Volk_runtime now does self-initialization. You can call volk_xxx_a16() just like in volk.c.

The first run of each function does the rank_archs call. Subsequent calls proceed with no overhead.
volk_init is still being generated but not used at all.

3 files changed, 9 insertions, 14 deletions

diff --git a/volk/include/volk/make_runtime.py b/volk/include/volk/make_runtime.py
index 645b3aaee..91e703b24 100644
--- a/volk/include/volk/make_runtime.py
+++ b/volk/include/volk/make_runtime.py
@@ -4,7 +4,7 @@ from volk_regexp import *
 
 
 
-def make_runtime(funclist) :
+def make_runtime(funclist, arglist) :
     tempstring = "";
     tempstring = tempstring + '/*this file is auto generated by volk_register.py*/\n';
 
@@ -18,14 +18,8 @@ def make_runtime(funclist) :
 
     tempstring = tempstring + '\n';
 
-    tempstring = tempstring + "struct VOLK_RUNTIME {\n";
-
     for i in range(len(funclist)):
-        tempstring = tempstring + replace_volk.sub("p", funclist[i]) + " " + funclist[i] + ";\n";
-    tempstring = tempstring + "};\n\n";
-
-    tempstring = tempstring + "struct VOLK_RUNTIME* get_volk_runtime();\n\n"
-    tempstring = tempstring + "\nvoid volk_runtime_init();\n";
+        tempstring = tempstring + "extern void (*" + funclist[i] + ")(" + arglist[i] + ");\n"
 
     tempstring = tempstring + emit_epilog();
     tempstring = tempstring + "#endif /*INCLUDED_VOLK_RUNTIME*/\n";
diff --git a/volk/include/volk/make_runtime_c.py b/volk/include/volk/make_runtime_c.py
index 070df9ba7..0519dddf8 100644
--- a/volk/include/volk/make_runtime_c.py
+++ b/volk/include/volk/make_runtime_c.py
@@ -3,7 +3,7 @@ import string
 from volk_regexp import *
 
 
-def make_runtime_c(funclist, taglist, arched_arglist, retlist, my_arglist, fcountlist) :
+def make_runtime_c(funclist, taglist, arched_arglist, retlist, my_arglist, fcountlist, my_argtypelist) :
     tempstring = "";
     tempstring = tempstring + '/*this file is auto generated by volk_register.py*/';
 
@@ -18,8 +18,6 @@ def make_runtime_c(funclist, taglist, arched_arglist, retlist, my_arglist, fcoun
     for func in funclist:
         tempstring = tempstring + "#include<volk/" + func + ".h>\n" ;
     tempstring = tempstring + '\n';
-
-    tempstring = tempstring + "struct VOLK_RUNTIME volk_runtime;\n";
     
     for i in range(len(funclist)): 
         tempstring = tempstring + "static const " + replace_volk.sub("p", funclist[i]) + " " + funclist[i] + "_archs[] = {\n";
@@ -42,6 +40,9 @@ def make_runtime_c(funclist, taglist, arched_arglist, retlist, my_arglist, fcoun
 
 
         tempstring = tempstring + retlist[i] + "default_acquire_" + funclist[i] + replace_arch.sub("", arched_arglist[i]) + '\n';
-        tempstring = tempstring + "volk_runtime." + funclist[i] + " = " + funclist[i] + "_archs[volk_rank_archs(" + funclist[i] + "_arch_defs, volk_get_lvarch())];\n" + "return " + funclist[i] + "_archs[volk_rank_archs(" + funclist[i] + "_arch_defs, volk_get_lvarch())](" + my_arglist[i] + ");" + '\n}\n';
+        tempstring = tempstring + "    %s = %s_archs[volk_rank_archs(%s_arch_defs, volk_get_lvarch())];\n" % (funclist[i], funclist[i], funclist[i])
+        tempstring = tempstring + "    %s(%s);\n}\n\n" % (funclist[i], my_arglist[i])
+        
+        tempstring = tempstring + "%s(*%s)(%s) = &default_acquire_%s;\n\n" % (retlist[i], funclist[i], my_argtypelist[i], funclist[i])
     
     return tempstring;
diff --git a/volk/include/volk/volk_register.py b/volk/include/volk/volk_register.py
index 460a11fab..b719042df 100755
--- a/volk/include/volk/volk_register.py
+++ b/volk/include/volk/volk_register.py
@@ -241,13 +241,13 @@ outfile_config_fixed.close();
 outfile_c.write( make_c(functions, taglist, arched_arglist, retlist, my_arglist, fcountlist));
 outfile_c.close();
 
-outfile_runtime_c.write(make_runtime_c(functions, taglist, arched_arglist, retlist, my_arglist, fcountlist));
+outfile_runtime_c.write(make_runtime_c(functions, taglist, arched_arglist, retlist, my_arglist, fcountlist, my_argtypelist));
 outfile_runtime_c.close();
 
 outfile_init_c.write(make_init_c(functions, filearchs));
 outfile_init_c.close();
 
-outfile_runtime.write(make_runtime(functions));
+outfile_runtime.write(make_runtime(functions, my_argtypelist));
 outfile_runtime.close();
 
 outfile_typedefs.write(make_typedefs(functions, retlist, my_argtypelist));