diff --git a/hipamd/tests/hipify-clang/run_test.bat b/hipamd/tests/hipify-clang/run_test.bat
index 58563be2c7..8b10bc8c2e 100644
--- a/hipamd/tests/hipify-clang/run_test.bat
+++ b/hipamd/tests/hipify-clang/run_test.bat
@@ -8,12 +8,13 @@ set HIPIFY=%1
 set IN_FILE=%2
 set TMP_FILE=%3
 set CUDA_ROOT=%4
+set ROC=%5
 
 set all_args=%*
-call set clang_args=%%all_args:*%5=%%
-set clang_args=%5%clang_args%
+call set clang_args=%%all_args:*%6=%%
+set clang_args=%6%clang_args%
 
-%HIPIFY% -o=%TMP_FILE% %IN_FILE% %CUDA_ROOT% -- %clang_args%
+%HIPIFY% -o=%TMP_FILE% %IN_FILE% %CUDA_ROOT% %ROC% -- %clang_args%
 if errorlevel 1 (echo      Error: hipify-clang.exe failed with exit code: %errorlevel% && exit /b %errorlevel%)
 
 findstr /v /r /c:"[ ]*//[ ]*[CHECK*|RUN]" %TMP_FILE% | %FILE_CHECK% %IN_FILE%
diff --git a/hipamd/tests/hipify-clang/run_test.sh b/hipamd/tests/hipify-clang/run_test.sh
index eb3e822549..357976558e 100755
--- a/hipamd/tests/hipify-clang/run_test.sh
+++ b/hipamd/tests/hipify-clang/run_test.sh
@@ -10,8 +10,9 @@ HIPIFY=$1
 IN_FILE=$2
 TMP_FILE=$3
 CUDA_ROOT=$4
-shift 4
+ROC=$5
+shift 5
 
 # Remaining args are the ones to forward to clang proper.
 
-$HIPIFY -o=$TMP_FILE $IN_FILE $CUDA_ROOT -- $@ && cat $TMP_FILE | sed -Ee 's|//.+|// |g' | FileCheck $IN_FILE
+$HIPIFY -o=$TMP_FILE $IN_FILE $CUDA_ROOT $ROC -- $@ && cat $TMP_FILE | sed -Ee 's|//.+|// |g' | FileCheck $IN_FILE
diff --git a/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/cublas_0_based_indexing.cu b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/cublas_0_based_indexing.cu
index af24737dbc..69812c98b0 100644
--- a/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/cublas_0_based_indexing.cu
+++ b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/cublas_0_based_indexing.cu
@@ -1,5 +1,6 @@
 // RUN: %run_test hipify "%s" "%t" %hipify_args %clang_args
 
+// CHECK: #include <hip/hip_runtime.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <math.h>
@@ -33,6 +34,7 @@ int main(void) {
   }
   // cublasInit is not supported yet
   cublasInit();
+  // cublasAlloc is not supported yet
   stat = cublasAlloc(M*N, sizeof(*a), (void**)&devPtrA);
   // CHECK: if (stat != HIPBLAS_STATUS_SUCCESS) {
   if (stat != CUBLAS_STATUS_SUCCESS) {
diff --git a/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_0_based_indexing_rocblas.cu b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_0_based_indexing_rocblas.cu
new file mode 100644
index 0000000000..c61b9c8bd5
--- /dev/null
+++ b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_0_based_indexing_rocblas.cu
@@ -0,0 +1,81 @@
+// RUN: %run_test hipify "%s" "%t" %hipify_args "-roc" %clang_args
+
+// CHECK: #include <hip/hip_runtime.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+// CHECK: #include "rocblas.h"
+#include "cublas.h"
+#define M 6
+#define N 5
+#define IDX2C(i,j,ld) (((j)*(ld))+(i))
+static __inline__ void modify(float *m, int ldm, int n, int p, int q, float
+  alpha, float beta) {
+  // CHECK: rocblas_sscal(n - p, alpha, &m[IDX2C(p, q, ldm)], ldm);
+  // CHECK: rocblas_sscal(ldm - p, beta, &m[IDX2C(p, q, ldm)], 1);
