Replace GCproto reference with bytecode PC in GCfuncL.

3 files changed, 32 insertions, 30 deletions

diff --git a/src/buildvm_x86.dasc b/src/buildvm_x86.dasc
index e09c37e9..90d80dbf 100644
--- a/src/buildvm_x86.dasc
+++ b/src/buildvm_x86.dasc
@@ -305,6 +305,8 @@
 #define DISPATCH_GL(field)      (GG_DISP2G + (int)offsetof(global_State, field))
 #define DISPATCH_J(field)       (GG_DISP2J + (int)offsetof(jit_State, field))
 |
+#define PC2PROTO(field)  ((int)offsetof(GCproto, field)-(int)sizeof(GCproto))
+|
 |// Decrement hashed hotcount and trigger trace recorder if zero.
 |.macro hotloop, reg
 |  mov reg, PC
@@ -406,11 +408,10 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  // RA = new base, RB = LFUNC, RC = nargs+1, (BASE = old base), PC = return
   |  // DISPATCH initialized
   |  mov BASE, RA
-  |  mov PROTO:RB, LFUNC:RB->pt
   |  mov [BASE-4], PC                   // Store caller PC.
-  |  movzx RA, byte PROTO:RB->framesize
-  |  lea PC, [PROTO:RB+sizeof(GCproto)]
-  |  mov KBASE, PROTO:RB->k
+  |  mov PC, LFUNC:RB->pc
+  |  movzx RA, byte [PC+PC2PROTO(framesize)]
+  |  mov KBASE, [PC+PC2PROTO(k)]
   |  mov L:RB, SAVE_L
   |  lea RA, [BASE+RA*8]                // Top of frame.
   |  lea RC, [BASE+NARGS:RC*8-4]        // Points to tag of 1st free slot.
@@ -437,16 +438,16 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  lea PC, [NARGS:RC*8+FRAME_VARG]
   |  lea BASE, [RA+PC-FRAME_VARG]
   |  mov [BASE-8], LFUNC:RB             // Store copy of LFUNC.
-  |  mov PROTO:RB, LFUNC:RB->pt
   |  mov [BASE-4], PC                   // Store delta + FRAME_VARG.
-  |  movzx PC, byte PROTO:RB->framesize
-  |  lea KBASE, [BASE+PC*8]
-  |  mov L:PC, SAVE_L
+  |  mov PC, LFUNC:RB->pc
+  |  movzx RB, byte [PC+PC2PROTO(framesize)]
+  |  lea KBASE, [BASE+RB*8]
+  |  mov L:RB, SAVE_L
   |  lea RC, [BASE+4]
-  |  cmp KBASE, L:PC->maxstack
+  |  cmp KBASE, L:RB->maxstack
   |  ja ->gate_lv_growstack             // Need to grow stack.
-  |  movzx PC, byte PROTO:RB->numparams
-  |  test PC, PC
+  |  movzx RB, byte [PC+PC2PROTO(numparams)]
+  |  test RB, RB
   |  jz >2
   |1:  // Copy fixarg slots up.
   |  add RA, 8
@@ -458,12 +459,11 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  mov [RC], KBASE
   |  add RC, 8
   |  mov dword [RA-4], LJ_TNIL          // Clear old fixarg slot (help the GC).
-  |  sub PC, 1
+  |  sub RB, 1
   |  jnz <1
   |2:
-  |  movzx RA, byte PROTO:RB->framesize
-  |  lea PC, [PROTO:RB+sizeof(GCproto)]
-  |  mov KBASE, PROTO:RB->k
+  |  movzx RA, byte [PC+PC2PROTO(framesize)]
+  |  mov KBASE, [PC+PC2PROTO(k)]
   |  lea RA, [BASE+RA*8]
   |  jmp <9
   |
@@ -663,6 +663,7 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  jmp >1
   |
   |->gate_lv_growstack:                 // Grow stack for vararg Lua function.
+  //XXX
   |  sub RC, 8
   |  mov BASE, RA
   |  mov RA, KBASE
@@ -863,8 +864,8 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  mov RA, dword [RB-16]
   |.endif
   |  mov LFUNC:KBASE, [BASE-8]
-  |  mov PROTO:KBASE, LFUNC:KBASE->pt
-  |  mov KBASE, PROTO:KBASE->k
+  |  mov KBASE, LFUNC:KBASE->pc
+  |  mov KBASE, [KBASE+PC2PROTO(k)]
   |  // BASE = base, RC = result, RB = meta base
   |  jmp RAa                            // Jump to continuation.
   |
@@ -2719,8 +2720,8 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |->vm_exit_interp:
 #if LJ_HASJIT
   |  mov LFUNC:KBASE, [BASE-8]
-  |  mov PROTO:KBASE, LFUNC:KBASE->pt
-  |  mov KBASE, PROTO:KBASE->k
