diff --git a/src/codegen/llvm/codegen_amdgpu.cc b/src/codegen/llvm/codegen_amdgpu.cc
index 4769efdb0405f870202516d7dcb803a13fc8dadd..146880b7dd89f3470188aed500cf81e8e3f2d124 100644
--- a/src/codegen/llvm/codegen_amdgpu.cc
+++ b/src/codegen/llvm/codegen_amdgpu.cc
@@ -136,7 +136,7 @@ runtime::Module BuildAMDGPU(Array<LoweredFunc> funcs, std::string target) {
 ) >= 4 &&
         target.substr(0, 4) == "rocm");
   llvm::TargetMachine* tm = \
-    GetLLVMTargetMachine("-mtriple=amdgcn-amd-amdhsa-hcc -mcpu=gfx900" + \
+    GetLLVMTargetMachine("-mtriple=amdgcn-amd-amdhsa-hcc -mcpu=gfx803" + \
     target.substr(4, target.length() - 4));
 
   std::unique_ptr<CodeGenAMDGPU> cg(new CodeGenAMDGPU());
diff --git a/src/runtime/rocm/rocm_module.cc b/src/runtime/rocm/rocm_module.cc
index 2839e10945f8a2ac03d6bea0b69a358931c8a137..98ec6f63d74e93f72bacdea422e83af30aff6b36 100644
--- a/src/runtime/rocm/rocm_module.cc
+++ b/src/runtime/rocm/rocm_module.cc
@@ -152,7 +152,7 @@ class ROCMWrappedFunc {
 
     ThreadWorkLoad wl = thread_axis_cfg_.Extract(args);
     void* config[] = {
-      HIP_LAUNCH_PARAM_BUFFER_POINTER, &packed_args,
+      HIP_LAUNCH_PARAM_BUFFER_POINTER, packed_args,
       HIP_LAUNCH_PARAM_BUFFER_SIZE, &packed_nbytes,
       HIP_LAUNCH_PARAM_END
     };