static uint8_t __attribute__((aligned(64))) ppu_bitreverse_lut[256] = {
#   define R2(n)     n,     n + 2*64,     n + 1*64,     n + 3*64
#   define R4(n) R2(n), R2(n + 2*16), R2(n + 1*16), R2(n + 3*16)
#   define R6(n) R4(n), R4(n + 2*4 ), R4(n + 1*4 ), R4(n + 3*4 )
    R6(0), R6(2), R6(1), R6(3)
};
#undef R2
#undef R4
#undef R6

static void ppu_reset(struct nes_state *state) {
	struct ppu_state *restrict ppu = &state->ppu;
	memset(ppu, 0, sizeof(struct ppu_state));
}


__attribute__((always_inline, hot))
static inline 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;

	uint8_t * restrict src = ppu->oam;
	uint8_t * restrict dst = ppu->secondary_oam;
	for(uint8_t i = 0; i < 64; i++) {
		uint8_t y = src[0];
		int32_t row = (int32_t)ppu->scanline - y;

		if(row >= 0 && row < sprite_height) {
			if(n < 8) {
				dst[0] = src[0];
				dst[1] = src[1];
				dst[2] = src[2];
				dst[3] = src[3];
				ppu->sprite_indexes[n] = i;

				dst += 4;
				n++;
			} else {
				ppu->reg_status |= 0x20;
				break;
			}
		}
		src += 4;
	}

	ppu->sprite_count = n;
}

__attribute__((always_inline, hot))
static inline void ppu_fetch_sprite_patterns(struct nes_state *state) {
	struct ppu_state *restrict ppu = &state->ppu;
	uint32_t addr;
	uint32_t bank;
	uint8_t lsb;
	uint8_t msb;

	uint8_t * restrict s = ppu->secondary_oam;
	uint8_t height = (ppu->reg_ctrl & 0x20) ? 16 : 8;

	for(uint8_t i = 0; i < ppu->sprite_count; i++) {
		uint8_t y = s[0], tile = s[1], attr = s[2], x = s[3];
		uint8_t row = ppu->scanline - y;

		row = (attr & 0x80) ? height - 1 - row : row;

		if(height == 16) {
			bank = (tile & 1) << 12;
			tile &= 0xfe;
			if(row >= 8) {
				tile++;
				row -= 8;
			}
		} else {
			bank = (ppu->reg_ctrl & 0x08) << 9;
		}
		addr = bank + tile * 16 + row;

		if(attr & 0x40) {
			lsb = ppu_bitreverse_lut[state->mapper.chr_read(state, addr)];
			msb = ppu_bitreverse_lut[state->mapper.chr_read(state, addr + 8)];
		} else {
			lsb = state->mapper.chr_read(state, addr);
			msb = state->mapper.chr_read(state, addr + 8);
		}

		ppu->sprite_shift_lo[i] = lsb;
		ppu->sprite_shift_hi[i] = msb;

		ppu->sprite_positions[i] = x;
		ppu->sprite_priorities[i] = attr & 0x20;
		s += 4;
	}
}

__attribute__((always_inline, hot))
static inline void ppu_render_pixel(struct nes_state *state) {
	struct ppu_state *restrict ppu = &state->ppu;

	uint32_t x = ppu->dot - 1;
	uint32_t y = ppu->scanline;

	uint16_t bit = 0x8000 >> ppu->fine_x;

	uint8_t bg_pixel = 0;
	uint8_t bg_palette = 0;
	uint8_t sp_pixel = 0;
	uint8_t sp_palette = 0;
	uint8_t sp_prio = 0;
	uint8_t sp_zero = 0;

	uint8_t bg_mask = (ppu->reg_mask & 0x08) ? 0xff : 0x00;
	uint8_t sp_mask = (ppu->reg_mask & 0x10) ? 0xff : 0x00;

	// uint8_t left_mask = (x >= 8) ? 0xff : 0x00;
	// bg_mask &= ((ppu->reg_mask & 0x02) ? 0xff : 0x00) | left_mask;
	// sp_mask &= ((ppu->reg_mask & 0x04) ? 0xff : 0x00) | left_mask;


	// Background
	uint8_t p0 = !!(ppu->bg_shift_pattern_low & bit);
	uint8_t p1 = !!(ppu->bg_shift_pattern_high & bit);
	uint8_t a0 = !!(ppu->bg_shift_attrib_low & bit);
	uint8_t a1 = !!(ppu->bg_shift_attrib_high & bit);

	bg_pixel   = ((p1 << 1) | p0) & bg_mask;
	bg_palette = ((a1 << 1) | a0) & bg_mask;

	// Sprite
	for(uint8_t i = 0; i < ppu->sprite_count; i++) {
		if(ppu->sprite_positions[i]) continue;

		uint8_t lo = ppu->sprite_shift_lo[i];
		uint8_t hi = ppu->sprite_shift_hi[i];
		sp_pixel = (((hi & 0x80) >> 6) | ((lo & 0x80) >> 7)) & sp_mask;

