Add __restrict__ to jit kernel params (#1300)

Add restrict to jit kernel params (#1300)
2781ccd8 · turneram · GitHub · 2bbb50c4 · 2781ccd8
Unverified Commit 2781ccd8 authored Jul 11, 2022 by turneram Committed by GitHub Jul 11, 2022
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 2 deletions

src/targets/gpu/compile_hip_code_object.cpp src/targets/gpu/compile_hip_code_object.cpp +2 -2

No files found.
--- a/src/targets/gpu/compile_hip_code_object.cpp
+++ b/src/targets/gpu/compile_hip_code_object.cpp
@@ -51,9 +51,9 @@ static const char* const make_tensor_template = R"__migraphx__(
 template<>
 struct make_tensor<${n}>
 {
-    static __device__ auto apply(void* p)
+    static __device__ auto apply(void* __restrict__ p)
    {
-        return make_tensor_view(reinterpret_cast<${type}*>(p), make_shape(${lens}, ${strides}));
+        return make_tensor_view(reinterpret_cast<${type}* __restrict__>(p), make_shape(${lens}, ${strides}));
    }
 };
 )__migraphx__";