Better benchmarking, some small optimizations

9 files changed, 211 insertions, 150 deletions

diff --git a/Bench.sh b/Bench.sh
index 4a26ab5..157ad57 100755
--- a/Bench.sh
+++ b/Bench.sh
@@ -1,11 +1,45 @@
-#. ~/.local/bin/dev
+#!/usr/bin/env bash
 ./build.sh clean
 ./build.sh profile
 ./mknes
 ./build.sh profile_release
-taskset -c 1 chrt -f 99 -- perf stat -- ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -- ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -- ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -e stalled-cycles-frontend:u -e L1-icache-loads -e L1-icache-load-misses -- ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -e stalled-cycles-frontend:u -e L1-icache-loads -e L1-icache-load-misses -- ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -e stalled-cycles-frontend:u -e L1-icache-loads -e L1-icache-load-misses -- ./mknes
+
+runs=10
+frames=4096
+events="cycles,instructions,task-clock"
+tmp=$(mktemp)
+
+taskset -c 1 ./mknes
+
+> "$tmp"
+for i in $(seq 1 $runs); do
+  taskset -c 1 chrt -f 99 perf stat -x, -e $events -- ./mknes 2>>"$tmp"
+done
+
+
+awk -F, -v F="$frames" '
+  $3=="cycles"        { c[++nc]=$1/F }
+  $3=="instructions"  { i[++ni]=$1/F }
+#   $3=="task-clock"    { t[++nt]=$1 }  # milliseconds NOTE(peter): changed to nanoseconds...
+  $3=="task-clock" { t[++nt]=$1/1000000 }
+
+  END {
+    for(k=1;k<=nc;k++) sumc+=c[k]; mc=sumc/nc
+    for(k=1;k<=ni;k++) sumi+=i[k]; mi=sumi/ni
+    for(k=1;k<=nt;k++) sumt+=t[k]; mt=sumt/nt
+
+    for(k=1;k<=nc;k++) sdc+=(c[k]-mc)^2; sdc=sqrt(sdc/(nc-1))
+    for(k=1;k<=ni;k++) sdi+=(i[k]-mi)^2; sdi=sqrt(sdi/(ni-1))
+    for(k=1;k<=nt;k++) sdt+=(t[k]-mt)^2; sdt=sqrt(sdt/(nt-1))
+
+    ms_per_frame = mt / F
+    fps = F / (mt / 1000)
+
+    printf "IPC (insn/cycle) = %.3f\n", mi/mc
+    printf "cycles/frame  mean=%.0f  sd=%.0f  relSD=%.3f%%  n=%d\n", mc, sdc, 100*sdc/mc, nc
+    printf "insn/frame    mean=%.0f  sd=%.0f  relSD=%.3f%%  n=%d\n", mi, sdi, 100*sdi/mi, ni
+    printf "time (ms)     mean=%.3f  sd=%.3f  relSD=%.3f%%  n=%d\n", mt, sdt, 100*sdt/mt, nt
+    printf "FPS (frames/second) = %.2f\n", fps
+    printf "ms/frame     = %.6f\n", ms_per_frame
+  }' "$tmp"
+
diff --git a/build.sh b/build.sh
index b8e4fa9..a8b26e8 100755
--- a/build.sh
+++ b/build.sh
@@ -53,17 +53,18 @@ case "$BUILD_TYPE" in
 		;;
 	"release")
 		# CFLAGS+="-s -Wl,--strip-all -O2 "
-		CFLAGS+="-g -O2 "
+		CFLAGS+=" -O2 "
 		;;
 	"profile")
-		CFLAGS+="-O2 -fprofile-generate -ftest-coverage "
+		CFLAGS+="-O2 -fprofile-generate -ftest-coverage -DBENCHMARK "
 		;;
 	"profile_release")
 		# CFLAGS+="-s -Wl,--strip-all -O2 -fprofile-use "
-		CFLAGS+="-g -O2 -fprofile-use "
+		CFLAGS+="-g -O2 -fprofile-use -DBENCHMARK "
 		;;
 	"debug")
-		CFLAGS+="-g -O0 -DTIMER_DEBUG "
+		CFLAGS+="-g -O0 "
+# -DTIMER_DEBUG "
 		LDFLAGS+="-fno-pie -no-pie "
 		;;
 	"coverage")
diff --git a/mknes.c b/mknes.c
index a6d224d..f8850b0 100644
--- a/mknes.c
+++ b/mknes.c
@@ -37,8 +37,8 @@
 #define CIRAM_SIZE	0x1000
 #define CHR_RAM_SIZE	0x4000
 
