optimized, but bug with rasterdemos

10 files changed, 174 insertions, 283 deletions

diff --git a/Bench.sh b/Bench.sh
index 14208cd..4a26ab5 100755
--- a/Bench.sh
+++ b/Bench.sh
@@ -3,6 +3,9 @@
 ./build.sh profile
 ./mknes
 ./build.sh profile_release
-taskset -c 1 chrt -f 99 -- perf stat -C 1  ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -C 1 ./mknes
-taskset -c 1 chrt -f 99 -- perf stat -C 1 ./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 -- ./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
diff --git a/build.sh b/build.sh
index b854c3e..b8e4fa9 100755
--- a/build.sh
+++ b/build.sh
@@ -52,17 +52,18 @@ case "$BUILD_TYPE" in
 #  -pg # for gprof
 		;;
 	"release")
-		CFLAGS+="-s -Wl,--strip-all -O2 "
+		# CFLAGS+="-s -Wl,--strip-all -O2 "
+		CFLAGS+="-g -O2 "
 		;;
 	"profile")
 		CFLAGS+="-O2 -fprofile-generate -ftest-coverage "
 		;;
 	"profile_release")
 		# CFLAGS+="-s -Wl,--strip-all -O2 -fprofile-use "
-		CFLAGS+="-O2 -fprofile-use "
+		CFLAGS+="-g -O2 -fprofile-use "
 		;;
 	"debug")
-		CFLAGS+="-g -O0 "
+		CFLAGS+="-g -O0 -DTIMER_DEBUG "
 		LDFLAGS+="-fno-pie -no-pie "
 		;;
 	"coverage")
@@ -90,8 +91,8 @@ set -e
 ) &
 
 # Build Windows version
-# (
-# 	$WIN_CC $CFLAGS ${PROJECT_NAME}.c -o ${PROJECT_NAME}.exe $INCLUDE_PATHS $WINDOWS_INCLUDE $LDFLAGS $WINDOWS_LIBS
-# ) &
+#(
+#	$WIN_CC $CFLAGS ${PROJECT_NAME}.c -o ${PROJECT_NAME}.exe $INCLUDE_PATHS $WINDOWS_INCLUDE $LDFLAGS $WINDOWS_LIBS
+#) &
 
 wait
diff --git a/linux_timer.c b/linux_timer.c
index dce0a5b..9a9d296 100644
--- a/linux_timer.c
+++ b/linux_timer.c
@@ -81,7 +81,6 @@ static void *timer_thread_func(void *arg) {
 
 #ifdef TIMER_DEBUG
 		if(t->last_wait_start.tv_sec) {
-			int64_t total_frame_time_ns = timespec_diff_ns(&now, &t->last_wait_start);
 			int64_t overshoot_ns = timespec_diff_ns(&now, &t->next_deadline);
 			if(overshoot_ns < 0) overshoot_ns = 0;
 
diff --git a/mknes.c b/mknes.c
index 37225ff..a6d224d 100644
--- a/mknes.c
+++ b/mknes.c
@@ -44,7 +44,7 @@ static void audio_callback(int16_t *data, size_t frames) { }
 
 #define FRAME_INTERVAL_NS (1000000000ULL / 60.0988)
 
-#define DEBUG_PRINT printf
+#define DEBUG_PRINT(fmt, ...) printf(fmt, ##__VA_ARGS__)
 #ifdef _WIN32
 #include "win32_timer.c"
 #else
@@ -158,7 +158,6 @@ static uint32_t frames;		// debug information
 
 // #include "smb_tas.h"	// REMOVE ME
 
-
 // NES core
 #include "mknes_mapper.h"
 #include "mknes.h"
@@ -172,7 +171,7 @@ static uint32_t frames;		// debug information
 
 // struct nes_state nstate;
 
