Nitpick run now allocates N and shifts double* to actually compute all of N

Author: lukas-valenta-tul

test/nitpick/nitpick_run.cc

@@ -1,5 +1,6 @@
 #include "nitpick_include.hh"
 #include <chrono>
+#include "scalar_node.hh"
 // only get us the variables
 #define NITPICK_IDE_IGNORE
 
@@ -10,11 +11,14 @@ using namespace bparser;
 #include NITPICK_GEN_FILE
 
 int main() {
+	const uint N = 1048576; //2^20 //1073741824U; //2^30
 
 	const uint max_vec_size = 2000;
-	const size_t buffer_size = sizeof(double) * max_vec_size + 500000;
+	const size_t buffer_size = sizeof(double) * N * 64;
 
-	const uint N = 1073741824U; //2^30
+	std::cout << "Buffer size: " << buffer_size << std::endl;
+	int simd_size = bparser::get_simd_size();
+	std::cout << "SIMD size: " << simd_size << std::endl;
 
 	void* buffer = ::operator new(buffer_size);
 	ExprCase exprcase(max_vec_size, (void*)buffer, buffer_size);
@@ -25,31 +29,61 @@ int main() {
 		Throw() << "No expression was set!";
 	}
 
-	std::vector<uint> subset = { 0, 1, 2, 3, 4,5,6,7,8,9,10,11,12 }; //ctverice doubluu //TODO: autocreate from vec_size
 	std::vector<uint> vec_sizes({ 16U, 64U, 256U, 1024U });
 
 	PatchArenaPtr arena = exprcase.get_patch_arena();
 	for (uint vec_size : vec_sizes) {
 
 		uint n_repeats = N / vec_size;
-		
+
 		std::cout << "Running \"" << exprcase.get_expression() << "\" " << n_repeats << "x for vec_size: " << vec_size << std::endl;
-		exprcase.allocate(vec_size);
+
+		//exprcase.allocate(vec_size);
+		exprcase.allocate(N);
+
+		Parser& p = exprcase.get_parser();
+		p.compile();
+		auto nodes = p.result_array().elements();
+		std::vector<ScalarNodePtr> value_nodes = {};
+		std::vector<ScalarNodePtr> val_copy_nodes = {};
+		for (uint i = 0; i < nodes.size(); i++) {
+			ScalarNodePtr n = nodes[i];
+			if (n->result_storage == value) {
+				value_nodes.push_back(n);
+			}
+			else if (n->result_storage == value_copy) {
+				val_copy_nodes.push_back(n);
+			}
+		}
+
+		auto dangerous_shift = [&](uint a) {
+			for (auto n : value_nodes) {
+				n->values_ += a;
+			}
+			for (auto n : val_copy_nodes) {
+				details::ValueCopyNode& vcn = dynamic_cast<details::ValueCopyNode&>(*n);
+				vcn.source_ptr_ += a;
+			}
+		};
 
 		//Get autogenerated dag;
 		ExpressionDAG se = gen(exprcase.get_node_map());
+		//I would expect vec_size instead of max_vec_size but that certainly does not work because the processor decides to go out of program reach and run memory -LV
+		ProcessorBase* processor = ProcessorBase::create_processor(se, max_vec_size, simd_size, arena); //get arena from ExprCase
 
-		ProcessorBase* processor = ProcessorBase::create_processor(se, max_vec_size, bparser::get_simd_size(), arena); //get arena from ExprCase
-		processor->set_subset(subset);
+		std::vector<uint> subset(std::ceil(vec_size / (float)simd_size)); //ex: 32/4 = 8 => 
+		std::iota(subset.begin(), subset.end(), 1);					// => {1, 2, 3, 4, 5, 6, 7, 8}
+		processor->set_subset(subset); //all subsets
 
 		auto start_time = std::chrono::high_resolution_clock::now();
 		for (uint i_rep = 0; i_rep < n_repeats; i_rep++) {
 			processor->run();
+			dangerous_shift(vec_size);
 		}
 		auto end_time = std::chrono::high_resolution_clock::now();
 
 		double time = std::chrono::duration_cast<std::chrono::duration<double>>(end_time - start_time).count();
-		std::cout << "Finished in " << time << std::endl;
+		std::cout << std::fixed << "Finished in " << time << std::defaultfloat << std::endl;
 
 
 		exprcase.deallocate(); //processor is allocated in arena