erts, kernel: Add os:perf_counter function

The perf_counter is a very very cheap and high resolution timer
that can be used to timestamp system events. It does not have
monoticity guarantees, but should on most OS's expose a monotonous
time.

A special instruction has been created for this counter to further
speed up fetching it.

OTP-12908

5 files changed, 119 insertions, 2 deletions

diff --git a/erts/emulator/beam/beam_emu.c b/erts/emulator/beam/beam_emu.c
index 4d7b00b032..a0979c7b0e 100644
--- a/erts/emulator/beam/beam_emu.c
+++ b/erts/emulator/beam/beam_emu.c
@@ -4895,6 +4895,38 @@ do {						\
      }
  }
 
+ /* This is optimised as an instruction because
+    it has to be very very fast */
+ OpCase(i_perf_counter): {
+    BeamInstr* next;
+    ErtsSysHrTime ts;
+    PreFetch(0, next);
+
+    sys_perf_counter(&ts);
+
+    if (IS_SSMALL(ts)) {
+        r(0) = make_small((Sint)ts);
+    } else {
+        TestHeap(ERTS_SINT64_HEAP_SIZE(ts),0);
+        r(0) = make_big(HTOP);
+#if defined(ARCH_32) || HALFWORD_HEAP
+        if (ts >= (((Uint64) 1) << 32)) {
+            *HTOP = make_pos_bignum_header(2);
+            BIG_DIGIT(HTOP, 0) = (Uint) (ts & ((Uint) 0xffffffff));
+            BIG_DIGIT(HTOP, 1) = (Uint) ((ts >> 32) & ((Uint) 0xffffffff));
+            HTOP += 3;
+        }
+        else
+#endif
+        {
+                *HTOP = make_pos_bignum_header(1);
+                BIG_DIGIT(HTOP, 0) = (Uint) ts;
+                HTOP += 2;
+        }
+    }
+    NextPF(0, next);
+ }
+
  OpCase(i_debug_breakpoint): {
      HEAVY_SWAPOUT;
      I = call_error_handler(c_p, I-3, reg, am_breakpoint);
diff --git a/erts/emulator/beam/bif.tab b/erts/emulator/beam/bif.tab
index 1b8ae8cef5..4efc055aaf 100644
--- a/erts/emulator/beam/bif.tab
+++ b/erts/emulator/beam/bif.tab
@@ -170,6 +170,7 @@ bif erts_internal:term_type/1
 bif erts_internal:map_hashmap_children/1
 
 bif erts_internal:time_unit/0
+bif erts_internal:perf_counter_unit/0
 
 bif erts_internal:is_system_process/1
 
@@ -369,6 +370,7 @@ bif os:getpid/0
 bif os:timestamp/0
 bif os:system_time/0
 bif os:system_time/1
+bif os:perf_counter/0
 
 #
 # Bifs in the erl_ddll module (the module actually does not exist)
diff --git a/erts/emulator/beam/erl_time.h b/erts/emulator/beam/erl_time.h
index 93a0d556bf..446adcf4af 100644
--- a/erts/emulator/beam/erl_time.h
+++ b/erts/emulator/beam/erl_time.h
@@ -142,6 +142,8 @@ erts_time_unit_conversion(Uint64 value,
 			  Uint32 from_time_unit,
 			  Uint32 to_time_unit);
 
+ErtsSysHrTime erts_perf_counter_unit(void);
+
 #if ERTS_GLB_INLINE_INCL_FUNC_DEF
 
 ERTS_GLB_INLINE Uint64
diff --git a/erts/emulator/beam/erl_time_sup.c b/erts/emulator/beam/erl_time_sup.c
index 668dbb5daa..6509c6a805 100644
--- a/erts/emulator/beam/erl_time_sup.c
+++ b/erts/emulator/beam/erl_time_sup.c
@@ -1749,7 +1749,7 @@ erts_get_monotonic_time(ErtsSchedulerData *esdp)
 {
     ErtsMonotonicTime mtime = time_sup.r.o.get_time();
     update_last_mtime(esdp, mtime);
-    return mtime;    
+    return mtime;
 }
 
 ErtsMonotonicTime
@@ -2435,9 +2435,81 @@ BIF_RETTYPE os_system_time_0(BIF_ALIST_0)
     BIF_RET(make_time_val(BIF_P, stime));
 }
 