-static void framebuffer_callback(int32_t width, int32_t height, float aspect_ratio) {
+static void framebuffer_callback(struct mkfw_state *mkfw_window, int32_t width, int32_t height, float aspect_ratio) {
 	// state.screen_width = width;
 	// state.screen_height = height;
 	// state.viewport.x = 0;
@@ -254,7 +253,7 @@ 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");
@@ -291,7 +290,7 @@ int main(int argc, char **argv) {
 	mapper_setup(nstate);
 	cpu_reset(nstate);
 
-#if 1
+#if 0
 	for(uint32_t i = 0; i < 0x5000; ++i) {
 		while(!nstate->ppu.frame_ready) {
 			// PROFILE_NAMED("nes emulator");
@@ -300,6 +299,8 @@ int main(int argc, char **argv) {
 		nstate->ppu.frame_ready = 0;
 		frames++;
 	}
+
+
 	// for(size_t i = 0; i < 256; ++i) {
 	// 	printf("instr %2.2x: %lld\n", i, instr_count[i]);
 	// }
@@ -307,18 +308,18 @@ int main(int argc, char **argv) {
 #else
 
 //   WINDOW SETUP
-	mkfw_init(WINDOW_WIDTH, WINDOW_HEIGHT);
-	mkfw_set_window_title("mknes");
-	mkfw_set_framebuffer_size_callback(framebuffer_callback);
-	mkfw_set_swapinterval(0);
+	struct mkfw_state *window = mkfw_init(WINDOW_WIDTH, WINDOW_HEIGHT);
+	mkfw_set_window_title(window, "mknes");
+	mkfw_set_framebuffer_size_callback(window, framebuffer_callback);
+	mkfw_set_swapinterval(window, 0);
 
 	gl_loader();
 	opengl_setup();
 	// setup_render_targets();
 	change_resolution(BUFFER_WIDTH, BUFFER_HEIGHT);
-	mkfw_show_window();
+	mkfw_show_window(window);
 
-	mkfw_set_window_min_size_and_aspect(WINDOW_WIDTH, WINDOW_HEIGHT, 4.f, 3.f);
+	mkfw_set_window_min_size_and_aspect(window, WINDOW_WIDTH, WINDOW_HEIGHT, 4.f, 3.f);
 
 
 	timer_init();
@@ -326,22 +327,22 @@ int main(int argc, char **argv) {
 
 	uint8_t running = true;
 
-	while(running && !mkfw_should_close()) {
-		mkfw_pump_messages();
+	while(running && !mkfw_should_close(window)) {
+		mkfw_pump_messages(window);
 
 		// // Check for ESC key
-		if(mkfw_is_key_pressed(MKS_KEY_ESCAPE)) running = false;
+		if(mkfw_is_key_pressed(window, MKS_KEY_ESCAPE)) running = false;
 
 		// Joypad input
 		uint8_t input = 0;
-		if(keyboard_state[MKS_KEY_X]) { input |= (1 << 0); }
-		if(keyboard_state[MKS_KEY_Z]) { input |= (1 << 1); }
-		if(keyboard_state[MKS_KEY_SPACE]) { input |= (1 << 2); }
-		if(keyboard_state[MKS_KEY_RETURN]) { input |= (1 << 3); }
-		if(keyboard_state[MKS_KEY_UP]) { input |= (1 << 4); }
-		if(keyboard_state[MKS_KEY_DOWN]) { input |= (1 << 5); }
-		if(keyboard_state[MKS_KEY_LEFT]) { input |= (1 << 6); }
-		if(keyboard_state[MKS_KEY_RIGHT]) { input |= (1 << 7); }
+		if(window->keyboard_state[MKS_KEY_X]) { input |= (1 << 0); }
+		if(window->keyboard_state[MKS_KEY_Z]) { input |= (1 << 1); }
+		if(window->keyboard_state[MKS_KEY_SPACE]) { input |= (1 << 2); }
+		if(window->keyboard_state[MKS_KEY_RETURN]) { input |= (1 << 3); }
+		if(window->keyboard_state[MKS_KEY_UP]) { input |= (1 << 4); }
+		if(window->keyboard_state[MKS_KEY_DOWN]) { input |= (1 << 5); }
+		if(window->keyboard_state[MKS_KEY_LEFT]) { input |= (1 << 6); }
+		if(window->keyboard_state[MKS_KEY_RIGHT]) { input |= (1 << 7); }
 		nstate->ppu.input[0] = input;
 
