diff --git a/main.c b/main.c
index f0442b6..b6d1f46 100644
--- a/main.c
+++ b/main.c
@@ -53,6 +53,7 @@ int main(int argc, char **argv) {
 	int frame = 0;
 	int press = 0;
 	int wait = 0;
+	int store = 0;
 
 	while (1) {
 		if (DEBUG) printf("0x%x ", state.program_counter);
@@ -66,10 +67,7 @@ int main(int argc, char **argv) {
 			press = 0;
 		}
 
-		clock_t start = clock() / (CLOCKS_PER_SEC/1000);
-
 		// opcodes are 16 bits
-		word opcode = (state.memory[state.program_counter++] << 8) | (state.memory[state.program_counter++]);
 
 		SDL_PollEvent(&e);
 		if (e.type == SDL_QUIT) break;
@@ -91,13 +89,15 @@ int main(int argc, char **argv) {
 			for (int i = 0; i < 0x10; ++i) {
 				if (state.input[i] != 0)
 				{
-					state.registers[0xF] = i;
+					state.registers[store] = i;
 					break;
 				}
 			}
 			wait = 0;
 		}
 
+		word opcode = (state.memory[state.program_counter++] << 8) | (state.memory[state.program_counter++]);
+
 		switch (opcode & 0xF000) 
 		{
 			case 0x0000: // screen clear or function return 
@@ -284,9 +284,11 @@ int main(int argc, char **argv) {
 					byte data = state.memory[state.address_I + line];
 					for (int xpos = 0; xpos < 8; ++xpos)
 					{
+						int ypos = y + line;
+						if (ypos > 63) ypos -= 64;
 						if (!(data & (1 << xpos))) continue;
-						if (state.screen[x + (7 - xpos)][y + line]) state.registers[0xF] = 1;
-						state.screen[x + (7 - xpos)][y + line] ^= 1;
+						if (state.screen[x + (7 - xpos)][ypos]) state.registers[0xF] = 1;
+						state.screen[x + (7 - xpos)][ypos] ^= 1;
 					}
 				}
 
@@ -329,6 +331,7 @@ int main(int argc, char **argv) {
 
 					case 0x0A:
 						wait = 1;
+						store = reg;
 						break;
 
 					case 0x15:
@@ -393,7 +396,6 @@ int main(int argc, char **argv) {
 
 		SDL_RenderPresent(renderer);
 
-		clock_t stop = clock() / (CLOCKS_PER_SEC/1000);
 		SDL_Delay(1);
 		++frame;
 		// SDL_DestroyWindow(window);