVENTIONS |
339 | =head2 BASIC CONVENTIONS |
313 | |
340 | |
… | |
… | |
530 | @OpenCL::Image1D::ISA = |
557 | @OpenCL::Image1D::ISA = |
531 | @OpenCL::Image1DArray::ISA = |
558 | @OpenCL::Image1DArray::ISA = |
532 | @OpenCL::Image1DBuffer::ISA = OpenCL::Image::; |
559 | @OpenCL::Image1DBuffer::ISA = OpenCL::Image::; |
533 | |
560 | |
534 | @OpenCL::UserEvent::ISA = OpenCL::Event::; |
561 | @OpenCL::UserEvent::ISA = OpenCL::Event::; |
|
|
562 | |
|
|
563 | @OpenCL::MappedBuffer::ISA = |
|
|
564 | @OpenCL::MappedImage::ISA = OpenCL::Mapped::; |
535 | } |
565 | } |
536 | |
566 | |
537 | =head2 THE OpenCL PACKAGE |
567 | =head2 THE OpenCL PACKAGE |
538 | |
568 | |
539 | =over 4 |
569 | =over 4 |
… | |
… | |
1128 | |
1158 | |
1129 | Creates a new OpenCL::Program object from the given source code. |
1159 | Creates a new OpenCL::Program object from the given source code. |
1130 | |
1160 | |
1131 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateProgramWithSource.html> |
1161 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateProgramWithSource.html> |
1132 | |
1162 | |
|
|
1163 | =item ($program, \@status) = $ctx->program_with_binary (\@devices, \@binaries) |
|
|
1164 | |
|
|
1165 | Creates a new OpenCL::Program object from the given binaries. |
|
|
1166 | |
|
|
1167 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clCreateProgramWithBinary.html> |
|
|
1168 | |
|
|
1169 | Example: clone an existing program object that contains a successfully |
|
|
1170 | compiled program, no matter how useless this is. |
|
|
1171 | |
|
|
1172 | my $clone = $ctx->program_with_binary ([$prog->devices], [$prog->binaries]); |
|
|
1173 | |
1133 | =item $packed_value = $ctx->info ($name) |
1174 | =item $packed_value = $ctx->info ($name) |
1134 | |
1175 | |
1135 | See C<< $platform->info >> for details. |
1176 | See C<< $platform->info >> for details. |
1136 | |
1177 | |
1137 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetContextInfo.html> |
1178 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clGetContextInfo.html> |
… | |
… | |
1207 | |
1248 | |
1208 | =item $ev = $queue->write_buffer_rect (OpenCL::Memory buf, cl_bool blocking, $buf_x, $buf_y, $buf_z, $host_x, $host_y, $host_z, $width, $height, $depth, $buf_row_pitch, $buf_slice_pitch, $host_row_pitch, $host_slice_pitch, $data, $wait_events...) |
1249 | =item $ev = $queue->write_buffer_rect (OpenCL::Memory buf, cl_bool blocking, $buf_x, $buf_y, $buf_z, $host_x, $host_y, $host_z, $width, $height, $depth, $buf_row_pitch, $buf_slice_pitch, $host_row_pitch, $host_slice_pitch, $data, $wait_events...) |
1209 | |
1250 | |
1210 | http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueWriteBufferRect.html |
1251 | http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueWriteBufferRect.html |
1211 | |
1252 | |
|
|
1253 | =item $ev = $queue->copy_buffer_to_image ($src_buffer, $dst_image, $src_offset, $dst_x, $dst_y, $dst_z, $width, $height, $depth, $wait_events...) |
|
|
1254 | |
|
|
1255 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueCopyBufferToImage.html> |
|
|
1256 | |
1212 | =item $ev = $queue->read_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $slice_pitch, $data, $wait_events...) |
1257 | =item $ev = $queue->read_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $slice_pitch, $data, $wait_events...) |
1213 | |
1258 | |
1214 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueCopyBufferRect.html> |