 		// Run NES emulation for one frame
@@ -373,13 +374,13 @@ int main(int argc, char **argv) {
 			dst += BUFFER_WIDTH;
 		}
 
-		mkfw_update_keyboard_state();
-		mkfw_update_modifier_state();
-		mkfw_update_mouse_state();
+		mkfw_update_keyboard_state(window);
+		mkfw_update_modifier_state(window);
+		mkfw_update_mouse_state(window);
 
 		// Render and swap buffers
 		render_frame();
-		mkfw_swap_buffers();
+		mkfw_swap_buffers(window);
 		timer_wait(timer);
 	}
 
@@ -388,7 +389,7 @@ int main(int argc, char **argv) {
 	timer_destroy(timer);
 	// free_nes_state(&nstate);
 	timer_shutdown();
-	mkfw_cleanup();
+	mkfw_cleanup(window);
 #endif
 
 	return 0;
diff --git a/mknes.h b/mknes.h
index c41d7af..d21a9ad 100644
--- a/mknes.h
+++ b/mknes.h
@@ -60,7 +60,7 @@ struct ppu_state {
 		uint8_t position;
 		uint8_t priority;
 		uint8_t palette;
-	} __attribute__((packed, aligned(64))) sprites[8];
+	} __attribute__((packed)) sprites[8] __attribute__((aligned(64)));
 
 	uint8_t input[2];						// 40 - Controller 1 & 2
 	uint8_t input_latch[2];				// 42 - Latched inputs after strobe
@@ -151,83 +151,3 @@ __attribute__((aligned(4096))) static uint32_t nes_palette[65] = {
 	0xdee086ff, 0xc6ec87ff, 0xb2f29dff, 0xa7f0c3ff, 0xa8e7f0ff, 0xacacacff, 0x000000ff, 0x000000ff,
 	0xffffffff	// one extra for debug-coloring...
 };
-
-struct remake_state {
-	struct { int32_t x, y, w, h; } viewport;
-	int32_t mouse_dx;
-	int32_t mouse_dy;
-
-	float contrast;
-	float saturation;
-	float brightness;
-	float tone_data[4];
-
-	// OpenGL Objects
-	GLuint shader_program;
-	GLuint persistence_program;
-	// GLuint upscale_program;
-	GLuint upscale_warp_program;
-	GLuint bloom_extract_program;
-	GLuint bloom_blur_program;
-	GLuint bloom_warp_program;
-	GLuint bloom_composite_program;
-
-	GLuint texture;
-	GLuint persistence_texture;
-	GLuint persistence_output_texture;
-	GLuint crt_output_texture;
-	GLuint bloom_texture;
-	GLuint bloom_temp_texture;
-	GLuint bloom_warped_texture;
-	GLuint upscaled_source_texture;
-
-	GLuint persistence_fbo;
-	GLuint upscaled_source_fbo;
-	GLuint crt_fbo;
-	GLuint bloom_fbo;
-	GLuint bloom_temp_fbo;
-	GLuint bloom_warp_fbo;
-
-	GLuint vao;
-	GLuint vbo;
-	GLuint ebo;
-
-	// CRT Shader Uniforms
-	GLuint uniform_resolution;
-	GLuint uniform_src_image_size;
-	GLuint uniform_brightness;
-	GLuint uniform_tone;
-	GLuint uniform_crt_emulation;
-	GLuint uniform_apply_mask;
-	GLuint uniform_sampler_location;
-
-	// Bloom Shader Uniforms
-	GLuint bloom_uniform_threshold;
-	GLuint bloom_uniform_sampler;
-	GLuint blur_uniform_horizontal;
-	GLuint blur_uniform_sampler;
-	GLuint composite_uniform_bloom_strength;
-	GLuint composite_uniform_crt_sampler;
-	GLuint composite_uniform_bloom_sampler;
-
-	// Bloom settings
-	float bloom_threshold;
-	float bloom_strength;
-	uint32_t bloom_width;
-	uint32_t bloom_height;
-
-	// Phosphor persistence
-	float persistence_decay;
-
-	// Rendering & Dynamic Resolution
-	uint32_t render_width;	// The actual remake resolution (e.g., 360)
-	uint32_t render_height;	// The actual remake resolution (e.g., 270)
-	uint32_t frame_number;
-	uint8_t running;
-	uint8_t toggle_crt_emulation;
-	uint8_t toggle_bloom;
-	uint8_t fullscreen;
-	uint8_t viewport_changed;  // Flag to signal render thread to recreate FBOs
-};
-
-// static struct remake_state state;
diff --git a/mknes_cpu.c b/mknes_cpu.c
index 3c0a3f2..0eccf03 100644
--- a/mknes_cpu.c
+++ b/mknes_cpu.c
@@ -6,6 +6,7 @@ static inline uint8_t pack_flags(struct cpu_state *cpu) {
 	return (cpu->n << 7) | (cpu->v << 6) | (1 << 5) | (cpu->d << 3) | (cpu->i << 2) | (cpu->z << 1) | cpu->c;
 }
 
