lj_vmprofile: Port & adapt revisions

Adapt to RaptorJIT style e.g. assume the feature is enabled without
needing the #ifdef.

Port improvement from Snabb LuaJIT branch: tracking GC time separately
for each trace.

src/lib_jit.c

@@ -200,11 +200,8 @@ LJLIB_CF(jit_opt_start)
 
 #include "lj_libdef.h"
 
-
 /* -- jit.vmprofile module  ----------------------------------------------- */
 
-#ifdef LUAJIT_VMPROFILE
-
 #define LJLIB_MODULE_jit_vmprofile
 
 LJLIB_CF(jit_vmprofile_start)
@@ -227,8 +224,6 @@ static int luaopen_jit_vmprofile(lua_State *L)
   return 1;
 }
 
-#endif
-
 /* -- JIT compiler initialization ----------------------------------------- */
 
 /* Default values for JIT parameters. */

src/lj_gc.c

@@ -648,6 +648,7 @@ int lj_gc_step(lua_State *L)
   global_State *g = G(L);
   GCSize lim;
   int32_t ostate = g->vmstate;
+  g->gcvmstate = ostate;
   setvmstate(g, GC);
   lim = (GCSTEPSIZE/100) * g->gc.stepmul;
   if (lim == 0)
@@ -698,6 +699,7 @@ void lj_gc_fullgc(lua_State *L)
 {
   global_State *g = G(L);
   int32_t ostate = g->vmstate;
+  g->gcvmstate = ostate;
   setvmstate(g, GC);
   if (g->gc.state <= GCSatomic) {  /* Caught somewhere in the middle. */
     setmref(g->gc.sweep, &g->gc.root);  /* Sweep everything (preserving it). */

src/lj_obj.h

@@ -516,7 +516,8 @@ typedef struct global_State {
   lua_Alloc allocf;	/* Memory allocator. */
   void *allocd;		/* Memory allocator data. */
   GCState gc;		/* Garbage collector. */
-  volatile int32_t vmstate;  /* VM state or current JIT code trace number. */
+  volatile int32_t vmstate;   /* VM state or current JIT code trace number. */
+  volatile int32_t gcvmstate; /* Previous VM state (only when state is GC). */
   SBuf tmpbuf;		/* Temporary string buffer. */
   GCstr strempty;	/* Empty string. */
   uint8_t stremptyz;	/* Zero terminator of empty string. */

src/lj_vmprofile.c

@@ -6,8 +6,6 @@
 #define lj_vmprofile_c
 #define LUA_CORE
 
-#ifdef LUAJIT_VMPROFILE
-
 #define _GNU_SOURCE 1
 #include <stdio.h>
 #include <assert.h>
@@ -42,7 +40,7 @@ int vmprofile_get_profile_size() {
 void vmprofile_set_profile(void *counters) {
   profile = (VMProfile*)counters;
   profile->magic = 0x1d50f007;
-  profile->major = 1;
+  profile->major = 2;
   profile->minor = 0;
 }
 
@@ -54,16 +52,19 @@ static void vmprofile_signal(int sig, siginfo_t *si, void *data)
   if (profile != NULL) {
     lua_State *L = gco2th(gcref(state.g->cur_L));
     int vmstate = state.g->vmstate;
+    int trace = ~vmstate == LJ_VMST_GC ? state.g->gcvmstate : vmstate;
     /* Not in a trace */
-    if (vmstate < 0) {
+    if (trace < 0) {
       profile->vm[~vmstate]++;
     } else {
-      int bucket = vmstate > LJ_VMPROFILE_TRACE_MAX ? 0 : vmstate;
+      int bucket = trace > LJ_VMPROFILE_TRACE_MAX ? 0 : trace;
       VMProfileTraceCount *count = &profile->trace[bucket];
-      GCtrace *T = traceref(L2J(L), (TraceNo)vmstate);
+      GCtrace *T = traceref(L2J(L), (TraceNo)trace);
       intptr_t ip = (intptr_t)((ucontext_t*)data)->uc_mcontext.gregs[REG_RIP];
       ptrdiff_t mcposition = ip - (intptr_t)T->mcode;
-      if ((mcposition < 0) || (mcposition >= T->szmcode)) {
+      if (~vmstate == LJ_VMST_GC) {
+        count->gc++;
+      } else if ((mcposition < 0) || (mcposition >= T->szmcode)) {
         count->other++;
       } else if ((T->mcloop != 0) && (mcposition >= T->mcloop)) {
         count->loop++;
@@ -110,4 +111,3 @@ LUA_API void luaJIT_vmprofile_stop(lua_State *L)
   stop_timer();
 }
 
-#endif

src/lj_vmprofile.h

@@ -19,12 +19,13 @@ typedef struct VMProfileTraceCount {
   VMProfileCount head;          /* Head of the trace (non-looping part) */
   VMProfileCount loop;          /* Loop of the trace */
   VMProfileCount other;         /* Outside the trace mcode (unidentified) */
+  VMProfileCount gc;            /* Garbage collection from this trace. */
 } VMProfileTraceCount;
 
 /* Complete set of counters for VM and traces. */
 typedef struct VMProfile {
   uint32_t magic;               /* 0x1d50f007 */
-  uint16_t major, minor;        /* 1, 0 */
+  uint16_t major, minor;        /* 2, 0 */
   VMProfileCount vm[LJ_VMST__MAX];
   VMProfileTraceCount trace[LJ_VMPROFILE_TRACE_MAX+1];
 } VMProfile;