Merge pull request #38 from 01org/develop

Develop

Makefile

@@ -186,7 +186,6 @@ else ifeq ($(arch),knl)
  FB_TARGET  	?=       512
  def_block_args	?=	-b 96
  def_block_threads ?=	8
- streaming_stores  ?= 	0
  SUB_BLOCK_LOOP_INNER_MODS  ?=	prefetch(L1)
 
 else ifeq ($(arch),skx)
@@ -224,6 +223,7 @@ else ifeq ($(arch),snb)
 else ifeq ($(arch),intel64)
 
  ISA		?=	-xHOST
+ GCXX_ISA       ?=      -march=native
  FB_TARGET	?=	cpp
 
 else
@@ -233,7 +233,7 @@ $(error Architecture not recognized; use arch=knl, knc, skx, hsw, ivb, snb, or i
 endif # arch-specific.
 
 # general defaults for vars if not set above.
-streaming_stores	?= 	1
+streaming_stores	?= 	0
 omp_par_for		?=	omp parallel for
 omp_region_schedule	?=	dynamic,1
 omp_block_schedule	?=	static,1
@@ -282,13 +282,13 @@ else
 endif
 LD		:=	$(CXX)
 MAKE		:=	make
-CXXOPT		:=	-O3
+CXXOPT		?=	-O3
 CXXFLAGS        +=   	-g -std=c++11 -Wall $(CXXOPT)
 OMPFLAGS	+=	-fopenmp 
 LFLAGS          +=      -lrt
-FB_CXX    	:=       g++  # faster than icpc for the foldBuilder.
+FB_EXEC		:=	bin/foldBuilder.exe
+FB_CXX    	?=	g++  # faster than icpc for the foldBuilder.
 FB_CXXFLAGS 	+=	-g -O0 -std=c++11 -Wall  # low opt to reduce compile time.
-EXTRA_FB_CXXFLAGS =
 FB_FLAGS   	+=	-st $(stencil) -cluster $(cluster) -fold $(fold)
 ST_MACRO_FILE	:=	stencil_macros.hpp
 ST_CODE_FILE	:=	stencil_code.hpp
@@ -382,6 +382,28 @@ else # not Intel compiler
 
 endif # compiler.
 
+# Compile with model_cache=1 or 2 to check prefetching.
+ifeq ($(model_cache),1)
+ MACROS       	+=      MODEL_CACHE=1
+ OMPFLAGS	=	-qopenmp-stubs
+else ifeq ($(model_cache),2)
+ MACROS       	+=      MODEL_CACHE=2
+ OMPFLAGS	=	-qopenmp-stubs
+endif
+
+# Add in OMP flags and user-added flags.
+CXXFLAGS	+=	$(OMPFLAGS) $(EXTRA_CXXFLAGS)
+
+# Some file names.
+TAG			:=	$(stencil).$(arch)
+STENCIL_BASES		:=	stencil_main stencil_calc realv_grids utils
+STENCIL_OBJS		:=	$(addprefix src/,$(addsuffix .$(TAG).o,$(STENCIL_BASES)))
+STENCIL_CXX		:=	$(addprefix src/,$(addsuffix .$(TAG).i,$(STENCIL_BASES)))
+EXEC_NAME		:=	bin/yask.$(TAG).exe
+MAKE_REPORT_FILE	:=	make-report.$(TAG).txt
+CXXFLAGS_FILE		:=	cxx-flags.$(TAG).txt
+LFLAGS_FILE		:=	ld-flags.$(TAG).txt
+
 # gen-loops.pl args:
 
 # Rank loops break up the whole rank into smaller regions.  In order for
@@ -446,40 +468,24 @@ HALO_LOOP_OUTER_VARS	?=	wv,xv,yv,zv
 HALO_LOOP_CODE		?=	$(HALO_LOOP_OUTER_MODS) loop($(HALO_LOOP_OUTER_VARS)) \
 				$(HALO_LOOP_INNER_MODS) { calc(halo(t)); }
 
-# compile with model_cache=1 or 2 to check prefetching.
-ifeq ($(model_cache),1)
- MACROS       	+=      MODEL_CACHE=1
- OMPFLAGS	=	-qopenmp-stubs
-else ifeq ($(model_cache),2)
- MACROS       	+=      MODEL_CACHE=2
- OMPFLAGS	=	-qopenmp-stubs
-endif
-
-CXXFLAGS	+=	$(OMPFLAGS) $(EXTRA_CXXFLAGS)
+#### Targets and rules ####
 