+__attribute__((always_inline))
 static inline void unpack_flags(struct cpu_state *cpu, uint8_t value) {
 	cpu->n = (value >> 7) & 1;
 	cpu->v = (value >> 6) & 1;
@@ -15,6 +16,8 @@ static inline void unpack_flags(struct cpu_state *cpu, uint8_t value) {
 	cpu->c = value & 1;
 }
 
+
+__attribute__((always_inline))
 static inline void update_zn(struct cpu_state *cpu, uint8_t result) {
 	cpu->z = (result == 0);
 	cpu->n = (result & 0x80) != 0;
@@ -63,6 +66,7 @@ static inline void do_irq(struct nes_state *state) {
 	cpu->i = 1;
 }
 
+__attribute__((always_inline))
 static inline void check_interrupts(struct nes_state *state) {
 	struct cpu_state * restrict cpu = &state->cpu;
 
@@ -86,7 +90,17 @@ static inline void cpu_reset(struct nes_state *state) {
 
 static inline void cpu_tick(struct nes_state *state) {
 	struct cpu_state * restrict cpu = &state->cpu;
-	check_interrupts(state);
+	// check_interrupts(state);
+	if(state->cpu.nmi_pending) {
+		state->cpu.nmi_pending = 0;
+		do_nmi(state);
+	}
+	if(state->cpu.irq_pending && cpu->i == 0) {
+		state->cpu.irq_pending = 0;
+		do_irq(state);
+	}
+
+
 
 	// printf("%4.4x: ", cpu->pc);
 	uint8_t opcode = memory_read(state, cpu->pc++);
diff --git a/mknes_cpu_opcodes.c b/mknes_cpu_opcodes.c
index 0254b5c..b2fafd5 100644
--- a/mknes_cpu_opcodes.c
+++ b/mknes_cpu_opcodes.c
@@ -1,6 +1,7 @@
 
 
 // ADC
+__attribute__((hot, always_inline))
 static inline void adc(struct cpu_state *cpu, uint8_t value) {
 #ifdef ENABLE_DECIMAL_MODE
 	if(cpu->d) {
diff --git a/mknes_memory.c b/mknes_memory.c
index a27fa04..7b74424 100644
--- a/mknes_memory.c
+++ b/mknes_memory.c
@@ -1,8 +1,9 @@
+
 __attribute__((hot))
 static inline uint8_t memory_read(struct nes_state *state, uint32_t offset) {
 	state->cpu.cycles++;
-	ppu_tick(state);
 	apu_tick(state);
+	ppu_tick(state);
 
 	if(offset <= 0x1fff) {
 		return state->ram[offset & 0x07ff];
@@ -20,11 +21,11 @@ static inline uint8_t memory_read(struct nes_state *state, uint32_t offset) {
 		return apu_read4015(state);
 	}
 
