Merge branch 'sg_as_poc' of github.com:vladimirlaz/llvm into sg_as_poc

Author: vladimirlaz

sycl/doc/extensions/SubGroupAlgorithms/SYCL_INTEL_sub_group_algorithms.asciidoc

@@ -135,7 +135,7 @@ The load and store sub-group functions enable developers to assert that all work
 |Function|Description
 
 |+template <typename T> T load(sub_group sg, const T *src)+
-|Load contiguous data from _src_.  Returns one element per work-item, corresponding to the memory location at _src_ + +get_local_id()+. The value of _src_ must be the same for all work-items in the sub-group. The address space information is deduced autmaticaly. Only pointers to global and local address spaces are fully valid. Passing pointer in private address space will cause assertion. Other address spaces are casted to global with potentially undefined behaviour.
+|Load contiguous data from _src_.  Returns one element per work-item, corresponding to the memory location at _src_ + +get_local_id()+. The value of _src_ must be the same for all work-items in the sub-group. The address space information is deduced automatically. Only pointers to global and local address spaces are valid. Passing a pointer to private address space will cause an assertion. Other address spaces are cast to global with potentially undefined behavior.
 
 |+template <typename T, access::address_space Space> T load(sub_group sg, const multi_ptr<T,Space> src)+
 |Load contiguous data from _src_.  Returns one element per work-item, corresponding to the memory location at _src_ + +get_local_id()+. The value of _src_ must be the same for all work-items in the sub-group.  _Space_ must be +access::address_space::global_space+ or +access::address_space::local_space+.
@@ -144,7 +144,7 @@ The load and store sub-group functions enable developers to assert that all work
 |Load contiguous data from _src_.  Returns _N_ elements per work-item, corresponding to the _N_ memory locations at _src_ + +i+ * +get_max_local_range()+ + +get_local_id()+ for +i+ between 0 and _N_. The value of _src_ must be the same for all work-items in the sub-group.  _Space_ must be +access::address_space::global_space+ or +access::address_space::local_space+.
 
 |+template <typename T> void store(sub_group sg, T *dst, const T& x)+
-|Store contiguous data to _dst_.  The value of _x_ from each work-item is written to the memory location at _dst_ + +get_local_id()+. The value of _dst_ must be the same for all work-items in the sub-group. The address space information is deduced autmaticaly. Only pointers to global and local address spaces are fully valid. Passing pointer in private address space will cause assertion. Other address spaces are casted to global with potentially undefined behaviour.
+|Store contiguous data to _dst_.  The value of _x_ from each work-item is written to the memory location at _dst_ + +get_local_id()+. The value of _dst_ must be the same for all work-items in the sub-group. The address space information is deduced automatically. Only pointers to global and local address spaces are valid. Passing a pointer to private address space will cause an assertion. Other address spaces are cast to global with potentially undefined behavior.
 
 |+template <typename T, access::address_space Space> void store(sub_group sg, multi_ptr<T,Space> dst, const T& x)+
 |Store contiguous data to _dst_.  The value of _x_ from each work-item is written to the memory location at _dst_ + +get_local_id()+. The value of _dst_ must be the same for all work-items in the sub-group.  _Space_ must be +access::address_space::global_space+ or +access::address_space::local_space+.