Fix ROCm 6.4 warnings (#47)

* Removed __AMDGCN_WAVEFRONT_SIZE

* Added unit test to validate WF_SIZE

src/containers/array_impl.hpp

@@ -26,6 +26,7 @@
 #include <cassert>
 
 #include "array.hpp"
+#include "../constants.hpp"
 
 namespace rocshmem {
 
@@ -148,8 +149,8 @@ __device__ void Array<TYPE>::zero_thread_dump() {
 template <typename TYPE>
 __device__ void Array<TYPE>::any_thread_dump() {
   Identity id{};
-  for (int i = 0; i < __AMDGCN_WAVEFRONT_SIZE; i++) {
-    if ((id.local_thread_id() % __AMDGCN_WAVEFRONT_SIZE) == i) {
+  for (int i = 0; i < WF_SIZE; i++) {
+    if ((id.local_thread_id() % WF_SIZE) == i) {
       while (atomicCAS(GLOBAL_DEVICE_PRINT_LOCK, 0, 1) == 1) {
       }
 

src/containers/share_strategy.cpp

@@ -21,14 +21,10 @@
  *****************************************************************************/
 
 #include "share_strategy.hpp"
+#include "../constants.hpp"
 
 namespace rocshmem {
 
-/*
- * @brief The warp (wave-front) size.
- */
-static constexpr uint64_t WARP_SIZE = __AMDGCN_WAVEFRONT_SIZE;
-
 __device__ uint64_t Block::lane_id() {
   /*
    * amd_detail/device_functions.h::__lane_id()
@@ -202,7 +198,7 @@ __device__ uint64_t Block::active_logical_lane_id() {
 }
 
 __device__ uint64_t Block::broadcast_up(uint64_t fetch_value) {
-  for (unsigned i = 0; i < WARP_SIZE; i++) {
+  for (unsigned i = 0; i < WF_SIZE; i++) {
     uint64_t temp = __shfl_up(fetch_value, i);
     if (temp) {
       fetch_value = temp;

src/reverse_offload/context_ro_device.cpp

@@ -470,9 +470,9 @@ __device__ uint64_t active_logical_lane_id() {
 }
 
 __device__ uint64_t broadcast_lds(bool lowest_active, uint64_t value) {
-  constexpr size_t SIZE = 1024 / __AMDGCN_WAVEFRONT_SIZE;
+  constexpr size_t SIZE = 1024 / WF_SIZE;
   __shared__ uint64_t value_per_warp[SIZE];
-  auto wavefront_number {get_flat_block_id() / __AMDGCN_WAVEFRONT_SIZE};
+  auto wavefront_number {get_flat_block_id() / WF_SIZE};
   if (lowest_active) {
     value_per_warp[wavefront_number] = value;
     __threadfence_block();
@@ -481,7 +481,7 @@ __device__ uint64_t broadcast_lds(bool lowest_active, uint64_t value) {
 }
 
 __device__ uint64_t broadcast_shfl_up(uint64_t value) {
-  for (unsigned i{0}; i < __AMDGCN_WAVEFRONT_SIZE; i++) {
+  for (unsigned i{0}; i < WF_SIZE; i++) {
     uint64_t temp{__shfl_up(value, i)};
     if (temp) {
       value = temp;