-uint32_t buffer[BUFFER_WIDTH * BUFFER_HEIGHT] __attribute__((section(".bss"), aligned(4096)));
-uint32_t display_buffer[BUFFER_WIDTH * BUFFER_HEIGHT] __attribute__((section(".bss"), aligned(4096)));
+static uint32_t buffer[BUFFER_WIDTH * BUFFER_HEIGHT] __attribute__((section(".bss"), aligned(4096)));
+static uint32_t display_buffer[BUFFER_WIDTH * BUFFER_HEIGHT] __attribute__((section(".bss"), aligned(4096)));
 
 static void audio_callback(int16_t *data, size_t frames) { }
 
@@ -54,6 +54,14 @@ static void audio_callback(int16_t *data, size_t frames) { }
 // #include "audio.c"
 #include "incbin.h"
 
+#ifdef BENCHMARK
+// Embed the ROM for benchmarking to eliminate file I/O overhead
+// Uncomment the ROM you want to benchmark:
+INCBIN_BYTES(benchmark_rom, "data/Life Force (USA).nes");
+// INCBIN_BYTES(benchmark_rom, "data/0000/Super Mario Bros. (World) (HVC-SM).nes");
+// INCBIN_BYTES(benchmark_rom, "data/0003/Gradius (USA).zip");
+#endif
+
 #include "platform_gl_loader.c"
 
 
@@ -152,7 +160,7 @@ struct main_state state __attribute__((aligned(64)));
 #include "mkfw.h"
 #include "platform_opengl.c"
 
-
+static size_t sprite_counts[8];
 static uint32_t frames;		// debug information
 // static int32_t tas_frame_count;
 
@@ -235,12 +243,17 @@ int main(int argc, char **argv) {
 	memset(nstate, 0, sizeof(struct nes_state));
 
 	ppu_reset(nstate);
+#ifdef BENCHMARK
+	// Use embedded ROM for consistent benchmarking without file I/O overhead
+	ines2_load_from_memory(nstate, benchmark_rom, INCBIN_SIZE(benchmark_rom));
+
+#else
 	// ines2_load(nstate, "data/0000/10-Yard Fight (USA, Europe).nes");
 	// ines2_load(nstate, "data/0000/Balloon Fight (USA).nes");
 	// ines2_load(nstate, "data/0000/Excitebike (Japan, USA).nes");
 	// ines2_load(nstate, "data/0000/Ice Climber (USA, Europe, Korea).nes");
 	// ines2_load(nstate, "data/0000/Kung Fu (Japan, USA).nes");
-	ines2_load(nstate, "data/0000/Super Mario Bros. (World) (HVC-SM).nes");
+	// ines2_load(nstate, "data/0000/Super Mario Bros. (World) (HVC-SM).nes");
 	// ines2_load(nstate, "data/Super Mario Bros. (W) (V1.0) [!].nes");
 	// ines2_load(nstate, "data/Super Mario Bros. (JU) [!].nes");
 	// ines2_load(nstate, "data/0000/Urban Champion (World).nes");
@@ -253,7 +266,7 @@ int main(int argc, char **argv) {
 	// ines2_load(nstate, "data/0000/raster_demos/RasterTest3.NES");
 	// ines2_load(nstate, "data/0000/raster_demos/RasterTest3a.NES");
 	// ines2_load(nstate, "data/0000/raster_demos/RasterTest3b.NES");
-	ines2_load(nstate, "data/0000/raster_demos/RasterTest3c.NES");
+	// ines2_load(nstate, "data/0000/raster_demos/RasterTest3c.NES");
 	// ines2_load(nstate, "data/0000/raster_demos/RasterTest3d.NES");
 	// ines2_load(nstate, "data/0000/raster_demos/RasterTest3e.NES");
 	// ines2_load(nstate, "data/0000/NEStress.NES");
@@ -262,7 +275,7 @@ 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).nes");	// 2002
 
 	// ines2_load(nstate, "data/0003/Flipull - An Exciting Cube Game (Japan) (En).zip");
 	// ines2_load(nstate, "data/0003/Friday the 13th (USA).zip");
@@ -286,12 +299,13 @@ int main(int argc, char **argv) {
 
 	// ines2_load(nstate, "data/Blaster Master (USA).zip");	// mapper 1
 	// ines2_load(nstate, "AccuracyCoin.nes");	// mapper 1
+#endif
 
 	mapper_setup(nstate);
 	cpu_reset(nstate);
 
