[gomp4] More deferral of partitioning to target
Nathan Sidwell
nathan@acm.org
Wed Oct 14 12:57:00 GMT 2015
I've committed this to gomp4 branch.
It is another step towards deferring partitioned execution choices to the
target compiler -- though sadly not the last step.
At early omp lowering, we now attach partitioning flags to the HEAD_MARK
function I introduced yesterday, and adjust the partition enter/exit sequence to
not be level specific.
At the oacc_device_lower stage, the loop structure is augmented to hold up to 3
levels of head & tail markers, allowing a single loop to be partitioned over up
to 3 axes. Once constructed we then iterate of the marker sequences replacing
the dummy axis argument of the appropriate builtins with the specific chosen axis.
The next step is to iterate over the body doing the same for the loop
abstraction builtin.
nathan
-------------- next part --------------
A non-text attachment was scrubbed...
Name: gomp4-loop-ht.patch
Type: text/x-patch
Size: 15080 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20151014/7c977a4a/attachment.bin>
More information about the Gcc-patches
mailing list