Revert "Revert adoption of CUDA indexing in general - this can only work with later versions of the compiler, just like module based dispatch, and thus must be guarded against usage in earlier (e.g. 1.6) versions."

This reverts commit d2fd1f5

src/device_util.cpp

@@ -45,8 +45,8 @@ __device__ void *__hip_hc_malloc(size_t size)
     {
         return (void*)nullptr;
     }
-    uint32_t totalThreads = blockDim.x * gridDim.x * blockDim.y * gridDim.y * blockDim.z * gridDim.z;
-    uint32_t currentWorkItem = threadIdx.x + blockDim.x * blockIdx.x;
+    uint32_t totalThreads = hipBlockDim_x * hipGridDim_x * hipBlockDim_y * hipGridDim_y * hipBlockDim_z * hipGridDim_z;
+    uint32_t currentWorkItem = hipThreadIdx_x + hipBlockDim_x * hipBlockIdx_x;
 
     uint32_t numHeapsPerWorkItem = NUM_PAGES / totalThreads;
     uint32_t heapSizePerWorkItem = SIZE_OF_HEAP / totalThreads;

src/hip_memory.cpp

@@ -1307,9 +1307,9 @@ namespace
     __global__
     void hip_fill_n(RandomAccessIterator f, N n, T value)
     {
-        const uint32_t grid_dim = gridDim.x * blockDim.x;
+        const uint32_t grid_dim = hipGridDim_x * hipBlockDim_x;
 
-        size_t idx = blockIdx.x * block_dim + threadIdx.x;
+        size_t idx = hipBlockIdx_x * block_dim + hipThreadIdx_x;
         while (idx < n) {
             __builtin_memcpy(
                 reinterpret_cast<void*>(&f[idx]),