feat: Update nes/gbc/sms/genesis emulators to use shared memory (#99)

* Add `shared_memory` component for sharing memory between the emulators more easily
* Refactored all emulators to use new shared memory for majority of state and data. This means each emulator component now has less than 5k of D/IRAM in use that is not from shared memory.
* Update to have better clear screen function that wont have any race conditions and wont destroy buffers. Update video task to have simple clear screen function if passed nullptr.
* Update cmake lists to use newer version of cmake
* Update to hid roms from rom list if their emulator is not enabled for easier / faster iteration and testing when adding / modifying emulators
* Update espp to latest

Shared memory means that it is not easier to add more emulators (such as msx and doom) since the emulators no longer fill up the internal ram. This does require in some cases modifying the emulators to leverage the shared memory, but this is a one-time operation and actually helps ensure the emulator maintains high performance by keeping its state within internal memory as much as possible, instead of in offboard PSRAM.

Build and run `main` on box-3 and ensure all emulators work for multiple sessions of different roms in sequence. Also ensure that saving state and loading state works for each emulator.

Author: finger563

CMakeLists.txt

@@ -1,6 +1,6 @@
 # The following lines of boilerplate have to be in your project's CMakeLists
 # in this exact order for cmake to work correctly
-cmake_minimum_required(VERSION 3.5)
+cmake_minimum_required(VERSION 3.15)
 
 include($ENV{IDF_PATH}/tools/cmake/project.cmake)
 

components/box-emu/include/box-emu.hpp

@@ -151,6 +151,7 @@ class BoxEmu : public espp::BaseComponent {
   /////////////////////////////////////////////////////////////////////////////
 
   bool initialize_video();
+  void clear_screen();
   void display_size(size_t width, size_t height);
   void native_size(size_t width, size_t height, int pitch = -1);
   void palette(const uint16_t *palette, size_t size = 256);

components/box-emu/src/box-emu.cpp

@@ -467,6 +467,11 @@ bool BoxEmu::initialize_video() {
   return true;
 }
 
+void BoxEmu::clear_screen() {
+  static int buffer = 0;
+  xQueueSend(video_queue_, &buffer, 10);
+}
+
 void BoxEmu::display_size(size_t width, size_t height) {
   display_width_ = width;
   display_height_ = height;
@@ -717,13 +722,29 @@ bool BoxEmu::video_task_callback(std::mutex &m, std::condition_variable& cv, boo
     return false;
   }
   static constexpr int num_lines_to_write = num_rows_in_framebuffer;
-  auto &box = Bsp::get();
+  static auto &box = Bsp::get();
+  static const uint16_t *vram0 = (uint16_t*)box.vram0();
+  static const uint16_t *vram1 = (uint16_t*)box.vram1();
   static uint16_t vram_index = 0; // has to be static so that it persists between calls
   const int _x_offset = x_offset();
   const int _y_offset = y_offset();
   const uint16_t* _palette = palette();
-  uint16_t *vram0 = (uint16_t*)box.vram0();
-  uint16_t *vram1 = (uint16_t*)box.vram1();
+  // special case: if _frame_ptr is null, then we ignore all palette, scaling,
+  // etc. and simply fill the screen with 0
+  if (_frame_ptr == nullptr) {
+    for (int y=0; y<lcd_height(); y+= num_lines_to_write) {
+      Pixel* _buf = (Pixel*)((uint32_t)vram0 * (vram_index ^ 0x01) + (uint32_t)vram1 * vram_index);
+      vram_index = vram_index ^ 0x01;
+      int num_lines = std::min<int>(num_lines_to_write, lcd_height()-y);
+      // memset the buffer to 0
+      memset(_buf, 0, lcd_width() * num_lines * sizeof(Pixel));
+      box.write_lcd_frame(0, y + _y_offset, lcd_width(), num_lines, (uint8_t*)&_buf[0]);
+    }
+
+    // now return
+    return false;
+  }
+
   if (is_native()) {
     if (has_palette()) {
       for (int y=0; y<display_height_; y+= num_lines_to_write) {

components/espp

@@ -3,7 +3,7 @@ idf_component_register(
   SRC_DIRS "src" "gnuboy/src"
   PRIV_INCLUDE_DIRS "gnuboy/include"
   # LDFRAGMENTS "linker.lf"
-  REQUIRES "box-emu" "statistics"
+  REQUIRES "box-emu" "statistics" "shared_memory"
   )
 
 target_compile_options(${COMPONENT_LIB} PRIVATE -Wno-misleading-indentation -Wno-implicit-fallthrough -Wno-unused-function -Wno-unused-variable -Wno-discarded-qualifiers)

components/gbc/CMakeLists.txt

@@ -25,7 +25,7 @@ struct cpu
 	int snd;
 };
 
-extern struct cpu cpu;
+extern struct cpu *cpu;
 
 
 void cpu_timers(int cnt);

components/gbc/gnuboy/include/gnuboy/cpu.h

@@ -14,33 +14,33 @@
 #define W(r) ((r).w[LO])
 #define DW(r) ((r).d)
 
