workingHEADmaster

6 files changed, 82 insertions, 59 deletions

diff --git a/build.sh b/build.sh
index d2cb7b2..057cd6a 100755
--- a/build.sh
+++ b/build.sh
@@ -3,11 +3,13 @@
 # Use project-local GCC if available, otherwise system GCC
 TOOLCHAIN_GCC="./toolchain/gcc-15.2.0/bin/gcc"
 if [ -f "${TOOLCHAIN_GCC}" ]; then
-    CC="${TOOLCHAIN_GCC}"
+	CC="${TOOLCHAIN_GCC}"
 else
-    CC=gcc
+	CC=gcc
 fi
 
+CC=gcc
+
 # Set the project name here
 PROJECT_NAME="mknes"  # Change this for each new project
 
@@ -38,7 +40,7 @@ INCLUDE_PATHS="-Ibase -I../mkfw "
 
 # Linux-specific includes and libraries
 LINUX_INCLUDE=" "
-LINUX_LIBS="-lXi -lX11 -lGL -lm -ldl -pthread -larchive -lSDL2"
+LINUX_LIBS="-lXi -lX11 -lGL -lm -ldl -pthread -larchive "
 
 # Windows-specific includes and libraries
 WINDOWS_INCLUDE=""
diff --git a/linux_timer.c b/linux_timer.c
index 9a9d296..6a86fdd 100644
--- a/linux_timer.c
+++ b/linux_timer.c
@@ -101,6 +101,10 @@ static void *timer_thread_func(void *arg) {
 
 static struct timer_handle *timer_new(uint64_t interval_ns) {
 	struct timer_handle *t = calloc(1, sizeof(struct timer_handle));
+	if(!t) {
+		printf("ERROR: Could not allocate a timer!\n");
+		exit(1);
+	}
 
 	t->interval_ns = interval_ns;
 	clock_gettime(CLOCK_MONOTONIC_RAW, &t->next_deadline);
diff --git a/mknes.c b/mknes.c
index 8169fae..044a612 100644
--- a/mknes.c
+++ b/mknes.c
@@ -273,7 +273,17 @@ int main(int argc, char **argv) {
 	init_opcode_ud_lut();
 	// protect_opcode_lut();
 
-	struct nes_state *nstate = aligned_alloc(4096, (sizeof(struct nes_state) + 4095) & ~4095);
+	// Add this line to /etc/sysctl.conf or create a new file /etc/sysctl.d/99-hugepages.conf:
+	// vm.nr_hugepages = 20
+	// Then apply it:
+	// sudo sysctl -p /etc/sysctl.d/99-hugepages.conf
+
+	size_t nstate_size = (sizeof(struct nes_state) + 4095) & ~4095; // Round up to page size
+	struct nes_state *nstate = mmap(0, nstate_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE | MAP_LOCKED, -1, 0);
+	if (nstate == MAP_FAILED) {
+		perror("mmap nstate failed");
+		exit(1);
+	}
 
 #ifdef BENCHMARK
 	// Run benchmark with configurable parameters
@@ -299,9 +309,9 @@ int main(int argc, char **argv) {
 	// 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/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");
@@ -342,8 +352,8 @@ int main(int argc, char **argv) {
 	// ines2_load(nstate, "data/2002/Best of the Best - Championship Karate (USA).zip");
 
 	// ines2_load(nstate, "data/0001/Kid Icarus (UE) (V1.1) [!].nes");
-	ines2_load(nstate, "data/0001/Metroid (U) [!].nes");
-	ines2_load(nstate, "data/0001/Legend of Zelda, The (U) (V1.1) [!].nes");
+	// ines2_load(nstate, "data/0001/Metroid (U) [!].nes");
+	// ines2_load(nstate, "data/0001/Legend of Zelda, The (U) (V1.1) [!].nes");
 
 	// ines2_load(nstate, "data/Blaster Master (USA).zip");	// mapper 1
 	// ines2_load(nstate, "AccuracyCoin.nes");	// mapper 1
@@ -408,6 +418,7 @@ int main(int argc, char **argv) {
 		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;
+		// nstate->ppu.input[0] = tas_input[tas_frame++];
 
 		// Run NES emulation for one frame
 		while(!nstate->ppu.frame_ready) {
@@ -422,14 +433,13 @@ int main(int argc, char **argv) {
 		// 	}
 		// 	nstate->ppu.open_bus = v;
 		// }
-		// nstate->ppu.input[0] = tas_input[tas_frame++];
 
 		frames++;
 
 		// Dump state every frame starting from 2400
-		if(frames >= 2400 && frames <= 3100) {
-			dump_state(nstate);
-		}
+		// if(frames >= 2400 && frames <= 3100) {
+		// 	dump_state(nstate);
+		// }
 
 		// Convert NES pixels to display buffer
 		uint32_t * restrict dst = display_buffer;
@@ -456,15 +466,14 @@ int main(int argc, char **argv) {
 	}
 
 	// printf("total frames: %6d  total cycles: %12llu\n", frames, (unsigned long long)nstate->cpu.cycles);
-	printf("state dumps created: %zu\n", state_dump_count);
+	// printf("state dumps created: %zu\n", state_dump_count);
 
 	if(state_dump_file) {
 		fclose(state_dump_file);
 	}
 
 	timer_destroy(timer);
-	free(nstate);
-	// free_nes_state(&nstate);
+	munmap(nstate, nstate_size);
 	timer_shutdown();
 	mkfw_cleanup(window);
 
diff --git a/mknes.h b/mknes.h
index 944f762..8972897 100644
--- a/mknes.h
+++ b/mknes.h
@@ -69,14 +69,14 @@ struct ppu_state {
 		uint8_t palette;
 	} __attribute__((packed)) sprites[8];	// 32 -> +40
 
