Update starch to upstream commit f76162205fa082d4c9a75b89eb4295e03f5f68fd

starch/example/generated/benchmark.c

@@ -425,15 +425,15 @@ static void starch_benchmark_usage(const char *argv0)
         "                     (default: benchmark all functions)\n"
         "\n"
         "Supported flavors:   "
-#ifdef STARCH_FLAVOR_GENERIC
-          "generic "
-#endif
 #ifdef STARCH_FLAVOR_ARMV7A_VFPV3
           "armv7a_vfpv3 "
 #endif
 #ifdef STARCH_FLAVOR_ARMV7A_VFPV4
           "armv7a_vfpv4 "
 #endif
+#ifdef STARCH_FLAVOR_GENERIC
+          "generic "
+#endif
 #ifdef STARCH_FLAVOR_X86_64_AVX
           "x86_64_avx "
 #endif

starch/example/generated/dispatcher.c

@@ -73,12 +73,6 @@ void starch_subtract_n_set_wisdom (const char * const * received_wisdom)
 
 starch_subtract_n_regentry starch_subtract_n_registry[] = {
   
-#ifdef STARCH_MIX_GENERIC
-    { 0, "generic_generic", "generic", starch_subtract_n_generic_generic, NULL },
-    { 1, "unroll_4_generic", "generic", starch_subtract_n_unroll_4_generic, NULL },
-    { 2, "bad_implementation_generic", "generic", starch_subtract_n_bad_implementation_generic, NULL },
-#endif /* STARCH_MIX_GENERIC */
-  
 #ifdef STARCH_MIX_ARM
     { 0, "neon_intrinsics_armv7a_vfpv4", "armv7a_vfpv4", starch_subtract_n_neon_intrinsics_armv7a_vfpv4, supports_neon_vfpv4 },
     { 1, "neon_intrinsics_armv7a_vfpv3", "armv7a_vfpv3", starch_subtract_n_neon_intrinsics_armv7a_vfpv3, supports_neon_vfpv3 },
@@ -93,6 +87,12 @@ starch_subtract_n_regentry starch_subtract_n_registry[] = {
     { 10, "bad_implementation_generic", "generic", starch_subtract_n_bad_implementation_generic, NULL },
 #endif /* STARCH_MIX_ARM */
   
+#ifdef STARCH_MIX_GENERIC
+    { 0, "generic_generic", "generic", starch_subtract_n_generic_generic, NULL },
+    { 1, "unroll_4_generic", "generic", starch_subtract_n_unroll_4_generic, NULL },
+    { 2, "bad_implementation_generic", "generic", starch_subtract_n_bad_implementation_generic, NULL },
+#endif /* STARCH_MIX_GENERIC */
+  
 #ifdef STARCH_MIX_X86_64
     { 0, "generic_x86_64_avx2", "x86_64_avx2", starch_subtract_n_generic_x86_64_avx2, supports_x86_avx2 },
     { 1, "generic_x86_64_avx", "x86_64_avx", starch_subtract_n_generic_x86_64_avx, supports_x86_avx },
@@ -161,12 +161,6 @@ void starch_subtract_n_aligned_set_wisdom (const char * const * received_wisdom)
 
 starch_subtract_n_aligned_regentry starch_subtract_n_aligned_registry[] = {
   
-#ifdef STARCH_MIX_GENERIC
-    { 0, "generic_generic", "generic", starch_subtract_n_generic_generic, NULL },
-    { 1, "unroll_4_generic", "generic", starch_subtract_n_unroll_4_generic, NULL },
-    { 2, "bad_implementation_generic", "generic", starch_subtract_n_bad_implementation_generic, NULL },
-#endif /* STARCH_MIX_GENERIC */
-  
 #ifdef STARCH_MIX_ARM
     { 0, "generic_armv7a_vfpv4_aligned", "armv7a_vfpv4", starch_subtract_n_aligned_generic_armv7a_vfpv4, supports_neon_vfpv4 },
     { 1, "unroll_4_armv7a_vfpv4_aligned", "armv7a_vfpv4", starch_subtract_n_aligned_unroll_4_armv7a_vfpv4, supports_neon_vfpv4 },
@@ -189,6 +183,12 @@ starch_subtract_n_aligned_regentry starch_subtract_n_aligned_registry[] = {
     { 18, "bad_implementation_generic", "generic", starch_subtract_n_bad_implementation_generic, NULL },
 #endif /* STARCH_MIX_ARM */
   
