[PATCH 6/6] Add generic hp-timing support

[Adhemerval Zanella <adhemerval dot zanella at linaro dot org>]

From: Wilco Dijkstra <Wilco.Dijkstra@arm.com>

Add missing generic hp_timing support.  It uses clock_gettime (CLOCK_MONOTONIC)
which has unspecified starting time, nano-second accuracy, and should faster on
architectures that implementes the symbol as vDSO.

Checked on aarch64-linux-gnu, x86_64-linux-gnu, and i686-linux-gnu. I also
checked the builds for all afected ABIs.

	* benchtests/Makefile (USE_CLOCK_GETTIME) Remove.
	* benchtests/README: Update description.
	* benchtests/bench-timing.h: Default to hp-timing.
	* sysdeps/generic/hp-timing.h (HP_TIMING_DIFF, HP_TIMING_ACCUM_NT,
	HP_TIMING_PRINT): Remove.
	(HP_TIMING_NOW): Add generic implementation.
	(hp_timing_t): Change to uint64_t.
---
 benchtests/Makefile         |  6 -----
 benchtests/README           |  7 +-----
 benchtests/bench-timing.h   | 44 +++++++------------------------------
 sysdeps/generic/hp-timing.h | 25 +++++++++++++--------
 4 files changed, 25 insertions(+), 57 deletions(-)

diff --git a/benchtests/Makefile b/benchtests/Makefile
index e1907753b4..50cb7cd2e8 100644
--- a/benchtests/Makefile
+++ b/benchtests/Makefile
@@ -127,17 +127,11 @@ endif
 
 CPPFLAGS-nonlib += -DDURATION=$(BENCH_DURATION) -D_ISOMAC
 
-# Use clock_gettime to measure performance of functions.  The default is to use
-# HP_TIMING if it is available.
-ifdef USE_CLOCK_GETTIME
-CPPFLAGS-nonlib += -DUSE_CLOCK_GETTIME
-else
 # On x86 processors, use RDTSCP, instead of RDTSC, to measure performance
 # of functions.  All x86 processors since 2010 support RDTSCP instruction.
 ifdef USE_RDTSCP
 CPPFLAGS-nonlib += -DUSE_RDTSCP
 endif
-endif
 
 DETAILED_OPT :=
 
diff --git a/benchtests/README b/benchtests/README
index aaf0b659e2..c4f03fd872 100644
--- a/benchtests/README
+++ b/benchtests/README
@@ -27,12 +27,7 @@ BENCH_DURATION.
 
 The benchmark suite does function call measurements using architecture-specific
 high precision timing instructions whenever available.  When such support is
-not available, it uses clock_gettime (CLOCK_PROCESS_CPUTIME_ID).  One can force
-the benchmark to use clock_gettime by invoking make as follows:
-
-  $ make USE_CLOCK_GETTIME=1 bench
-
-Again, one must run `make bench-clean' before changing the measurement method.
+not available, it uses clock_gettime (CLOCK_MONOTONIC).
 
 On x86 processors, RDTSCP instruction provides more precise timing data
 than RDTSC instruction.  All x86 processors since 2010 support RDTSCP
diff --git a/benchtests/bench-timing.h b/benchtests/bench-timing.h
index 93fe379f99..e213dec3fd 100644
--- a/benchtests/bench-timing.h
+++ b/benchtests/bench-timing.h
@@ -18,49 +18,21 @@
 
 #undef attribute_hidden
 #define attribute_hidden
+#define __clock_gettime clock_gettime
 #include <hp-timing.h>
 #include <stdint.h>
 
-#if HP_TIMING_INLINE && !defined USE_CLOCK_GETTIME
-# define GL(x) _##x
-# define GLRO(x) _##x
+#define GL(x) _##x
+#define GLRO(x) _##x
 typedef hp_timing_t timing_t;
 
-# define TIMING_TYPE "hp_timing"
+#define TIMING_TYPE "hp_timing"
 
-# define TIMING_INIT(res) ({ (res) = 1; })
+#define TIMING_INIT(res) ({ (res) = 1; })
 
-# define TIMING_NOW(var) HP_TIMING_NOW (var)
-# define TIMING_DIFF(diff, start, end) HP_TIMING_DIFF ((diff), (start), (end))
-# define TIMING_ACCUM(sum, diff) HP_TIMING_ACCUM_NT ((sum), (diff))
-
-#else
-
-#include <time.h>
-typedef uint64_t timing_t;
-
-# define TIMING_TYPE "clock_gettime"
-
-/* Measure the resolution of the clock so we can scale the number of
-   benchmark iterations by this value.  */
-# define TIMING_INIT(res) \
-({									      \
-  struct timespec start;						      \
-  clock_getres (CLOCK_PROCESS_CPUTIME_ID, &start);			      \
-  (res) = start.tv_nsec;					      \
-})
-
-# define TIMING_NOW(var) \
-({									      \
-  struct timespec tv;							      \
-  clock_gettime (CLOCK_PROCESS_CPUTIME_ID, &tv);			      \
-  (var) = (uint64_t) (tv.tv_nsec + (uint64_t) 1000000000 * tv.tv_sec);	      \
-})
-
-# define TIMING_DIFF(diff, start, end) (diff) = (end) - (start)
-# define TIMING_ACCUM(sum, diff) (sum) += (diff)
-
-#endif
+#define TIMING_NOW(var) HP_TIMING_NOW (var)
+#define TIMING_DIFF(diff, start, end) HP_TIMING_DIFF ((diff), (start), (end))
+#define TIMING_ACCUM(sum, diff) HP_TIMING_ACCUM_NT ((sum), (diff))
 
 #define TIMING_PRINT_MEAN(d_total_s, d_iters) \
   printf ("\t%g", (d_total_s) / (d_iters))
diff --git a/sysdeps/generic/hp-timing.h b/sysdeps/generic/hp-timing.h
index 278998d2c2..2528279558 100644
--- a/sysdeps/generic/hp-timing.h
+++ b/sysdeps/generic/hp-timing.h
@@ -20,16 +20,23 @@
 #ifndef _HP_TIMING_H
 #define _HP_TIMING_H	1
 
-/* There are no generic definitions for the times.  We could write something
-   using the `gettimeofday' system call where available but the overhead of
-   the system call might be too high.  */
+#include <time.h>
+#include <stdint.h>
+#include <hp-timing-common.h>
 
-/* Provide dummy definitions.  */
+/* It should not be used for ld.so.  */
 #define HP_TIMING_INLINE	(0)
-typedef int hp_timing_t;
-#define HP_TIMING_NOW(var)
-#define HP_TIMING_DIFF(Diff, Start, End)
-#define HP_TIMING_ACCUM_NT(Sum, Diff)
-#define HP_TIMING_PRINT(Buf, Len, Val)
+
+typedef uint64_t hp_timing_t;
+
+/* The clock_gettime (CLOCK_MONOTONIC) has unspecified starting time,
+   nano-second accuracy, and for some architectues is implemented as
+   vDSO symbol.  */
+#define HP_TIMING_NOW(var) \
+({								\
+  struct timespec tv;						\
+  __clock_gettime (CLOCK_MONOTONIC, &tv);			\
+  (var) = (tv.tv_nsec + UINT64_C(1000000000) * tv.tv_sec);	\
+})
 
 #endif	/* hp-timing.h */
-- 
2.17.1