Merge 'master' into 'amd-master'

Change-Id: I539a79195161876b0a6eae952f64af85d4c47b28

cmake/FindHIP.cmake

@@ -265,7 +265,7 @@ endmacro()
 set(HIP_HIPCC_INCLUDE_ARGS_USER "")
 macro(HIP_INCLUDE_DIRECTORIES)
     foreach(dir ${ARGN})
-        list(APPEND HIP_HIPCC_INCLUDE_ARGS_USER -I${dir})
+        list(APPEND HIP_HIPCC_INCLUDE_ARGS_USER $<$<BOOL:${dir}>:-I${dir}>)
     endforeach()
 endmacro()
 
@@ -373,7 +373,7 @@ macro(HIP_PREPARE_TARGET_COMMANDS _target _format _generated_files _source_files
     list(REMOVE_DUPLICATES _hip_include_directories)
     if(_hip_include_directories)
         foreach(dir ${_hip_include_directories})
-            list(APPEND HIP_HIPCC_INCLUDE_ARGS -I${dir})
+            list(APPEND HIP_HIPCC_INCLUDE_ARGS $<$<BOOL:${dir}>:-I${dir}>)
         endforeach()
     endif()
 

include/hip/nvcc_detail/hip_complex.h

@@ -83,6 +83,10 @@ __device__ __host__ static inline hipDoubleComplex hipCsub(hipDoubleComplex p, h
     return cuCsub(p, q);
 }
 
+__device__ __host__ static inline hipDoubleComplex hipCmul(hipDoubleComplex p, hipDoubleComplex q) {
+    return cuCmul(p, q);
+}
+
 __device__ __host__ static inline hipDoubleComplex hipCdiv(hipDoubleComplex p, hipDoubleComplex q) {
     return cuCdiv(p, q);
 }

src/hip_memory.cpp

@@ -1668,6 +1668,7 @@ hipError_t hipMemcpy2DAsync(void* dst, size_t dpitch, const void* src, size_t sp
     void *pinnedPtr=NULL;
     void *actualSrc = (void*)src;
     void *actualDest = dst;
+    stream = ihipSyncAndResolveStream(stream);
     if(kind == hipMemcpyHostToDevice ) {
         if(getLockedPointer((void*)src, spitch, &pinnedPtr) == hipSuccess ){
             isLocked = 1;
@@ -1679,9 +1680,12 @@ hipError_t hipMemcpy2DAsync(void* dst, size_t dpitch, const void* src, size_t sp
             actualDest = pinnedPtr;
           }
     }
+#if 0
     if((width == dpitch) && (width == spitch)) {
             hip_internal::memcpyAsync(dst, src, width*height, kind, stream);
-    } else {
+    } else
+#endif
+   {
         try {
             if(!isLocked){
                 for (int i = 0; i < height; ++i)