about summary refs log tree commit diff
path: root/benchtests/scripts/bench.py
diff options
context:
space:
mode:
authorWilco Dijkstra <wdijkstr@arm.com>2017-08-17 16:27:20 +0100
committerWilco Dijkstra <wdijkstr@arm.com>2017-08-17 16:27:20 +0100
commitd4505b895f2aa577a6b974ab4bd0a1fd0db60ceb (patch)
tree5ae35fc88eda5f2ad45b63f4617f4a87d470d885 /benchtests/scripts/bench.py
parent34d6a3cbf2be45aa039a7eb9f0084a4b710437b8 (diff)
downloadglibc-d4505b895f2aa577a6b974ab4bd0a1fd0db60ceb.tar.gz
glibc-d4505b895f2aa577a6b974ab4bd0a1fd0db60ceb.tar.xz
glibc-d4505b895f2aa577a6b974ab4bd0a1fd0db60ceb.zip
Add math benchmark latency test
This patch further improves math function benchmarking by adding a latency
test in addition to throughput.  This enables more accurate comparisons of the
math functions. The latency test works by creating a dependency on the previous
iteration: func_res = F (func_res * zero + input[i]). The multiply by zero
avoids changing the input.

It reports reciprocal throughput and latency in nanoseconds (depending on the
timing header used) and max/min throughput in iterations per second:

   "workload-spec2006.wrf": {
    "reciprocal-throughput": 100,
    "latency": 200,
    "max-throughput": 1.0e+07,
    "min-throughput": 5.0e+06
   }

	* benchtests/bench-skeleton.c (main): Add support for
	latency benchmarking.
	* benchtests/scripts/bench.py: Add support for latency benchmarking.
Diffstat (limited to 'benchtests/scripts/bench.py')
-rwxr-xr-xbenchtests/scripts/bench.py20
1 files changed, 14 insertions, 6 deletions
diff --git a/benchtests/scripts/bench.py b/benchtests/scripts/bench.py
index 8c1c9eeb2b..33dc7b9e9f 100755
--- a/benchtests/scripts/bench.py
+++ b/benchtests/scripts/bench.py
@@ -45,7 +45,7 @@ DEFINES_TEMPLATE = '''
 # variant is represented by the _VARIANT structure.  The ARGS structure
 # represents a single set of arguments.
 STRUCT_TEMPLATE = '''
-#define CALL_BENCH_FUNC(v, i) %(func)s (%(func_args)s)
+#define CALL_BENCH_FUNC(v, i, x) %(func)s (x %(func_args)s)
 
 struct args
 {
@@ -84,7 +84,9 @@ EPILOGUE = '''
 #define RESULT(__v, __i) (variants[(__v)].in[(__i)].timing)
 #define RESULT_ACCUM(r, v, i, old, new) \\
         ((RESULT ((v), (i))) = (RESULT ((v), (i)) * (old) + (r)) / ((new) + 1))
-#define BENCH_FUNC(i, j) ({%(getret)s CALL_BENCH_FUNC (i, j);})
+#define BENCH_FUNC(i, j) ({%(getret)s CALL_BENCH_FUNC (i, j, );})
+#define BENCH_FUNC_LAT(i, j) ({%(getret)s CALL_BENCH_FUNC (i, j, %(latarg)s);})
+#define BENCH_VARS %(defvar)s
 #define FUNCNAME "%(func)s"
 #include "bench-skeleton.c"'''
 
@@ -122,17 +124,23 @@ def gen_source(func, directives, all_vals):
     # If we have a return value from the function, make sure it is
     # assigned to prevent the compiler from optimizing out the
     # call.
+    getret = ''
+    latarg = ''
+    defvar = ''
+
     if directives['ret']:
         print('static %s volatile ret;' % directives['ret'])
-        getret = 'ret = '
-    else:
-        getret = ''
+        print('static %s zero __attribute__((used)) = 0;' % directives['ret'])
+        getret = 'ret = func_res = '
+        # Note this may not work if argument and result type are incompatible.
+        latarg = 'func_res * zero +'
+        defvar = '%s func_res = 0;' % directives['ret']
 
     # Test initialization.
     if directives['init']:
         print('#define BENCH_INIT %s' % directives['init'])
 
-    print(EPILOGUE % {'getret': getret, 'func': func})
+    print(EPILOGUE % {'getret': getret, 'func': func, 'latarg': latarg, 'defvar': defvar })
 
 
 def _print_arg_data(func, directives, all_vals):