+  |  mov KBASE, LFUNC:KBASE->pc
+  |  mov KBASE, [KBASE+PC2PROTO(k)]
   |  mov dword [DISPATCH+DISPATCH_GL(jit_L)], 0
   |  set_vmstate INTERP
   |  ins_next
@@ -4454,8 +4455,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     |  not RDa
     |  lea BASE, [BASE+RD*8]
     |  mov LFUNC:KBASE, [BASE-8]
-    |  mov PROTO:KBASE, LFUNC:KBASE->pt
-    |  mov KBASE, PROTO:KBASE->k
+    |  mov KBASE, LFUNC:KBASE->pc
+    |  mov KBASE, [KBASE+PC2PROTO(k)]
     |  jmp <4
     |
     |7:  // Tailcall from a vararg function.
@@ -4492,8 +4493,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     |  ins_AB_  // RA = base, RB = nresults+1, (RC = 1)
     |  mov LFUNC:RC, [BASE-8]
     |  lea RA, [BASE+RA*8]
-    |  mov PROTO:RC, LFUNC:RC->pt
-    |  movzx RC, byte PROTO:RC->numparams
+    |  mov RC, LFUNC:RC->pc
+    |  movzx RC, byte [RC+PC2PROTO(numparams)]
     |  mov TMP1, KBASE                  // Need one more free register.
     |  lea KBASE, [BASE+RC*8+(8+FRAME_VARG)]
     |  sub KBASE, [BASE-4]
@@ -4618,8 +4619,8 @@ static void build_ins(BuildCtx *ctx, BCOp op, int defop, int cmov, int sse)
     |  not RAa                          // Note: ~RA = -(RA+1)
     |  lea BASE, [BASE+RA*8]            // base = base - (RA+1)*8
     |  mov LFUNC:KBASE, [BASE-8]
-    |  mov PROTO:KBASE, LFUNC:KBASE->pt
-    |  mov KBASE, PROTO:KBASE->k
+    |  mov KBASE, LFUNC:KBASE->pc
+    |  mov KBASE, [KBASE+PC2PROTO(k)]
     |  ins_next
     |
     |6:  // Fill up results with nil.
diff --git a/src/lj_func.c b/src/lj_func.c
index adc8c039..7d130176 100644
--- a/src/lj_func.c
+++ b/src/lj_func.c
@@ -112,8 +112,8 @@ GCfunc *lj_func_newL(lua_State *L, GCproto *pt, GCtab *env)
   fn->l.gct = ~LJ_TFUNC;
   fn->l.ffid = FF_LUA;
   fn->l.nupvalues = cast_byte(pt->sizeuv);
-  /* NOBARRIER: The GCfunc is new (marked white). */
-  setgcref(fn->l.pt, obj2gco(pt));
+  /* NOBARRIER: Really a setgcref. But the GCfunc is new (marked white). */
+  setmref(fn->l.pc, proto_bc(pt));
   setgcref(fn->l.env, obj2gco(env));
   fn->l.gate = (pt->flags & PROTO_IS_VARARG) ? lj_gate_lv : lj_gate_lf;
   return fn;
@@ -137,7 +137,7 @@ GCfunc *lj_func_newL_gc(lua_State *L, GCproto *pt, GCfuncL *parent)
     GCupval *uv;
     if ((v & 0x8000)) {
       uv = func_finduv(L, base + (v & 0xff));
-      uv->dhash = (uint32_t)(uintptr_t)gcref(parent->pt) ^ (v << 24);
+      uv->dhash = (uint32_t)(uintptr_t)mref(parent->pc, char) ^ (v << 24);
     } else {
       uv = &gcref(puv[v])->uv;
     }
diff --git a/src/lj_obj.h b/src/lj_obj.h
index f243a75b..a6637954 100644
--- a/src/lj_obj.h
+++ b/src/lj_obj.h
@@ -429,7 +429,7 @@ typedef struct GCfuncC {
 
 typedef struct GCfuncL {
   GCfuncHeader;
-  GCRef pt;             /* Link to prototype this function is based on. */
+  MRef pc;              /* Start PC (and GCproto reference). */
   GCRef uvptr[1];       /* Array of _pointers_ to upvalue objects (GCupval). */
 } GCfuncL;
 
@@ -443,7 +443,8 @@ typedef union GCfunc {
 #define isluafunc(fn)   ((fn)->c.ffid == FF_LUA)
 #define iscfunc(fn)     ((fn)->c.ffid == FF_C)
 #define isffunc(fn)     ((fn)->c.ffid > FF_C)
-#define funcproto(fn)   check_exp(isluafunc(fn), &gcref((fn)->l.pt)->pt)
+#define funcproto(fn) \
+  check_exp(isluafunc(fn), (GCproto *)(mref((fn)->l.pc, char)-sizeof(GCproto)))
 #define sizeCfunc(n)    (sizeof(GCfuncC)-sizeof(TValue)+sizeof(TValue)*(n))
 #define sizeLfunc(n)    (sizeof(GCfuncL)-sizeof(GCRef)+sizeof(GCRef)*(n))