1259 | C<$row_pitch> (and C<$slice_pitch>) can be C<0>, in which case the OpenCL |
1215 | |
1260 | module uses the image width (and height) to supply default values. |
1216 | =item $ev = $queue->copy_buffer_to_image ($src_buffer, $dst_image, $src_offset, $dst_x, $dst_y, $dst_z, $width, $height, $depth, $wait_events...) |
|
|
1217 | |
1261 | |
1218 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueReadImage.html> |
1262 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueReadImage.html> |
1219 | |
1263 | |
1220 | =item $ev = $queue->write_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $slice_pitch, $data, $wait_events...) |
1264 | =item $ev = $queue->write_image ($src, $blocking, $x, $y, $z, $width, $height, $depth, $row_pitch, $slice_pitch, $data, $wait_events...) |
1221 | |
1265 | |
|
|
1266 | C<$row_pitch> (and C<$slice_pitch>) can be C<0>, in which case the OpenCL |
|
|
1267 | module uses the image width (and height) to supply default values. |
1222 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueWriteImage.html> |
1268 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueWriteImage.html> |
1223 | |
1269 | |
1224 | =item $ev = $queue->copy_image ($src_image, $dst_image, $src_x, $src_y, $src_z, $dst_x, $dst_y, $dst_z, $width, $height, $depth, $wait_events...) |
1270 | =item $ev = $queue->copy_image ($src_image, $dst_image, $src_x, $src_y, $src_z, $dst_x, $dst_y, $dst_z, $width, $height, $depth, $wait_events...) |
1225 | |
1271 | |
1226 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueCopyImage.html> |
1272 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueCopyImage.html> |
… | |
… | |
1330 | =item $command_queue_properties = $command_queue->properties |
1376 | =item $command_queue_properties = $command_queue->properties |
1331 | |
1377 | |
1332 | Calls C<clGetCommandQueueInfo> with C<CL_QUEUE_PROPERTIES> and returns the result. |
1378 | Calls C<clGetCommandQueueInfo> with C<CL_QUEUE_PROPERTIES> and returns the result. |
1333 | |
1379 | |
1334 | =for gengetinfo end command_queue |
1380 | =for gengetinfo end command_queue |
|
|
1381 | |
|
|
1382 | =back |
|
|
1383 | |
|
|
1384 | =head3 MEMORY MAPPED BUFFERS |
|
|
1385 | |
|
|
1386 | OpenCL allows you to map buffers and images to host memory (read: perl |
|
|
1387 | scalars). This is done much like reading or copying a buffer, by enqueuing |
|
|
1388 | a map or unmap operation on the command queue. |
|
|
1389 | |
|
|
1390 | The map operations return an C<OpenCL::Mapped> object - see L<THE |
|
|
1391 | OpenCL::Mapped CLASS> section for details on what to do with these |
|
|
1392 | objects. |
|
|
1393 | |
|
|
1394 | The object will be unmapped automatically when the mapped object is |
|
|
1395 | destroyed (you can use a barrier to make sure the unmap has finished, |
|
|
1396 | before using the buffer in a kernel), but you can also enqueue an unmap |
|
|
1397 | operation manually. |
|
|
1398 | |
|
|
1399 | =over 4 |
|
|
1400 | |
|
|
1401 | =item $mapped_buffer = $queue->map_buffer ($buf, $blocking=1, $map_flags=OpenCL::MAP_READ|OpenCL::MAP_WRITE, $offset=0, $size=undef, $wait_events...) |
|
|
1402 | |
|
|
1403 | Maps the given buffer into host memory and returns an |
|
|
1404 | C<OpenCL::MappedBuffer> object. If C<$size> is specified as undef, then |
|
|
1405 | the map will extend to the end of the buffer. |
|
|
1406 | |
|
|
1407 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueMapBuffer.html> |
|
|
