From e3a442e6ec782fcf994023976d95635a368b7e8e Mon Sep 17 00:00:00 2001
From: Luke Gorrie <luke@snabb.co>
Date: Mon, 17 Oct 2016 22:37:50 +0200
Subject: [PATCH] LJ_TRACEPROFILE: Added jit.traceprofile.vmstats()

This function returns counters for total events tallied in each VM
state. Events during traces are aggregated into separate buckets for
non-loop code, loop code, and other code (FFI calls).

Can be used to print summaries like this:

    Interp   C (LJ)       GC     Exit   Record      Opt      ASM  Nonloop     Loop      FFI
    0.000%   0.150%   0.000%   0.000%   0.000%   0.000%   0.000%  23.802%  27.246%  48.802%
---
 src/lib_jit.c         | 22 ++++++++++++++++++++++
 src/lj_traceprofile.c | 22 +++++++++++++---------
 src/lj_traceprofile.h |  9 +++++++++
 src/luajit.h          |  2 ++
 4 files changed, 46 insertions(+), 9 deletions(-)

diff --git a/src/lib_jit.c b/src/lib_jit.c
index 045ce11e..2d4b2af9 100644
--- a/src/lib_jit.c
+++ b/src/lib_jit.c
@@ -33,6 +33,7 @@
 #include "lj_vm.h"
 #include "lj_vmevent.h"
 #include "lj_lib.h"
+#include "lj_def.h"
 
 #include "luajit.h"
 
@@ -650,6 +651,27 @@ LJLIB_CF(jit_traceprofile_tracestats)
   return 0;
 }
 
+LJLIB_CF(jit_traceprofile_vmstats)
+{
+  uint64_t *vmstats = luaJIT_traceprofile_vmstats(L);
+  if (vmstats == NULL) {
+    return 0;
+  } else {
+    setint64V(L->top++, vmstats[LJ_VMST_INTERP]);
+    setint64V(L->top++, vmstats[LJ_VMST_C]);
+    setint64V(L->top++, vmstats[LJ_VMST_GC]);
+    setint64V(L->top++, vmstats[LJ_VMST_EXIT]);
+    setint64V(L->top++, vmstats[LJ_VMST_RECORD]);
+    setint64V(L->top++, vmstats[LJ_VMST_OPT]);
+    setint64V(L->top++, vmstats[LJ_VMST_ASM]);
+    setint64V(L->top++, vmstats[LJ_TRACEPROF_VMST_TRACE_NONLOOP]);
+    setint64V(L->top++, vmstats[LJ_TRACEPROF_VMST_TRACE_LOOP]);
+    setint64V(L->top++, vmstats[LJ_TRACEPROF_VMST_TRACE_OTHER]);
+    setint64V(L->top++, vmstats[LJ_TRACEPROF_VMST_TOTAL]);
+    return 11;
+  }
+}
+
 LJLIB_CF(jit_traceprofile_start)
 {
   int interval = lj_lib_checkint(L, 1);
diff --git a/src/lj_traceprofile.c b/src/lj_traceprofile.c
index 7ba0ef3b..dc998d95 100644
--- a/src/lj_traceprofile.c
+++ b/src/lj_traceprofile.c
@@ -23,16 +23,9 @@
 #include "lj_trace.h"
 #include "lj_traceprofile.h"
 
-/* vmstate extended to represent running in a trace (any one) */
-enum {
-  LJ_TRACEPROF_VMST_TRACE = LJ_VMST__MAX,
-  LJ_TRACEPROF_VMST__MAX
-};
-
 typedef struct TraceProfileState {
   global_State *g;                  /* VM state that started the profiler. */
   struct sigaction oldsa;
-  int events;                       /* Number of events (signals) handled. */
   uint64_t vmstate[LJ_TRACEPROF_VMST__MAX]; /* Counter per VM state */
 } TraceProfileState;
 
