FFI: Fix and optimize recording of cdata[cdata].

2 files changed, 41 insertions, 4 deletions

diff --git a/src/lj_crecord.c b/src/lj_crecord.c
index d689b81b..c93cece3 100644
--- a/src/lj_crecord.c
+++ b/src/lj_crecord.c
@@ -503,10 +503,17 @@ void LJ_FASTCALL recff_cdata_index(jit_State *J, RecordFFData *rd)
     IRType t;
     if (ctype_isenum(ctk->info)) ctk = ctype_child(cts, ctk);
     if (ctype_isinteger(ctk->info) && (t = crec_ct2irt(ctk)) != IRT_CDATA) {
-      idx = emitir(IRT(IR_ADD, IRT_PTR), idx, lj_ir_kintp(J, sizeof(GCcdata)));
-      idx = emitir(IRT(IR_XLOAD, t), idx, 0);
-      if (!LJ_64 && (t == IRT_I64 || t == IRT_U64)) {
-        idx = emitconv(idx, IRT_INT, t, 0);
+      if (ctk->size == 8) {
+        idx = emitir(IRT(IR_FLOAD, t), idx, IRFL_CDATA_INT64);
+      } else {
+        idx = emitir(IRT(IR_ADD, IRT_PTR), idx,
+                     lj_ir_kintp(J, sizeof(GCcdata)));
+        idx = emitir(IRT(IR_XLOAD, t), idx, 0);
+      }
+      if (LJ_64 && ctk->size < sizeof(intptr_t) && !(ctk->info & CTF_UNSIGNED))
+        idx = emitconv(idx, IRT_INTP, IRT_INT, IRCONV_SEXT);
+      if (!LJ_64 && ctk->size > sizeof(intptr_t)) {
+        idx = emitconv(idx, IRT_INTP, t, 0);
         lj_needsplit(J);
       }
       goto integer_key;
diff --git a/src/lj_opt_fold.c b/src/lj_opt_fold.c
index 495b784d..471a4b29 100644
--- a/src/lj_opt_fold.c
+++ b/src/lj_opt_fold.c
@@ -897,6 +897,16 @@ LJFOLDF(simplify_conv_i64_num)
   return NEXTFOLD;
 }
 
+LJFOLD(CONV CONV IRCONV_INT_I64)  /* _INT */
+LJFOLD(CONV CONV IRCONV_INT_U64)  /* _INT */
+LJFOLDF(simplify_conv_int_i64)
+{
+  PHIBARRIER(fleft);
+  if ((fleft->op2 & IRCONV_SRCMASK) == IRT_INT)
+    return fleft->op1;
+  return NEXTFOLD;
+}
+
 /* Shortcut TOBIT + IRT_NUM <- IRT_INT/IRT_U32 conversion. */
 LJFOLD(TOBIT CONV KNUM)
 LJFOLDF(simplify_tobit_conv)
@@ -956,6 +966,26 @@ LJFOLDF(simplify_conv_sext)
   return NEXTFOLD;
 }
 
+/* Strength reduction of narrowing. */
+LJFOLD(CONV ADD IRCONV_INT_I64)
+LJFOLD(CONV SUB IRCONV_INT_I64)
+LJFOLD(CONV MUL IRCONV_INT_I64)
+LJFOLD(CONV ADD IRCONV_INT_U64)
+LJFOLD(CONV SUB IRCONV_INT_U64)
+LJFOLD(CONV MUL IRCONV_INT_U64)
+LJFOLDF(simplify_conv_narrow)
+{
+  IROp op = (IROp)fleft->o;
+  IRRef op1 = fleft->op1, op2 = fleft->op2, mode = fins->op2;
+  PHIBARRIER(fleft);
+  op1 = emitir(IRTI(IR_CONV), op1, mode);
+  op2 = emitir(IRTI(IR_CONV), op2, mode);
+  fins->ot = IRTI(op);
+  fins->op1 = op1;
+  fins->op2 = op2;
+  return RETRYFOLD;
+}
+
 /* Special CSE rule for CONV. */
 LJFOLD(CONV any any)
 LJFOLDF(cse_conv)