2
0
mirror of https://github.com/boostorg/compute.git synced 2026-02-18 01:52:20 +00:00
Commit Graph

1177 Commits

Author SHA1 Message Date
Kyle Lutz
5c1b688cca Merge pull request #640 from haahh/pr_system_platforms_fix
Fix system::platforms()
2016-08-15 18:36:07 -07:00
Jakub Szuppe
90861687d1 Fix system::platforms()
The 2nd call to clGetPlatformsIDs crashed when there
were no OpenCL platforms because then count variable was 0
and clGetPlatformIDs specification says: "If platforms is
not NULL, the num_entries must be greater than zero".
2016-08-12 11:58:49 +02:00
Kyle Lutz
882277f735 Merge pull request #638 from boostorg/fix-warnings
Fix warnings
2016-08-06 11:42:38 -07:00
Kyle Lutz
05ab43ed58 Cast index argument to cl_uint in kernel::get_arg_info()
This casts the 'index' argument in kernel::get_arg_info() to cl_uint
in order to match the signature for clGetKernelArgInfo(). This fixes
the "conversion from 'size_t' to 'cl_uint', possible loss of data"
compiler warning.
2016-08-03 20:16:19 -07:00
Kyle Lutz
6688e92f07 Change size_t argument to uint_ in opengl_enqueue_*_gl_objects() functions
This changes the opengl_enqueue_*_gl_objects() functions to take their
'num_objects' argument as 'uint_' instead of 'size_t'. This fixes the
"conversion from 'size_t' to 'cl_uint', possible loss of data" compiler
warnings from MSVC.
2016-07-30 18:28:41 -07:00
Kyle Lutz
c4b3793be1 Ignore unused arguments in kernel::set_arg_svm_ptr()
This ignores unused arguments in the kernel::set_arg_svm_ptr()
when compiled without OpenCL 2.0 support. This fixes the "unused
parameter" compiler warning.
2016-07-30 18:19:26 -07:00
Kyle Lutz
75daaf67d1 Remove const from return type of context_error::get_private_info_size()
This removes the const specifier from the return type of the
context_error::get_private_info_size() method. This fixes the "type
qualifiers ignored on function return type" compiler warning.
2016-07-30 18:14:56 -07:00
Kyle Lutz
b8fdabab19 Merge pull request #636 from dPavelDev/develop
Fixed bug in opengl_create_shared_context
2016-07-27 21:14:38 -07:00
Kyle Lutz
b4f8494ec2 Merge pull request #634 from haahh/pr_scan_on_cpu
Scan for multicore CPUs
2016-07-27 21:13:05 -07:00
dPavelDev
e978396b21 Fixed bug in opengl_create_shared_context 2016-07-27 14:36:57 +03:00
Jakub Szuppe
979ec52351 Add more tests for scan 2016-07-25 22:37:17 +02:00
Jakub Szuppe
790dcab387 Fix benchmark for std::partial_sum()
In partial_sum benchmark for STL std::parial_sum() should not
be run in-place since benchmarks for other libs are not run
in-place.
2016-07-25 22:37:17 +02:00
Jakub Szuppe
32dc57bbf9 Add scan for multicore CPUs 2016-07-25 22:37:17 +02:00
Jakub Szuppe
cd5bb606ec Rename scan_on_cpu to serial_scan 2016-07-25 15:28:19 +02:00
Kyle Lutz
1d2f8cfbd9 Merge pull request #633 from haahh/pr_find_extrema_cpu
find_extrema for CPUs
2016-07-20 22:23:51 -07:00
Kyle Lutz
813a63c4ca Merge pull request #632 from haahh/pr_various_test_fixes
Various fixes
2016-07-20 22:16:03 -07:00
Jakub Szuppe
06124180dc Fix find_extrema_with_reduce()
It fixes wrong type of variable representing index.
2016-07-17 13:24:07 +02:00
Jakub Szuppe
934257b97d Add more tests for find_extrema algorithms 2016-07-17 13:24:07 +02:00
Jakub Szuppe
4c21633bf9 Fix vector copy ctor when using non-default context
Now correct queue (from correct context) is used in copy
constructor of boost::compute::vector.
2016-07-17 00:01:47 +02:00
Jakub Szuppe
2af23fa7e9 Change various tests to work with non-default context
Now all tests work even if Context::context, Context::device,
Context::queue are not default context, device and queue. This
is required for developing better tests in the future.

