[PATCH] D46295: [ubsan] Make the C++ code here not cast floating point values outside the range of an integer type to that integer type.

[Chandler Carruth via Phabricator via llvm-commits]

chandlerc created this revision.
chandlerc added a reviewer: asbirlea.
Herald added subscribers: mcrosier, sanjoy.

Instead, change the random value function to cast to the destination
type first and do the devision by 8 afterward. When the destination is
floating point this should be roughly equivalent, but when it is an
integer type it will end up truncating stuff down rather than ending up
outside the range.


Repository:
  rT test-suite

https://reviews.llvm.org/D46295

Files:
  Bitcode/Regression/vector_widen/driver.cpp
  Bitcode/simd_ops/simd_ops.cpp


Index: Bitcode/simd_ops/simd_ops.cpp
===================================================================
--- Bitcode/simd_ops/simd_ops.cpp
+++ Bitcode/simd_ops/simd_ops.cpp
@@ -24,7 +24,7 @@
 
 template<typename T>
 T rand_value() {
-    return (T)(rand() * 0.125) - 100;
+    return (T)((T)rand() / 8) - 100;
 }
 
 // Even on android, we want errors to stdout
Index: Bitcode/Regression/vector_widen/driver.cpp
===================================================================
--- Bitcode/Regression/vector_widen/driver.cpp
+++ Bitcode/Regression/vector_widen/driver.cpp
@@ -2,7 +2,7 @@
 
 template<typename T>
 T rand_value() {
-    return (T)(rand() * 0.125) - 100;
+    return (T)((T)rand() / 8) - 100;
 }
 
 template<typename T>


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D46295.144654.patch
Type: text/x-patch
Size: 735 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180501/2986cad3/attachment.bin>