-STENCIL_BASES		:=	stencil_main stencil_calc realv_grids utils
-STENCIL_OBJS		:=	$(addprefix src/,$(addsuffix .$(arch).o,$(STENCIL_BASES)))
-STENCIL_CXX		:=	$(addprefix src/,$(addsuffix .$(arch).i,$(STENCIL_BASES)))
-STENCIL_EXEC_NAME	:=	stencil.$(arch).exe
-MAKE_REPORT_FILE	:=	make-report.txt
-CXXFLAGS_FILE		:=	cxx-flags.txt
-LFLAGS_FILE		:=	ld-flags.txt
-
-all:	$(STENCIL_EXEC_NAME) $(MAKE_REPORT_FILE)
+all:	$(EXEC_NAME) $(MAKE_REPORT_FILE)
 	echo $(CXXFLAGS) > $(CXXFLAGS_FILE)
 	echo $(LFLAGS) > $(LFLAGS_FILE)
 	@cat $(MAKE_REPORT_FILE)
-	@echo $(STENCIL_EXEC_NAME) "has been built."
+	@echo $(EXEC_NAME) "has been built. Use bin/yask.sh to run it."
 
-$(MAKE_REPORT_FILE): $(STENCIL_EXEC_NAME)
+$(MAKE_REPORT_FILE): $(EXEC_NAME)
 	@echo MAKEFLAGS="\"$(MAKEFLAGS)"\" > $@ 2>&1
 	$(MAKE) -j1 $(CODE_STATS) echo-settings >> $@ 2>&1
 
 echo-settings:
 	@echo
-	@echo "Build environment for" $(STENCIL_EXEC_NAME) on `date`
-	@echo arch=$(arch)
+	@echo "Build environment for" $(EXEC_NAME) on `date`
+	@echo host=`hostname`
 	@echo stencil=$(stencil)
+	@echo arch=$(arch)
 	@echo def_thread_divisor=$(def_thread_divisor)
 	@echo def_block_threads=$(def_block_threads)
 	@echo def_rank_args=$(def_rank_args)
@@ -509,7 +515,10 @@ echo-settings:
 	@echo ISA=$(ISA)
 	@echo OMPFLAGS="\"$(OMPFLAGS)\""
 	@echo EXTRA_CXXFLAGS="\"$(EXTRA_CXXFLAGS)\""
+	@echo CXX=$(CXX)
+	@echo CXXOPT=$(CXXOPT)
 	@echo CXXFLAGS="\"$(CXXFLAGS)\""
+	@$(CXX) -v; $(CXX_VER_CMD)
 	@echo RANK_LOOP_OPTS="\"$(RANK_LOOP_OPTS)\""
 	@echo RANK_LOOP_OUTER_MODS="\"$(RANK_LOOP_OUTER_MODS)\""
 	@echo RANK_LOOP_OUTER_VARS="\"$(RANK_LOOP_OUTER_VARS)\""
@@ -536,44 +545,41 @@ echo-settings:
 	@echo HALO_LOOP_OUTER_VARS="\"$(HALO_LOOP_OUTER_VARS)\""
 	@echo HALO_LOOP_INNER_MODS="\"$(HALO_LOOP_INNER_MODS)\""
 	@echo HALO_LOOP_CODE="\"$(HALO_LOOP_CODE)\""
-	@echo CXX=$(CXX)
-	@echo CXXOPT=$(CXXOPT)
-	@$(CXX) -v; $(CXX_VER_CMD)
 
 code_stats:
 	@echo
 	@echo "Code stats for stencil computation:"
-	./get-loop-stats.pl -t='sub_block_loops' *.s
+	bin/get-loop-stats.pl -t='sub_block_loops' *.s
 
-$(STENCIL_EXEC_NAME): $(STENCIL_OBJS)
+$(EXEC_NAME): $(STENCIL_OBJS)
 	$(LD) -o $@ $(STENCIL_OBJS) $(CXXFLAGS) $(LFLAGS)
 
 preprocess: $(STENCIL_CXX)
 
