Updated docs for ROCm 7.x.x (#239)

Co-authored-by: Aurelien Bouteiller <aurelien.bouteiller@amd.com>
Co-authored-by: yugang-amd <yugang.wang@amd.com>

include/rocshmem/rocshmem.hpp

@@ -80,9 +80,9 @@ __host__ void rocshmem_init(void);
 /**
  * @brief Query rocSHMEM context from host API
  *
- * @param[out] ctx      Returns ROCSHMEM_CTX_DEFAULT device pointer that users 
+ * @param[out] ctx      Returns ROCSHMEM_CTX_DEFAULT device pointer that users
  *                      can query from one instance of rocshmem host library and
- *                      use use later for dynamic module initialization in 
+ *                      use use later for dynamic module initialization in
  *                      kernel bitcode device library in the same application
  */
 __host__ void * rocshmem_get_device_ctx();
@@ -91,14 +91,15 @@ __host__ void * rocshmem_get_device_ctx();
  * @brief Query rocSHMEM remote symmetric heap pointer
  *
  * @param[in]  dest     local symmetric heap allocation pointer for current pe/device
- * 
+ *
  * @param[in]  pe       remote PE
- * 
+ *
  * @param[out] ptr      Returns remote symmetric heap device pointer from host-side API.
  *                      This can be used to issue load/store from custom kernels
  *                      instead of using rocshmem device side get/put APIs for RMA operations.
  */
-__host__ void *rocshmem_ptr(void *dest, int pe);
+__host__ void* rocshmem_ptr(const void *dest, int pe);
+__device__ ATTR_NO_INLINE void* rocshmem_ptr(const void *dest, int pe);
 
 #if defined(HAVE_EXTERNAL_MPI)
 /**

include/rocshmem/rocshmem_COLL.hpp

@@ -775,28 +775,6 @@ __device__ ATTR_NO_INLINE void rocshmem_ctx_sync_wave(
 __device__ ATTR_NO_INLINE void rocshmem_ctx_sync_wg(
     rocshmem_ctx_t ctx, rocshmem_team_t team);
 
-/**
- * @brief Query a local pointer to a symmetric data object on the
- * specified \pe . Returns an address that may be used to directly reference
- * dest on the specified \pe. This address can be accesses with LD/ST ops.
- *
- * Can be called per thread with no performance penalty.
- */
-__device__ ATTR_NO_INLINE void *rocshmem_ptr(const void *dest, int pe);
-
-/**
- * @brief Make all uncacheable GPU data visible to other agents in the sytem.
- *
- * This only works for data that was explicitly allocated uncacheable on the
- * GPU!
- *
- * Can be called per thread with no performance penalty.
- *
- * @param[in] GPU-side handle.
- *
- * @return void
- */
-
 }  // namespace rocshmem
 
 #endif  // LIBRARY_INCLUDE_ROCSHMEM_COLL_HPP