Low-overhead profiler, part 4: JIT compiler support.

src/Makefile.dep

@@ -170,12 +170,12 @@ lj_parse.o: lj_parse.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_vm.h lj_vmevent.h
 lj_profile.o: lj_profile.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h \
  lj_buf.h lj_gc.h lj_str.h lj_frame.h lj_bc.h lj_debug.h lj_dispatch.h \
- lj_jit.h lj_ir.h lj_profile.h luajit.h
+ lj_jit.h lj_ir.h lj_trace.h lj_traceerr.h lj_profile.h luajit.h
 lj_record.o: lj_record.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_meta.h lj_frame.h lj_bc.h \
- lj_ctype.h lj_gc.h lj_ff.h lj_ffdef.h lj_ir.h lj_jit.h lj_ircall.h \
- lj_iropt.h lj_trace.h lj_dispatch.h lj_traceerr.h lj_record.h \
- lj_ffrecord.h lj_snap.h lj_vm.h
+ lj_ctype.h lj_gc.h lj_ff.h lj_ffdef.h lj_debug.h lj_ir.h lj_jit.h \
+ lj_ircall.h lj_iropt.h lj_trace.h lj_dispatch.h lj_traceerr.h \
+ lj_record.h lj_ffrecord.h lj_snap.h lj_vm.h
 lj_snap.o: lj_snap.c lj_obj.h lua.h luaconf.h lj_def.h lj_arch.h lj_gc.h \
  lj_tab.h lj_state.h lj_frame.h lj_bc.h lj_ir.h lj_jit.h lj_iropt.h \
  lj_trace.h lj_dispatch.h lj_traceerr.h lj_snap.h lj_target.h \

src/lj_asm.c

@@ -1565,6 +1565,7 @@ static void asm_ir(ASMState *as, IRIns *ir)
   case IR_PHI: asm_phi(as, ir); break;
   case IR_HIOP: asm_hiop(as, ir); break;
   case IR_GCSTEP: asm_gcstep(as, ir); break;
+  case IR_PROF: asm_prof(as, ir); break;
 
   /* Guarded assertions. */
   case IR_LT: case IR_GE: case IR_LE: case IR_GT:

src/lj_asm_arm.h

@@ -1915,6 +1915,16 @@ static void asm_hiop(ASMState *as, IRIns *ir)
 #endif
 }
 
+/* -- Profiling ----------------------------------------------------------- */
+
+static void asm_prof(ASMState *as, IRIns *ir)
+{
+  UNUSED(ir);
+  asm_guardcc(as, CC_NE);
+  emit_n(as, ARMI_TST|ARMI_K12|HOOK_PROFILE, RID_TMP);
+  emit_lsptr(as, ARMI_LDRB, RID_TMP, (void *)&J2G(as->J)->hookmask);
+}
+
 /* -- Stack handling ------------------------------------------------------ */
 
 /* Check Lua stack size for overflow. Use exit handler as fallback. */

src/lj_asm_mips.h

@@ -1562,6 +1562,17 @@ static void asm_hiop(ASMState *as, IRIns *ir)
 #endif
 }
 
+/* -- Profiling ----------------------------------------------------------- */
+
+static void asm_prof(ASMState *as, IRIns *ir)
+{
+  UNUSED(ir);
+  asm_guard(as, MIPSI_BNE, RID_TMP, RID_ZERO);
+  emit_tsi(as, MIPSI_ANDI, RID_TMP, RID_TMP, HOOK_PROFILE);
+  emit_lsglptr(as, MIPSI_LBU, RID_TMP,
+	       (int32_t)offsetof(global_State, hookmask));
+}
+
 /* -- Stack handling ------------------------------------------------------ */
 
 /* Check Lua stack size for overflow. Use exit handler as fallback. */

src/lj_asm_ppc.h

@@ -1738,6 +1738,17 @@ static void asm_hiop(ASMState *as, IRIns *ir)
 #endif
 }
 
+/* -- Profiling ----------------------------------------------------------- */
+
+static void asm_prof(ASMState *as, IRIns *ir)
+{
+  UNUSED(ir);
+  asm_guardcc(as, CC_NE);
+  emit_asi(as, PPCI_ANDIDOT, RID_TMP, RID_TMP, HOOK_PROFILE);
+  emit_lsglptr(as, PPCI_LBZ, RID_TMP,
+	       (int32_t)offsetof(global_State, hookmask));
+}
+
 /* -- Stack handling ------------------------------------------------------ */
 