+#ifdef STARCH_MIX_GENERIC
+    { 0, "generic_generic", "generic", starch_subtract_n_generic_generic, NULL },
+    { 1, "unroll_4_generic", "generic", starch_subtract_n_unroll_4_generic, NULL },
+    { 2, "bad_implementation_generic", "generic", starch_subtract_n_bad_implementation_generic, NULL },
+#endif /* STARCH_MIX_GENERIC */
+  
 #ifdef STARCH_MIX_X86_64
     { 0, "generic_x86_64_avx2_aligned", "x86_64_avx2", starch_subtract_n_aligned_generic_x86_64_avx2, supports_x86_avx2 },
     { 1, "unroll_4_x86_64_avx2_aligned", "x86_64_avx2", starch_subtract_n_aligned_unroll_4_x86_64_avx2, supports_x86_avx2 },

starch/example/generated/starch.h

@@ -5,12 +5,6 @@
 
 /* mixes */
 
-/* Generic build, compiler defaults only */
-#ifdef STARCH_MIX_GENERIC
-#define STARCH_FLAVOR_GENERIC
-#define STARCH_MIX_ALIGNMENT 1
-#endif /* STARCH_MIX_GENERIC */
-
 /* ARM */
 #ifdef STARCH_MIX_ARM
 #define STARCH_FLAVOR_ARMV7A_VFPV4
@@ -19,6 +13,12 @@
 #define STARCH_MIX_ALIGNMENT 16
 #endif /* STARCH_MIX_ARM */
 
+/* Generic build, compiler defaults only */
+#ifdef STARCH_MIX_GENERIC
+#define STARCH_FLAVOR_GENERIC
+#define STARCH_MIX_ALIGNMENT 1
+#endif /* STARCH_MIX_GENERIC */
+
 /* x64-64 */
 #ifdef STARCH_MIX_X86_64
 #define STARCH_FLAVOR_X86_64_AVX2
@@ -71,14 +71,6 @@ void starch_subtract_n_aligned_set_wisdom( const char * const * received_wisdom
 
 /* flavors and prototypes */
 
-#ifdef STARCH_FLAVOR_GENERIC
-void starch_subtract_n_generic_generic ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
-void starch_subtract_n_unroll_4_generic ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
-void starch_subtract_n_bad_implementation_generic ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
-#endif /* STARCH_FLAVOR_GENERIC */
-
-int starch_read_wisdom (const char * path);
-
 #ifdef STARCH_FLAVOR_ARMV7A_VFPV3
 int supports_neon_vfpv3 (void);
 void starch_subtract_n_generic_armv7a_vfpv3 ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
@@ -107,6 +99,14 @@ void starch_subtract_n_aligned_neon_intrinsics_armv7a_vfpv4 ( const uint16_t * a
 
 int starch_read_wisdom (const char * path);
 
+#ifdef STARCH_FLAVOR_GENERIC
+void starch_subtract_n_generic_generic ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
+void starch_subtract_n_unroll_4_generic ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
+void starch_subtract_n_bad_implementation_generic ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );
+#endif /* STARCH_FLAVOR_GENERIC */
+
+int starch_read_wisdom (const char * path);
+
 #ifdef STARCH_FLAVOR_X86_64_AVX
 int supports_x86_avx (void);
 void starch_subtract_n_generic_x86_64_avx ( const uint16_t * arg0, unsigned arg1, uint16_t arg2, uint16_t * arg3 );

starch/starch.py

@@ -8,8 +8,9 @@ import sys
 import re
 import os
 import mako.lookup
+from functools import total_ordering
 
-from typing import Optional, Union, Iterable, Sequence, MutableSequence, Mapping, MutableMapping, FrozenSet
+from typing import Optional, Union, Iterable, Sequence, MutableSequence, Mapping, MutableMapping, FrozenSet, Dict, List
 
 class Feature(object):
     """Feature represents a type of code that can only be built with
@@ -35,7 +36,7 @@ themselves using the STARCH_IMPL_REQUIRES macro."""
     def macro(self) -> str:
         return 'STARCH_FEATURE_' + self.name.upper()
 
-
+@total_ordering
 class BuildFlavor(object):
     """BuildFlavor models code built with specific compiler flags.
 Shared implementation code will be built multiple times, once per flavor.
@@ -89,7 +90,13 @@ intrinsics)"""
     def cflags(self) -> str:
         return ' '.join(self.compile_flags)
 
+    def __lt__(self, other: object) -> bool:
+        if not isinstance(other, BuildFlavor):
+            return NotImplemented
+        return self.name < other.name
 
+
+@total_ordering
 class Function(object):
     """A user-callable function that will be dispatched to
 one of the many possible implementations based on runtime feature
@@ -175,6 +182,11 @@ support."""
     def benchmark_verify_symbol(self) -> str:
         return self.gen.sym(self.name + '_benchmark_verify')
 
+    def __lt__(self, other: object) -> bool:
+        if not isinstance(other, Function):
+            return NotImplemented
+        return self.name < other.name
+
 
 class FunctionImpl(object):
     """A possible implementation of a function, not built in any particular way yet."""
@@ -210,6 +222,7 @@ class FunctionImpl(object):
         return self.gen.sym(self.function.name + '_' + self.name + '_' + flavor.name)
 
 
+@total_ordering
 class SourceFile(object):
     """A scanned source file that contains implementation code."""
 