-src/stencil_rank_loops.hpp: gen-loops.pl Makefile
-	./$< -output $@ $(RANK_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_RANK_LOOP_OPTS) "$(RANK_LOOP_CODE)"
+src/stencil_rank_loops.hpp: bin/gen-loops.pl Makefile
+	$< -output $@ $(RANK_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_RANK_LOOP_OPTS) "$(RANK_LOOP_CODE)"
 
-src/stencil_region_loops.hpp: gen-loops.pl Makefile
-	./$< -output $@ $(REGION_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_REGION_LOOP_OPTS) "$(REGION_LOOP_CODE)"
+src/stencil_region_loops.hpp: bin/gen-loops.pl Makefile
+	$< -output $@ $(REGION_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_REGION_LOOP_OPTS) "$(REGION_LOOP_CODE)"
 
-src/stencil_block_loops.hpp: gen-loops.pl Makefile
-	./$< -output $@ $(BLOCK_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_BLOCK_LOOP_OPTS) "$(BLOCK_LOOP_CODE)"
+src/stencil_block_loops.hpp: bin/gen-loops.pl Makefile
+	$< -output $@ $(BLOCK_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_BLOCK_LOOP_OPTS) "$(BLOCK_LOOP_CODE)"
 
-src/stencil_sub_block_loops.hpp: gen-loops.pl Makefile
-	./$< -output $@ $(SUB_BLOCK_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_SUB_BLOCK_LOOP_OPTS) "$(SUB_BLOCK_LOOP_CODE)"
+src/stencil_sub_block_loops.hpp: bin/gen-loops.pl Makefile
+	$< -output $@ $(SUB_BLOCK_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_SUB_BLOCK_LOOP_OPTS) "$(SUB_BLOCK_LOOP_CODE)"
 
-src/stencil_halo_loops.hpp: gen-loops.pl Makefile
-	./$< -output $@ $(HALO_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_HALO_LOOP_OPTS) "$(HALO_LOOP_CODE)"
+src/stencil_halo_loops.hpp: bin/gen-loops.pl Makefile
+	$< -output $@ $(HALO_LOOP_OPTS) $(EXTRA_LOOP_OPTS) $(EXTRA_HALO_LOOP_OPTS) "$(HALO_LOOP_CODE)"
 
-src/layout_macros.hpp: gen-layouts.pl
-	./$< -m > $@
+src/layout_macros.hpp: bin/gen-layouts.pl
+	$< -m > $@
 
-src/layouts.hpp: gen-layouts.pl
-	./$< -d > $@
+src/layouts.hpp: bin/gen-layouts.pl
+	$< -d > $@
 
 # Compile the stencil compiler.
 # TODO: move this to its own makefile.
-foldBuilder: src/foldBuilder/*.*pp src/foldBuilder/stencils/*.*pp $(FB_STENCIL_LIST)
+$(FB_EXEC): src/foldBuilder/*.*pp src/foldBuilder/stencils/*.*pp $(FB_STENCIL_LIST)
 	$(FB_CXX) $(FB_CXXFLAGS) -Isrc/foldBuilder/stencils -o $@ src/foldBuilder/*.cpp $(EXTRA_FB_CXXFLAGS)
 
 $(FB_STENCIL_LIST): src/foldBuilder/stencils/*.hpp
@@ -584,8 +590,8 @@ $(FB_STENCIL_LIST): src/foldBuilder/stencils/*.hpp
 
 # Run the stencil compiler and post-process its output files.
 # Use the gmake pattern-rule trick to specify simultaneous targets.
-%/$(ST_MACRO_FILE) %/$(ST_CODE_FILE): foldBuilder
-	./$< $(FB_FLAGS) $(EXTRA_FB_FLAGS) \
+%/$(ST_MACRO_FILE) %/$(ST_CODE_FILE): $(FB_EXEC)
+	$< $(FB_FLAGS) $(EXTRA_FB_FLAGS) \
 	  -pm $*/$(ST_MACRO_FILE) -p$(FB_TARGET) $*/$(ST_CODE_FILE)
 	echo >> $*/$(ST_MACRO_FILE)
 	echo '// Settings from YASK Makefile' >> $*/$(ST_MACRO_FILE)
@@ -599,20 +605,21 @@ $(FB_STENCIL_LIST): src/foldBuilder/stencils/*.hpp
 headers: $(GEN_HEADERS) $(FB_STENCIL_LIST)
 	@ echo 'Header files generated.'
 