-#if 0
-	for(uint32_t i = 0; i < 0x5000; ++i) {
+#ifdef BENCHMARK
+	for(uint32_t i = 0; i < 0x1000; ++i) {
 		while(!nstate->ppu.frame_ready) {
 			// PROFILE_NAMED("nes emulator");
 			cpu_tick(nstate);
@@ -300,10 +314,14 @@ int main(int argc, char **argv) {
 		frames++;
 	}
 
+	// for(size_t i = 0; i < 9; ++i) {
+	// 	printf("count %d: %lld\n", i, sprite_counts[i]);
+	// }
 
 	// for(size_t i = 0; i < 256; ++i) {
 	// 	printf("instr %2.2x: %lld\n", i, instr_count[i]);
 	// }
+
 	return 0;
 #else
 
diff --git a/mknes.h b/mknes.h
index d21a9ad..e9d3e0e 100644
--- a/mknes.h
+++ b/mknes.h
@@ -1,17 +1,27 @@
 
 
 
-#define PPU_CTRL_BG_TILE_SELECT     0x10
-#define PPU_CTRL_SPRITE_TILE_SELECT 0x08
-
-// Define constants for PPU control and mask bits
+// Define constants for PPU control register bits ($2000)
 #define PPU_CTRL_NMI               0x80
 #define PPU_CTRL_SPRITE_HEIGHT     0x20
+#define PPU_CTRL_BG_TILE_SELECT    0x10
 #define PPU_CTRL_SPRITE_TILE       0x08
 #define PPU_CTRL_VRAM_INCREMENT    0x04
 
-#define PPU_MASK_SHOW_BG           0x08
+// Define constants for PPU mask register bits ($2001)
 #define PPU_MASK_SHOW_SPRITES      0x10
+#define PPU_MASK_SHOW_BG           0x08
+
+// Define constants for PPU status register bits ($2002)
+#define PPU_STATUS_VBLANK          0x80
+#define PPU_STATUS_SPRITE_ZERO_HIT 0x40
+#define PPU_STATUS_SPRITE_OVERFLOW 0x20
+
+// Define constants for sprite attribute bits
+#define SPRITE_ATTR_FLIP_VERTICAL  0x80
+#define SPRITE_ATTR_FLIP_HORIZONTAL 0x40
+#define SPRITE_ATTR_PRIORITY       0x20
+#define SPRITE_ATTR_PALETTE_MASK   0x03
 
 // Define mirroring modes
 #define MIRROR_HORIZONTAL 0
@@ -45,13 +55,11 @@ struct ppu_state {
 	uint8_t vram_read_buffer;			// 28
 	uint8_t open_bus;						// 29
 
-	uint8_t sprite_count;				// 30
-	uint8_t overflow_scheduled_dot;	// 31
-	uint8_t palette[32];					// 32
+	uint8_t sprite_count;				// 30 - Number of sprites in secondary OAM (0-8)
+	uint8_t palette[32];					// 31
 
 	// NOTE(peter): CACHELINE 2
 	uint8_t secondary_oam[32] __attribute__((aligned(64)));
-	uint8_t temp_secondary_oam[32];
 
 	// NOTE(peter): CACHELINE 3
 	struct sprite_data {
@@ -68,8 +76,6 @@ struct ppu_state {
 	uint8_t input_strobe;				// 46 - Control bit (0 or 1)
 	uint8_t frame_ready;					// 47
 	uint8_t sprite_zero_in_range;		// 48 - Boolean: is sprite 0 in range (will always be slot 0 if true)
-	uint8_t sprite_count_next;			// 49 - Sprite count for next scanline
-	// 15 bytes left.
 
 	// NOTE(peter): CACHELINE 4
 	uint8_t oam[256] __attribute__((aligned(64)));
diff --git a/mknes_cpu_opcodes.c b/mknes_cpu_opcodes.c
index b2fafd5..6b67f77 100644
--- a/mknes_cpu_opcodes.c
+++ b/mknes_cpu_opcodes.c
@@ -1758,6 +1758,7 @@ static void opcode_ror_absx(struct nes_state *state) {
 
 
 // SBC
+__attribute__((hot, always_inline))
 static inline void sbc(struct cpu_state * restrict cpu, uint8_t value) {
 #ifdef ENABLE_DECIMAL_MODE
 	if(cpu->d) {
diff --git a/mknes_ines2.c b/mknes_ines2.c
index 29ceb71..686eb38 100644
--- a/mknes_ines2.c
+++ b/mknes_ines2.c
@@ -1,9 +1,11 @@
+#ifndef BENCHMARK
 #define USE_LIBARCHIVE
 
 #ifdef USE_LIBARCHIVE
 #include <archive.h>
 #include <archive_entry.h>
 #endif
+#endif
 