		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);
		break;
	}

	// Final pixel composition
	uint8_t palette_index = 0;
	uint8_t bg_index = (bg_palette << 2) + bg_pixel;
	uint8_t sp_index = (sp_palette << 2) + sp_pixel;
	uint8_t selector = (bg_pixel ? 2 : 0) | (sp_pixel ? 1 : 0);

	switch(selector) {
		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;		// NOTE(peter): Sprite zero hit!
			palette_index = (sp_prio) ? bg_index : 0x10 | sp_index;
		} break;
	}

	state->pixels[y * 256 + x] = ppu->palette[palette_index];		// NOTE(peter): Add color_emphasis bits (expand palette to 8x).
}

__attribute__((hot, flatten))
static void ppu_tick(struct nes_state *state) {
	struct ppu_state *restrict ppu = &state->ppu;


	uint32_t dot = ppu->dot;
	uint32_t scanline = ppu->scanline;
	uint8_t rendering = (ppu->reg_mask & 0x18);

	for(uint32_t ppu_loops = 0; ppu_loops < 3; ++ppu_loops) {

		if(LIKELY(rendering)) {
			switch(scanline) {
				case 0 ... 239: {
					switch(dot) {
						case 256: {

							if((ppu->vram_addr & 0x7000) != 0x7000) {
								ppu->vram_addr += 0x1000;
							} else {
								ppu->vram_addr &= ~0x7000;
								uint32_t y = (ppu->vram_addr & 0x03e0) >> 5;
								if(y == 29) {
									y = 0;
									ppu->vram_addr ^= 0x0800;
								} else if(y == 31) {
									y = 0;
								} else {
									y++;
								}
								ppu->vram_addr = (ppu->vram_addr & ~0x03e0) | (y << 5);
							}
						}
							__attribute__((fallthrough));

						case 1 ... 255:		// fallthrough: this is 1->256
							ppu_render_pixel(state);
							__attribute__((fallthrough));

						case 321 ... 336: {	// fallthrough: the code below has to run  1->256 + 321->336
							// Rendering and tile fetch goes here
							if(ppu->reg_mask & 0x10) {
								for(uint32_t i = 0; i < ppu->sprite_count; i++) {
									if(ppu->sprite_positions[i] > 0) {
										ppu->sprite_positions[i]--;
									} else {
										ppu->sprite_shift_lo[i] <<= 1;
										ppu->sprite_shift_hi[i] <<= 1;
									}
								}
							}

							ppu->bg_shift_pattern_low <<= 1;
							ppu->bg_shift_pattern_high <<= 1;
							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);
									ppu->bg_next_tile_id = state->mapper.ciram_read(state, nt_addr);
								} break;

								case 3: {
									uint32_t attr_addr = 0x23c0 | (ppu->vram_addr & 0x0c00) | ((ppu->vram_addr >> 4) & 0x38) | ((ppu->vram_addr >> 2) & 0x07);
									uint8_t attr = state->mapper.ciram_read(state, attr_addr & 0x0fff);
									uint8_t shift = ((ppu->vram_addr >> 4) & 4) | (ppu->vram_addr & 2);
									ppu->bg_next_tile_attrib = (attr >> shift) & 3;
								} break;

								case 5: {
									uint32_t base = (ppu->reg_ctrl & 0x10) ? 0x1000 : 0x0000;
									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;
									ppu->bg_next_tile_lsb = state->mapper.chr_read(state, addr_lsb);
								} break;

								case 7: {
									uint32_t base = (ppu->reg_ctrl & 0x10) ? 0x1000 : 0x0000;
									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;
									ppu->bg_next_tile_msb =  state->mapper.chr_read(state, addr_msb);
								} break;

								case 0: {
									ppu->bg_shift_pattern_low = (ppu->bg_shift_pattern_low & 0xff00) | ppu->bg_next_tile_lsb;
									ppu->bg_shift_pattern_high = (ppu->bg_shift_pattern_high & 0xff00) | ppu->bg_next_tile_msb;

									uint8_t a = ppu->bg_next_tile_attrib;
									ppu->bg_shift_attrib_low = (ppu->bg_shift_attrib_low & 0xff00) | ((a & 1) ? 0xff : 0x00);
									ppu->bg_shift_attrib_high = (ppu->bg_shift_attrib_high & 0xff00) | ((a & 2) ? 0xff : 0x00);

									if((ppu->vram_addr & 0x001f) == 31) {
										ppu->vram_addr &= ~0x001f;
										ppu->vram_addr ^= 0x0400;
									} else {
										ppu->vram_addr++;
									}
								} break;
							}
						} break;

						case 257: {			// Sprite evaluation and horizontal vram transfer
							ppu->vram_addr = (ppu->vram_addr & ~0x041f) | (ppu->temp_addr & 0x041f);
							ppu_evaluate_sprites(state);
						} break;

						case 340: {			// sprite fetch pattern
							ppu_fetch_sprite_patterns(state);
						} break;
					}