-%.$(arch).o: %.cpp src/*.hpp src/foldBuilder/*.hpp $(GEN_HEADERS)
+%.$(TAG).o: %.cpp src/*.hpp src/foldBuilder/*.hpp $(GEN_HEADERS)
 	$(CXX) $(CXXFLAGS) -c -o $@ $<
 
-%.$(arch).i: %.cpp src/*.hpp src/foldBuilder/*.hpp $(GEN_HEADERS)
+%.$(TAG).i: %.cpp src/*.hpp src/foldBuilder/*.hpp $(GEN_HEADERS)
 	$(CXX) $(CXXFLAGS) -E $< > $@
 
 tags:
 	rm -f TAGS ; find . -name '*.[ch]pp' | xargs etags -C -a
 
 clean:
-	rm -fv src/*.[io] *.optrpt src/*.optrpt *.s $(GEN_HEADERS) $(MAKE_REPORT_FILE)
+	rm -fv src/*.[io] *.optrpt */*.optrpt *.s $(GEN_HEADERS) $(MAKE_REPORT_FILE)
 
 realclean: clean
-	rm -fv stencil*.exe foldBuilder TAGS $(MAKE_REPORT_FILE) $(CXXFLAGS_FILE) $(LFLAGS_FILE) $(FB_STENCIL_LIST)
+	rm -fv bin/yask*.exe make-report*.txt cxx-flags*.txt ld-flags.*txt $(FB_EXEC) $(FB_STENCIL_LIST) TAGS
+	rm -fv stencil*.exe stencil-tuner-summary.csh stencil-tuner.pl gen-layouts.pl gen-loops.pl get-loop-stats.pl
 	find . -name '*~' | xargs -r rm -v
 
 help:

stencil-tuner-summary.csh → bin/yask-tuner-summary.csh

@@ -26,15 +26,15 @@
 # Purpose: find best result from each GA search csv file.
 
 if ( "-$1" == "-" ) then
-    if ( `echo stencil-tuner*.csv | wc -l` > 0 ) then
-        $0 stencil-tuner*.csv
+    if ( `echo yask-tuner*.csv | wc -l` > 0 ) then
+        $0 yask-tuner*.csv
     else
-        echo "usage: $0 <csv-file(s) from stencil-tuner>"
+        echo "usage: $0 <csv-file(s) from yask-tuner>"
     endif
     exit
 endif
 
-echo "Summary of stencil-tuner results:"
+echo "Summary of yask-tuner results:"
 head -n1 $1
 foreach f ($*)
     echo '=========='

stencil-tuner.pl → bin/yask-tuner.pl

@@ -97,7 +97,7 @@ sub usage {
       " -sde               Run binary on SDE (for testing only).\n".
       " -makePrefix=<CMD>  Prefix make command with <CMD>.\n".
       " -makeArgs=<ARGS>   Pass additional <ARGS> to make command.\n".
-      " -runArgs=<ARGS>    Pass additional <ARGS> to stencil-run command.\n".
+      " -runArgs=<ARGS>    Pass additional <ARGS> to bin/yask.sh command.\n".
       " -ranks=<N>         Number of ranks to use on host (x-dimension only).\n".
       "\nstencil options:\n".
       " -stencil=<NAME>    Specify stencil: iso3dfd, 3axis, 9axis, 3plane, cube, ave, awp, ... (required).\n".
@@ -120,7 +120,7 @@ sub usage {
       " -dw=<N>            Set size of 'w' dim to <N> (only for 4D problems).\n".
       " -mem=<N>-<M>           Set allowable est. memory usage between <N> and <M> GiB (default is $minGB-$maxGB).\n".
       " -maxVecsInCluster=<N>  Maximum vectors allowed in cluster (default is $maxVecsInCluster).\n".
-      " -noPrefetch        Disable any prefetching (shortcut for '-pfdl1=0 -pfdl2=0').\n".
+      " -noPrefetch        Disable any prefetching (shortcut for '-pfd_l1=0 -pfd_l2=0').\n".
       " -noFolding         Allow only 1D vectorization (in any direction).\n".
       " -zLoop             Force inner loop in 'z' direction.\n".
       " -zLayout           Force inner memory layout in 'z' direction.\n".
@@ -209,8 +209,8 @@ sub usage {
     $stencil = $1;
   }
   elsif ($opt eq '-noprefetch') {
-    $geneRanges{$autoKey.'pfdl1'} = [ 0 ];
-    $geneRanges{$autoKey.'pfdl2'} = [ 0 ];
+    $geneRanges{$autoKey.'pfd_l1'} = [ 0 ];
+    $geneRanges{$autoKey.'pfd_l2'} = [ 0 ];
   }
   elsif ($opt =~ '^-maxvecsincluster=(\d+)$') {
     $maxVecsInCluster = $1;
@@ -304,7 +304,7 @@ sub usage {
 my $hostStr = defined $host ? $host : hostname();
 my $timeStamp=`date +%Y-%m-%d_%H-%M-%S`;
 chomp $timeStamp;
-my $outFile = "stencil-tuner$searchTypeStr.$stencil.$arch.$hostStr.$timeStamp.csv";
+my $outFile = "yask-tuner$searchTypeStr.$stencil.$arch.$hostStr.$timeStamp.csv";
 print "Output will be saved in '$outFile'.\n";
 $outFile = '/dev/null' if $checking;
 