 // iNES header fields
 #define INES_HEADER_SIZE	16
@@ -30,7 +32,7 @@
 #define MIRROR_VERTICAL		1
 #define MIRROR_FOUR_SCREEN	2
 
-
+#ifndef BENCHMARK
 static uint8_t *ines2_read_entire_file(const char *path, size_t *out_size) {
 	FILE *f = fopen(path, "rb");
 	if(!f) return 0;
@@ -99,32 +101,16 @@ uint8_t *ines2_unzip_file_to_memory(const char *zip_path, size_t *out_size) {
 	return buffer;
 }
 #endif
-
-static int ines2_load(struct nes_state *state, const char *path) {
-	uint8_t *data = 0;
-	size_t size = 0;
-
-#ifdef USE_LIBARCHIVE
-	if(strstr(path, ".zip")) {
-		data = ines2_unzip_file_to_memory(path, &size);
-	} else {
-		data = ines2_read_entire_file(path, &size);
-	}
-#else
-	if(strstr(path, ".zip")) {
-		fprintf(stderr, "ZIP support not compiled in. Please use .nes files directly.\n");
-		return -1;
-	}
-	data = ines2_read_entire_file(path, &size);
 #endif
 
+// Load ROM from memory buffer (for embedded ROMs)
+static int ines2_load_from_memory(struct nes_state *state, const uint8_t *data, size_t size) {
 	if(!data || size < INES_HEADER_SIZE) {
-		free(data);
 		return -1;
 	}
 
-	uint8_t *ptr = data;
-	uint8_t *header = ptr; ptr += INES_HEADER_SIZE;
+	const uint8_t *ptr = data;
+	const uint8_t *header = ptr; ptr += INES_HEADER_SIZE;
 
 	uint8_t prg_lsb = header[INES_PRG_SIZE_LSB];
 	uint8_t chr_lsb = header[INES_CHR_SIZE_LSB];
@@ -168,9 +154,38 @@ static int ines2_load(struct nes_state *state, const char *path) {
 		memcpy(state->chr_rom, ptr, chr_size);
 	}
 
-	free(data);
 	return 0;
 }
 
+#ifndef BENCHMARK
+static int ines2_load(struct nes_state *state, const char *path) {
+	uint8_t *data = 0;
+	size_t size = 0;
+
+#ifdef USE_LIBARCHIVE
+	if(strstr(path, ".zip")) {
+		data = ines2_unzip_file_to_memory(path, &size);
+	} else {
+		data = ines2_read_entire_file(path, &size);
+	}
+#else
+	if(strstr(path, ".zip")) {
+		fprintf(stderr, "ZIP support not compiled in. Please use .nes files directly.\n");
+		return -1;
+	}
+	data = ines2_read_entire_file(path, &size);
+#endif
+
+	if(!data || size < INES_HEADER_SIZE) {
+		free(data);
+		return -1;
+	}
+
+	int result = ines2_load_from_memory(state, data, size);
+	free(data);
+	return result;
+}
+#endif
+
 
 
diff --git a/mknes_memory.c b/mknes_memory.c
index 7b74424..49546b2 100644
--- a/mknes_memory.c
+++ b/mknes_memory.c
@@ -5,6 +5,8 @@ static inline uint8_t memory_read(struct nes_state *state, uint32_t offset) {
 	apu_tick(state);
 	ppu_tick(state);
 
+	if(offset >= 0x10000) __builtin_unreachable();
+
 	if(offset <= 0x1fff) {
 		return state->ram[offset & 0x07ff];
 	}
@@ -55,55 +57,56 @@ static inline void memory_write(struct nes_state *state, uint32_t offset, uint8_
 	apu_tick(state);
 	ppu_tick(state);
 
-	switch(offset) {
-		case 0x0000 ... 0x1fff: {
-			state->ram[offset & 0x07ff] = value;
-		} break;
+	if(offset >= 0x10000) __builtin_unreachable();
 
+	if(offset <= 0x1fff) {
+		state->ram[offset & 0x07ff] = value;
+		return;
+	}
 
-		case 0x2000 ... 0x3fff: {
-			ppu_write(state, offset, value);
-		} break;
+	if(offset >= 0x8000) {
+		state->mapper_function.prg_rom_write(state, offset, value);
+		return;
+	}
 
-		case 0x4000 ... 0x4017: {
-			switch(offset) {
-				case 0x4014: {
-					ppu_dma_4014(state, value);
-				} break;
+	if(offset >= 0x2000 && offset <= 0x3fff) {
+		ppu_write(state, offset, value);
+		return;
+	}
 
