reverted rewrite of ppu, optimized what functions should be forced inline, gained ~2.5% performance

1 files changed, 10 insertions, 12 deletions

diff --git a/ppu.c b/ppu.c
index cd549d1..48ccbfc 100644
--- a/ppu.c
+++ b/ppu.c
@@ -26,8 +26,8 @@ static void ppu_reset(struct nes_state *state) {
 }
 
 
-__attribute__((always_inline, hot))
-static inline void ppu_evaluate_sprites(struct nes_state *state) {
+__attribute__((hot))
+static void ppu_evaluate_sprites(struct nes_state *state) {
 	struct ppu_state *restrict ppu = &state->ppu;
 	uint8_t sprite_height = (ppu->reg_ctrl & 0x20) ? 16 : 8;
 	uint8_t n = 0;
@@ -59,8 +59,8 @@ static inline void ppu_evaluate_sprites(struct nes_state *state) {
 	ppu->sprite_count = n;
 }
 
-__attribute__((always_inline, hot))
-static inline void ppu_fetch_sprite_patterns(struct nes_state *state) {
+__attribute__((hot))
+static void ppu_fetch_sprite_patterns(struct nes_state *state) {
 	struct ppu_state *restrict ppu = &state->ppu;
 	uint32_t addr;
 	uint32_t bank;
@@ -126,8 +126,8 @@ static inline void ppu_render_pixel(struct nes_state *state) {
 	uint8_t left_bg      = ppu->reg_mask & 0x02;
 	uint8_t left_sp      = ppu->reg_mask & 0x04;
 
-	uint8_t bg_mask = (show_bg && (left_bg || x >= 8)) ? 0xff : 0x00;
-	uint8_t sp_mask = (show_sprites && (left_sp || x >= 8)) ? 0xff : 0x00;
+	uint8_t bg_mask = (show_bg && (left_bg || x & ~7)) ? 0xff : 0x00;
+	uint8_t sp_mask = (show_sprites && (left_sp || x & ~7)) ? 0xff : 0x00;
 
 	// Background
 	uint8_t p0 = !!(ppu->bg_shift_pattern_low & bit);
@@ -149,8 +149,8 @@ static inline void ppu_render_pixel(struct nes_state *state) {
 		if(!sp_pixel) continue;
 
 		sp_palette = ppu->secondary_oam[i * 4 + 2] & 3;
-		sp_prio    = ppu->sprite_priorities[i];
-		sp_zero    = (ppu->sprite_indexes[i] == 0);
+		sp_prio = ppu->sprite_priorities[i];
+		sp_zero = (ppu->sprite_indexes[i] == 0);
 		break;
 	}
 
@@ -167,8 +167,6 @@ static inline void ppu_render_pixel(struct nes_state *state) {
 		case 3: {
 			ppu->reg_status |= (sp_zero && x < 255) ? 0x40 : 0;		// NOTE(peter): Sprite zero hit!
 			palette_index = (sp_prio) ? bg_index : 0x10 | sp_index;
-// printf("sprite zero hit: scanline: %3.3d  dot: %3.3d\n", ppu->scanline, ppu->dot);
-
 		} break;
 	}
 
@@ -397,8 +395,8 @@ static void ppu_tick(struct nes_state *state) {
 		}
 
 		if(UNLIKELY(scanline == 241) && dot == 1) {
-			static int32_t tas_frame = 0;
-			state->input[0] = tas_input[tas_frame++];
+			// static int32_t tas_frame = 0;
+			// state->input[0] = tas_input[tas_frame++];
 
 			ppu->reg_status |= 0x80;
 			if(ppu->reg_ctrl & 0x80) {