[Mlir-commits] [mlir] 3c448c2 - [mlir][sparse] Updating checkedMul to use intrinsics.

[wren romano]

Author: wren romano
Date: 2022-11-16T16:22:58-08:00
New Revision: 3c448c2cef43c754b48022d99b4801d1de9e7af2

URL: https://github.com/llvm/llvm-project/commit/3c448c2cef43c754b48022d99b4801d1de9e7af2
DIFF: https://github.com/llvm/llvm-project/commit/3c448c2cef43c754b48022d99b4801d1de9e7af2.diff

LOG: [mlir][sparse] Updating checkedMul to use intrinsics.

Depends On D138149

Reviewed By: aartbik

Differential Revision: https://reviews.llvm.org/D138154

Added: 
    

Modified: 
    mlir/include/mlir/ExecutionEngine/SparseTensor/ArithmeticUtils.h

Removed: 
    


################################################################################
diff  --git a/mlir/include/mlir/ExecutionEngine/SparseTensor/ArithmeticUtils.h b/mlir/include/mlir/ExecutionEngine/SparseTensor/ArithmeticUtils.h
index 50aa93e90e04..ab5764c6b1e5 100644
--- a/mlir/include/mlir/ExecutionEngine/SparseTensor/ArithmeticUtils.h
+++ b/mlir/include/mlir/ExecutionEngine/SparseTensor/ArithmeticUtils.h
@@ -131,16 +131,22 @@ template <typename To, typename From>
   return static_cast<To>(x);
 }
 
-// TODO: would be better to use various architectures' intrinsics to
-// detect the overflow directly, instead of doing the assertion beforehand
-// (which requires an expensive division).
-//
 /// A version of `operator*` on `uint64_t` which guards against overflows
 /// (when assertions are enabled).
 inline uint64_t checkedMul(uint64_t lhs, uint64_t rhs) {
+  // If assertions are enabled and we have the intrinsic, then use it to
+  // avoid the expensive division.  If assertions are disabled, then don't
+  // bother with intrinsics (to avoid any possible slowdown vs `operator*`).
+#if !defined(NDEBUG) && __has_builtin(__builtin_mul_overflow)
+  uint64_t result;
+  bool overflowed = __builtin_mul_overflow(lhs, rhs, &result);
+  assert(!overflowed && "Integer overflow");
+  return result;
+#else
   assert((lhs == 0 || rhs <= std::numeric_limits<uint64_t>::max() / lhs) &&
          "Integer overflow");
   return lhs * rhs;
+#endif
 }
 
 } // namespace detail