| 1 |
Revision history for Perl extension OpenCL. |
| 2 |
|
| 3 |
TODO: http://www.khronos.org/registry/cl/sdk/1.1/docs/man/xhtml/clSetMemObjectDestructorCallback.html |
| 4 |
TODO: some convenience methods maybe? |
| 5 |
TODO: OpenCL::context |
| 6 |
TODO: remove set_image23d |
| 7 |
TODO: http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clCompileProgram.html |
| 8 |
TODO: http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clLinkProgram.html |
| 9 |
TODO: http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clCreateProgramWithBuiltInKernels.html |
| 10 |
TODO: http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clCreateSubDevices.html |
| 11 |
TODO: http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clEnqueueMigrateMemObjects.html |
| 12 |
TODO: http://www.khronos.org/registry/cl/sdk/1.2/docs/man/xhtml/clGetKernelArgInfo.html (also generic accessor) |
| 13 |
|
| 14 |
- OpenCL 1.1 is now effectively feature complete. |
| 15 |
- work around a bug in the nvidia implementation, where build fails with |
| 16 |
INVALID_BINARY if the compiler runs succesfully but ptxas fails. |
| 17 |
- explicitly check for 0 size dimensions in nd_range_kernel, as nvidia |
| 18 |
crashes (and OpenCL probably allows this behaviour). |
| 19 |
- actually report clSetKernelArg errors in ("OpenCL::Kernel::setf. |
| 20 |
- setf properly can now properly set local arguments ('z'). |
| 21 |
- using undef to pass null memory buffers now works in set_xxx. |
| 22 |
- release_gl_texture was missing. |
| 23 |
- fix/improve/balance gl sharing example. |
| 24 |
- added non-opengl sharing opengl example. |
| 25 |
- fixed some documentation links. |
| 26 |
- fixed a bug where enqueue operations with optional parameters at the |
| 27 |
end not specified would crash. |
| 28 |
- implemented clEnqueueMapBuffer/Image/UnmapMemObject, and a "mapped object" |
| 29 |
abstraction. |
| 30 |
- both context_from_type methods now properly pass the notify callback. |
| 31 |
- $platform->context now properly passes CL_CONTEXT_PLATFORM again. |
| 32 |
- implement clCreateProgramWithBinary. |
| 33 |
- speed up object constructions by not looking up stashes. |
| 34 |
- use libebc for fine-tuning. |
| 35 |
- reduce codesize considerably by disabling inlining on some functions. |
| 36 |
- speed up many method calls on objects with no subclasses. |
| 37 |
|
| 38 |
0.98 Sat Apr 28 15:31:46 CEST 2012 |
| 39 |
- INCOMPATIBLE CHANGE: $prog->build now wants an array of devices, or undef. |
| 40 |
- INCOMPATIBLE CHANGE: $queue->enqueue_xxx methods have been renamed to just |
| 41 |
$queue->xxx, the enqueue_ prefix has been deprecated. |
| 42 |
- INCOMPATIBLE CHANGE: use ->set_image to set kernel args, ->image[23]d have |
| 43 |
been deprecated. |
| 44 |
- new $kernel->setf function to set kernel arguments in one go. |
| 45 |
- (optional, automatic) anyevent integration. |
| 46 |
- implement an event queue, use Async::Interrupt for communication. |
| 47 |
- implement context notification callbacks (untested). |
| 48 |
- provide a default context notification function that prints to stderr. |
| 49 |
- implement event callback notifications. |
| 50 |
- implement build program notifications. |
| 51 |
- implement an asynchronous program build function. |
| 52 |
- $ctx->build_program now distinguishes between wrong build invocations |
| 53 |
and build failures. |
| 54 |
|
| 55 |
0.97 Tue Apr 24 16:56:21 CEST 2012 |
| 56 |
- created $ctx->build_program utility function. |
| 57 |
- enqueue_acquire/release_gl_objects did not return an event object. |
| 58 |
- OpenCL::errno did not return the correct value. |
| 59 |
- allow undef values in event wait lists, for convenience. |
| 60 |
- implement $program->kernels_in_program. |
| 61 |
- add opencl 1.2 constants. |
| 62 |
- implement $ctx->image, $ctx->gl_texture, $platform->unload_compiler, |
| 63 |
$queue->enqueue_fill_buffer/image, all untested (opencl 1.2). |
| 64 |
- port image2d and image3d to the opencl 1.2 API. |
| 65 |
- port enqueue_marker/barrier to the opencl 1.2 API, and croak |
| 66 |
