Merge branch 'master' of github.com:RRZE-HPC/kerncraft

examples/kernels/himeno.c

@@ -4,9 +4,9 @@ float gosa, ss, s0, omega;
 float a[4][L][M][N], b[3][L][M][N], c[3][L][M][N];
 float p[L][M][N], bnd[L][M][N], wrk1[L][M][N], wrk2[L][M][N];
 
-for(int i=1 ; i<L; i++)
-  for(int j=1 ; j<M ; j++)
-    for(int k=1 ; k<N ; k++){
+for(int i=1 ; i<L-1; i++)
+  for(int j=1 ; j<M-1; j++)
+    for(int k=1 ; k<N-1; k++){
           s0 = a[0][i][j][k]*p[i+1][j][k]
              + a[1][i][j][k]*p[i][j+1][k]
              + a[2][i][j][k]*p[i][j][k+1]

kerncraft/__init__.py

@@ -1,5 +1,5 @@
 """Kerncraft static analytical performance modeling framework and tool."""
-__version__ = '0.5.9'
+__version__ = '0.5.10'
 
 
 def get_header_path():

kerncraft/headers/dummy.c

@@ -1,2 +1,2 @@
-void dummy(double* a) {}
-int var_false = 0;
+void dummy(void* a) {}
+int var_false = 0;

kerncraft/kernel.py

@@ -1016,7 +1016,7 @@ def as_code(self, type_='iaca'):
         # add dummy function declaration
         decl = c_ast.Decl('dummy', [], [], [], c_ast.FuncDecl(
             c_ast.ParamList([c_ast.Typename(None, [], c_ast.PtrDecl(
-                [], c_ast.TypeDecl(None, [], c_ast.IdentifierType(['double']))))]),
+                [], c_ast.TypeDecl(None, [], c_ast.IdentifierType(['void']))))]),
             c_ast.TypeDecl('dummy', [], c_ast.IdentifierType(['void']))),
             None, None)
         ast.ext.insert(0, decl)

kerncraft/models/benchmark.py

@@ -17,6 +17,7 @@
     from itertools import zip_longest
 except ImportError:
     from itertools import izip_longest as zip_longest
+from pprint import pprint
 
 import six
 import sympy
@@ -272,6 +273,7 @@ def perfctr(self, cmd, group='MEM', cpu='S0:0', code_markers=True, pin=True):
     def analyze(self):
         """Run analysis."""
         bench = self.kernel.build(verbose=self.verbose > 1)
+        element_size = self.kernel.datatypes_size[self.kernel.datatype]
 
         # Build arguments to pass to command:
         args = [bench] + [six.text_type(s) for s in list(self.kernel.constants.values())]
@@ -325,7 +327,6 @@ def analyze(self):
                         event_counter_results[sym] = measured_ctrs[event][r]
 
             # Analytical metrics needed for futher calculation
-            element_size = self.kernel.datatypes_size[self.kernel.datatype]
             elements_per_cacheline = float(self.machine['cacheline size']) // element_size
             total_iterations = self.kernel.iteration_length() * repetitions
             total_cachelines = total_iterations/elements_per_cacheline
@@ -381,14 +382,13 @@ def analyze(self):
 
         self.results['Runtime (per repetition) [s]'] = time_per_repetition
         # TODO make more generic to support other (and multiple) constant names
-        # TODO support SP (devide by 4 instead of 8.0)
         iterations_per_repetition = reduce(
             operator.mul,
             [self.kernel.subs_consts(max_-min_)/self.kernel.subs_consts(step)
              for idx, min_, max_, step in self.kernel._loop_stack],
             1)
         self.results['Iterations per repetition'] = iterations_per_repetition
-        iterations_per_cacheline = float(self.machine['cacheline size'])/8.0
+        iterations_per_cacheline = float(self.machine['cacheline size'])/element_size
         cys_per_repetition = time_per_repetition*float(self.machine['clock'])
         self.results['Runtime (per cacheline update) [cy/CL]'] = \
             (cys_per_repetition/iterations_per_repetition)*iterations_per_cacheline
@@ -405,6 +405,9 @@ def analyze(self):
 
     def report(self, output_file=sys.stdout):
         """Report gathered analysis data in human readable form."""
+        if self.verbose > 1:
+            pprint(self.results)
+
         if self.verbose > 0:
             print('Runtime (per repetition): {:.2g} s'.format(
                       self.results['Runtime (per repetition) [s]']),
@@ -430,7 +433,7 @@ def report(self, output_file=sys.stdout):
                   file=output_file)
         print('', file=output_file)
 
-        if self.no_phenoecm:
+        if not self.no_phenoecm:
             print("Data Transfers:")
             print("{:^8} |".format("cache"), end='')
             for metrics in self.results['data transfers'].values():