-				case 0x4016: {
-					// joypad strobe
-					uint8_t s = value & 1;
+	if(offset == 0x4014) {
+		ppu_dma_4014(state, value);
+		return;
+	}
 
-					// if(s) {
-						uint8_t prev = state->ppu.input_strobe;
-						state->ppu.input_strobe = s;
+	if(offset == 0x4016) {
+		// joypad strobe
+		uint8_t s = value & 1;
 
-						if(prev == 1 && (s) == 0) {
-							// state->ppu.input[0] = tas_input[tas_frame_count];
+		// if(s) {
+			uint8_t prev = state->ppu.input_strobe;
+			state->ppu.input_strobe = s;
 
-							state->ppu.input_latch[0] = state->ppu.input[0];
-							state->ppu.input_latch[1] = state->ppu.input[1];
-							state->ppu.input_bit[0] = 0;
-							state->ppu.input_bit[1] = 0;
-						}
-					// }
-				} break;
+			if(prev == 1 && (s) == 0) {
+				// state->ppu.input[0] = tas_input[tas_frame_count];
 
-				default: {
-					apu_write(state, offset, value);
-				} break;
+				state->ppu.input_latch[0] = state->ppu.input[0];
+				state->ppu.input_latch[1] = state->ppu.input[1];
+				state->ppu.input_bit[0] = 0;
+				state->ppu.input_bit[1] = 0;
 			}
-		} break;
-
-		case 0x6000 ... 0x7fff: {
-			state->mapper_function.prg_ram_write(state, offset, value);
-		} break;
+		// }
+		return;
+	}
 
-		case 0x8000 ... 0xffff: {
-			state->mapper_function.prg_rom_write(state, offset, value);
-		} break;
+	if(offset >= 0x4000 && offset <= 0x4017) {
+		apu_write(state, offset, value);
+		return;
+	}
 
+	if(offset >= 0x6000 && offset <= 0x7fff) {
+		state->mapper_function.prg_ram_write(state, offset, value);
+		return;
 	}
 }
 
diff --git a/mknes_ppu.c b/mknes_ppu.c
index 92b22fc..fcaf681 100644
--- a/mknes_ppu.c
+++ b/mknes_ppu.c
@@ -27,39 +27,32 @@ static void ppu_reset(struct nes_state *state) {
 __attribute__((hot, flatten))
 static inline void ppu_evaluate_sprites(struct nes_state *state, uint32_t scanline) {
 	struct ppu_state *restrict ppu = &state->ppu;
-	uint8_t sprite_height = (ppu->reg_ctrl & 0x20) ? 16 : 8;
+	uint8_t sprite_height = (ppu->reg_ctrl & PPU_CTRL_SPRITE_HEIGHT) ? 16 : 8;
 	uint8_t n = 0;
 	uint8_t sprite_zero_found = 0;
 
 	uint8_t * restrict src = ppu->oam;
-	uint8_t * restrict dst = ppu->temp_secondary_oam;  // Write to temp buffer
+	uint8_t * restrict dst = ppu->secondary_oam;
 
 	for(uint8_t i = 0; i < 64; i++, src += 4) {
-		uint8_t y = src[0];
-		int32_t row = (int32_t)scanline - y;
-
-		if(row >= 0 && row < sprite_height) {
+		uint32_t row = scanline - src[0];
+		if(row < sprite_height) {
 			if(n < 8) {
 				dst[0] = src[0];
 				dst[1] = src[1];
 				dst[2] = src[2];
 				dst[3] = src[3];
-
-				if(i == 0) {
-					sprite_zero_found = 1;  // Sprite 0 is in range, will be in slot 0
-				}
-
 				dst += 4;
+				sprite_zero_found |= (i == 0);
 				n++;
-
 			} else {
-				ppu->reg_status |= 0x20;  // Set overflow immediately
+				ppu->reg_status |= PPU_STATUS_SPRITE_OVERFLOW;
 				break;
 			}
 		}
 	}
 	ppu->sprite_zero_in_range = sprite_zero_found;
-	ppu->sprite_count_next = n;
+	ppu->sprite_count = n;
 }
 