if the user requests something that 1.1 cannot do when compiled |
| 67 |
against 1.1. This slightly changes the API for enqueue_barrier, |
| 68 |
which now checks the context it is called in. |
| 69 |
- add ->format accessor for CL_IMAGHE_FORMAT. |
| 70 |
- $object->id function to access internal object id. |
| 71 |
|
| 72 |
0.96 Fri Apr 20 22:23:58 CEST 2012 |
| 73 |
- ->enqueue_write_buffer actually called enqueue_read_buffer. d'oh. |
| 74 |
- new $kernel->set_local function, to size local buffer arguments. |
| 75 |
- use finish, not flush, in examples. |
| 76 |
|
| 77 |
0.95 Thu Apr 19 21:25:32 CEST 2012 |
| 78 |
- add lengthy opengl sharing example. |
| 79 |
- support property lists in CreateContext* calls. |
| 80 |
- support some opengl sharing extensions |
| 81 |
(gl_buffer, gl_texture[23]d, gl_renderbuffer). |
| 82 |
- support some cl_khr_gl_sharing extensions. |
| 83 |
- add some extension constants (cl_gl.h cl_ext.h, cl_d3d10.h). |
| 84 |
- auto-generate error codes (and include gl+d3d extensions). |
| 85 |
|
| 86 |
0.92 Mon Apr 16 08:39:21 CEST 2012 |
| 87 |
- work around some bugs in amd's opencl implementation (INVALID_HOST_PTR |
| 88 |
when value deosn't matter). |
| 89 |
- minor documentation fix (patch by Maik Hentsche). |
| 90 |
|
| 91 |
0.91 Sun Dec 11 00:05:02 CET 2011 |
| 92 |
- introduce $buffer_obj->sub_buffer_region method and a new class, OpenCL::BufferObj |
| 93 |
for a subbuffer is a buffer without sub_buffer methods! |
| 94 |
- support apple using a different location for opencl headers - untested. |
| 95 |
- be more strict when passing event lists. |
| 96 |
|
| 97 |
0.59 Tue Nov 22 11:44:32 CET 2011 |
| 98 |
- provide $program->binaries method. |
| 99 |
- provide $queue->enqueue_read_buffer_rect method. |
| 100 |
- provide $queue->enqueue_write_buffer_rect method. |
| 101 |
- coalesce like-typed accessor methods into the same xs function and |
| 102 |
multiple ALIASes - this reduces the OpenCL codesize from 162kb to |
| 103 |
132kb on my system. |
| 104 |
- minor doc improvements. |
| 105 |
|
| 106 |
0.55 Sun Nov 20 11:19:25 CET 2011 |
| 107 |
- semi-automatically generate query methods for almost all clGet*Info names. |
| 108 |
- provide cl_ext.h constants. |
| 109 |
- provide OpenCL::Image->image_info method. |
| 110 |
- provide OpenCL::Event->profiling_info method. |
| 111 |
- provide OpenCL::Kernel->work_group_info method. |
| 112 |
|
| 113 |
0.15 Thu Nov 17 07:22:11 CET 2011 |
| 114 |
- include slice_pitch in enqueue_write_image (can be 0). |
| 115 |
- check length in enqueue_write_image. |
| 116 |
- properly calculate size in enqueue_read_image. |
| 117 |
- likewise, include row_pitch and slice_pitch in image2d and image3d, |
| 118 |
which must be nonzero right now, too, sorry. |
| 119 |
- allow undef or missing data scalar for image2d and image3d. |
| 120 |
- fix enqueue_copy_image prototype. |
| 121 |
- fix formatting. |
| 122 |
|
| 123 |
0.14 Thu Nov 17 04:04:23 CET 2011 |
| 124 |
- major incompatibilities due to API cleanup, the API should now be |
| 125 |
mostly stable, if incomplete. |
| 126 |
- fix prerequisites. |
| 127 |
- replaced all _simple functions by "full" implementations, with some |
| 128 |
parameters marked for future expansion - that means their API |
| 129 |
should be stable from now on: |
| 130 |
OpenCL::context_from_type_simple => OpenCL::context_from_type |
| 131 |
$platform->context_from_type_simple => $platform->context_from_type |
| 132 |
$dev->context_simple => $platform->context |
| 133 |
$ctx->command_queue_simple became => $ctx->queue |
| 134 |
- add OpenCL::context function. |
| 135 |
- officially require C99. |
| 136 |
- errno value is unspecified after a successful call. |
| 137 |
|
| 138 |
0.03 Wed Nov 16 01:44:26 CET 2011 |
| 139 |
- add some reference documentation for all functions. |
| 140 |
- add user event support. |
| 141 |
- implement event wait lists. |
| 142 |
|
| 143 |
0.02 Tue Nov 15 22:15:47 CET 2011 |
| 144 |
- first opencl kernel executed, some example code. |
| 145 |
|
| 146 |
0.01 Tue Nov 15 05:59:00 CET 2011 |
| 147 |
- cloned from Convert::Scalar. |
| 148 |
|
