optimization of ppu

3 files changed, 5 insertions, 4 deletions

diff --git a/mknes.c b/mknes.c
index 77bdd45..a2f764c 100644
--- a/mknes.c
+++ b/mknes.c
@@ -146,7 +146,7 @@ int main(int argc, char **argv) {
 	// ines2_load(nstate, "data/0000/Xevious - The Avenger (USA).zip");
 	// ines2_load(nstate, "data/tv.nes");
 
-	// ines2_load(nstate, "data/Life Force (USA).zip");	// 2002
+	ines2_load(nstate, "data/Life Force (USA).zip");	// 2002
 
 	// ines2_load(nstate, "data/0003/Flipull - An Exciting Cube Game (Japan) (En).zip");
 	// ines2_load(nstate, "data/0003/Friday the 13th (USA).zip");
diff --git a/mknes.h b/mknes.h
index 27d4f6e..8d12e95 100644
--- a/mknes.h
+++ b/mknes.h
@@ -45,10 +45,10 @@ struct ppu_state {
 	uint8_t vram_read_buffer;
 	uint8_t open_bus;
 
-	uint8_t sprite_zero_hit_possible;
 	uint8_t sprite_count;
 	uint8_t palette[32];
 
+// NOTE(peter): one byte left on cacheline
 	// NOTE(peter): CACHELINE 2 start here!
 
 	uint8_t sprite_indexes[8];
diff --git a/ppu.c b/ppu.c
index 570f3dd..e0fce44 100644
--- a/ppu.c
+++ b/ppu.c
@@ -103,7 +103,7 @@ static inline void ppu_fetch_sprite_patterns(struct nes_state *state) {
 	}
 }
 
-__attribute__((hot, flatten))
+__attribute__((always_inline, hot, flatten))
 static inline void ppu_render_pixel(struct nes_state *state) {
 	struct ppu_state *restrict ppu = &state->ppu;
 
@@ -171,6 +171,7 @@ static inline void ppu_render_pixel(struct nes_state *state) {
 }
 
 __attribute__((hot, flatten))
+__attribute__((optimize("no-jump-tables")))
 static inline void ppu_tick(struct nes_state *state) {
 	struct ppu_state *restrict ppu = &state->ppu;
 
@@ -178,7 +179,7 @@ static inline void ppu_tick(struct nes_state *state) {
 	uint32_t scanline = ppu->scanline;
 	uint8_t rendering = (ppu->reg_mask & 0x18);
 
-	for(uint32_t ppu_loops = 0; ppu_loops < 3; ++ppu_loops) {
+	for(uint8_t ppu_loops = 0; ppu_loops < 3; ++ppu_loops) {
 
 		if(rendering) {
 			switch(scanline) {