@@ -220,7 +233,13 @@ class SourceFile(object):
         self.path = path
         self.impls = []
 
+    def __lt__(self, other: object) -> bool:
+        if not isinstance(other, SourceFile):
+            return NotImplemented
+        return self.path < other.path
 
+
+@total_ordering
 class BuildMix(object):
     """A combination of build flavors that make up one possible way of building all
 the code. The output of a mix is a library that dispatches functions within the
@@ -254,6 +273,12 @@ specified first."""
     def function_wisdom(self, function) -> Sequence[str]:
         return self.wisdom.get(function, [])
 
+    def __lt__(self, other: object) -> bool:
+        if not isinstance(other, BuildMix):
+            return NotImplemented
+        return self.name < other.name
+
+
 class Generator(object):
     functions: MutableMapping[str, Function]
     features: MutableMapping[str, Feature]
@@ -380,8 +405,8 @@ class Generator(object):
             return key
         return self.flavors[key]                        
 
-    def load_wisdom(self, path: str) -> Mapping[str,Sequence[str]]:
+    def load_wisdom(self, path: str) -> Mapping[Function,Sequence[str]]:
-        results: Mapping[Function,Sequence[str]] = {}
+        results: Dict[Function,List[str]] = {}
 
         try:
             f = open(path, 'r')
@@ -420,7 +445,7 @@ class Generator(object):
         if wisdom_file:
             resolved_wisdom = self.load_wisdom(wisdom_file)
         else:
-            resolved_wisdom = dict( (self.get_function(name), values) for name,values in wisdom.items() )
+            resolved_wisdom = dict( (self.get_function(name), list(values)) for name,values in wisdom.items() )
         self.mixes[name] = BuildMix(name, description, resolved_flavors, resolved_wisdom)
 
     def sym(self, symbol: str) -> str:

starch/templates/benchmark.c.template

@@ -115,7 +115,7 @@ static bool starch_benchmark_flavor_in_list(const char *flavor, const starch_ben
 }
 
 <% functions_to_benchmark = [f for f in gen.functions.values() if f.benchmark] %>
-% for function in functions_to_benchmark:
+% for function in sorted(functions_to_benchmark):
 /* prototypes for benchmark helpers provided by user code */
 void ${function.benchmark_symbol} (void);
 % if function.benchmark_verify:
@@ -252,7 +252,7 @@ static void starch_benchmark_run_${function.name}( ${function.declaration_arglis
 #define STARCH_BENCHMARK_ALLOC(_count, _type) ((_type *) starch_benchmark_aligned_alloc(1, alignof(_type), (_count) * sizeof(_type)))
 #define STARCH_BENCHMARK_FREE(_ptr) starch_benchmark_aligned_free(_ptr)
 
-% for source in gen.benchmark_files:
+% for source in sorted(gen.benchmark_files):
 #include "${os.path.relpath(source.path, current_dir)}"
 % endfor
 
@@ -278,13 +278,13 @@ static void starch_benchmark_run_${function.name}( ${function.declaration_arglis
 #define STARCH_BENCHMARK_ALLOC(_count, _type) ((_type *) starch_benchmark_aligned_alloc(STARCH_MIX_ALIGNMENT, alignof(_type), (_count) * sizeof(_type)))
 #define STARCH_BENCHMARK_FREE(_ptr) starch_benchmark_aligned_free(_ptr)
 
-% for source in gen.benchmark_files:
+% for source in sorted(gen.benchmark_files):
   % if any( (function.aligned and function.benchmark == source) for function in gen.functions.values() ):
 #include "${os.path.relpath(source.path, current_dir)}"
   % endif
 % endfor
 