+  cublasSscal(n - p, alpha, &m[IDX2C(p, q, ldm)], ldm);
+  cublasSscal(ldm - p, beta, &m[IDX2C(p, q, ldm)], 1);
+}
+int main(void) {
+  int i, j;
+  // CHECK: rocblas_status stat;
+  cublasStatus stat;
+  float* devPtrA;
+  float* a = 0;
+  a = (float *)malloc(M * N * sizeof(*a));
+  if (!a) {
+    printf("host memory allocation failed");
+    return EXIT_FAILURE;
+  }
+  for (j = 0; j < N; j++) {
+    for (i = 0; i < M; i++) {
+      a[IDX2C(i, j, M)] = (float)(i * M + j + 1);
+    }
+  }
+  // cublasInit is not supported yet
+  cublasInit();
+  // cublasAlloc is not supported yet
+  stat = cublasAlloc(M*N, sizeof(*a), (void**)&devPtrA);
+  // CHECK: if (stat != rocblas_status_success) {
+  if (stat != CUBLAS_STATUS_SUCCESS) {
+    printf("device memory allocation failed");
+    // cublasShutdown is not supported yet
+    cublasShutdown();
+    return EXIT_FAILURE;
+  }
+  // CHECK: stat = rocblas_set_matrix(M, N, sizeof(*a), a, M, devPtrA, M);
+  stat = cublasSetMatrix(M, N, sizeof(*a), a, M, devPtrA, M);
+  // CHECK: if (stat != rocblas_status_success) {
+  if (stat != CUBLAS_STATUS_SUCCESS) {
+    printf("data download failed");
+    // cublasFree is not supported yet
+    cublasFree(devPtrA);
+    // cublasShutdown is not supported yet
+    cublasShutdown();
+    return EXIT_FAILURE;
+  }
+  modify(devPtrA, M, N, 1, 2, 16.0f, 12.0f);
+  // CHECK: stat = rocblas_get_matrix(M, N, sizeof(*a), devPtrA, M, a, M);
+  stat = cublasGetMatrix(M, N, sizeof(*a), devPtrA, M, a, M);
+  // CHECK: if (stat != rocblas_status_success) {
+  if (stat != CUBLAS_STATUS_SUCCESS) {
+    printf("data upload failed");
+    // cublasFree is not supported yet
+    cublasFree(devPtrA);
+    // cublasShutdown is not supported yet
+    cublasShutdown();
+    return EXIT_FAILURE;
+  }
+  // cublasFree is not supported yet
+  cublasFree(devPtrA);
+  // cublasShutdown is not supported yet
+  cublasShutdown();
+  for (j = 0; j < N; j++) {
+    for (i = 0; i < M; i++) {
+      printf("%7.0f", a[IDX2C(i, j, M)]);
+    }
+    printf("\n");
+  }
+  free(a);
+  return EXIT_SUCCESS;
+}
diff --git a/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_1_based_indexing_rocblas.cu b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_1_based_indexing_rocblas.cu
new file mode 100644
index 0000000000..0202e2f7fa
--- /dev/null
+++ b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_1_based_indexing_rocblas.cu
@@ -0,0 +1,90 @@
+// RUN: %run_test hipify "%s" "%t" %hipify_args "-roc" %clang_args
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+// CHECK: #include <hip/hip_runtime.h>
+#include <cuda_runtime.h>
+// CHECK: #include "rocblas.h"
+#include "cublas_v2.h"
+#define M 6
+#define N 5
+#define IDX2F(i,j,ld) ((((j)-1)*(ld))+((i)-1))
+// CHECK: static __inline__ void modify(rocblas_handle handle, float *m, int ldm, int
+static __inline__ void modify(cublasHandle_t handle, float *m, int ldm, int
+  n, int p, int q, float alpha, float beta) {
+  // CHECK: rocblas_sscal(handle, n - p + 1, &alpha, &m[IDX2F(p, q, ldm)], ldm);
