FFI: Optimize snapshots for cdata comparisons.

4 files changed, 23 insertions, 8 deletions

diff --git a/src/Makefile.dep b/src/Makefile.dep
index b869fdff..de46c57e 100644
--- a/src/Makefile.dep
+++ b/src/Makefile.dep
@@ -75,9 +75,9 @@ lj_cparse.o: lj_cparse.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_ctype.h lj_cparse.h lj_frame.h \
  lj_bc.h lj_vm.h lj_char.h
 lj_crecord.o: lj_crecord.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
- lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_ctype.h lj_gc.h lj_cparse.h \
- lj_cconv.h lj_clib.h lj_ir.h lj_jit.h lj_iropt.h lj_trace.h \
- lj_dispatch.h lj_bc.h lj_traceerr.h lj_ffrecord.h lj_crecord.h
+ lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_frame.h lj_bc.h lj_ctype.h \
+ lj_gc.h lj_cparse.h lj_cconv.h lj_clib.h lj_ir.h lj_jit.h lj_iropt.h \
+ lj_trace.h lj_dispatch.h lj_traceerr.h lj_ffrecord.h lj_crecord.h
 lj_ctype.o: lj_ctype.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_gc.h lj_err.h lj_errmsg.h lj_str.h lj_tab.h lj_ctype.h
 lj_dispatch.o: lj_dispatch.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
diff --git a/src/lj_crecord.c b/src/lj_crecord.c
index e5413fb3..b4bfd0c2 100644
--- a/src/lj_crecord.c
+++ b/src/lj_crecord.c
@@ -13,6 +13,7 @@
 #include "lj_err.h"
 #include "lj_str.h"
 #include "lj_tab.h"
+#include "lj_frame.h"
 #include "lj_ctype.h"
 #include "lj_cparse.h"
 #include "lj_cconv.h"
@@ -835,6 +836,15 @@ void LJ_FASTCALL recff_cdata_arith(jit_State *J, RecordFFData *rd)
     err_type:
       lj_trace_err(J, LJ_TRERR_BADTYPE);
     }
+    /* Fixup cdata comparisons, too. Avoids some cdata escapes. */
+    if (J->postproc == LJ_POST_FIXGUARD && frame_iscont(J->L->base-1)) {
+      const BCIns *pc = frame_contpc(J->L->base-1) - 1;
+      if (bc_op(*pc) <= BC_ISNEP) {
+        setframe_pc(&J2G(J)->tmptv, pc);
+        J2G(J)->tmptv.u32.lo = ((tref_istrue(tr) ^ bc_op(*pc)) & 1);
+        J->postproc = LJ_POST_FIXCOMP;
+      }
+    }
     J->base[0] = tr;
   }
 }
diff --git a/src/lj_jit.h b/src/lj_jit.h
index 38970fc7..2f3df2a0 100644
--- a/src/lj_jit.h
+++ b/src/lj_jit.h
@@ -111,6 +111,7 @@ typedef enum {
 /* Post-processing action. */
 typedef enum {
   LJ_POST_NONE,         /* No action. */
+  LJ_POST_FIXCOMP,      /* Fixup comparison and emit pending guard. */
   LJ_POST_FIXGUARD      /* Fixup and emit pending guard. */
 } PostProc;
 
diff --git a/src/lj_record.c b/src/lj_record.c
index 94bd0ed4..d8c67499 100644
--- a/src/lj_record.c
+++ b/src/lj_record.c
@@ -1355,10 +1355,10 @@ static void rec_comp_prep(jit_State *J)
 }
 
 /* Fixup comparison. */
-static void rec_comp_fixup(jit_State *J, int cond)
+static void rec_comp_fixup(jit_State *J, const BCIns *pc, int cond)
 {
-  BCIns jmpins = J->pc[1];
-  const BCIns *npc = J->pc + 2 + (cond ? bc_j(jmpins) : 0);
+  BCIns jmpins = pc[1];
+  const BCIns *npc = pc + 2 + (cond ? bc_j(jmpins) : 0);
   SnapShot *snap = &J->cur.snap[J->cur.nsnap-1];
   /* Set PC to opposite target to avoid re-recording the comp. in side trace. */
   J->cur.snapmap[snap->mapofs + snap->nent] = SNAP_MKPC(npc);
@@ -1383,6 +1383,10 @@ void lj_record_ins(jit_State *J)
   /* Perform post-processing action before recording the next instruction. */
   if (LJ_UNLIKELY(J->postproc != LJ_POST_NONE)) {
     switch (J->postproc) {
+    case LJ_POST_FIXCOMP:  /* Fixup comparison. */
+      pc = frame_pc(&J2G(J)->tmptv);
+      rec_comp_fixup(J, pc, (!tvistruecond(&J2G(J)->tmptv2) ^ (bc_op(*pc)&1)));
+      /* fallthrough */
     case LJ_POST_FIXGUARD:  /* Fixup and emit pending guard. */
       if (!tvistruecond(&J2G(J)->tmptv2)) {
         BCReg s;
@@ -1505,7 +1509,7 @@ void lj_record_ins(jit_State *J)
         break;
       }
       emitir(IRTG(irop, ta), ra, rc);
-      rec_comp_fixup(J, ((int)op ^ irop) & 1);
+      rec_comp_fixup(J, J->pc, ((int)op ^ irop) & 1);
     }
     break;
 
@@ -1529,7 +1533,7 @@ void lj_record_ins(jit_State *J)
         rec_mm_equal(J, &ix, (int)op);
         break;
       }
-      rec_comp_fixup(J, ((int)op & 1) == !diff);
+      rec_comp_fixup(J, J->pc, ((int)op & 1) == !diff);
     }
     break;