-	if(offset == 0x4016 || offset == 0x4017) {
+	if(offset >= 0x4016 && offset <= 0x4017) {
 		uint32_t index = offset & 1;
 		uint8_t value = (state->ppu.input_latch[index] >> state->ppu.input_bit[index]) & 1;
 		state->ppu.input_bit[index]++;
-		return value | 0x40; // Bit 6 open bus high, bit 7 low
+		return value | 0x40;	// Bit 6 open bus high, bit 7 low
 	}
 
 	if(offset >= 0x6000 && offset <= 0x7fff) {
@@ -34,65 +35,76 @@ static inline uint8_t memory_read(struct nes_state *state, uint32_t offset) {
 	return 0;
 }
 
-
-__attribute__((hot))
+__attribute__((always_inline, hot))
 static inline uint8_t memory_read_dummy(struct nes_state *state, uint32_t offset) {
 	state->cpu.cycles++;
-	ppu_tick(state);
 	apu_tick(state);
+	ppu_tick(state);
+
+	uint8_t result = 0;
 
 	if(offset >= 0x2000 && offset < 0x4000) {
-		return ppu_read(state, offset);
+		result = ppu_read(state, offset);
 	}
-	return 0;
+	return result;
 }
 
-
 __attribute__((hot, optimize("no-jump-tables")))
 static inline void memory_write(struct nes_state *state, uint32_t offset, uint8_t value) {
 	state->cpu.cycles++;
-	ppu_tick(state);
 	apu_tick(state);
+	ppu_tick(state);
 
-	if(offset <= 0x1fff) {
-		state->ram[offset & 0x07ff] = value;
-	}
+	switch(offset) {
+		case 0x0000 ... 0x1fff: {
+			state->ram[offset & 0x07ff] = value;
+		} break;
 
-	if(offset >= 0x2000 && offset <= 0x3fff) {
-		ppu_write(state, offset, value);
-	}
 
-	if(offset >= 0x4000 && offset <= 0x4017) {
-		if(offset == 0x4014) {
-			ppu_dma_4014(state, value);
-		}
-
-		if(offset == 0x4016) {
-			uint8_t s = value & 1;
-			uint8_t prev = state->ppu.input_strobe;
-			state->ppu.input_strobe = s;
-
-			if(prev == 1 && s == 0) {
-					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;
+		case 0x2000 ... 0x3fff: {
+			ppu_write(state, offset, value);
+		} break;
+
+		case 0x4000 ... 0x4017: {
+			switch(offset) {
+				case 0x4014: {
+					ppu_dma_4014(state, value);
+				} break;
+
+				case 0x4016: {
+					// joypad strobe
+					uint8_t s = value & 1;
+
+					// if(s) {
+						uint8_t prev = state->ppu.input_strobe;
+						state->ppu.input_strobe = s;
+
+						if(prev == 1 && (s) == 0) {
+							// state->ppu.input[0] = tas_input[tas_frame_count];
+
+							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;
+
+				default: {
+					apu_write(state, offset, value);
+				} break;
 			}
-		}
+		} break;
 
-		if(offset != 0x4014 && offset != 0x4016) {
-			apu_write(state, offset, value);
-		}
-	}
+		case 0x6000 ... 0x7fff: {
+			state->mapper_function.prg_ram_write(state, offset, value);
+		} break;
 
-	if(offset >= 0x6000 && offset <= 0x7fff) {
-		state->mapper_function.prg_ram_write(state, offset, value);
-	}
+		case 0x8000 ... 0xffff: {
+			state->mapper_function.prg_rom_write(state, offset, value);
+		} break;
 
