]> git.ipfire.org Git - thirdparty/gcc.git/blobdiff - libgomp/ChangeLog
Update GCC to autoconf 2.69, automake 1.15.1 (PR bootstrap/82856).
[thirdparty/gcc.git] / libgomp / ChangeLog
index a3ed6ff2a20689dca5f0051226777e6916d88007..43e27013a70000957699c5e09b4b448e5ff2e898 100644 (file)
@@ -1,3 +1,200 @@
+2018-10-31  Joseph Myers  <joseph@codesourcery.com>
+
+       PR bootstrap/82856
+       * Makefile.am: Include multilib.am
+       (AUTOMAKE_OPTIONS): Add info-in-builddir.
+       (CLEANFILES): Remove libgomp.info.
+       * configure.ac: Remove AC_PREREQ.
+       * testsuite/Makefile.am (RUNTEST): Remove quotes.
+       * Makefile.in, aclocal.m4, configure, testsuite/Makefile.in:
+       Regenerate.
+
+2018-10-29  Joseph Myers  <joseph@codesourcery.com>
+           Julian Brown  <julian@codesourcery.com>
+
+       * testsuite/libgomp.oacc-c++/this.C: New.
+
+2018-09-18  Cesar Philippidis  <cesar@codesourcery.com>
+
+       * plugin/plugin-nvptx.c (struct cuda_map): New.
+       (struct ptx_stream): Replace d, h, h_begin, h_end, h_next, h_prev,
+       h_tail with (cuda_map *) map.
+       (cuda_map_create): New function.
+       (cuda_map_destroy): New function.
+       (map_init): Update to use a linked list of cuda_map objects.
+       (map_fini): Likewise.
+       (map_pop): Likewise.
+       (map_push): Likewise.  Return CUdeviceptr instead of void.
+       (init_streams_for_device): Remove stales references to ptx_stream
+       members.
+       (select_stream_for_async): Likewise.
+       (nvptx_exec): Update call to map_init.
+
+2018-09-09  Cesar Philippidis  <cesar@codesourcery.com>
+            Julian Brown  <julian@codesourcery.com>
+
+       PR middle-end/86336
+       * testsuite/libgomp.oacc-c++/non-scalar-data.C: Remove XFAIL.
+
+2018-08-21  Nicolas Koenig  <koenigni@gcc.gnu.org>
+       Thomas Koenig <tkoenig@gcc.gnu.org>
+
+       PR fortran/25829
+       * testsuite/libgomp.fortran/async_io_1.f90: New test.
+       * testsuite/libgomp.fortran/async_io_2.f90: New test.
+       * testsuite/libgomp.fortran/async_io_3.f90: New test.
+       * testsuite/libgomp.fortran/async_io_4.f90: New test.
+       * testsuite/libgomp.fortran/async_io_5.f90: New test.
+       * testsuite/libgomp.fortran/async_io_6.f90: New test.
+       * testsuite/libgomp.fortran/async_io_7.f90: New test.
+
+2018-08-13  Cesar Philippidis  <cesar@codesourcery.com>
+           Tom de Vries  <tdevries@suse.de>
+
+       PR target/85590
+       * plugin/cuda/cuda.h (CUoccupancyB2DSize): New typedef.
+       (cuOccupancyMaxPotentialBlockSize): Declare.
+       * plugin/cuda-lib.def (cuOccupancyMaxPotentialBlockSize): New
+       CUDA_ONE_CALL_MAYBE_NULL.
+       * plugin/plugin-nvptx.c (CUDA_VERSION < 6050): Define
+       CUoccupancyB2DSize and declare
+       cuOccupancyMaxPotentialBlockSize.
+       (nvptx_exec): Use cuOccupancyMaxPotentialBlockSize to set the
+       default num_gangs and num_workers when the driver supports it.
+
+2018-08-08  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/cuda-lib.def (cuLinkAddData_v2, cuLinkCreate_v2): Declare using
+       CUDA_ONE_CALL_MAYBE_NULL.
+       * plugin/plugin-nvptx.c (cuLinkAddData, cuLinkCreate): Undef and declare.
+       (cuLinkAddData_v2, cuLinkCreate_v2): Declare.
+       (link_ptx): Fall back to cuLinkAddData/cuLinkCreate if the _v2 versions
+       are not found.
+
+2018-08-08  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/cuda-lib.def (cuGetErrorString): Use CUDA_ONE_CALL_MAYBE_NULL.
+       * plugin/plugin-nvptx.c (cuda_error): Handle if cuGetErrorString is not
+       present.
+
+2018-08-08  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c
+       (CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR): Define.
+       (nvptx_open_device): Use
+       CU_DEVICE_ATTRIBUTE_MAX_REGISTERS_PER_MULTIPROCESSOR.
+
+2018-08-08  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (cuda_error): Move declaration of cuGetErrorString ...
+       (cuGetErrorString): ... here.  Guard with CUDA_VERSION < 6000.
+
+2018-08-07  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (DO_PRAGMA): Define.
+       (struct cuda_lib_s): Add def/undef of CUDA_ONE_CALL_MAYBE_NULL.
+       (init_cuda_lib): Add new param to CUDA_ONE_CALL_1.  Add arg to
+       corresponding call in CUDA_ONE_CALL.  Add def/undef of
+       CUDA_ONE_CALL_MAYBE_NULL.
+       (CUDA_CALL_EXISTS): Define.
+
+2018-08-07  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (struct cuda_lib_s, init_cuda_lib): Put
+       CUDA_ONE_CALL defines right before the cuda-lib.def include, and the
+       corresponding undefs right after.
+
+2018-08-04  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/configfrag.ac: For --without-cuda-driver, set
+       CUDA_DRIVER_INCLUDE and CUDA_DRIVER_LIB to no.  Handle
+       CUDA_DRIVER_INCLUDE == no and CUDA_DRIVER_LIB == no.
+       * configure: Regenerate.
+
+2018-08-02  Tom de Vries  <tdevries@suse.de>
+
+       PR target/86660
+       * testsuite/libgomp.oacc-c++/routine-1-auto.C: Remove -fno-exceptions.
+       * testsuite/libgomp.oacc-c++/routine-1-template-auto.C: Same.
+       * testsuite/libgomp.oacc-c++/routine-1-template-trailing-return-type.C:
+       Same.
+       * testsuite/libgomp.oacc-c++/routine-1-template.C: Same.
+       * testsuite/libgomp.oacc-c++/routine-1-trailing-return-type.C: Same.
+       * testsuite/libgomp.oacc-c-c++-common/routine-1.c: Same.
+
+2018-08-01  Cesar Philippidis  <cesar@codesourcery.com>
+           Thomas Schwinge <thomas@codesourcery.com>
+
+       * config/nvptx/oacc-parallel.c: Truncate.
+
+2018-08-01  Cesar Philippidis  <cesar@codesourcery.com>
+           James Norris <jnorris@codesourcery.com>
+
+       * plugin/plugin-nvptx.c (struct map): Removed.
+       (map_init, map_pop): Remove use of struct map.
+       (map_push): Likewise and change argument list.
+       * testsuite/libgomp.oacc-c-c++-common/mapping-1.c: New
+
+2018-08-01  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/cuda-lib.def: New file.  Factor out of ...
+       * plugin/plugin-nvptx.c (CUDA_CALLS): ... here.
+       (struct cuda_lib_s, init_cuda_lib): Include cuda-lib.def instead of
+       using CUDA_CALLS.
+
+2018-07-31  Andre Vieira  <andre.simoesdiasvieira@arm.com>
+
+       Revert 'AsyncI/O patch committed'.
+       2018-07-25  Nicolas Koenig  <koenigni@gcc.gnu.org>
+               Thomas Koenig <tkoenig@gcc.gnu.org>
+
+       PR fortran/25829
+       * testsuite/libgomp.fortran/async_io_1.f90: New test.
+       * testsuite/libgomp.fortran/async_io_2.f90: New test.
+       * testsuite/libgomp.fortran/async_io_3.f90: New test.
+       * testsuite/libgomp.fortran/async_io_4.f90: New test.
+       * testsuite/libgomp.fortran/async_io_5.f90: New test.
+       * testsuite/libgomp.fortran/async_io_6.f90: New test.
+       * testsuite/libgomp.fortran/async_io_7.f90: New test.
+
+2018-07-30  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (MIN, MAX): Redefine.
+       (nvptx_exec): Ensure worker and vector default dims don't exceed
+       targ_fn->max_threads_per_block.
+
+2018-07-30  Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (struct ptx_device): Add default_dims field.
+       (nvptx_open_device): Init default_dims for device.
+       (nvptx_exec): Use default_dims from device.
+
+2018-07-26  Jakub Jelinek  <jakub@redhat.com>
+
+       PR testsuite/86660
+       * testsuite/libgomp.c++/for-15.C (results): Include it in
+       omp declare target region.
+       (main): Use map (always, tofrom: results) instead of
+       map (tofrom: results).
+
+       PR middle-end/86660
+       * testsuite/libgomp.c/pr86660.c: New test.
+
+2018-07-26  Cesar Philippidis  <cesar@codesourcery.com>
+           Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (nvptx_exec): Error if the hardware doesn't have
+       sufficient resources to launch a kernel, and give a hint on how to fix
+       it.
+
+2018-07-26  Cesar Philippidis  <cesar@codesourcery.com>
+           Tom de Vries  <tdevries@suse.de>
+
+       * plugin/plugin-nvptx.c (struct ptx_device): Add warp_size,
+       max_threads_per_block and max_threads_per_multiprocessor fields.
+       (nvptx_open_device): Initialize new fields.
+       (nvptx_exec): Use num_sms, and new fields.
+
 2018-07-26  Tom de Vries  <tdevries@suse.de>
 