@@ -43,7 +36,7 @@ static void traceprofile_signal(int sig, siginfo_t *si, void *data)
   global_State *g = state.g;
   intptr_t ip = (intptr_t)((ucontext_t*)data)->uc_mcontext.gregs[REG_RIP];
   int st = g->vmstate;
-  state.events++;
+  state.vmstate[LJ_TRACEPROF_VMST_TOTAL]++;
   assert((st >= 0) || (~st < LJ_VMST__MAX));
   if (st >= 0) {
     lua_State *L = gco2th(gcref(g->cur_L));
@@ -54,13 +47,15 @@ static void traceprofile_signal(int sig, siginfo_t *si, void *data)
     if ((rel_ip >= 0) && (rel_ip < T->szmcode)) {
       if (rel_ip < T->mcloop) {
 	T->prof.nonloop++;	/* Sample is in non-loop mcode. */
+	state.vmstate[LJ_TRACEPROF_VMST_TRACE_NONLOOP]++;
       } else {
 	T->prof.loop++;		/* Sample is in loop mcode. */
+	state.vmstate[LJ_TRACEPROF_VMST_TRACE_LOOP]++;
       }
     } else {
       T->prof.other++;		/* Sample is outside the trace mcode. */
+      state.vmstate[LJ_TRACEPROF_VMST_TRACE_OTHER]++;
     }
-    state.vmstate[LJ_TRACEPROF_VMST_TRACE]++;
   } else {
     state.vmstate[~st]++;
   }
@@ -100,5 +95,14 @@ LUA_API void luaJIT_traceprofile_stop(lua_State *L)
   traceprofile_stop_timer();
 }
 
+LUA_API uint64_t *luaJIT_traceprofile_vmstats(lua_State *L)
+{
+  if (state.g && L == gco2th(gcref(state.g->cur_L))) {
+    return state.vmstate;
+  } else {
+    return NULL;
+  }
+}
+
 #endif
 
diff --git a/src/lj_traceprofile.h b/src/lj_traceprofile.h
index 152f7e32..38bfde78 100644
--- a/src/lj_traceprofile.h
+++ b/src/lj_traceprofile.h
@@ -6,6 +6,15 @@
 #ifndef _LJ_TRACEPROFILE_H
 #define _LJ_TRACEPROFILE_H
 
+/* vmstate extended to represent running in a trace (any one) */
+typedef enum {
+  LJ_TRACEPROF_VMST_TRACE_NONLOOP = LJ_VMST__MAX,
+  LJ_TRACEPROF_VMST_TRACE_LOOP,
+  LJ_TRACEPROF_VMST_TRACE_OTHER,
+  LJ_TRACEPROF_VMST_TOTAL,
+  LJ_TRACEPROF_VMST__MAX
+} TraceProfileVMState;
+
 typedef struct TraceProfile {
   uint64_t nonloop;	/* Samples taken in non-loop mcode */
   uint64_t loop;	/* Samples taken in loop mcode */
diff --git a/src/luajit.h b/src/luajit.h
index a1ba4b24..8974d0a9 100644
--- a/src/luajit.h
+++ b/src/luajit.h
@@ -29,6 +29,7 @@
 #define _LUAJIT_H
 
 #include "lua.h"
+#include "lj_def.h"
 
 #define LUAJIT_VERSION		"LuaJIT 2.1.0-beta2"
 #define LUAJIT_VERSION_NUM	20100  /* Version 2.1.0 = 02.01.00. */
@@ -76,6 +77,7 @@ LUA_API const char *luaJIT_profile_dumpstack(lua_State *L, const char *fmt,
 /* Trace profiling API. */
 LUA_API void luaJIT_traceprofile_start(lua_State *L, int interval);
 LUA_API void luaJIT_traceprofile_stop(lua_State *L);
+LUA_API uint64_t *luaJIT_traceprofile_vmstats(lua_State *L);
 
 
 /* Enforce (dynamic) linker error for version mismatches. Call from main. */