 /* Check Lua stack size for overflow. Use exit handler as fallback. */

src/lj_asm_x86.h

@@ -2348,6 +2348,16 @@ static void asm_hiop(ASMState *as, IRIns *ir)
 #endif
 }
 
+/* -- Profiling ----------------------------------------------------------- */
+
+static void asm_prof(ASMState *as, IRIns *ir)
+{
+  UNUSED(ir);
+  asm_guardcc(as, CC_NE);
+  emit_i8(as, HOOK_PROFILE);
+  emit_rma(as, XO_GROUP3b, XOg_TEST, &J2G(as->J)->hookmask);
+}
+
 /* -- Stack handling ------------------------------------------------------ */
 
 /* Check Lua stack size for overflow. Use exit handler as fallback. */

src/lj_ir.h

@@ -40,6 +40,7 @@
   _(USE,	S , ref, ___) \
   _(PHI,	S , ref, ref) \
   _(RENAME,	S , ref, lit) \
+  _(PROF,	S , ___, ___) \
   \
   /* Constants. */ \
   _(KPRI,	N , ___, ___) \

src/lj_jit.h

@@ -396,6 +396,12 @@ typedef struct jit_State {
   size_t szallmcarea;	/* Total size of all allocated mcode areas. */
 
   TValue errinfo;	/* Additional info element for trace errors. */
+
+#if LJ_HASPROFILE
+  GCproto *prev_pt;	/* Previous prototype. */
+  BCLine prev_line;	/* Previous line. */
+  int prof_mode;	/* Profiling mode: 0, 'f', 'l'. */
+#endif
 }
 #if LJ_TARGET_ARM
 LJ_ALIGN(16)		/* For DISPATCH-relative addresses in assembler part. */

src/lj_opt_fold.c

@@ -2285,6 +2285,17 @@ LJFOLDF(barrier_tnew_tdup)
   return DROPFOLD;
 }
 
+/* -- Profiling ----------------------------------------------------------- */
+
+LJFOLD(PROF any any)
+LJFOLDF(prof)
+{
+  IRRef ref = J->chain[IR_PROF];
+  if (ref+1 == J->cur.nins)  /* Drop neighbouring IR_PROF. */
+    return ref;
+  return EMITFOLD;
+}
+
 /* -- Stores and allocations ---------------------------------------------- */
 
 /* Stores and allocations cannot be folded or passed on to CSE in general.

src/lj_profile.c

@@ -14,6 +14,10 @@
 #include "lj_frame.h"
 #include "lj_debug.h"
 #include "lj_dispatch.h"
+#if LJ_HASJIT
+#include "lj_jit.h"
+#include "lj_trace.h"
+#endif
 #include "lj_profile.h"
 
 #include "luajit.h"
@@ -218,13 +222,20 @@ LUA_API void luaJIT_profile_start(lua_State *L, const char *mode,
   ProfileState *ps = &profile_state;
   int interval = LJ_PROFILE_INTERVAL_DEFAULT;
   while (*mode) {
-    switch (*mode++) {
+    int m = *mode++;
+    switch (m) {
     case 'i':
       interval = 0;
       while (*mode >= '0' && *mode <= '9')
 	interval = interval * 10 + (*mode++ - '0');
       if (interval <= 0) interval = 1;
       break;
+#if LJ_HASJIT
+    case 'l': case 'f':
+      L2J(L)->prof_mode = m;
+      lj_trace_flushall(L);
+      break;
+#endif
     default:  /* Ignore unknown mode chars. */
       break;
     }
@@ -251,6 +262,10 @@ LUA_API void luaJIT_profile_stop(lua_State *L)
     profile_timer_stop(ps);
     g->hookmask &= ~HOOK_PROFILE;
     lj_dispatch_update(g);
+#if LJ_HASJIT
+    G2J(g)->prof_mode = 0;
+    lj_trace_flushall(L);
+#endif
     lj_buf_free(g, &ps->sb);
     setmref(ps->sb.b, NULL);
     setmref(ps->sb.e, NULL);

src/lj_record.c

@@ -20,6 +20,9 @@
 #endif
 #include "lj_bc.h"
 #include "lj_ff.h"
+#if LJ_HASPROFILE
+#include "lj_debug.h"
+#endif
 #include "lj_ir.h"
 #include "lj_jit.h"
 #include "lj_ircall.h"
@@ -579,6 +582,52 @@ static void rec_loop_jit(jit_State *J, TraceNo lnk, LoopEvent ev)
   }  /* Side trace continues across a loop that's left or not entered. */
 }
 