+  // CHECK: rocblas_sscal(handle, ldm - p + 1, &beta, &m[IDX2F(p, q, ldm)], 1);
+  cublasSscal(handle, n - p + 1, &alpha, &m[IDX2F(p, q, ldm)], ldm);
+  cublasSscal(handle, ldm - p + 1, &beta, &m[IDX2F(p, q, ldm)], 1);
+}
+int main(void) {
+  // CHECK: hipError_t cudaStat;
+  // CHECK: rocblas_status stat;
+  // CHECK: rocblas_handle handle;
+  cudaError_t cudaStat;
+  cublasStatus_t stat;
+  cublasHandle_t handle;
+  int i, j;
+  float* devPtrA;
+  float* a = 0;
+  a = (float *)malloc(M * N * sizeof(*a));
+  if (!a) {
+    printf("host memory allocation failed");
+    return EXIT_FAILURE;
+  }
+  for (j = 1; j <= N; j++) {
+    for (i = 1; i <= M; i++) {
+      a[IDX2F(i, j, M)] = (float)((i - 1) * M + j);
+    }
+  }
+  // CHECK: cudaStat = hipMalloc((void**)&devPtrA, M*N * sizeof(*a));
+  cudaStat = cudaMalloc((void**)&devPtrA, M*N * sizeof(*a));
+  // CHECK: if (cudaStat != hipSuccess) {
+  if (cudaStat != cudaSuccess) {
+    printf("device memory allocation failed");
+    return EXIT_FAILURE;
+  }
+  // CHECK: stat = rocblas_create_handle(&handle);
+  stat = cublasCreate(&handle);
+  // CHECK: if (stat != rocblas_status_success) {
+  if (stat != CUBLAS_STATUS_SUCCESS) {
+    printf("CUBLAS initialization failed\n");
+    return EXIT_FAILURE;
+  }
+  // CHECK: stat = rocblas_set_matrix(M, N, sizeof(*a), a, M, devPtrA, M);
+  stat = cublasSetMatrix(M, N, sizeof(*a), a, M, devPtrA, M);
+  // CHECK: if (stat != rocblas_status_success) {
+  if (stat != CUBLAS_STATUS_SUCCESS) {
+    printf("data download failed");
+    // CHECK: hipFree(devPtrA);
+    // CHECK: rocblas_destroy_handle(handle);
+    cudaFree(devPtrA);
+    cublasDestroy(handle);
+    return EXIT_FAILURE;
+  }
+  modify(handle, devPtrA, M, N, 2, 3, 16.0f, 12.0f);
+  // CHECK: stat = rocblas_get_matrix(M, N, sizeof(*a), devPtrA, M, a, M);
+  stat = cublasGetMatrix(M, N, sizeof(*a), devPtrA, M, a, M);
+  // CHECK: if (stat != rocblas_status_success) {
+  if (stat != CUBLAS_STATUS_SUCCESS) {
+    printf("data upload failed");
+    // CHECK: hipFree(devPtrA);
+    // CHECK: rocblas_destroy_handle(handle);
+    cudaFree(devPtrA);
+    cublasDestroy(handle);
+    return EXIT_FAILURE;
+  }
+  // CHECK: hipFree(devPtrA);
+  // CHECK: rocblas_destroy_handle(handle);
+  cudaFree(devPtrA);
+  cublasDestroy(handle);
+  for (j = 1; j <= N; j++) {
+    for (i = 1; i <= M; i++) {
+      printf("%7.0f", a[IDX2F(i, j, M)]);
+    }
+    printf("\n");
+  }
+  free(a);
+  return EXIT_SUCCESS;
+}
diff --git a/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_sgemm_matrix_multiplication_rocblas.cu b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_sgemm_matrix_multiplication_rocblas.cu
new file mode 100644
index 0000000000..8e35f28f0e
--- /dev/null
+++ b/hipamd/tests/hipify-clang/unit_tests/libraries/cuBLAS/rocBLAS/cublas_sgemm_matrix_multiplication_rocblas.cu
@@ -0,0 +1,108 @@
+// RUN: %run_test hipify "%s" "%t" %hipify_args "-roc" %clang_args
+
+#include <stdio.h>
+#include <stdlib.h>
+// CHECK: #include <hip/hip_runtime.h>
+#include <cuda_runtime.h>
+// CHECK: #include "rocblas.h"