-	if(offset >= 0x8000) {
-		state->mapper_function.prg_rom_write(state, offset, value);
 	}
-
 }
 
 __attribute__((hot, flatten))
diff --git a/mknes_ppu.c b/mknes_ppu.c
index 64d6821..92b22fc 100644
--- a/mknes_ppu.c
+++ b/mknes_ppu.c
@@ -25,7 +25,7 @@ static void ppu_reset(struct nes_state *state) {
 }
 
 __attribute__((hot, flatten))
-static inline void ppu_evaluate_sprites(struct nes_state *state) {
+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 n = 0;
@@ -36,7 +36,7 @@ static inline void ppu_evaluate_sprites(struct nes_state *state) {
 
 	for(uint8_t i = 0; i < 64; i++, src += 4) {
 		uint8_t y = src[0];
-		int32_t row = (int32_t)ppu->scanline - y;
+		int32_t row = (int32_t)scanline - y;
 
 		if(row >= 0 && row < sprite_height) {
 			if(n < 8) {
@@ -63,7 +63,7 @@ static inline void ppu_evaluate_sprites(struct nes_state *state) {
 }
 
 __attribute__((hot))
-static inline void ppu_fetch_sprite_patterns(struct nes_state *state, uint32_t scanline) {
+static inline void ppu_fetch_sprite_patterns(struct nes_state * restrict state, uint32_t scanline) {
 	struct ppu_state *restrict ppu = &state->ppu;
 
 	uint8_t * restrict sec_oam = ppu->secondary_oam;
@@ -110,8 +110,8 @@ static inline void ppu_fetch_sprite_patterns(struct nes_state *state, uint32_t s
 }
 
 
-__attribute__((always_inline, hot))
-static inline void ppu_render_pixel(struct nes_state *state, uint32_t x, uint32_t y) {
+__attribute__((always_inline, hot, optimize("no-jump-tables")))
+static inline void ppu_render_pixel(struct nes_state * restrict state, uint32_t x, uint32_t y) {
 	struct ppu_state *restrict ppu = &state->ppu;
 
 	uint16_t bit = 0x8000 >> ppu->fine_x;
@@ -138,100 +138,42 @@ static inline void ppu_render_pixel(struct nes_state *state, uint32_t x, uint32_
 	uint8_t bg_pixel   = ((p1 << 1) | p0) & bg_mask;
 	uint8_t bg_palette = ((a1 << 1) | a0) & bg_mask;
 
