This is a slightly-updated version of the following patch, but this
time tested (with the aid of a series of patches implementing PTX
support from Bernd Schmidt), and against the gomp4 branch:

https://gcc.gnu.org/ml/gcc-patches/2014-09/msg02022.html

Results (at least for the parts where the middle-end support is on the
branch already) are comparable with our local development branch.

Many of Jakub's initial review comments from the mainline version of
the patch have not yet been addressed, but I have a couple of bits ready
as follow-up patches and will be posting those shortly also. I plan to
address the remainder of the issues directly on the gomp4 branch, if
possible.

OK to apply (to the gomp4 branch)?

Thanks,

Julian

ChangeLog

xxxx-xx-xx  Nathan Sidwell  <nathan@codesourcery.com>
	    James Norris  <jnorris@codesourcery.com>
	    Thomas Schwinge  <thomas@codesourcery.com>
	    Tom de Vries  <tom@codesourcery.com>
	    Julian Brown  <julian@codesourcery.com>

    include/
    * gomp-constants.h: New file.

    libgomp/
    * Makefile.am (AM_CPPFLAGS): Search in ../include also.
    (libgomp_plugin_nvptx_version_info, libgomp_plugin_nvptx_la_SOURCES)
    (libgomp_plugin_nvptx_la_CPPFLAGS, libgomp_plugin_nvptx_la_LDFLAGS)
    (libgomp_plugin_nvptx_la_LIBADD)
    (libgomp_plugin_nvptx_la_LIBTOOLFLAGS): Set variables if
    PLUGIN_NVPTX is defined.
    (toolexeclib_LTLIBRARIES): Add nonshm-host
    and (conditionally) nvidia plugins.
    (libgomp_plugin_nonshm_host_version_info)
    (libgomp_plugin_nonshm_host_la_SOURCES)
    (libgomp_plugin_nonshm_host_la_CPPFLAGS)
    (libgomp_plugin_nonshm_host_la_LDFLAGS)
    (libgomp_plugin_nonshm_host_la_LIBTOOLFLAGS): Set variables.
    (libgomp_la_SOURCES): Add oacc-parallel.c, splay-tree.c,
    oacc-host.c, oacc-init.c, oacc-mem.c, oacc-async.c, oacc-plugin.c,
    oacc-cuda.c, libgomp-plugin.c.
    (nodist_libsubinclude_HEADERS): Add openacc.h,
    ../include/gomp-constants.h.
    * Makefile.in: Regenerate.
    * config.h.in: Regenerate.
    * configure.ac: Add TODOs for OpenACC in various places.
    (CUDA_DRIVER_CPPFLAGS, CUDA_DRIVER_LDFLAGS): Initialize.
    (--with-cuda-driver, --with-cuda-driver-include)
    (--with-cuda-driver-lib, --enable-offload-targets): Implement new
    options.
    (PLUGIN_NVPTX, PLUGIN_NVPTX_CPPFLAGS, PLUGIN_NVPTX_LDFLAGS)
    (PLUGIN_NVPTX_LIBS): Initialize variables.
    * configure: Regenerate.
    * env.c (target.h): Include.
    (goacc_device_num, goacc_device_type): New globals.
    (goacc_parse_device_num, goacc_parse_device_type): New functions.
    (initialize_env): Parse GCC_ACC_NOTIFY, ACC_DEVICE_TYPE,
    ACC_DEVICE_NUM environment variables.
    * error.c (gomp_verror, gomp_vfatal, gomp_vnotify, gomp_notify):
    New functions.
    (gomp_fatal): Make global.
    * libgomp.h (stdarg.h): Include.
    (struct gomp_memory_mapping): Forward declaration.
    (struct gomp_task_icv): Add acc_notify_var member.
    (goacc_device_num, goacc_device_type): Add extern declarations.
    (gomp_vnotify, gomp_notify, gomp_verror, gomp_vfatal): Add
    prototypes.
    (gomp_init_targets_once): Add prototype.
    * libgomp.map (OACC_2.0): New symbol version. Add public acc_*
    interface functions.
    (PLUGIN_1.0): New symbol version. Add gomp plugin interface
    functions.
    * libgomp_g.h (GOACC_kernels, GOACC_parallel): Update prototypes.
    (GOACC_wait): Add prototype.
    * target.c (limits.h, stdbool.h, stdlib.h): Don't include.
    (oacc-plugin.h, gomp-constants.h, stdio.h, assert.h): Include.
    (splay_tree_node, splay_tree, splay_tree_key, target_mem_desc)
    (splay_tree_key_s, enum target_type, gomp_device_descr): Don't
    declare here.
    (splay-tree.h): Include.
    (target.h): Include.
    (splay_compare): Change linkage to hidden not static.
    (gomp_init_targets_once): New function.
    (gomp_get_num_devices): Use above.
    (dump_mappings): New function (for debugging).
    (get_kind): New function.
    (gomp_map_vars): Add gomp_memory_mapping (mm), is_openacc
    parameters. Change KINDS to void *. Use lock from memory map
    not device. Use macros from gomp-constants.h instead of
    hard-coded values. Support OpenACC-specific mappings.
    (gomp_copy_from_async): New function.
    (gomp_unmap_vars): Add DO_COPYFROM argument. Only copy memory
    back from device if it is true. Use lock from memory map not
    device.
    (gomp_update): Add mm, is_openacc args. Use lock from
    memory map not device. Use macros from gomp-constants.h not
    hard-coded values.
    (gomp_register_image_for_device): Add forward
    declaration.
    (GOMP_offload_register): Change TARGET_DATA type to
    void **. Check realloc result.
    (gomp_init_device): Change linkage to hidden not static. Tweak mem
    map location.
    (gomp_fini_device): New function.
    (GOMP_target): Adjust lazy initialization, check target
    capabilities for OpenMP 4.0 support. Add locking around splay tree
    lookup. Add new arg to gomp_unmap_vars call.
    (GOMP_target_data): Tweak lazy initialization. Add new args to
    gomp_map_vars, gomp_unmap_vars calls.
    (GOMP_target_update): Tweak lazy initialization. Add new args to
    gomp_update call.
    (gomp_load_plugin_for_device): Initialize device_fini and
    OpenACC-specific plugin hooks.
    (gomp_register_images_for_device): Rename to...
    (gomp_register_image_for_device): This, and register a single
    device only, and only if it has not already had images registered.
    (gomp_find_available_plugins): Rearrange to fix plugin loading and
    initialization for OpenACC. Prefer a device with
    TARGET_CAP_OPENMP_400 for OpenMP.
    * target.h: New file.
    * splay-tree.h: Move bulk of implementation to...
    * splay-tree.c: New file.
    * libgomp-plugin.c: New file.
    * libgomp-plugin.h: New file.
    * oacc-async.c: New file.
    * oacc-cuda.c: New file.
    * oacc-host.c: New file.
    * oacc-init.c: New file.
    * oacc-mem.c: New file.
    * oacc-parallel.c: New file.
    * oacc-plugin.c: New file.
    * plugin-nvptx.c: New file.
    * oacc-int.h: New file.
    * openacc.f90: New file.
    * openacc.h: New file.
    * openacc_lib.h: New file.