+/* -- Record profiler hook checks ----------------------------------------- */
+
+#if LJ_HASPROFILE
+
+/* Need to insert profiler hook check? */
+static int rec_profile_need(jit_State *J, GCproto *pt, const BCIns *pc)
+{
+  GCproto *ppt;
+  lua_assert(J->prof_mode == 'f' || J->prof_mode == 'l');
+  if (!pt)
+    return 0;
+  ppt = J->prev_pt;
+  J->prev_pt = pt;
+  if (pt != ppt && ppt) {
+    J->prev_line = -1;
+    return 1;
+  }
+  if (J->prof_mode == 'l') {
+    BCLine line = lj_debug_line(pt, proto_bcpos(pt, pc));
+    BCLine pline = J->prev_line;
+    J->prev_line = line;
+    if (pline != line)
+      return 1;
+  }
+  return 0;
+}
+
+static void rec_profile_ins(jit_State *J, const BCIns *pc)
+{
+  if (J->prof_mode && rec_profile_need(J, J->pt, pc)) {
+    emitir(IRTG(IR_PROF, IRT_NIL), 0, 0);
+    lj_snap_add(J);
+  }
+}
+
+static void rec_profile_ret(jit_State *J)
+{
+  if (J->prof_mode == 'f') {
+    emitir(IRTG(IR_PROF, IRT_NIL), 0, 0);
+    J->prev_pt = NULL;
+    lj_snap_add(J);
+  }
+}
+
+#endif
+
 /* -- Record calls and returns -------------------------------------------- */
 
 /* Specialize to the runtime value of the called function or its prototype. */
@@ -1770,6 +1819,10 @@ void lj_record_ins(jit_State *J)
   rec_check_ir(J);
 #endif
 
+#if LJ_HASPROFILE
+  rec_profile_ins(J, pc);
+#endif
+
   /* Keep a copy of the runtime values of var/num/str operands. */
 #define rav	(&ix.valv)
 #define rbv	(&ix.tabv)
@@ -2074,6 +2127,9 @@ void lj_record_ins(jit_State *J)
     rc = (BCReg)(J->L->top - J->L->base) - ra + 1;
     /* fallthrough */
   case BC_RET: case BC_RET0: case BC_RET1:
+#if LJ_HASPROFILE
+    rec_profile_ret(J);
+#endif
     lj_record_ret(J, ra, (ptrdiff_t)rc-1);
     break;
 
@@ -2303,6 +2359,10 @@ void lj_record_setup(jit_State *J)
     if (1 + J->pt->framesize >= LJ_MAX_JSLOTS)
       lj_trace_err(J, LJ_TRERR_STACKOV);
   }
+#if LJ_HASPROFILE
+  J->prev_pt = NULL;
+  J->prev_line = -1;
+#endif
 #ifdef LUAJIT_ENABLE_CHECKHOOK
   /* Regularly check for instruction/line hooks from compiled code and
   ** exit to the interpreter if the hooks are set.

src/lj_trace.c

@@ -766,17 +766,20 @@ int LJ_FASTCALL lj_trace_exit(jit_State *J, void *exptr)
   if (errcode)
     return -errcode;  /* Return negated error code. */
 
-  lj_vmevent_send(L, TEXIT,
-    lj_state_checkstack(L, 4+RID_NUM_GPR+RID_NUM_FPR+LUA_MINSTACK);
-    setintV(L->top++, J->parent);
-    setintV(L->top++, J->exitno);
-    trace_exit_regs(L, ex);
-  );
+  if (!(LJ_HASPROFILE && (G(L)->hookmask & HOOK_PROFILE)))
+    lj_vmevent_send(L, TEXIT,
+      lj_state_checkstack(L, 4+RID_NUM_GPR+RID_NUM_FPR+LUA_MINSTACK);
+      setintV(L->top++, J->parent);
+      setintV(L->top++, J->exitno);
+      trace_exit_regs(L, ex);
+    );
 
   pc = exd.pc;
   cf = cframe_raw(L->cframe);
   setcframe_pc(cf, pc);
-  if (G(L)->gc.state == GCSatomic || G(L)->gc.state == GCSfinalize) {
+  if (LJ_HASPROFILE && (G(L)->hookmask & HOOK_PROFILE)) {
+    /* Just exit to interpreter. */
+  } else if (G(L)->gc.state == GCSatomic || G(L)->gc.state == GCSfinalize) {
     if (!(G(L)->hookmask & HOOK_GC))
       lj_gc_step(L);  /* Exited because of GC: drive GC forward. */
   } else {