        * testsuite/libgomp.oacc-fortran/lib-12.f90: Move acc_async_test calls
        * config/nvptx/affinity.c: Delete to use fallback implementation.
 
 2016-11-23  Alexander Monakov  <amonakov@ispras.ru>
-            Jakub Jelinek  <jakub@redhat.com>
+           Jakub Jelinek  <jakub@redhat.com>
            Dmitry Melnik  <dm@ispras.ru>
 
        * Makefile.am (libgomp_la_SOURCES): Add atomic.c, icv.c, icv-device.c.
        * testsuite/libgomp.hsa.c/tiling-2.c: Likewise.
 
 2016-11-23  Martin Liska  <mliska@suse.cz>
-            Martin Jambor  <mjambor@suse.cz>
+           Martin Jambor  <mjambor@suse.cz>
 
        * plugin/hsa.h: New file.
        * plugin/hsa_ext_finalize.h: New file.
        * testsuite/Makefile.in: Likewise.
 
 2016-11-15  Martin Jambor  <mjambor@suse.cz>
-            Alexander Monakov  <amonakov@ispras.ru>
+           Alexander Monakov  <amonakov@ispras.ru>
 
        * testsuite/libgomp.fortran/examples-4/device-1.f90 (e_57_1): Add
        mapping clauses to target constructs.
 
 2016-05-16  Martin Jambor  <mjambor@suse.cz>
 
-        * testsuite/libgomp.hsa.c/complex-align-2.c: New test.
+       * testsuite/libgomp.hsa.c/complex-align-2.c: New test.
 
 2016-05-02  Nathan Sidwell  <nathan@codesourcery.com>