Fix underflow error in calculate_dim_order() (#5498)

SS-JIA · facebook-github-bot · commit 28c9a1dfeeb2 · 2024-09-19T14:52:24.000-07:00
Summary: Pull Request resolved: #5498 ## Context FIx an underflow error. The issue is that `ndim` is unsigned, therefore if it is `0` when it gets subtracted by one it underflows. Not sure why this wasn't caught by CI tests before. ghstack-source-id: 243610987 exported-using-ghexport Reviewed By: jorgep31415 Differential Revision: D63050011 fbshipit-source-id: d5ea57c00d3b5c4adb577c74d7a10c7840643ae3
diff --git a/backends/vulkan/runtime/api/containers/Tensor.cpp b/backends/vulkan/runtime/api/containers/Tensor.cpp
@@ -21,7 +21,8 @@ std::vector<int64_t> calculate_dim_order(
     return {0};
   }
   std::vector<int64_t> dim_order(ndim);
-  int64_t last_dim = ndim - 1 - packed_dim;
+  // Explicitly convert ndim to signed to prevent underflow
+  int64_t last_dim = int64_t(ndim) - 1 - packed_dim;
 
   int64_t cur_dim = 0;
   for (int d = 0; d < ndim; ++d) {

Original file line number	Diff line number	Diff line change
`@@ -21,7 +21,8 @@ std::vector<int64_t> calculate_dim_order(`
`21`	`21`	`return {0};`
`22`	`22`	`}`
`23`	`23`	`std::vector<int64_t> dim_order(ndim);`
`24`		`- int64_t last_dim = ndim - 1 - packed_dim;`
	`24`	`+ // Explicitly convert ndim to signed to prevent underflow`
	`25`	`+ int64_t last_dim = int64_t(ndim) - 1 - packed_dim;`
`25`	`26`
`26`	`27`	`int64_t cur_dim = 0;`
`27`	`28`	`for (int d = 0; d < ndim; ++d) {`