Fix bitselect for float/double types v2
authorTom Stellard <thomas.stellard@amd.com>
Thu, 5 Mar 2015 15:31:05 +0000 (15:31 +0000)
committerTom Stellard <thomas.stellard@amd.com>
Thu, 5 Mar 2015 15:31:05 +0000 (15:31 +0000)
commitadfd96f742089dad11231fa7b7654f56140a3afb
tree5a3216716092fd4ca9b618f0886b80e4e34cb9ec
parentde05f10de27a05796c4e5b7283fbba7e757afb03
Fix bitselect for float/double types v2

We need to reinterpret float/double types as uint/ulong in order to
perform the bitwise operations.

This has been tested with piglit, OpenCV, and the ocl conformance tests.

v2:
  - Use vector operations rather than splitting vectors into scalar
    components.

Reviewed-by: Aaron Watry <awatry@gmail.com>
llvm-svn: 231373
libclc/generic/include/clc/relational/bitselect.h
libclc/generic/include/clc/relational/bitselect.inc [new file with mode: 0644]
libclc/generic/lib/SOURCES
libclc/generic/lib/relational/bitselect.cl [new file with mode: 0644]
libclc/generic/lib/relational/bitselect.inc [new file with mode: 0644]