diff --git a/lib/Support/BigIntSupport.cpp b/lib/Support/BigIntSupport.cpp
index d9b0fe35b64..2453fd0bdb9 100644
--- a/lib/Support/BigIntSupport.cpp
+++ b/lib/Support/BigIntSupport.cpp
@@ -1741,10 +1741,6 @@ multiply(MutableBigIntRef dst, ImmutableBigIntRef lhs, ImmutableBigIntRef rhs) {
 
 namespace {
 namespace div_rem {
-static uint32_t getResultSize(ImmutableBigIntRef lhs, ImmutableBigIntRef rhs) {
-  return std::max(lhs.numDigits, rhs.numDigits) + 1;
-}
-
 static OperationStatus compute(
     MutableBigIntRef quoc,
     MutableBigIntRef rem,
@@ -1868,7 +1864,14 @@ static OperationStatus compute(
 } // namespace
 
 uint32_t divideResultSize(ImmutableBigIntRef lhs, ImmutableBigIntRef rhs) {
-  return div_rem::getResultSize(lhs, rhs);
+  if (isNegative(lhs) && compare(rhs, -1) == 0) {
+    // In this (and only this) case, we can end up with more digits than we started with
+    // Examples: -(2n**63n)/-1n, -(2n**127n)/-1n
+    // We avoid this in general case as it makes division much slower
+    return std::max(lhs.numDigits, rhs.numDigits) + 1;
+  }
+
+  return std::max(lhs.numDigits, rhs.numDigits);
 }
 
 OperationStatus
@@ -1879,7 +1882,7 @@ divide(MutableBigIntRef dst, ImmutableBigIntRef lhs, ImmutableBigIntRef rhs) {
 }
 
 uint32_t remainderResultSize(ImmutableBigIntRef lhs, ImmutableBigIntRef rhs) {
-  return div_rem::getResultSize(lhs, rhs);
+  return divideResultSize(lhs, rhs); // We currently expect them to be equivalent in div_rem::compute()
 }
 
 OperationStatus remainder(