1408 | |
|
|
1409 | Example: map the buffer $buf fully and replace the first 4 bytes by "abcd", then unmap. |
|
|
1410 | |
|
|
1411 | { |
|
|
1412 | my $mapped = $queue->map_buffer ($buf, 1, OpenCL::MAP_WRITE); |
|
|
1413 | substr $$mapped, 0, 4, "abcd"; |
|
|
1414 | } # asynchronously unmap because $mapped is destroyed |
|
|
1415 | |
|
|
1416 | =item $mapped_image = $queue->map_image ($img, $blocking=1, $map_flags=OpenCL::MAP_READ|OpenCL::MAP_WRITE, $x=0, $y=0, $z=0, $width=undef, $height=undef, $depth=undef, $wait_events...) |
|
|
1417 | |
|
|
1418 | Maps the given image area into host memory and return an |
|
|
1419 | C<OpenCL::MappedImage> object. |
|
|
1420 | |
|
|
1421 | If any of C<$width>, C<$height> and/or C<$depth> are C<undef> then they |
|
|
1422 | will be replaced by the maximum possible value. |
|
|
1423 | |
|
|
1424 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clEnqueueMapImage.html> |
|
|
1425 | |
|
|
1426 | Example: map an image (with OpenCL::UNSIGNED_INT8 channel type) and set |
|
|
1427 | the first channel of the leftmost column to 5, then explicitly unmap |
|
|
1428 | it. You are not necessarily meant to do it this way, this example just |
|
|
1429 | shows you the accessors to use :) |
|
|
1430 | |
|
|
1431 | my $mapped = $queue->map_image ($image, 1, OpenCL::MAP_WRITE); |
|
|
1432 | |
|
|
1433 | $mapped->set ($_ * $mapped->row_pitch, pack "C", 5) |
|
|
1434 | for 0..$image->height; |
|
|
1435 | |
|
|
1436 | $mapped->unmap;. |
|
|
1437 | $mapped->wait; # only needed for out of order queues normally |
|
|
1438 | |
|
|
1439 | =item $ev = $queue->unmap ($mapped, $wait_events...) |
|
|
1440 | |
|
|
1441 | Unmaps the data from host memory. You must not call any methods that |
|
|
1442 | modify the data, or modify the data scalar directly, after calling this |
|
|
1443 | method. |
|
|
1444 | |
|
|
1445 | The mapped event object will always be passed as part of the |
|
|
1446 | $wait_events. The mapped event object will be replaced by the new event |
|
|
1447 | object that this request creates. |
1335 | |
1448 | |
1336 | =back |
1449 | =back |
1337 | |
1450 | |
1338 | =head2 THE OpenCL::Memory CLASS |
1451 | =head2 THE OpenCL::Memory CLASS |
1339 | |
1452 | |
… | |
… | |
1883 | |
1996 | |
1884 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetUserEventStatus.html> |
1997 | L<http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetUserEventStatus.html> |
1885 | |
1998 | |
1886 | =back |
1999 | =back |
1887 | |
2000 | |
|
|
2001 | =head2 THE OpenCL::Mapped CLASS |
|
|
2002 | |
|
|
2003 | This class represents objects mapped into host memory. They are |
|
|
2004 | represented by a blessed string scalar. The string data is the mapped |
|
|
2005 | memory area, that is, if you read or write it, then the mapped object is |
|
|
2006 | accessed directly. |
|
|
2007 | |
|
|
2008 | You must only ever use operations that modify the string in-place - for |
|
|
2009 | example, a C<substr> that doesn't change the length, or maybe a regex that |
|
|
2010 | doesn't change the length. Any other operation might cause the data to be |
|
|
2011 | copied. |
|
|
2012 | |
|
|
2013 | When the object is destroyed it will enqueue an implicit unmap operation |
|
|
2014 | on the queue that was used to create it. |
|
|
2015 | |
|
|
2016 | Keep in mind that you I<need> to unmap (or destroy) mapped objects before |
|
|
2017 | OpenCL sees the changes, even if some implementations don't need this |
