profiling and ~64x perf improvements

.gitignore

@@ -3,3 +3,4 @@ obj
 build/
 test/
 shuffle3-*
+profiling/

Makefile

@@ -16,6 +16,8 @@ RELEASE_LDFLAGS?= -O3 -flto
 DEBUG_CFLAGS?= -g -O0
 DEBUG_LDFLAGS?= -O0
 
+STRIP=strip
+
 OBJ = $(addprefix obj/,$(SRC:.c=.o))
 
 .PHONY: release
@@ -35,7 +37,7 @@ $(BUILD)/$(PROJECT)-release: CFLAGS+= $(RELEASE_CFLAGS)
 $(BUILD)/$(PROJECT)-release: LDFLAGS+= $(RELEASE_LDFLAGS)
 $(BUILD)/$(PROJECT)-release: $(OBJ)
 	$(CC) $^ $(CFLAGS) -o $@ $(LDFLAGS)
-	strip $@
+	$(STRIP) $@
 
 $(BUILD)/$(PROJECT)-debug: CFLAGS+= $(DEBUG_CFLAGS)
 $(BUILD)/$(PROJECT)-debug: LDFLAGS+= $(DEBUG_LDFLAGS)

lean/Makefile

@@ -18,6 +18,8 @@ CFLAGS   += $(COMMON_FLAGS) --std=gnu11
 CXXFLAGS += $(COMMON_FLAGS) --std=gnu++20 -fno-exceptions
 LDFLAGS  += -lfmt
 
+STRIP=strip
+
 RELEASE_CFLAGS?=   -O3 -flto $(OPT_FLAGS)
 RELEASE_CXXFLAGS?= -O3 -flto $(CXX_OPT_FLAGS)
 RELEASE_LDFLAGS?=  -O3 -flto
@@ -58,7 +60,7 @@ $(PROJECT)-release: CXXFLAGS += $(RELEASE_CXXFLAGS)
 $(PROJECT)-release: LDFLAGS += $(RELEASE_LDFLAGS)
 $(PROJECT)-release: $(OBJ)
 	$(CXX) $^ $(CXXFLAGS) -o $@ $(LDFLAGS)
-	strip $@
+	$(STRIP) $@
 
 $(PROJECT)-debug: CFLAGS+= $(DEBUG_CFLAGS)
 $(PROJECT)-debug: CXXFLAGS += $(DEBUG_CXXFLAGS)

lean/include/shuffle.hpp

@@ -23,7 +23,7 @@ namespace rng {
 	inline void unshuffle(R& rng, span<T> span)
 	{
 		if(!span.size()) return;
-		std::vector<std::size_t> rng_values;
+		std::vector<std::size_t> rng_values(span.size());
 
 		fmt::print(" -> unshuffling {} objects...", span.size());
 		for(std::size_t i=span.size()-1;i>0;i--)

lean/src/main.c

@@ -13,37 +13,9 @@
 
 _Static_assert(sizeof(float)==sizeof(uint32_t), "float is not 32 bits");
 
-struct prog_args {
-	int argc;
-	char ** argv;
-};
-
-/*
-static void shuffle_file(const char* filename)
-{
-	panic("unimplemented");
-}
-
-static void unshuffle_file(const char* filename)
-{
-	panic("unimplemented");
-}*/
-
-//XXX: Not using this anymore 
-static void* map_callback(mmap_t map, void* user)
-{
-	struct prog_args args = *(struct prog_args*)user;
-
-	dprintf("fd %d mapped (sz %lu) (ptr %p)\n", map.fd, map.len, map.ptr);	
-
-
-	return NULL;
-}
-
-
 int main(int argc, char** argv)
 {
-	struct prog_args args = {.argc = argc, .argv = argv};
+	//struct prog_args args = {.argc = argc, .argv = argv};
 	
 	//rng_test();
 	
@@ -62,26 +34,3 @@ int main(int argc, char** argv)
 	return 0;
 }
 
-#ifdef _TEST
-static void do_reinterpret_test()
-{
-	char* string = "Hello world.. how are you?????";
-	size_t string_sz = strlen(string);
-	size_t outsz;
-
-	uint64_t* string_as_longs = bytes_to_long((uint8_t*)string, string_sz, &outsz);
-
-	printf("%s (%lu) -> u64 %lu\n", string, string_sz, outsz);
-	for (register int i=0;i<outsz;i++) {
-		printf(" %lu\n", string_as_longs[i]);
-	}
-	printf("OK!\n\n");
-
-	float* string_as_floats = bytes_to_float((uint8_t*)string, string_sz, &outsz);
-	printf("%s (%lu) -> f32 %lu\n", string, string_sz, outsz);
-	for (register int i=0;i<outsz;i++) {
-		printf(" %f\n", string_as_floats[i]);
-	}
-	printf("OK!\n\n");
-}
-#endif

lean/src/work.cpp

@@ -17,12 +17,16 @@ std::tuple<T, T> minmax_t(const span<T>& array, Fn keep)
 {
 	T highest;
 	T lowest;
+	bool init=false;
 	for(std::size_t i=0;i<array.size();i++)
 	{
 		if(!keep(array[i])) continue;
 
 		auto item = array[i];
-		if(!i) lowest = highest = item;
+		if(!init) {
+			init = true;
+			lowest = highest = item;
+		}
 		else if(item < lowest) lowest = item;
 		else if(item > highest) highest = item;
 	}