-	// Sprite
-
-	if(sp_mask && ppu->sprite_count) {
-		switch(ppu->sprite_count) {
-			case 8: {
-				if(!ppu->sprites[7].position) {
-					sp_pixel = (((ppu->sprites[7].shift_hi & 0x80) >> 6) | ((ppu->sprites[7].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[7].priority;
-						sp_palette = ppu->sprites[7].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 7: {
-				if(!ppu->sprites[6].position) {
-					sp_pixel = (((ppu->sprites[6].shift_hi & 0x80) >> 6) | ((ppu->sprites[6].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[6].priority;
-						sp_palette = ppu->sprites[6].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 6: {
-				if(!ppu->sprites[5].position) {
-					sp_pixel = (((ppu->sprites[5].shift_hi & 0x80) >> 6) | ((ppu->sprites[5].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[5].priority;
-						sp_palette = ppu->sprites[5].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 5: {
-				if(!ppu->sprites[4].position) {
-					sp_pixel = (((ppu->sprites[4].shift_hi & 0x80) >> 6) | ((ppu->sprites[4].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[4].priority;
-						sp_palette = ppu->sprites[4].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 4: {
-				if(!ppu->sprites[3].position) {
-					sp_pixel = (((ppu->sprites[3].shift_hi & 0x80) >> 6) | ((ppu->sprites[3].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[3].priority;
-						sp_palette = ppu->sprites[3].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 3: {
-				if(!ppu->sprites[2].position) {
-					sp_pixel = (((ppu->sprites[2].shift_hi & 0x80) >> 6) | ((ppu->sprites[2].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[2].priority;
-						sp_palette = ppu->sprites[2].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 2: {
-				if(!ppu->sprites[1].position) {
-					sp_pixel = (((ppu->sprites[1].shift_hi & 0x80) >> 6) | ((ppu->sprites[1].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[1].priority;
-						sp_palette = ppu->sprites[1].palette;
-						break;
-					}
-				}
-			} __attribute__((fallthrough));
-
-			case 1: {
-				if(!ppu->sprites[0].position) {
-					sp_pixel = (((ppu->sprites[0].shift_hi & 0x80) >> 6) | ((ppu->sprites[0].shift_lo & 0x80) >> 7));
-					if(sp_pixel) {
-						sp_prio = ppu->sprites[0].priority;
-						sp_palette = ppu->sprites[0].palette;
-						sp_zero = ppu->sprite_zero_in_range;  // Only slot 0 can be sprite 0
-						break;
-					}
-				}
-			}
-		}
+	// Sprites
+#define SPRITE_STEP(N) do {														\
+	if(!ppu->sprites[(N)].position) {											\
+		sp_pixel = (((ppu->sprites[(N)].shift_hi & 0x80) >> 6) | ((ppu->sprites[(N)].shift_lo & 0x80) >> 7));	\
+		if(sp_pixel) {																	\
+			sp_prio    = ppu->sprites[(N)].priority;							\
+			sp_palette = ppu->sprites[(N)].palette;							\
+			if((N) == 0) {																\
+				sp_zero = ppu->sprite_zero_in_range;							\
+			}																				\
+			goto sprite_done;															\
+		}																					\
+	}																						\
+} while (0)
+
+	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 == 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;
+
+sprite_8:	SPRITE_STEP(7);
+sprite_7:	SPRITE_STEP(6);
+sprite_6:	SPRITE_STEP(5);
+sprite_5:	SPRITE_STEP(4);
+sprite_4:	SPRITE_STEP(3);
+sprite_3:	SPRITE_STEP(2);
+sprite_2:	SPRITE_STEP(1);
+sprite_1:	SPRITE_STEP(0);
 	}
+sprite_done:;
+
 
 	// Final pixel composition
 	uint8_t bg_index = (bg_palette << 2) + bg_pixel;
@@ -266,7 +208,7 @@ static void ppu_tick(struct nes_state *state) {
 			if(scanline <= 239) {
 
 				if(dot == 65) {
-					ppu_evaluate_sprites(state);  // Evaluate sprites early, sets overflow immediately
+					ppu_evaluate_sprites(state, scanline);  // Evaluate sprites early, sets overflow immediately
 				}
 
 				if(dot >= 1 && dot <= 256) {
@@ -371,7 +313,7 @@ stupid:			if(ppu->reg_mask & 0x10) {
 
 			if(scanline == 261) {
 				if(dot == 65) {
-					ppu_evaluate_sprites(state);  // Evaluate sprites early
+					ppu_evaluate_sprites(state, scanline);  // Evaluate sprites early
 				}
 
 				if(dot >= 1 && dot <= 256) {
@@ -424,6 +366,7 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 					ppu->bg_shift_attrib_low <<= 1;
 					ppu->bg_shift_attrib_high <<= 1;
 
+
 					switch(dot % 8) {
 						case 1: {
 							uint32_t nt_addr = 0x2000 | (ppu->vram_addr & 0x0fff);
@@ -512,9 +455,7 @@ stupid2:			if(ppu->reg_mask & 0x10) {
 		if(state->mapper_function.tick) {
 			state->mapper_function.tick(state);
 		}
-
-		ppu->dot = dot;
-		ppu->scanline = scanline;
-
 	}
+	ppu->dot = dot;
+	ppu->scanline = scanline;
 }
diff --git a/win32_timer.c b/win32_timer.c
index c89000f..250861b 100644
--- a/win32_timer.c
+++ b/win32_timer.c
@@ -18,8 +18,6 @@ struct timer_handle {
 