 __attribute__((hot))
@@ -68,8 +61,8 @@ static inline void ppu_fetch_sprite_patterns(struct nes_state * restrict state,
 
 	uint8_t * restrict sec_oam = ppu->secondary_oam;
 	uint8_t ctrl = ppu->reg_ctrl;
-	uint8_t sprite_height = (ctrl & 0x20) ? 16 : 8;
-	uint32_t sprite_pattern_table_base = (ctrl & 0x08) << 9;
+	uint8_t sprite_height = (ctrl & PPU_CTRL_SPRITE_HEIGHT) ? 16 : 8;
+	uint32_t sprite_pattern_table_base = (ctrl & PPU_CTRL_SPRITE_TILE) << 9;
 
 	for(uint8_t i = 0; i < ppu->sprite_count; i++, sec_oam += 4) {
 		uint8_t y = sec_oam[0];
@@ -78,9 +71,10 @@ static inline void ppu_fetch_sprite_patterns(struct nes_state * restrict state,
 		uint8_t x = sec_oam[3];
 
 		uint32_t row = scanline - y;
-		row = (attr & 0x80) ? sprite_height - 1 - row : row;
+		row = (attr & SPRITE_ATTR_FLIP_VERTICAL) ? sprite_height - 1 - row : row;
 
-		uint32_t bank, addr;
+		uint32_t bank;
+		uint32_t addr;
 		if(sprite_height == 16) {
 			bank = (tile & 1) << 12;
 			tile &= 0xfe;
@@ -97,15 +91,15 @@ static inline void ppu_fetch_sprite_patterns(struct nes_state * restrict state,
 		uint8_t val_lo = state->mapper_function.chr_read(state, addr);
 		uint8_t val_hi = state->mapper_function.chr_read(state, addr + 8);
 
-		uint8_t rev = -(!!(attr & 0x40));
+		uint8_t rev = -(!!(attr & SPRITE_ATTR_FLIP_HORIZONTAL));
 		uint8_t lsb = (rev & ppu_bitreverse_lut[val_lo]) | (~rev & val_lo);
 		uint8_t msb = (rev & ppu_bitreverse_lut[val_hi]) | (~rev & val_hi);
 
 		ppu->sprites[i].shift_lo = lsb;
 		ppu->sprites[i].shift_hi = msb;
 		ppu->sprites[i].position = x;
-		ppu->sprites[i].priority = attr & 0x20;
-		ppu->sprites[i].palette = attr & 0x3;
+		ppu->sprites[i].priority = attr & SPRITE_ATTR_PRIORITY;
+		ppu->sprites[i].palette = attr & SPRITE_ATTR_PALETTE_MASK;
 	}
 }
 
@@ -121,10 +115,11 @@ static inline void ppu_render_pixel(struct nes_state * restrict state, uint32_t
 	uint8_t sp_prio = 0;
 	uint8_t sp_zero = 0;
 
-	uint8_t show_bg      = ppu->reg_mask & 0x08;
-	uint8_t show_sprites = ppu->reg_mask & 0x10;
-	uint8_t left_bg      = ppu->reg_mask & 0x02;
-	uint8_t left_sp      = ppu->reg_mask & 0x04;
+	uint8_t mask_reg = ppu->reg_mask;  // Single load
+	uint8_t show_bg = mask_reg & PPU_MASK_SHOW_BG;
+	uint8_t show_sprites = mask_reg & PPU_MASK_SHOW_SPRITES;
+	uint8_t left_bg = mask_reg & 0x02;
+	uint8_t left_sp = mask_reg & 0x04;
 
 	uint8_t bg_mask = (show_bg && (left_bg || x & ~7)) ? 0xff : 0x00;
 	uint8_t sp_mask = (show_sprites && (left_sp || x & ~7));// ? 0xff : 0x00;
@@ -153,15 +148,17 @@ static inline void ppu_render_pixel(struct nes_state * restrict state, uint32_t
 	}																						\
 } while (0)
 
+	// sprite_counts[ppu->sprite_count]++;
 	if(sp_mask && ppu->sprite_count > 0) {
-		if(ppu->sprite_count == 1) goto sprite_1;
 		if(ppu->sprite_count == 2) goto sprite_2;
+		if(ppu->sprite_count == 1) goto sprite_1;
 		if(ppu->sprite_count == 3) goto sprite_3;
 		if(ppu->sprite_count == 4) goto sprite_4;
 		if(ppu->sprite_count == 5) goto sprite_5;
 		if(ppu->sprite_count == 6) goto sprite_6;
-		if(ppu->sprite_count == 7) goto sprite_7;
 		if(ppu->sprite_count == 8) goto sprite_8;
+		if(ppu->sprite_count == 7) goto sprite_7;
+
 
 sprite_8:	SPRITE_STEP(7);
 sprite_7:	SPRITE_STEP(6);
@@ -172,8 +169,7 @@ sprite_3:	SPRITE_STEP(2);
 sprite_2:	SPRITE_STEP(1);
 sprite_1:	SPRITE_STEP(0);
 	}
-sprite_done:;
-
+sprite_done:
 
