1 files changed, 16 insertions, 9 deletions

diff --git a/src/lj_asm_ppc.h b/src/lj_asm_ppc.h
index 73942b8b..ec22e260 100644
--- a/src/lj_asm_ppc.h
+++ b/src/lj_asm_ppc.h
@@ -162,22 +162,24 @@ static Reg asm_fuseahuref(ASMState *as, IRRef ref, int32_t *ofsp, RegSet allow)
 
 /* Fuse XLOAD/XSTORE reference into load/store operand. */
 static void asm_fusexref(ASMState *as, PPCIns pi, Reg rt, IRRef ref,
-                         RegSet allow)
+                         RegSet allow, int32_t ofs)
 {
   IRIns *ir = IR(ref);
-  int32_t ofs = 0;
   Reg base;
   if (ra_noreg(ir->r) && mayfuse(as, ref)) {
     if (ir->o == IR_ADD) {
-      if (irref_isk(ir->op2) && (ofs = IR(ir->op2)->i, checki16(ofs))) {
+      int32_t ofs2;
+      if (irref_isk(ir->op2) && (ofs2 = ofs + IR(ir->op2)->i, checki16(ofs2))) {
+        ofs = ofs2;
         ref = ir->op1;
-      } else {
+      } else if (ofs == 0) {
         Reg right, left = ra_alloc2(as, ir, allow);
         right = (left >> 8); left &= 255;
         emit_fab(as, PPCI_LWZX | ((pi >> 20) & 0x780), rt, left, right);
         return;
       }
     } else if (ir->o == IR_STRREF) {
+      lua_assert(ofs == 0);
       ofs = (int32_t)sizeof(GCstr);
       if (irref_isk(ir->op2)) {
         ofs += IR(ir->op2)->i;
@@ -904,13 +906,13 @@ static void asm_xload(ASMState *as, IRIns *ir)
   lua_assert(!(ir->op2 & IRXLOAD_UNALIGNED));
   if (irt_isi8(ir->t))
     emit_as(as, PPCI_EXTSB, dest, dest);
-  asm_fusexref(as, asm_fxloadins(ir), dest, ir->op1, RSET_GPR);
+  asm_fusexref(as, asm_fxloadins(ir), dest, ir->op1, RSET_GPR, 0);
 }
 
-static void asm_xstore(ASMState *as, IRIns *ir)
+static void asm_xstore(ASMState *as, IRIns *ir, int32_t ofs)
 {
   IRIns *irb;
-  if (mayfuse(as, ir->op2) && (irb = IR(ir->op2))->o == IR_BSWAP &&
+  if (ofs == 0 && mayfuse(as, ir->op2) && (irb = IR(ir->op2))->o == IR_BSWAP &&
       ra_noreg(irb->r) && (irt_isint(ir->t) || irt_isu32(ir->t))) {
     /* Fuse BSWAP with XSTORE to stwbrx. */
     Reg src = ra_alloc1(as, irb->op1, RSET_GPR);
@@ -918,7 +920,7 @@ static void asm_xstore(ASMState *as, IRIns *ir)
   } else {
     Reg src = ra_alloc1(as, ir->op2, irt_isfp(ir->t) ? RSET_FPR : RSET_GPR);
     asm_fusexref(as, asm_fxstoreins(ir), src, ir->op1,
-                 rset_exclude(RSET_GPR, src));
+                 rset_exclude(RSET_GPR, src), ofs);
   }
 }
 
@@ -1743,6 +1745,11 @@ static void asm_hiop(ASMState *as, IRIns *ir)
     as->curins--;  /* Always skip the loword comparison. */
     asm_comp64(as, ir);
     return;
+  } else if ((ir-1)->o == IR_XSTORE) {
+    as->curins--;  /* Handle both stores here. */
+    asm_xstore(as, ir, 0);
+    asm_xstore(as, ir-1, 4);
+    return;
   }
   if (!usehi) return;  /* Skip unused hiword op for all remaining ops. */
   switch ((ir-1)->o) {
@@ -2035,7 +2042,7 @@ static void asm_ir(ASMState *as, IRIns *ir)
 
   case IR_ASTORE: case IR_HSTORE: case IR_USTORE: asm_ahustore(as, ir); break;
   case IR_FSTORE: asm_fstore(as, ir); break;
-  case IR_XSTORE: asm_xstore(as, ir); break;
+  case IR_XSTORE: asm_xstore(as, ir, 0); break;
 
   /* Allocations. */
   case IR_SNEW: case IR_XSNEW: asm_snew(as, ir); break;