 #ifdef TIMER_DEBUG
 	uint64_t last_wait_start_ns;
-	uint32_t overshoot_log[1000000];
-	uint32_t overshoot_index;
 #endif
 };
 
@@ -75,12 +73,19 @@ static DWORD WINAPI timer_thread_func(LPVOID arg) {
 	set_realtime_priority(&t->mmcss_handle);
 
 	while(t->running) {
+#ifdef TIMER_DEBUG
+		int64_t remaining_after_sleep_ns = -1;
+#endif
 		uint64_t now = qpc_now_ns(t->qpc_frequency);
 
 		if(now < t->next_deadline) {
 			uint64_t diff = t->next_deadline - now;
 			if(diff > SPIN_THRESHOLD_NS) {
 				timer_sleep(diff - SPIN_THRESHOLD_NS);
+#ifdef TIMER_DEBUG
+				now = qpc_now_ns(t->qpc_frequency);
+				remaining_after_sleep_ns = (int64_t)(t->next_deadline - now);
+#endif
 			}
 			while(qpc_now_ns(t->qpc_frequency) < t->next_deadline) {
 				_mm_pause();
@@ -92,9 +97,14 @@ static DWORD WINAPI timer_thread_func(LPVOID arg) {
 
 #ifdef TIMER_DEBUG
 		if(t->last_wait_start_ns > 0) {
-			uint64_t overshoot_ns = (now > t->next_deadline) ? (now - t->next_deadline) : 0;
-			t->overshoot_log[t->overshoot_index % 1000000] = (uint32_t)overshoot_ns;
-			t->overshoot_index++;
+			int64_t overshoot_ns = (int64_t)(now - t->next_deadline);
+			if(overshoot_ns < 0) overshoot_ns = 0;
+
+			if(remaining_after_sleep_ns >= 0) {
+				DEBUG_PRINT("[DEBUG] Woke up with %lld ns left. Overshoot: %5lld ns\n", remaining_after_sleep_ns, overshoot_ns);
+			} else {
+				DEBUG_PRINT("[DEBUG] No sleep. Overshoot: %lld ns\n", overshoot_ns);
+			}
 		}
 		t->last_wait_start_ns = now;
 #endif
@@ -131,7 +141,6 @@ static struct timer_handle *timer_new(uint64_t interval_ns) {
 
 #ifdef TIMER_DEBUG
 	t->last_wait_start_ns = 0;
-	t->overshoot_index = 0;
 #endif
 
 	t->event = CreateEvent(0, FALSE, FALSE, 0);
@@ -147,6 +156,8 @@ static uint32_t timer_wait(struct timer_handle *t) {
 
 static void timer_destroy(struct timer_handle *t) {
 	t->running = 0;
+
+	SetEvent(t->event);
 	WaitForSingleObject(t->timer_thread, INFINITE);
 	CloseHandle(t->timer_thread);
 	CloseHandle(t->event);
@@ -155,17 +166,5 @@ static void timer_destroy(struct timer_handle *t) {
 		AvRevertMmThreadCharacteristics(t->mmcss_handle);
 	}
 
-#ifdef TIMER_DEBUG
-	uint32_t threshold = 10000;  // 10µs
-	uint32_t overshoot_count = 0;
-	for(uint32_t i = 0; i < t->overshoot_index && i < 1000000; i++) {
-		if(t->overshoot_log[i] >= threshold) {
-			DEBUG_PRINT("Frame %u: overshoot %u ns\n", i, t->overshoot_log[i]);
-			overshoot_count++;
-		}
-	}
-	DEBUG_PRINT("Total frames: %u, Overshoots >= %u ns: %u\n", t->overshoot_index, threshold, overshoot_count);
-#endif
-
 	free(t);
 }