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/device/hipDeviceSynchronize.cpp

@@ -34,7 +34,7 @@ THE SOFTWARE.
 #define NUM_STREAMS 2
 
 __global__ void Iter(hipLaunchParm lp, int *Ad, int num){
-    int tx = threadIdx.x + blockIdx.x * blockDim.x;
+    int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
     // Kernel loop designed to execute very slowly... ... ...   so we can test timing-related behavior below
     if(tx == 0){
         for(int i = 0; i<num;i++){

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];
     }

tests/src/runtimeApi/module/hipModule.cpp

@@ -34,7 +34,7 @@ THE SOFTWARE.
 #define kernel_name "hello_world"
 
 __global__ void Cpy(hipLaunchParm lp, float *Ad, float* Bd){
-  int tx = threadIdx.x;
+  int tx = hipThreadIdx_x;
   Bd[tx] = Ad[tx];
 }
 

tests/src/runtimeApi/module/vcpy_kernel.cpp

@@ -24,7 +24,7 @@ THE SOFTWARE.
 
 extern "C" __global__ void hello_world(hipLaunchParm lp, float *a, float *b)
 {
-    int tx = threadIdx.x;
+    int tx = hipThreadIdx_x;
     b[tx] = a[tx];
 }
 

tests/src/runtimeApi/multiThread/hipMultiThreadStreams2.cpp

@@ -35,7 +35,7 @@ THE SOFTWARE.
 
 template<typename T>
 __global__ void Inc(hipLaunchParm lp, T *Array){
-int tx = threadIdx.x + blockIdx.x * blockDim.x;
+int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
 Array[tx] = Array[tx] + T(1);
 }
 

tests/src/runtimeApi/stream/hipAPIStreamDisable.cpp

@@ -29,7 +29,7 @@ THE SOFTWARE.
 const int NN = 1 << 21;
 
 __global__ void kernel(hipLaunchParm lp, float *x, float *y, int n){
-	int tid = threadIdx.x;
+	int tid = hipThreadIdx_x;
 	if(tid < 1){
 		for(int i=0;i<n;i++){
 			x[i] = sqrt(powf(3.14159,i));
@@ -39,7 +39,7 @@ __global__ void kernel(hipLaunchParm lp, float *x, float *y, int n){
 }
 
 __global__ void nKernel(hipLaunchParm lp, float *y){
-	int tid = threadIdx.x;
+	int tid = hipThreadIdx_x;
 	y[tid] = y[tid] + 1.0f;
 }
 

tests/src/runtimeApi/stream/hipAPIStreamEnable.cpp

@@ -31,7 +31,7 @@ THE SOFTWARE.
 const int NN = 1 << 21;
 
 __global__ void kernel(hipLaunchParm lp, float *x, float *y, int n){
-	int tid = threadIdx.x;
+	int tid = hipThreadIdx_x;
 	if(tid < 1){
 		for(int i=0;i<n;i++){
 			x[i] = sqrt(powf(3.14159,i));
@@ -41,7 +41,7 @@ __global__ void kernel(hipLaunchParm lp, float *x, float *y, int n){
 }
 
 __global__ void nKernel(hipLaunchParm lp, float *y){
-	int tid = threadIdx.x;
+	int tid = hipThreadIdx_x;
 	y[tid] = y[tid] + 1.0f;
 }
 

tests/src/runtimeApi/stream/hipNullStream.cpp

@@ -41,8 +41,8 @@ vectorADDRepeat(hipLaunchParm lp,
             size_t NELEM,
             int repeat)
 {
-    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 (int j=1; j<=repeat;j++) {
         for (size_t i=offset; i<NELEM; i+=stride) {

tests/src/runtimeApi/stream/hipStream.h

@@ -73,7 +73,7 @@ void D2H(T *Dst, T *Src, size_t size){
 
 template<typename T>
 __global__ void Inc(hipLaunchParm lp, T *In){
-int tx = threadIdx.x + blockIdx.x * blockDim.x;
+int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
 In[tx] = In[tx] + 1;
 }
 

tests/src/runtimeApi/synchronization/copy_coherency.cpp

@@ -102,8 +102,8 @@ MemcpyFunction g_moduleMemcpy("memcpyInt.hsaco", "memcpyIntKernel");
 __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;
     }
@@ -112,8 +112,8 @@ memsetIntKernel(int * ptr, const int val, size_t numElements)
 __global__ void
 memcpyIntKernel(int *dst, const int * src, 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];
     }

tests/src/runtimeApi/synchronization/memcpyInt.device.cpp

@@ -5,8 +5,8 @@
 extern "C" __global__ void
 memcpyIntKernel(hipLaunchParm lp, int *dst, const int * src, 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];
     }