Added support for 64-bit shifts, fix const-wide

Change-Id: I4823056d83652ecc7d3e391e905d480d73fab718
diff --git a/src/runtime_support.S b/src/runtime_support.S
new file mode 100644
index 0000000..46c2dd3
--- /dev/null
+++ b/src/runtime_support.S
@@ -0,0 +1,72 @@
+#if defined(__arm__)
+
+    .balign 4
+    .global art_shl_long
+art_shl_long:
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to ignore all but the low
+     * 6 bits.
+     * On entry:
+     *   r0: low word
+     *   r1: high word
+     *   r2: shift count
+     */
+    /* shl-long vAA, vBB, vCC */
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    bx      lr
+
+    .balign 4
+    .global art_shr_long
+art_shr_long:
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to ignore all but the low
+     * 6 bits.
+     * On entry:
+     *   r0: low word
+     *   r1: high word
+     *   r2: shift count
+     */
+    /* shr-long vAA, vBB, vCC */
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    bx      lr
+
+    .balign 4
+    .global art_ushr_long
+art_ushr_long:
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to ignore all but the low
+     * 6 bits.
+     * On entry:
+     *   r0: low word
+     *   r1: high word
+     *   r2: shift count
+     */
+    /* ushr-long vAA, vBB, vCC */
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    bx      lr
+
+#endif