[ubsan] Use pass_object_size info in bounds checks (compiler-rt)

[vsk]

Summary

This is a test update for the clang change in https://reviews.llvm.org/D40940

[george.burgess.iv]

If the answer to my question is "no, it'll just work," LGTM. Thanks!

test/ubsan/TestCases/Misc/bounds.cpp

9

Do we need extra RUN: lines/CHECK prefixes for these? I thought ubsan runtime errors terminated the program.

[vsk]

test/ubsan/TestCases/Misc/bounds.cpp

9

We don't -- this test doesn't use ubsan's trapping mode. I'll try not to fat-finger the command to generate diff context next time -- this would've been clearer with the compile line present.

[vsk]

Landed in r320129.

test/ubsan/TestCases/Misc/bounds.cpp

 // RUN: %run %t 0 3 0 2>&1 | FileCheck %s --check-prefix=CHECK-B-3
 // RUN: %run %t 0 0 4 2>&1 | FileCheck %s --check-prefix=CHECK-C-4
 
+int get_int(int *const p __attribute__((pass_object_size(0))), int i) {
+  // CHECK-A-2: bounds.cpp:[[@LINE+1]]:10: runtime error: index 2 out of bounds for type 'int *'
+  return p[i];
+}
+
+int get_double(double *const p __attribute__((pass_object_size(0))), int i) {
+  // CHECK-A-2: bounds.cpp:[[@LINE+1]]:10: runtime error: index 2 out of bounds for type 'double *'
+  return p[i];
+}
+
 int main(int argc, char **argv) {
+  int bar[2];
+  get_int(bar, argv[1][0] - '0');
+
+  double baz[2];
+  get_double(baz, argv[1][0] - '0');
+
   int arr[2][3][4] = {};
 
   return arr[argv[1][0] - '0'][argv[2][0] - '0'][argv[3][0] - '0'];