Merge branch 'privatestaging' of https://github.com/AMDComputeLibraries/HIP-privatestaging into privatestaging

hipamd/include/nvcc_detail/hip_runtime_api.h

@@ -55,6 +55,9 @@ hipMemcpyHostToHost
 #define hipHostAllocMapped cudaHostAllocMapped
 #define hipHostAllocWriteCombined cudaHostAllocWriteCombined
 
+#define hipHostRegisterPortable cudaHostRegisterPortable
+#define hipHostRegisterMapped cudaHostRegisterMapped
+
 typedef cudaEvent_t hipEvent_t;
 typedef cudaStream_t hipStream_t;
 //typedef cudaChannelFormatDesc hipChannelFormatDesc;
@@ -132,6 +135,14 @@ inline static hipError_t hipHostGetFlags(unsigned int* flagsPtr, void* hostPtr){
 	return hipCUDAErrorTohipError(cudaHostGetFlags(flagsPtr, hostPtr));
 }
 
+inline static hipError_t hipHostRegister(void* ptr, size_t size, unsigned int flags){
+	return hipCUDAErrorTohipError(cudaHostRegister(ptr, size, flags));
+}
+
+inline static hipError_t hipHostUnregister(void* ptr){
+	return hipCUDAErrorTohipError(cudaHostUnregister(ptr));
+}
+
 inline static hipError_t hipFreeHost(void* ptr) {
     return hipCUDAErrorTohipError(cudaFreeHost(ptr));
 }

hipamd/tests/src/CMakeLists.txt

@@ -129,6 +129,7 @@ make_hip_executable (hipMultiThreadStreams2 hipMultiThreadStreams2.cpp)
 make_hip_executable (hipHostAlloc hipHostAlloc.cpp)
 make_hip_executable (hipStreamL5 hipStreamL5.cpp)
 make_hip_executable (hipHostGetFlags hipHostGetFlags.cpp)
+make_hip_executable (hipHostRegister hipHostRegister.cpp)
 target_link_libraries(hipMathFunctionsHost m)
 
 make_test(hip_ballot " " )
@@ -152,7 +153,7 @@ make_test(hipMemcpy  " " )
 make_test(hipMemcpyAsync  " " )
 make_test(hipHostGetFlags " ")
 make_test(hipHcc  " " )
-
+make_test(hipHostRegister " ")
 make_test(hipStreamL5 " ")
 
 make_hipify_test(specialFunc.cu )

hipamd/tests/src/hipHostRegister.cpp

@@ -0,0 +1,30 @@
+#include"test_common.h"
+
+__global__ void Inc(hipLaunchParm lp, float *Ad){
+int tx = hipThreadIdx_x + hipBlockIdx_x * hipBlockDim_x;
+Ad[tx] = Ad[tx] + float(1);
+}
+
+int main(){
+	float *A, *Ad;
+	const size_t size = N * sizeof(float);
+	A = (float*)malloc(size);
+	HIPCHECK(hipHostRegister(A, size, 0));
+
+	for(int i=0;i<N;i++){
+		A[i] = float(1);
+	}
+
+	HIPCHECK(hipMalloc(&Ad, size));
+
+	HIPCHECK(hipMemcpy(Ad, A, size, hipMemcpyHostToDevice));
+
+	hipLaunchKernel(HIP_KERNEL_NAME(Inc), dim3(N/512), dim3(512), 0, 0, Ad);
+	HIPCHECK(hipDeviceSynchronize());
+
+	HIPCHECK(hipMemcpy(A, Ad, size, hipMemcpyDeviceToHost));
+
+	HIPASSERT(A[10] == 2.0f);
+	HIPCHECK(hipHostUnregister(A));
+	passed();
+}