				} break;

				case 261: {
					switch(dot) {
						case 256: {
							if((ppu->vram_addr & 0x7000) != 0x7000) {
								ppu->vram_addr += 0x1000;
							} else {
								ppu->vram_addr &= ~0x7000;
								uint32_t y = (ppu->vram_addr & 0x03e0) >> 5;
								if(y == 29) {
									y = 0;
									ppu->vram_addr ^= 0x0800;
								} else if(y == 31) {
									y = 0;
								} else {
									y++;
								}
								ppu->vram_addr = (ppu->vram_addr & ~0x03e0) | (y << 5);
							}
						}
							__attribute__((fallthrough));

						case 1 ... 255:		__attribute__((fallthrough));

						case 321 ... 336: {		// Rendering and tile fetch
							if(ppu->reg_mask & 0x10) {
								for(uint32_t i = 0; i < ppu->sprite_count; i++) {
									if(ppu->sprite_positions[i] > 0) {
										ppu->sprite_positions[i]--;
									} else {
										ppu->sprite_shift_lo[i] <<= 1;
										ppu->sprite_shift_hi[i] <<= 1;
									}
								}
							}

							ppu->bg_shift_pattern_low <<= 1;
							ppu->bg_shift_pattern_high <<= 1;
							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);
									ppu->bg_next_tile_id = state->mapper.ciram_read(state, nt_addr);
								} break;

								case 3: {
									uint32_t attr_addr = 0x23c0 | (ppu->vram_addr & 0x0c00) | ((ppu->vram_addr >> 4) & 0x38) | ((ppu->vram_addr >> 2) & 0x07);
									uint8_t attr = state->mapper.ciram_read(state, attr_addr & 0x0fff);
									uint8_t shift = ((ppu->vram_addr >> 4) & 4) | (ppu->vram_addr & 2);
									ppu->bg_next_tile_attrib = (attr >> shift) & 3;
								} break;

								case 5: {
									uint32_t base = (ppu->reg_ctrl & 0x10) ? 0x1000 : 0x0000;
									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;
									ppu->bg_next_tile_lsb = state->mapper.chr_read(state, addr_lsb);
								} break;

								case 7: {
									uint32_t base = (ppu->reg_ctrl & 0x10) ? 0x1000 : 0x0000;
									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;
									ppu->bg_next_tile_msb =  state->mapper.chr_read(state, addr_msb);
								} break;

								case 0: {
									ppu->bg_shift_pattern_low	= (ppu->bg_shift_pattern_low & 0xff00) | ppu->bg_next_tile_lsb;
									ppu->bg_shift_pattern_high	= (ppu->bg_shift_pattern_high & 0xff00) | ppu->bg_next_tile_msb;

									uint8_t a = ppu->bg_next_tile_attrib;
									ppu->bg_shift_attrib_low	= (ppu->bg_shift_attrib_low & 0xff00) | ((a & 1) ? 0xff : 0x00);
									ppu->bg_shift_attrib_high	= (ppu->bg_shift_attrib_high & 0xff00) | ((a & 2) ? 0xff : 0x00);

									if((ppu->vram_addr & 0x001f) == 31) {
										ppu->vram_addr &= ~0x001f;
										ppu->vram_addr ^= 0x0400;
									} else {
										ppu->vram_addr++;
									}

								} break;
							}
						} break;

						case 257: {					// Sprite evaluation and horizontal vram transfer
							ppu->vram_addr = (ppu->vram_addr & ~0x041f) | (ppu->temp_addr & 0x041f);
							ppu_evaluate_sprites(state);

						} break;

						case 280 ... 304: {		// Vertical vram transfer
							ppu->vram_addr = (ppu->vram_addr & ~0x7be0) | (ppu->temp_addr & 0x7be0);
						} break;

						case 340: {					// Sprite pattern fetch
							ppu_fetch_sprite_patterns(state);
						} break;

					}

				} break;
			}
		}

		if(UNLIKELY(scanline == 241) && dot == 1) {
			ppu->reg_status |= 0x80;
			if(ppu->reg_ctrl & 0x80) {
				state->nmi_pending = 1;
			}
		}

		if(UNLIKELY(scanline == 261) && dot == 1) {
			ppu->reg_status &= ~0x80;
			ppu->reg_status &= ~0x40;
		}

		dot++;
		if(dot > 340) {
			dot = 0;
			scanline++;
			if(scanline > 261) {
				scanline = 0;
				ppu->frame_ready = 1;
				ppu->even_frame = !ppu->even_frame;
			}
		}

		ppu->dot = dot;
		ppu->scanline = scanline;
	}
}