Note: Some tests may work only for default context/queue/device
since classes that they test work only for default context/q/d.
There are two solutions for this problem: either those tests run
on default queue (no matter what) or they does not run when
Context::context is not the default context. See test_string.cpp.
2016-07-16 20:30:28 +02:00
Jakub Szuppe
fb641de7cc Add operator= for vectors with different allocator
Now code: `vector<T> a(context); vector<T, OtherAlloc> b(context);
a = b;` works.
2016-07-16 20:14:14 +02:00
Jakub Szuppe
8ae5cc9117 Array should always use queue from its context
Until now boost::compute::array worked correctly only
for default context (global) since it was using default
queue (global) for all its operations. Now it uses queue
from its context.
2016-07-16 19:30:25 +02:00
Jakub Szuppe
b810d12aea Making sure serial_reduce_threshold is in the right range
Parameter serial_reduce_threshold for reducing on CPU
should be always greater or equal to the number of compute units
(cores) in the CPU.
2016-07-16 13:02:34 +02:00
Jakub Szuppe
04b6e8fba3 find_extrema() optimized for multicore CPUs 2016-07-16 13:02:34 +02:00
Kyle Lutz
d303097000 Merge pull request #626 from haahh/pr_transform_cpu
Transform for multicore CPUs
2016-07-11 20:16:53 -07:00
Kyle Lutz
6e3213bfcf Merge pull request #628 from haahh/pr_reduce_cpu
Reduce for CPU
2016-07-11 20:16:19 -07:00
Kyle Lutz
3bcc318609 Merge pull request #631 from thtrummer/fix-uint-typo
Fix build error due to possible typo (on macOS)
2016-07-11 20:01:43 -07:00
Kyle Lutz
b1d3e0af83 Merge pull request #629 from thtrummer/develop
Add missing include for std::cerr (on macOS)
2016-07-11 20:01:13 -07:00
Thomas Trummer
a800dd33a1 Fix build error due to possible typo (on macOS) 2016-07-10 19:54:45 +02:00
Thomas Trummer
68d373c38a Add missing include for std::cerr (on macOS) 2016-07-10 18:38:23 +02:00
Jakub Szuppe
a10e7d31f7 Disable CPU-optimized transform/copy_on_device() on Apple
Yet another bug on Apple OpenCL Platform.
2016-07-10 16:17:48 +02:00
Jakub Szuppe
4093481920 Remove unnecessary check 2016-07-10 13:56:39 +02:00
Jakub Szuppe
97f0225072 Fix ‘local_vals_arg’ may be uninitialized warning 2016-07-09 14:39:47 +02:00
Jakub Szuppe
585eb747a0 More tests for adjacent difference 2016-07-09 14:11:01 +02:00
Jakub Szuppe
9e2fc74826 Use serial_reduce for small inputs
For inputs smaller than <serial_reduce_threshold>
serial_reduce algorithm is used.
2016-07-09 14:06:06 +02:00
Jakub Szuppe
a085e6b3e1 Reduce for multicore CPUs 2016-07-09 14:06:06 +02:00
Jakub Szuppe
34c476c87a Tranform/copy on device optimized for CPUs 2016-07-06 19:23:30 +02:00
Kyle Lutz
a3f72e6191 Merge pull request #622 from haahh/pr_adjacent_diff_fix
Fix adjacent_difference when first == result
2016-07-05 16:52:40 -07:00
Jakub Szuppe
c6123c403c Remove unused function 2016-07-05 13:40:42 +02:00
Jakub Szuppe
92a16e459c Fix adjacent_difference when first == result
Situation where first == result was only covered
when binary function was not specified.
2016-07-01 21:25:46 +02:00
Kyle Lutz
e5a9017d69 Merge pull request #620 from haahh/pr_pocl_fix_llvm_apt
Temp. fix for llvm apt being down
2016-06-25 09:15:49 -07:00
Kyle Lutz
b3633eb54b Merge pull request #619 from haahh/pr_comp_sort
Comparison sort (for GPUs)
2016-06-23 20:24:38 -07:00
Jakub Szuppe
a721db2f72 Temp. fix for llvm apt being down
See https://github.com/travis-ci/travis-ci/issues/6120
2016-06-23 15:56:36 +02:00
Jakub Szuppe
5d31bd0f07 Add tests for stable comp. sort and sort-by-key 2016-06-22 17:06:12 +02:00
Jakub Szuppe
c47753c4fa Stable comparison sort, sort-by-key 2016-06-22 17:06:12 +02:00
Jakub Szuppe
7996382178 Use index instead of copying vals in bitonic sort-by-key
Instead of copying values in local memory in bitonic block
sort (when sorting by key), we save and copy local index
and copy value once at the end. This saves local memory.
2016-06-22 15:59:55 +02:00
Jakub Szuppe
1311de15a3 Add benchmark for comparison sort 2016-06-22 15:59:55 +02:00
Jakub Szuppe
cc1cc0e345 Add tests for comparison sort 2016-06-21 17:41:36 +02:00
Jakub Szuppe
8f033e72ee Add comparison sort and sort by key 2016-06-21 17:41:36 +02:00
Kyle Lutz
04e7dde220 Merge pull request #617 from haahh/pr_travis_ci_boost161
Add Travis CI build with latest Boost libs
2016-06-03 20:09:32 -07:00