gcd.diff

diff --git a/src/jit/graph.c b/src/jit/graph.c
index 1efcd66..77a8b66 100644
--- a/src/jit/graph.c
+++ b/src/jit/graph.c
@@ -1912,6 +1912,7 @@ start:
     case MVM_OP_sp_add_bi:
     case MVM_OP_sp_sub_bi:
     case MVM_OP_sp_mul_bi:
+    case MVM_OP_sp_gcd_bi:
     case MVM_OP_sp_unbox_bi:
     case MVM_OP_sp_takewrite_bi:
     case MVM_OP_sp_materialize_bi:
diff --git a/src/jit/x64/emit.dasc b/src/jit/x64/emit.dasc
index 9d37026..2441e76 100644
--- a/src/jit/x64/emit.dasc
+++ b/src/jit/x64/emit.dasc
@@ -2461,7 +2461,8 @@ void MVM_jit_emit_primitive(MVMThreadContext *tc, MVMJitCompiler *compiler, MVMJ
     }
     case MVM_OP_sp_add_bi:
     case MVM_OP_sp_sub_bi:
-    case MVM_OP_sp_mul_bi: {
+    case MVM_OP_sp_mul_bi:
+    case MVM_OP_sp_gcd_bi: {
         MVMint16 a = ins->operands[1].reg.orig;
         MVMint16 b = ins->operands[2].reg.orig;
         MVMint16 c = ins->operands[0].reg.orig;
@@ -2474,8 +2475,7 @@ void MVM_jit_emit_primitive(MVMThreadContext *tc, MVMJitCompiler *compiler, MVMJ
         | cmp TMP1d, MVM_BIGINT_32_FLAG;
         | jne >1;
 
-        /* Both smallint, so try to do the calculation. If it overflows, fall
-         * back to the slow path. */
+        /* Both smallint, so try to do the calculation. */
         | mov TMP4d, dword [WORK + 8 * a + 4]
         switch (op) {
             case MVM_OP_sp_add_bi:
@@ -2487,8 +2487,38 @@ void MVM_jit_emit_primitive(MVMThreadContext *tc, MVMJitCompiler *compiler, MVMJ
             case MVM_OP_sp_mul_bi:
                 | imul TMP4d, dword [WORK + 8 * b + 4]
                 break;
+            case MVM_OP_sp_gcd_bi:
+                /* Need absolute values first. We'll want the first value in
+                 * eax for the GCD itself, but need eax for the absolute too,
+                 * so process the second operand first and move it into ecx. */
+                | mov eax, dword [WORK + 8 * b + 4]
+                | cdq
+                | xor eax, edx
+                | sub eax, edx
+                | mov ecx, eax
+                | mov eax, TMP4d
+                | cdq
+                | xor eax, edx
+                | sub eax, edx
+
+                /* We now have the values to do the GCD on in eax and ecx (we
+                 * avoid ebx since that is WORK); only eax matters since it's
+                 * the implicit argument to div. */
+                |4:
+                | xor edx, edx
+                | div ecx
+                | mov eax, ecx
+                | mov ecx, edx
+                | test ecx, ecx
+                | jnz >4
+                | mov TMP4d, eax
+                break;
         }
-        | jo >1
+
+        /* If it overflows, fall back to the slow path. (Some operations
+         * always produce a smaller value than the input, so cannot.) */
+        if (op != MVM_OP_sp_gcd_bi)
+            | jo >1
 
         /* No overflow; store the result. */
         | mov dword [WORK + 8 * c], MVM_BIGINT_32_FLAG
@@ -2513,6 +2543,9 @@ void MVM_jit_emit_primitive(MVMThreadContext *tc, MVMJitCompiler *compiler, MVMJ
             case MVM_OP_sp_mul_bi:
                 | callp &MVM_bigint_fallback_mul;
                 break;
+            case MVM_OP_sp_gcd_bi:
+                | callp &MVM_bigint_fallback_gcd;
+                break;
         }
         |3:
         break;