 	// Final pixel composition
 	uint8_t bg_index = (bg_palette << 2) + bg_pixel;
@@ -187,7 +183,7 @@ sprite_done:;
 		case 0: { palette_index = 0;												} break;
 		case 1: { palette_index = 0x10 | sp_index;							} break;
 		case 2: { palette_index = bg_index;										} break;
-		case 3: { ppu->reg_status |= (sp_zero && x < 255) ? 0x40 : 0;	} break;		// NOTE(peter): Sprite zero hit!
+		case 3: { ppu->reg_status |= (sp_zero && x < 255) ? PPU_STATUS_SPRITE_ZERO_HIT : 0;	} break;		// NOTE(peter): Sprite zero hit!
 	}
 
 	state->pixels[y * 256 + x] = ppu->palette[palette_index];		// NOTE(peter): Add color_emphasis bits (expand palette to 8x).
@@ -199,7 +195,7 @@ static void ppu_tick(struct nes_state *state) {
 
 	uint32_t dot = ppu->dot;
 	uint32_t scanline = ppu->scanline;
-	uint8_t rendering = (ppu->reg_mask & 0x18);
+	uint8_t rendering = (ppu->reg_mask & (PPU_MASK_SHOW_SPRITES | PPU_MASK_SHOW_BG));
 
 	for(uint8_t ppu_loops = 0; ppu_loops < 3; ++ppu_loops) {
 
@@ -207,10 +203,6 @@ static void ppu_tick(struct nes_state *state) {
 
 			if(scanline <= 239) {
 
-				if(dot == 65) {
-					ppu_evaluate_sprites(state, scanline);  // Evaluate sprites early, sets overflow immediately
-				}
-
 				if(dot >= 1 && dot <= 256) {
 					if(dot == 256) {
 						if((ppu->vram_addr & 0x7000) != 0x7000) {
@@ -237,12 +229,11 @@ static void ppu_tick(struct nes_state *state) {
 
 				if(dot == 257) {
 					ppu->vram_addr = (ppu->vram_addr & ~0x041f) | (ppu->temp_addr & 0x041f);
-					memcpy(ppu->secondary_oam, ppu->temp_secondary_oam, 32);
-					ppu->sprite_count = ppu->sprite_count_next;
+					ppu_evaluate_sprites(state, scanline);
 				}
 
 				if(dot >= 321 && dot <= 336) {
-stupid:			if(ppu->reg_mask & 0x10) {
+stupid:			if(ppu->reg_mask & PPU_MASK_SHOW_SPRITES) {
 						for(uint32_t i = 0; i < ppu->sprite_count; i++) {
 							if(ppu->sprites[i].position > 0) {
 								ppu->sprites[i].position--;
@@ -272,7 +263,7 @@ stupid:			if(ppu->reg_mask & 0x10) {
 						} break;
 
 						case 5: {
-							uint32_t base = (ppu->reg_ctrl & 0x10) << 8;
+							uint32_t base = (ppu->reg_ctrl & PPU_CTRL_BG_TILE_SELECT) << 8;
 							uint32_t tile = ppu->bg_next_tile_id;
 							uint32_t fine_y = (ppu->vram_addr >> 12) & 7;
 							uint32_t addr_lsb = (base + tile * 16 + fine_y) & 0x1fff;
@@ -280,7 +271,7 @@ stupid:			if(ppu->reg_mask & 0x10) {
 						} break;
 
 						case 7: {
-							uint32_t base = (ppu->reg_ctrl & 0x10) << 8;
+							uint32_t base = (ppu->reg_ctrl & PPU_CTRL_BG_TILE_SELECT) << 8;
 							uint32_t tile = ppu->bg_next_tile_id;
 							uint32_t fine_y = (ppu->vram_addr >> 12) & 7;
 							uint32_t addr_msb = (base + tile * 16 + fine_y + 8) & 0x1fff;
@@ -312,9 +303,6 @@ stupid:			if(ppu->reg_mask & 0x10) {
 			}
 
 			if(scanline == 261) {
-				if(dot == 65) {
-					ppu_evaluate_sprites(state, scanline);  // Evaluate sprites early
-				}
 
 				if(dot >= 1 && dot <= 256) {
 					if(dot == 256) {
@@ -341,8 +329,6 @@ stupid:			if(ppu->reg_mask & 0x10) {
 