-BIF_RETTYPE os_system_time_1(BIF_ALIST_0)
+BIF_RETTYPE os_system_time_1(BIF_ALIST_1)
 {
     ErtsSystemTime stime = erts_os_system_time();
     BIF_RET(time_unit_conversion(BIF_P, BIF_ARG_1, stime, 0));
 }
 
+BIF_RETTYPE
+os_perf_counter_0(BIF_ALIST_0)
+{
+    ErtsSysHrTime pcounter;
+    sys_perf_counter(&pcounter);
+    BIF_RET(make_time_val(BIF_P, pcounter));
+}
+
+BIF_RETTYPE erts_internal_perf_counter_unit_0(BIF_ALIST_0)
+{
+    BIF_RET(make_time_val(BIF_P, SYS_PERF_COUNTER_UNIT));
+}
+
+/* What resolution to spin to in micro seconds */
+#define RESOLUTION 100
+/* How many iterations to spin */
+#define ITERATIONS 1
+/* How many significant figures to round to */
+#define SIGFIGS 3
+
+static ErtsSysHrTime perf_counter_unit = 0;
+
+static ErtsSysHrTime erts_calculate_perf_counter_unit(void);
+static ErtsSysHrTime erts_calculate_perf_counter_unit() {
+    int i;
+    ErtsSysHrTime pre, post;
+    double value = 0;
+    double round_factor;
+#if defined(HAVE_GETHRTIME) && defined(GETHRTIME_WITH_CLOCK_GETTIME)
+    struct timespec basetime,comparetime;
+#define __GETTIME(arg) clock_gettime(CLOCK_MONOTONIC,arg)
+#define __GETUSEC(arg) (arg.tv_nsec / 1000)
+#else
+    SysTimeval basetime,comparetime;
+#define __GETTIME(arg) sys_gettimeofday(arg)
+#define __GETUSEC(arg) arg.tv_usec
+#endif
+
+    for (i = 0; i < ITERATIONS; i++) {
+        /* Make sure usec just flipped over at current resolution */
+        __GETTIME(&basetime);
+        do {
+            __GETTIME(&comparetime);
+        } while ((__GETUSEC(basetime) / RESOLUTION) == (__GETUSEC(comparetime) / RESOLUTION));
+
+        sys_perf_counter(&pre);
+
+        __GETTIME(&basetime);
+        do {
+            __GETTIME(&comparetime);
+        } while ((__GETUSEC(basetime) / RESOLUTION) == (__GETUSEC(comparetime) / RESOLUTION));
+
+        sys_perf_counter(&post);
+
+        value += post - pre;
+    }
+    /* After this value is ticks per us */
+    value /= (RESOLUTION*ITERATIONS);
+
+    /* We round to 3 significant figures */
+    round_factor = pow(10.0, SIGFIGS - ceil(log10(value)));
+    value = ((ErtsSysHrTime)(value * round_factor + 0.5)) / round_factor;
+
+    /* convert to ticks per second */
+    return 1000000 * value;
+}
+
+ErtsSysHrTime erts_perf_counter_unit() {
+    if (perf_counter_unit == 0)
+        perf_counter_unit = erts_calculate_perf_counter_unit();
+    return perf_counter_unit;
+}
diff --git a/erts/emulator/beam/ops.tab b/erts/emulator/beam/ops.tab
index 081c4108a0..9e53b4bfcc 100644
--- a/erts/emulator/beam/ops.tab
+++ b/erts/emulator/beam/ops.tab
@@ -988,6 +988,13 @@ move Discarded x==0 | move Something x==0 => move Something x=0
 
 %endif
 
+call_ext u==0 u$func:os:perf_counter/0 => \
+    i_perf_counter
+call_ext_last u==0 u$func:os:perf_counter/0 D => \
+    i_perf_counter | deallocate_return D
+call_ext_only u==0 u$func:os:perf_counter/0 => \
+    i_perf_counter | return
+
 #
 # The general case for BIFs that have no special instructions.
 # A BIF used in the tail must be followed by a return instruction.
@@ -1027,6 +1034,8 @@ i_apply_fun_only
 
 i_hibernate
 
+i_perf_counter
+
 call_bif e
 
 #