@@ -504,8 +504,8 @@ sub usage {
 
      # prefetch distances for l1 and l2.
      # all non-pos numbers => no prefetching, so ~50% chance of being enabled.
-     [ -$maxPfdl1, $maxPfdl1, 1, 'pfdl1' ],
-     [ -$maxPfdl2, $maxPfdl2, 1, 'pfdl2' ],
+     [ -$maxPfdl1, $maxPfdl1, 1, 'pfd_l1' ],
+     [ -$maxPfdl2, $maxPfdl2, 1, 'pfd_l2' ],
 
      # other build options.
      [ 0, 100, 1, 'exprSize' ],          # expression-size threshold.
@@ -734,7 +734,7 @@ ($$)
 
   my $makeCmd = "$makePrefix make clean; ".
     "$makePrefix make -j all EXTRA_MACROS='$macros' ".
-    "arch=$arch real_bytes=$realBytes stencil=$stencil radius=$radius $margs $makeArgs";
+    "stencil=$stencil arch=$arch real_bytes=$realBytes radius=$radius $margs $makeArgs";
   $makeCmd = "echo 'build disabled'" if !$doBuild;
   return $makeCmd;
 }
@@ -744,14 +744,14 @@ ()
   my $exePrefix = 'time';
   $exePrefix .= " sde -$arch --" if $sde;
 
-  my $runCmd = "./stencil-run.sh";
+  my $runCmd = "bin/yask.sh";
   if (defined $mic) {
     $runCmd .= " -mic $mic";
   } else {
     $exePrefix .= " numactl -p 1" if $arch eq 'knl' && !$sde; # TODO: fix for cache mode.
     $runCmd .= " -host $host" if defined $host;
   }
-  $runCmd .= " -exe_prefix '$exePrefix' -arch $arch $runArgs";
+  $runCmd .= " -exe_prefix '$exePrefix' -stencil $stencil -arch $arch $runArgs";
   return $runCmd;
 }
 
@@ -1208,8 +1208,8 @@ sub fitness {
   my $thread_divisor_exp = readHash($h, 'thread_divisor_exp', 0);
   my $bthreads_exp = readHash($h, 'bthreads_exp', 0);
   my $layout = readHash($h, 'layout', 1);
-  my $pfdl1 = readHash($h, 'pfdl1', 1);
-  my $pfdl2 = readHash($h, 'pfdl2', 1);
+  my $pfdl1 = readHash($h, 'pfd_l1', 1);
+  my $pfdl2 = readHash($h, 'pfd_l2', 1);
   my $ompRegionSchedule = readHash($h, 'ompRegionSchedule', 1);
   my $ompBlockSchedule = readHash($h, 'ompBlockSchedule', 1);
 
@@ -1357,8 +1357,8 @@ sub fitness {
     # make sure pfld2 > pfld1.
     $pfdl2 = $pfdl1 + 1 if $pfdl1 >= $pfdl2;
   }
-  $macros .= " PFDL1=$pfdl1" if $pfdl1 > 0;
-  $macros .= " PFDL2=$pfdl2" if $pfdl2 > 0;
+  $mvars .= " pfd_l1=$pfdl1" if $pfdl1 > 0;
+  $mvars .= " pfd_l2=$pfdl2" if $pfdl2 > 0;
 
   # cluster & fold.
   $mvars .= " cluster=x=$cvs[0],y=$cvs[1],z=$cvs[2]";