 				if(dot == 257) {
 					ppu->vram_addr = (ppu->vram_addr & ~0x041f) | (ppu->temp_addr & 0x041f);
-					memcpy(ppu->secondary_oam, ppu->temp_secondary_oam, 32);
-					ppu->sprite_count = ppu->sprite_count_next;
 				}
 
 				if(dot >= 280 && dot <= 304) {
@@ -350,7 +336,7 @@ stupid:			if(ppu->reg_mask & 0x10) {
 				}
 
 				if(dot >= 321 && dot <= 336) {
-stupid2:			if(ppu->reg_mask & 0x10) {
+stupid2:			if(ppu->reg_mask & PPU_MASK_SHOW_SPRITES) {
 						for(uint32_t i = 0; i < ppu->sprite_count; i++) {
 							if(ppu->sprites[i].position > 0) {
 								ppu->sprites[i].position--;
@@ -381,7 +367,7 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 						} break;
 
 						case 5: {
-							uint32_t base = (ppu->reg_ctrl & 0x10) << 8;
+							uint32_t base = (ppu->reg_ctrl & PPU_CTRL_BG_TILE_SELECT) << 8;
 							uint32_t tile = ppu->bg_next_tile_id;
 							uint32_t fine_y = (ppu->vram_addr >> 12) & 7;
 							uint32_t addr_lsb = (base + tile * 16 + fine_y) & 0x1fff;
@@ -389,7 +375,7 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 						} break;
 
 						case 7: {
-							uint32_t base = (ppu->reg_ctrl & 0x10) << 8;
+							uint32_t base = (ppu->reg_ctrl & PPU_CTRL_BG_TILE_SELECT) << 8;
 							uint32_t tile = ppu->bg_next_tile_id;
 							uint32_t fine_y = (ppu->vram_addr >> 12) & 7;
 							uint32_t addr_msb = (base + tile * 16 + fine_y + 8) & 0x1fff;
@@ -413,10 +399,6 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 						} break;
 					}
 				}
-
-				if(dot == 340) {
-					ppu_fetch_sprite_patterns(state, scanline);
-				}
 			}
 
 		}
@@ -424,13 +406,14 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 		if(dot == 1) {
 			switch(scanline) {
 				case 241: {
-					ppu->reg_status |= 0x80;
-					state->cpu.nmi_pending = (ppu->reg_ctrl & 0x80);
+					ppu->reg_status |= PPU_STATUS_VBLANK;
+					state->cpu.nmi_pending = (ppu->reg_ctrl & PPU_CTRL_NMI);
 				} break;
 
 				case 261: {
-					ppu->reg_status &= ~0x80;
-					ppu->reg_status &= ~0x40;
+					ppu->reg_status &= ~PPU_STATUS_VBLANK;
+					ppu->reg_status &= ~PPU_STATUS_SPRITE_ZERO_HIT;
+					ppu->reg_status &= ~PPU_STATUS_SPRITE_OVERFLOW;
 				} break;
 			}
 		}
@@ -440,7 +423,7 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 			dot = 0;
 			scanline++;
 
-			if(scanline == 261 && !ppu->even_frame && (ppu->reg_mask & 0x18)) {
+			if(scanline == 261 && !ppu->even_frame && rendering) {
 				dot = 1;
 			}
 
diff --git a/mknes_ppu_registers.c b/mknes_ppu_registers.c
index a7aa351..b66f7e5 100644
--- a/mknes_ppu_registers.c
+++ b/mknes_ppu_registers.c
@@ -65,7 +65,7 @@ static inline void ppu_write(struct nes_state *state, uint32_t offset, uint8_t v
 				} break;
 			}
 
-			ppu->vram_addr += (ppu->reg_ctrl & 0x04) ? 32 : 1;
+			ppu->vram_addr += (ppu->reg_ctrl & PPU_CTRL_VRAM_INCREMENT) ? 32 : 1;
 		} break;
 
 	}
@@ -84,7 +84,7 @@ static inline uint8_t ppu_read(struct nes_state *state, uint32_t offset) {
 		case 2: {
 			result &= 0x1f;
 			result |= ppu->reg_status & 0xe0;
-			ppu->reg_status &= ~0x80;
+			ppu->reg_status &= ~PPU_STATUS_VBLANK;
 			ppu->write_latch = 0;
 		} break;
 
@@ -118,7 +118,7 @@ static inline uint8_t ppu_read(struct nes_state *state, uint32_t offset) {
 
 			}
 
-			ppu->vram_addr += (ppu->reg_ctrl & 0x04) ? 32 : 1;
+			ppu->vram_addr += (ppu->reg_ctrl & PPU_CTRL_VRAM_INCREMENT) ? 32 : 1;
 		} break;
 
 	}