-% for function in functions_to_benchmark:
+% for function in sorted(functions_to_benchmark):
 static void starch_benchmark_all_${function.name}(void)
 {
     fprintf(stderr, "==== ${function.name} ===\n");
@@ -330,14 +330,14 @@ static void starch_benchmark_usage(const char *argv0)
         "                     (default: benchmark all functions)\n"
         "\n"
         "Supported flavors:   "
-% for flavor in gen.flavors.values():
+% for flavor in sorted(gen.flavors.values()):
 #ifdef ${flavor.macro}
           "${flavor.name} "
 #endif
 % endfor
           "\n"
         "Supported functions: "
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
           "${function.name} "
 % endfor
           "\n", argv0);
@@ -425,7 +425,7 @@ int main(int argc, char **argv)
     }
 
     for (int i = optind; i < argc; ++i) {
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
         if (!strcmp(argv[i], "${function.name}")) {
             specific = 1;
 % if function.benchmark:
@@ -443,7 +443,7 @@ int main(int argc, char **argv)
     }
 
     if (!specific) {
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
   % if function.benchmark:
         starch_benchmark_all_${function.name}();
   % else:

starch/templates/dispatcher.c.template

@@ -22,7 +22,7 @@ static int starch_regentry_rank_compare (const void *l, const void *r)
     return left->rank - right->rank;
 }
 
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
 /* dispatcher / registry for ${function.name} */
 
 ${function.regentry_type} * ${function.select_symbol}() {
@@ -80,7 +80,7 @@ void ${function.set_wisdom_symbol} (const char * const * received_wisdom)
 }
 
 ${function.regentry_type} ${function.registry_symbol}[] = {
-% for mix in gen.mixes.values():
+% for mix in sorted(gen.mixes.values()):
   <%
     # gather all implementations for this mix, sort by wisdom
     def rank_key(value, wisdom=mix.function_wisdom(function)):
@@ -128,7 +128,7 @@ int ${gen.sym("read_wisdom")} (const char * path)
         return -1;
 
     /* reset all ranks to identify entries not listed in the wisdom file; we'll assign ranks at the end to produce a stable sort */
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
     int rank_${function.name} = 0;
     for (${function.regentry_type} *entry = ${function.registry_symbol}; entry->name; ++entry) {
         entry->rank = 0;
@@ -167,7 +167,7 @@ int ${gen.sym("read_wisdom")} (const char * path)
         *end = 0;
 
         /* try to find a matching registry entry */
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
         if (!strcmp(name, "${function.name}")) {
             for (${function.regentry_type} *entry = ${function.registry_symbol}; entry->name; ++entry) {
                 if (!strcmp(impl, entry->name)) {
@@ -188,7 +188,7 @@ int ${gen.sym("read_wisdom")} (const char * path)
     fclose(fp);
 
     /* assign ranks to unmatched items to (stable) sort them last; re-sort everything */
-% for function in gen.functions.values():
+% for function in sorted(gen.functions.values()):
     {
         ${function.regentry_type} *entry;
         for (entry = ${function.registry_symbol}; entry->name; ++entry) {

starch/templates/flavor.c.template

@@ -19,7 +19,7 @@
 #define STARCH_IMPL(_function,_impl) ${gen.symbol_prefix} ## _function ## _ ## _impl ## _ ## ${flavor.name}
 #define STARCH_IMPL_REQUIRES(_function,_impl,_feature) STARCH_IMPL(_function,_impl)
 
-% for source in gen.impl_files:
+% for source in sorted(gen.impl_files):
   % if any( ((impl.feature is None or impl.feature in flavor.features) and not impl.function.aligned) for impl in source.impls):
 #include "${os.path.relpath(source.path, current_dir)}"
   % endif
@@ -39,7 +39,7 @@
 #define STARCH_IMPL(_function,_impl) ${gen.symbol_prefix} ## _function ## _aligned_ ## _impl ## _ ## ${flavor.name}
 #define STARCH_IMPL_REQUIRES(_function,_impl,_feature) STARCH_IMPL(_function,_impl)
 
-% for source in gen.impl_files:
+% for source in sorted(gen.impl_files):
   % if any( ((impl.feature is None or impl.feature in flavor.features) and impl.function.aligned) for impl in source.impls):
 #include "${os.path.relpath(source.path, current_dir)}"
   % endif

starch/templates/starch.h.template

@@ -10,7 +10,7 @@
 
 /* mixes */
 
-% for mix in gen.mixes.values():
+% for mix in sorted(gen.mixes.values()):
 /* ${mix.description} */
 #ifdef ${mix.macro}
   % for flavor in mix.flavors:
@@ -51,7 +51,7 @@ void ${function.set_wisdom_symbol}( const char * const * received_wisdom );
 % endfor
 /* flavors and prototypes */
 
-% for flavor in gen.flavors.values():
+% for flavor in sorted(gen.flavors.values()):
 #ifdef ${flavor.macro}
   % if flavor.test_function is not None:
 int ${flavor.test_function} (void);