improved scratch space

header.cxx

@@ -5,6 +5,7 @@
 #include "./server/gc.h"
 __AQEXPORT__(void) __AQ_Init_GC__(Context* cxt) {
     GC::gc_handle = static_cast<GC*>(cxt->gc);
+    GC::scratch_space = nullptr;
 }
 
 #else // __AQ_USE_THREADEDGC__

reconstruct/ast.py

@@ -4,8 +4,8 @@ from enum import Enum, auto
 from typing import Dict, List, Optional, Set, Tuple, Union
 
 from engine.types import *
-from engine.utils import (base62alp, base62uuid, enlist, get_innermost,
-                          get_legal_name)
+from engine.utils import (base62alp, base62uuid, enlist, 
+                          get_innermost, get_legal_name)
 from reconstruct.storage import ColRef, Context, TableInfo
 
 class ast_node:
@@ -599,7 +599,7 @@ class groupby_c(ast_node):
             self.context.emitc(f'{c}.reserve({self.group}.size());')
             if col_tovec[i]: # and type(col_types[i]) is VectorT:
                 typename : Types = col_types[i] # .inner_type
-                self.context.emitc(f'auto buf_{c} = static_cast<{typename.cname} *>(malloc({self.total_sz} * sizeof({typename.cname})));')
+                self.context.emitc(f'auto buf_{c} = static_cast<{typename.cname} *>(calloc({self.total_sz}, sizeof({typename.cname})));')
                 tovec_columns.add(c)
         self.arr_len = 'arrlen_' + base62uuid(3)
         self.arr_values = 'arrvals_' + base62uuid(3)
@@ -618,6 +618,7 @@ class groupby_c(ast_node):
                 )
             preproc_scanner.finalize()
         
+        self.context.emitc(f'GC::scratch_space = GC::gc_handle ? &(GC::gc_handle->scratch) : nullptr;')
         # gscanner = scan(self, self.group, loop_style = 'for_each')
         gscanner = scan(self, self.arr_len)
         key_var = 'key_'+base62uuid(7)
@@ -683,6 +684,7 @@ class groupby_c(ast_node):
                 gscanner.add(f'{ce[0]}.emplace_back({get_var_names_ex(ex)});\n')
         
         gscanner.finalize()
+        self.context.emitc(f'GC::scratch_space = nullptr;')
         
         self.datasource.groupinfo = None
 

server/aggregations.h

@@ -13,7 +13,7 @@ size_t count(const VT<T>& v) {
 }
 
 template <class T>
-constexpr static inline size_t count(const T&) { return 1; }
+constexpr static size_t count(const T&) { return 1; }
 
 // TODO: Specializations for dt/str/none
 template<class T, template<typename ...> class VT>
@@ -441,14 +441,12 @@ inline decayed_t<VT, T> aggnext(const VT<T>& arr) {
 template<class T, template<typename ...> class VT>
 T last(const VT<T>& arr) {
 	if (!arr.size) return 0;
-	const uint32_t& len = arr.size;
 	return arr[arr.size - 1];
 }
 
 template<class T, template<typename ...> class VT>
 T first(const VT<T>& arr) {
 	if (!arr.size) return 0;
-	const uint32_t& len = arr.size;
 	return arr[0];
 }
 

server/gc.h

@@ -1,4 +1,5 @@
 #pragma once
+#include <atomic>
 
 class ScratchSpace {
 public:
@@ -35,10 +36,8 @@ public:
 
 
 #ifndef __AQ_USE_THREADEDGC__
-#include <atomic>
 class GC {
-private:;
-
+private:
 	size_t max_slots, 
 		   interval, forced_clean, 
 		   forceclean_timer = 0;
@@ -53,7 +52,6 @@ private:;
 	std::atomic<uint64_t> current_size;
 	volatile bool lock;
 	using gc_deallocator_t = void (*)(void*);
-	ScratchSpace scratch;
 	// maybe use volatile std::thread::id instead
 protected:
 	void acquire_lock();
@@ -64,6 +62,7 @@ protected:
 	void terminate_daemon();
 
 public:
+	ScratchSpace scratch;
 	void reg(void* v, uint32_t sz = 0xffffffff, 
 			void(*f)(void*) = free
 		);
@@ -92,6 +91,7 @@ public:
 	}
 
 	static GC* gc_handle;
+	static ScratchSpace *scratch_space;
 	template <class T>
 	static inline gc_deallocator_t _delete(T*) {
 		return [](void* v){

server/libaquery.cpp

@@ -468,6 +468,7 @@ void GC::reg(void* v, uint32_t sz, void(*f)(void*)) { //~ 40ns expected v. free
 #endif
 
 inline GC* GC::gc_handle = nullptr;
+inline ScratchSpace* GC::scratch_space = nullptr;
 
 void ScratchSpace::init(size_t initial_capacity) {
 	ret = nullptr;

server/vector_type.hpp

@@ -49,20 +49,25 @@ public:
 		this->container = vt.container;
 		// puts("move");
 		vt.size = vt.capacity = 0;	
-		vt.container = 0;
+		vt.container = nullptr;
 	}
 public:
 	_Ty* container;
 	uint32_t size, capacity;
 	typedef _Ty* iterator_t;
 	typedef std::conditional_t<is_cstr<_Ty>(), astring_view, _Ty> value_t;
-	vector_type(const uint32_t& size) : size(size), capacity(size) {
+	explicit vector_type(const uint32_t& size) : size(size), capacity(size) {
+		if (GC::scratch_space != nullptr) {
+			[[likely]]
+			container = (_Ty*)GC::scratch_space->alloc(size * sizeof(_Ty));
+		}
 		container = (_Ty*)malloc(size * sizeof(_Ty));
 		// TODO: calloc for objects. 
 	}
-	constexpr vector_type(std::initializer_list<_Ty> _l) {
+	explicit constexpr vector_type(std::initializer_list<_Ty> _l) {
 		size = capacity = _l.size();
-		_Ty* _container = this->container = (_Ty*)malloc(sizeof(_Ty) * _l.size());
+		this->container = (_Ty*)malloc(sizeof(_Ty) * capacity);
+		_Ty* _container = this->container;
 		for (const auto& l : _l) {
 			*(_container++) = l;
 		}
@@ -80,8 +85,9 @@ public:
 	constexpr vector_type(vector_type<_Ty>&& vt) noexcept : capacity(0) {
 		_move(std::move(vt));
 	}
-	vector_type(vectortype_cstorage vt) noexcept : capacity(vt.capacity), size(vt.size), container((_Ty*)vt.container) {
-		out(10);
+	explicit vector_type(vectortype_cstorage vt) noexcept : 
+		capacity(vt.capacity), size(vt.size), container((_Ty*)vt.container) {
+		// out(10);
 	};
 	// size >= capacity ==> readonly vector
 	constexpr vector_type(const uint32_t size, void* data) : 
@@ -499,7 +505,7 @@ public:
 	}
 
 	inline void hashtable_push(Key&& k, uint32_t i){
-		reversemap[i] = ankerl::unordered_dense::set<Key, Hash>::hashtable_push(std::forward<Key&&>(k));
+		reversemap[i] = ankerl::unordered_dense::set<Key, Hash>::hashtable_push(std::move(k));
 		++ht_base[reversemap[i]];
 	}