-	uint8_t input_strobe;				// 76 - Control bit (0 or 1)
-	uint8_t input[2];						// 73 - Controller 1 & 2
-	uint8_t input_latch[2];				// 74 - Latched inputs after strobe
-	uint8_t input_bit[2];				// 75 - Current bit position being shifted out
-	uint8_t frame_ready;					// 77
-	uint8_t sprite_zero_in_range;		// 78 - Boolean: is sprite 0 in range (will always be slot 0 if true)
-
-	uint8_t palette[32];					// 79
+	uint8_t input_strobe;				// 73 - Control bit (0 or 1)
+	uint8_t input[2];						// 74 - Controller 1 & 2
+	uint8_t input_latch[2];				// 76 - Latched inputs after strobe
+	uint8_t input_bit[2];				// 78 - Current bit position being shifted out
+	uint8_t frame_ready;					// 80
+	uint8_t sprite_zero_in_range;		// 81 - Boolean: is sprite 0 in range (will always be slot 0 if true)
+
+	uint8_t palette[32];					// 82
 
 	// NOTE(peter): CACHELINE 5
 	uint8_t oam[256] __attribute__((aligned(64)));
diff --git a/mknes_apu.c b/mknes_apu.c
index f1c27a8..8b9f45b 100644
--- a/mknes_apu.c
+++ b/mknes_apu.c
@@ -131,38 +131,38 @@ static inline void apu_tick(struct nes_state *state) {
 		}
 	}
 
-	if(apu->dmc_dma_enabled && apu->dmc_bytes_remaining > 0) {
-		apu->dmc_sample_timer++;
-		if(apu->dmc_sample_timer >= dmc_rate_table[apu->dmc_freq_index]) {
-			apu->dmc_sample_timer = 0;
-
-			uint8_t val = memory_read(state, apu->dmc_current_addr);
-			(void)val;
-
-			apu->dmc_current_addr++;
-			if(apu->dmc_current_addr == 0x0000) {
-				apu->dmc_current_addr = 0x8000;
-			}
-
-			apu->dmc_bytes_remaining--;
-
-			if(apu->dmc_bytes_remaining == 0) {
-				if(apu->dmc_loop_flag) {
-					apu->dmc_current_addr = 0xc000 + ((uint16_t)apu->dmc_sample_addr << 6);
-					apu->dmc_bytes_remaining = ((uint16_t)apu->dmc_sample_len << 4);
-				} else {
-					if(apu->dmc_irq_enable) {
-						apu->irq_pending = 1;
-						state->cpu.irq_pending = 1;
-					}
-					apu->dmc_dma_enabled = 0;
-				}
-			}
-
-			for(uint32_t i = 0; i < 4; i++) {
-				state->cpu.cycles++;
-				ppu_tick(state);
-			}
-		}
-	}
+	// if(apu->dmc_dma_enabled && apu->dmc_bytes_remaining > 0) {
+	// 	apu->dmc_sample_timer++;
+	// 	if(apu->dmc_sample_timer >= dmc_rate_table[apu->dmc_freq_index]) {
+	// 		apu->dmc_sample_timer = 0;
+
+	// 		uint8_t val = memory_read(state, apu->dmc_current_addr);
+	// 		(void)val;
+
+	// 		apu->dmc_current_addr++;
+	// 		if(apu->dmc_current_addr == 0x0000) {
+	// 			apu->dmc_current_addr = 0x8000;
+	// 		}
+
+	// 		apu->dmc_bytes_remaining--;
+
+	// 		if(apu->dmc_bytes_remaining == 0) {
+	// 			if(apu->dmc_loop_flag) {
+	// 				apu->dmc_current_addr = 0xc000 + ((uint16_t)apu->dmc_sample_addr << 6);
+	// 				apu->dmc_bytes_remaining = ((uint16_t)apu->dmc_sample_len << 4);
+	// 			} else {
+	// 				if(apu->dmc_irq_enable) {
+	// 					apu->irq_pending = 1;
+	// 					state->cpu.irq_pending = 1;
+	// 				}
+	// 				apu->dmc_dma_enabled = 0;
+	// 			}
+	// 		}
+
+	// 		for(uint32_t i = 0; i < 4; i++) {
+	// 			state->cpu.cycles++;
+	// 			ppu_tick(state);
+	// 		}
+	// 	}
+	// }
 }
diff --git a/mknes_sdl.c b/mknes_sdl.c
index 98776b5..f7d94c4 100644
--- a/mknes_sdl.c
+++ b/mknes_sdl.c
@@ -108,7 +108,15 @@ int main(int argc, char **argv) {
 	init_opcode_lut();
 	init_opcode_ud_lut();
 
-	struct nes_state *nstate = aligned_alloc(4096, (sizeof(struct nes_state) + 4095) & ~4095);
+	size_t nstate_size = (sizeof(struct nes_state) + 4095) & ~4095; // Round up to page size
+	struct nes_state *nstate = mmap(NULL, nstate_size,
+	                                PROT_READ | PROT_WRITE,
+	                                MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE | MAP_LOCKED,
+	                                -1, 0);
+	if (nstate == MAP_FAILED) {
+		perror("mmap nstate failed");
+		exit(1);
+	}
 
 #ifdef BENCHMARK
 	// Run benchmark with configurable parameters
@@ -354,7 +362,7 @@ int main(int argc, char **argv) {
 	for(int i = 0; i < 4; i++) {
 		free(debug_nt_buffer[i]);
 	}
-	free(nstate);
+	munmap(nstate, nstate_size);
 	timer_shutdown();
 
 	for(int i = 0; i < 4; i++) {