diff --git a/expression/builtin_cast_test.go b/expression/builtin_cast_test.go
index ca8f63cb353ba..4de94a8be3798 100644
--- a/expression/builtin_cast_test.go
+++ b/expression/builtin_cast_test.go
@@ -127,6 +127,26 @@ func (s *testEvaluatorSuite) TestCast(c *C) {
 	lastWarn = warnings[len(warnings)-1]
 	c.Assert(terror.ErrorEqual(types.ErrTruncatedWrongVal, lastWarn.Err), IsTrue, Commentf("err %v", lastWarn.Err))
 
+	// cast('125e342.83' as unsigned)
+	f = BuildCastFunction(ctx, &Constant{Value: types.NewDatum("125e342.83"), RetType: types.NewFieldType(mysql.TypeString)}, tp1)
+	res, err = f.Eval(chunk.Row{})
+	c.Assert(err, IsNil)
+	c.Assert(res.GetUint64() == 125, IsTrue)
+
+	warnings = sc.GetWarnings()
+	lastWarn = warnings[len(warnings)-1]
+	c.Assert(terror.ErrorEqual(types.ErrOverflow, lastWarn.Err), IsTrue, Commentf("err %v", lastWarn.Err))
+
+	// cast('1e9223372036854775807' as unsigned)
+	f = BuildCastFunction(ctx, &Constant{Value: types.NewDatum("1e9223372036854775807"), RetType: types.NewFieldType(mysql.TypeString)}, tp1)
+	res, err = f.Eval(chunk.Row{})
+	c.Assert(err, IsNil)
+	c.Assert(res.GetUint64() == 1, IsTrue)
+
+	warnings = sc.GetWarnings()
+	lastWarn = warnings[len(warnings)-1]
+	c.Assert(terror.ErrorEqual(types.ErrOverflow, lastWarn.Err), IsTrue, Commentf("err %v", lastWarn.Err))
+
 	// cast('18446744073709551616' as signed);
 	mask := ^mysql.UnsignedFlag
 	tp1.Flag &= uint(mask)
@@ -149,6 +169,26 @@ func (s *testEvaluatorSuite) TestCast(c *C) {
 	lastWarn = warnings[len(warnings)-1]
 	c.Assert(terror.ErrorEqual(types.ErrCastAsSignedOverflow, lastWarn.Err), IsTrue, Commentf("err %v", lastWarn.Err))
 
+	// cast('125e342.83' as signed)
+	f = BuildCastFunction(ctx, &Constant{Value: types.NewDatum("125e342.83"), RetType: types.NewFieldType(mysql.TypeString)}, tp1)
+	res, err = f.Eval(chunk.Row{})
+	c.Assert(err, IsNil)
+	c.Assert(res.GetInt64() == 125, IsTrue)
+
+	warnings = sc.GetWarnings()
+	lastWarn = warnings[len(warnings)-1]
+	c.Assert(terror.ErrorEqual(types.ErrOverflow, lastWarn.Err), IsTrue, Commentf("err %v", lastWarn.Err))
+
+	// cast('1e9223372036854775807' as signed)
+	f = BuildCastFunction(ctx, &Constant{Value: types.NewDatum("1e9223372036854775807"), RetType: types.NewFieldType(mysql.TypeString)}, tp1)
+	res, err = f.Eval(chunk.Row{})
+	c.Assert(err, IsNil)
+	c.Assert(res.GetInt64() == 1, IsTrue)
+
+	warnings = sc.GetWarnings()
+	lastWarn = warnings[len(warnings)-1]
+	c.Assert(terror.ErrorEqual(types.ErrOverflow, lastWarn.Err), IsTrue, Commentf("err %v", lastWarn.Err))
+
 	// create table t1(s1 time);
 	// insert into t1 values('11:11:11');
 	// select cast(s1 as decimal(7, 2)) from t1;
diff --git a/types/convert.go b/types/convert.go
index 82aac0547b04a..ab727258b76fd 100644
--- a/types/convert.go
+++ b/types/convert.go
@@ -236,13 +236,13 @@ func getValidIntPrefix(sc *stmtctx.StatementContext, str string) (string, error)
 	if err != nil {
 		return floatPrefix, errors.Trace(err)
 	}
-	return floatStrToIntStr(floatPrefix)
+	return floatStrToIntStr(sc, floatPrefix, str)
 }
 
 // floatStrToIntStr converts a valid float string into valid integer string which can be parsed by
 // strconv.ParseInt, we can't parse float first then convert it to string because precision will
 // be lost.
-func floatStrToIntStr(validFloat string) (string, error) {
+func floatStrToIntStr(sc *stmtctx.StatementContext, validFloat string, oriStr string) (string, error) {
 	var dotIdx = -1
 	var eIdx = -1
 	for i := 0; i < len(validFloat); i++ {
@@ -275,7 +275,8 @@ func floatStrToIntStr(validFloat string) (string, error) {
 	}
 	if exp > 0 && int64(intCnt) > (math.MaxInt64-int64(exp)) {
 		// (exp + incCnt) overflows MaxInt64.
-		return validFloat, ErrOverflow.GenByArgs("BIGINT", validFloat)
+		sc.AppendWarning(ErrOverflow.GenByArgs("BIGINT", oriStr))
+		return validFloat[:eIdx], nil
 	}
 	intCnt += exp
 	if intCnt <= 0 {
@@ -291,8 +292,9 @@ func floatStrToIntStr(validFloat string) (string, error) {
 		// convert scientific notation decimal number
 		extraZeroCount := intCnt - len(digits)
 		if extraZeroCount > 20 {
-			// Return overflow to avoid allocating too much memory.
-			return validFloat, ErrOverflow.GenByArgs("BIGINT", validFloat)
+			// Append overflow warning and return to avoid allocating too much memory.
+			sc.AppendWarning(ErrOverflow.GenByArgs("BIGINT", oriStr))
+			return validFloat[:eIdx], nil
 		}
 		validInt = string(digits) + strings.Repeat("0", extraZeroCount)
 	}
diff --git a/types/convert_test.go b/types/convert_test.go
index 2a70a0a2d9d9a..08657d627dae8 100644
--- a/types/convert_test.go
+++ b/types/convert_test.go
@@ -684,10 +684,15 @@ func (s *testTypeConvertSuite) TestGetValidFloat(c *C) {
 		_, err := strconv.ParseFloat(prefix, 64)
 		c.Assert(err, IsNil)
 	}
-	_, err := floatStrToIntStr("1e9223372036854775807")
-	c.Assert(terror.ErrorEqual(err, ErrOverflow), IsTrue, Commentf("err %v", err))
-	_, err = floatStrToIntStr("1e21")
-	c.Assert(terror.ErrorEqual(err, ErrOverflow), IsTrue, Commentf("err %v", err))
+	floatStr, err := floatStrToIntStr(sc, "1e9223372036854775807", "1e9223372036854775807")
+	c.Assert(err, IsNil)
+	c.Assert(floatStr, Equals, "1")
+	floatStr, err = floatStrToIntStr(sc, "125e342", "125e342.83")
+	c.Assert(err, IsNil)
+	c.Assert(floatStr, Equals, "125")
+	floatStr, err = floatStrToIntStr(sc, "1e21", "1e21")
+	c.Assert(err, IsNil)
+	c.Assert(floatStr, Equals, "1")
 }
 
 // TestConvertTime tests time related conversion.