+#include "cublas_v2.h"
+#define IDX2C(i,j,ld) (((j)*(ld))+(i))
+#define m 6
+#define n 4
+#define k 5
+int main(void) {
+  // CHECK: hipError_t cudaStat;
+  // CHECK: rocblas_status stat;
+  // CHECK: rocblas_handle handle;
+  cudaError_t cudaStat;
+  cublasStatus_t stat;
+  cublasHandle_t handle;
+  int i, j;
+  float * a;
+  float * b;
+  float * c;
+  a = (float *)malloc(m*k * sizeof(float));
+  b = (float *)malloc(k*n * sizeof(float));
+  c = (float *)malloc(m*n * sizeof(float));
+  int ind = 11;
+  for (j = 0; j<k; j++) {
+    for (i = 0; i<m; i++) {
+      a[IDX2C(i, j, m)] = (float)ind++;
+    }
+  }
+  printf("a:\n");
+  for (i = 0; i<m; i++) {
+    for (j = 0; j<k; j++) {
+      printf(" %5.0f", a[IDX2C(i, j, m)]);
+    }
+    printf("\n");
+  }
+  ind = 11;
+  for (j = 0; j<n; j++) {
+    for (i = 0; i<k; i++) {
+      b[IDX2C(i, j, k)] = (float)ind++;
+    }
+  }
+  printf("b:\n");
+  for (i = 0; i<k; i++) {
+    for (j = 0; j<n; j++) {
+      printf(" %5.0f", b[IDX2C(i, j, k)]);
+    }
+    printf("\n");
+  }
+  ind = 11;
+  for (j = 0; j<n; j++) {
+    for (i = 0; i<m; i++) {
+      c[IDX2C(i, j, m)] = (float)ind++;
+    }
+  }
+  printf("c:\n");
+  for (i = 0; i<m; i++) {
+    for (j = 0; j<n; j++) {
+      printf(" %5.0f", c[IDX2C(i, j, m)]);
+    }
+    printf("\n");
+  }
+  float * d_a;
+  float * d_b;
+  float * d_c;
+  // CHECK: cudaStat = hipMalloc((void **)& d_a, m*k * sizeof(*a));
+  // CHECK: cudaStat = hipMalloc((void **)& d_b, k*n * sizeof(*b));
+  // CHECK: cudaStat = hipMalloc((void **)& d_c, m*n * sizeof(*c));
+  cudaStat = cudaMalloc((void **)& d_a, m*k * sizeof(*a));
+  cudaStat = cudaMalloc((void **)& d_b, k*n * sizeof(*b));
+  cudaStat = cudaMalloc((void **)& d_c, m*n * sizeof(*c));
+  // CHECK: stat = rocblas_create_handle(&handle);
+  stat = cublasCreate(&handle);
+  // CHECK: stat = rocblas_set_matrix(m, k, sizeof(*a), a, m, d_a, m);
+  // CHECK: stat = rocblas_set_matrix(k, n, sizeof(*b), b, k, d_b, k);
+  // CHECK: stat = rocblas_set_matrix(m, n, sizeof(*c), c, m, d_c, m);
+  stat = cublasSetMatrix(m, k, sizeof(*a), a, m, d_a, m);
+  stat = cublasSetMatrix(k, n, sizeof(*b), b, k, d_b, k);
+  stat = cublasSetMatrix(m, n, sizeof(*c), c, m, d_c, m);
+  float al = 1.0f;
+  float bet = 1.0f;
+  // CHECK: stat = rocblas_sgemm(handle, rocblas_operation_none, rocblas_operation_none, m, n, k, &al, d_a, m, d_b, k, &bet, d_c, m);
+  stat = cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, m, n, k, &al, d_a, m, d_b, k, &bet, d_c, m);
+  // CHECK: stat = rocblas_get_matrix(m, n, sizeof(*c), d_c, m, c, m);
+  stat = cublasGetMatrix(m, n, sizeof(*c), d_c, m, c, m);
+  printf("c after Sgemm :\n");
+  for (i = 0; i<m; i++) {
+    for (j = 0; j<n; j++) {
+      printf(" %7.0f", c[IDX2C(i, j, m)]);
+    }
+    printf("\n");
+  }
+  // CHECK: hipFree(d_a);
+  // CHECK: hipFree(d_b);
+  // CHECK: hipFree(d_c);
+  // CHECK: rocblas_destroy_handle(handle);
+  cudaFree(d_a);
+  cudaFree(d_b);
+  cudaFree(d_c);
+  cublasDestroy(handle);
+  free(a);
+  free(b);
+  free(c);
+  return EXIT_SUCCESS;
+}