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

tests/src/runtimeApi/memory/hipHostGetFlags.cpp

@@ -33,7 +33,7 @@ THE SOFTWARE.
 #define SIZE LEN*sizeof(float)
 
 __global__ void Add(hipLaunchParm lp, float *Ad, float *Bd, float *Cd){
-int tx = threadIdx.x + blockIdx.x * blockDim.x;
+int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
 Cd[tx] = Ad[tx] + Bd[tx];
 }
 

tests/src/runtimeApi/memory/hipHostMalloc.cpp

@@ -33,13 +33,13 @@
 #define SIZE LEN*sizeof(float)
 
 __global__ void Add(float *Ad, float *Bd, float *Cd){
-    int tx = threadIdx.x + blockIdx.x * blockDim.x;
+    int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
     Cd[tx] = Ad[tx] + Bd[tx];
 }
 
 
 __global__ void Set(int *Ad, int val){
-    int tx = threadIdx.x + blockIdx.x * blockDim.x;
+    int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
     Ad[tx] = val;
 }
 

tests/src/runtimeApi/memory/hipHostRegister.cpp

@@ -29,7 +29,7 @@ THE SOFTWARE.
 #include<malloc.h>
 
 __global__ void Inc(hipLaunchParm lp, float *Ad){
-    int tx = threadIdx.x + blockIdx.x * blockDim.x;
+    int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
     Ad[tx] = Ad[tx] + float(1);
 }
 

tests/src/runtimeApi/memory/hipMemcpyAsync.cpp

@@ -70,8 +70,8 @@ template<typename T>
 __global__ void
 addK (hipLaunchParm lp, T *A, T K, size_t numElements)
 {
-    size_t offset = (blockIdx.x * blockDim.x + threadIdx.x);
-    size_t stride = blockDim.x * gridDim.x ;
+    size_t offset = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
+    size_t stride = hipBlockDim_x * hipGridDim_x ;
 
     for (size_t i=offset; i<numElements; i+=stride) {
         A[i] = A[i] + K;

tests/src/runtimeApi/memory/hipMemoryAllocateCoherent.cpp

@@ -31,7 +31,7 @@ THE SOFTWARE.
 
 __global__ void Kernel(hipLaunchParm lp,volatile float* hostRes)
 {
-    int tid = threadIdx.x + blockIdx.x * blockDim.x;
+    int tid = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
     hostRes[tid] = tid + 1;
     __threadfence_system();
     // expecting that the data is getting flushed to host here!

tests/src/runtimeApi/memory/p2p_copy_coherency.cpp

@@ -63,8 +63,8 @@ int enablePeers(int dev0, int dev1)
 __global__ void
 memsetIntKernel(int * ptr, const int val, size_t numElements)
 {
-    int gid = (blockIdx.x * blockDim.x + threadIdx.x);
-    int stride = blockDim.x * gridDim.x ;
+    int gid = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
+    int stride = hipBlockDim_x * hipGridDim_x ;
     for (size_t i= gid; i< numElements; i+=stride){
        ptr[i] = val;
     }
@@ -73,8 +73,8 @@ memsetIntKernel(int * ptr, const int val, size_t numElements)
 __global__ void
 memcpyIntKernel(const int * src, int* dst, size_t numElements)
 {
-    int gid = (blockIdx.x * blockDim.x + threadIdx.x);
-    int stride = blockDim.x * gridDim.x ;
+    int gid = (hipBlockIdx_x * hipBlockDim_x + hipThreadIdx_x);
+    int stride = hipBlockDim_x * hipGridDim_x ;
     for (size_t i= gid; i< numElements; i+=stride){
        dst[i] = src[i];
     }