|
|
2018 | sometimes. |
|
|
2019 | |
|
|
2020 | Example, replace the first two floats in the mapped buffer by 1 and 2. |
|
|
2021 | |
|
|
2022 | my $mapped = $queue->map_buffer ($buf, ... |
|
|
2023 | $mapped->event->wait; # make sure it's there |
|
|
2024 | |
|
|
2025 | # now replace first 8 bytes by new data, which is exactly 8 bytes long |
|
|
2026 | # we blindly assume device endianness to equal host endianness |
|
|
2027 | # (and of course, we assume iee 754 single precision floats :) |
|
|
2028 | substr $$mapped, 0, 8, pack "f*", 1, 2; |
|
|
2029 | |
|
|
2030 | =over 4 |
|
|
2031 | |
|
|
2032 | =item $ev = $mapped->unmap ($wait_events...) |
|
|
2033 | |
|
|
2034 | Unmaps the mapped memory object, using the queue originally used to create |
|
|
2035 | it, quite similarly to C<< $queue->unmap ($mapped, ...) >>. |
|
|
2036 | |
|
|
2037 | =item $bool = $mapped->mapped |
|
|
2038 | |
|
|
2039 | Returns whether the object is still mapped - true before an C<unmap> is |
|
|
2040 | enqueued, false afterwards. |
|
|
2041 | |
|
|
2042 | =item $ev = $mapped->event |
|
|
2043 | |
|
|
2044 | Return the event object associated with the mapped object. Initially, this |
|
|
2045 | will be the event object created when mapping the object, and after an |
|
|
2046 | unmap, this will be the event object that the unmap operation created. |
|
|
2047 | |
|
|
2048 | =item $mapped->wait |
|
|
2049 | |
|
|
2050 | Same as C<< $mapped->event->wait >> - makes sure no operations on this |
|
|
2051 | mapped object are outstanding. |
|
|
2052 | |
|
|
2053 | =item $bytes = $mapped->size |
|
|
2054 | |
|
|
2055 | Returns the size of the mapped area, in bytes. Same as C<length $$mapped>. |
|
|
2056 | |
|
|
2057 | =item $ptr = $mapped->ptr |
|
|
2058 | |
|
|
2059 | Returns the raw memory address of the mapped area. |
|
|
2060 | |
|
|
2061 | =item $mapped->set ($offset, $data) |
|
|
2062 | |
|
|
2063 | Replaces the data at the given C<$offset> in the memory area by the new |
|
|
2064 | C<$data>. This method is safer than direct manipulation of C<$mapped> |
|
|
2065 | because it does bounds-checking, but also slower. |
|
|
2066 | |
|
|
2067 | =item $data = $mapped->get ($offset, $length) |
|
|
2068 | |
|
|
2069 | Returns (without copying) a scalar representing the data at the given |
|
|
2070 | C<$offset> and C<$length> in the mapped memory area. This is the same as |
|
|
2071 | the following substr, except much slower; |
|
|
2072 | |
|
|
2073 | $data = substr $$mapped, $offset, $length |
|
|
2074 | |
|
|
2075 | =cut |
|
|
2076 | |
|
|
2077 | sub OpenCL::Mapped::get { |
|
|
2078 | substr ${$_[0]}, $_[1], $_[2] |
|
|
2079 | } |
|
|
2080 | |
|
|
2081 | =back |
|
|
2082 | |
|
|
2083 | =head2 THE OpenCL::MappedBuffer CLASS |
|
|
2084 | |
|
|
2085 | This is a subclass of OpenCL::Mapped, representing mapped buffers. |
|
|
2086 | |
|
|
2087 | =head2 THE OpenCL::MappedImage CLASS |
|
|
2088 | |
|
|
2089 | This is a subclass of OpenCL::Mapped, representing mapped images. |
|
|
2090 | |
|
|
2091 | =over 4 |
|
|
2092 | |
|
|
2093 | =item $bytes = $mapped->row_pitch |
|
|
2094 | |
|
|
2095 | =item $bytes = $mapped->slice_pitch |
|
|
2096 | |
|
|
2097 | Return the row or slice pitch of the image that has been mapped. |
|
|
2098 | |
|
|
2099 | =back |
|
|
2100 | |
|
|
2101 | |
1888 | =cut |
2102 | =cut |
1889 | |
2103 | |
1890 | 1; |
2104 | 1; |
1891 | |
2105 | |
1892 | =head1 AUTHOR |
2106 | =head1 AUTHOR |