x86: Conditionally use SSE3 for 64 bit conversions.

Reported by Gero Schwäricke. #1496 #1411

1 files changed, 20 insertions, 1 deletions

diff --git a/src/lj_obj.h b/src/lj_obj.h
index 96dc1e0d0..30991a7e9 100644
--- a/src/lj_obj.h
+++ b/src/lj_obj.h
@@ -1035,11 +1035,30 @@ LJ_ASMF LJ_CONSTF int64_t lj_vm_num2int_check(double x);
 ** The uint64_t conversion accepts the union of the unsigned + signed range.
 */
 LJ_ASMF LJ_CONSTF int64_t lj_vm_num2i64(double x);
-LJ_ASMF LJ_CONSTF int64_t lj_vm_num2u64(double x);
+LJ_ASMF LJ_CONSTF uint64_t lj_vm_num2u64(double x);
+
+#if LJ_TARGET_X86
+
+LJ_ASMF LJ_CONSTF int64_t lj_vm_num2i64_sse3(double x);
+LJ_ASMF LJ_CONSTF uint64_t lj_vm_num2u64_sse3(double x);
+LJ_ASMF int64_t (*lj_vm_num2i64_ptr)(double x);
+LJ_ASMF uint64_t (*lj_vm_num2u64_ptr)(double x);
+static LJ_AINLINE int64_t lj_num2i64(double x)
+{
+  return (*lj_vm_num2i64_ptr)(x);
+}
+static LJ_AINLINE uint64_t lj_num2u64(double x)
+{
+  return (*lj_vm_num2u64_ptr)(x);
+}
+
+#else
 
 #define lj_num2i64(x)           (lj_vm_num2i64((x)))
 #define lj_num2u64(x)           (lj_vm_num2u64((x)))
 
+#endif
+
 /* Lua BitOp conversion semantics use the 2^52 + 2^51 trick. */
 LJ_ASMF LJ_CONSTF int32_t lj_vm_tobit(double x);