-#define A HB(cpu.af)
-#define F LB(cpu.af)
-#define B HB(cpu.bc)
-#define C LB(cpu.bc)
-#define D HB(cpu.de)
-#define E LB(cpu.de)
-#define H HB(cpu.hl)
-#define L LB(cpu.hl)
-
-#define AF W(cpu.af)
-#define BC W(cpu.bc)
-#define DE W(cpu.de)
-#define HL W(cpu.hl)
-
-#define PC W(cpu.pc)
-#define SP W(cpu.sp)
-
-#define xAF DW(cpu.af)
-#define xBC DW(cpu.bc)
-#define xDE DW(cpu.de)
-#define xHL DW(cpu.hl)
-
-#define xPC DW(cpu.pc)
-#define xSP DW(cpu.sp)
-
-#define IMA cpu.ima
-#define IME cpu.ime
+#define A HB(cpu->af)
+#define F LB(cpu->af)
+#define B HB(cpu->bc)
+#define C LB(cpu->bc)
+#define D HB(cpu->de)
+#define E LB(cpu->de)
+#define H HB(cpu->hl)
+#define L LB(cpu->hl)
+
+#define AF W(cpu->af)
+#define BC W(cpu->bc)
+#define DE W(cpu->de)
+#define HL W(cpu->hl)
+
+#define PC W(cpu->pc)
+#define SP W(cpu->sp)
+
+#define xAF DW(cpu->af)
+#define xBC DW(cpu->bc)
+#define xDE DW(cpu->de)
+#define xHL DW(cpu->hl)
+
+#define xPC DW(cpu->pc)
+#define xSP DW(cpu->sp)
+
+#define IMA cpu->ima
+#define IME cpu->ime
 #define IF R_IF
 #define IE R_IE
 

components/gbc/gnuboy/include/gnuboy/cpuregs.h

@@ -13,7 +13,7 @@ struct vissprite
 	//byte pad[2]; //6
 };
 
-struct scan
+struct gbc_scan
 {
 	int bg[64];
 	int wnd[64];
@@ -45,7 +45,7 @@ struct lcd
 };
 
 extern struct lcd lcd;
-extern struct scan scan;
+extern struct gbc_scan *scan;
 
 
 void lcd_begin();

components/gbc/gnuboy/include/gnuboy/lcd.h

@@ -22,6 +22,6 @@ int sram_save();
 void state_load(int n);
 void state_save(int n);
 
-
+extern unsigned char *gbc_filebuf;
 
 #endif

components/gbc/gnuboy/include/gnuboy/loader.h

@@ -15,12 +15,6 @@
 #include "gnuboy/asm.h"
 #endif
 
-
-struct cpu cpu;
-
-
-
-
 #define ZFLAG(n) ( (n) ? 0 : FZ )
 #define HFLAG(n) ( (n) ? 0 : FH )
 #define CFLAG(n) ( (n) ? 0 : FC )
@@ -231,7 +225,7 @@ label: op(b); break;
 #define RET ( POP(PC) )
 
 #define EI ( IMA = 1 )
-#define DI ( cpu.halt = IMA = IME = 0 )
+#define DI ( cpu->halt = IMA = IME = 0 )
 
 
 
@@ -251,13 +245,13 @@ label: op(b); break;
 
 void cpu_reset()
 {
-	cpu.speed = 0;
-	cpu.halt = 0;
-	cpu.div = 0;
-	cpu.tim = 0;
+	cpu->speed = 0;
+	cpu->halt = 0;
+	cpu->div = 0;
+	cpu->tim = 0;
 	/* set lcdc ahead of cpu by 19us; see A */
 	/* FIXME: leave value at 0, use lcdc_trans() to actually send lcdc ahead */
-	cpu.lcdc = 40;
+	cpu->lcdc = 40;
 
 	IME = 0;
 	IMA = 0;
@@ -281,11 +275,11 @@ void cpu_reset()
 	handle differences in place */
 void div_advance(int cnt)
 {
-	cpu.div += (cnt<<1);
-	if (cpu.div >= 256)
+	cpu->div += (cnt<<1);
+	if (cpu->div >= 256)
 	{
-		R_DIV += (cpu.div >> 8);
-		cpu.div &= 0xff;
+		R_DIV += (cpu->div >> 8);
+		cpu->div &= 0xff;
 	}
 }
 
@@ -302,12 +296,12 @@ void timer_advance(int cnt)
 	if (!(R_TAC & 0x04)) return;
 
 	unit = ((-R_TAC) & 3) << 1;
-	cpu.tim += (cnt<<unit);
+	cpu->tim += (cnt<<unit);
 
-	if (cpu.tim >= 512)
+	if (cpu->tim >= 512)
 	{
-		tima = R_TIMA + (cpu.tim >> 9);
-		cpu.tim &= 0x1ff;
+		tima = R_TIMA + (cpu->tim >> 9);
+		cpu->tim &= 0x1ff;
 		if (tima >= 256)
 		{
 			hw_interrupt(IF_TIMER, IF_TIMER);
@@ -325,21 +319,21 @@ void timer_advance(int cnt)
 */
 inline void lcdc_advance(int cnt)
 {
-	cpu.lcdc -= cnt;
-	if (cpu.lcdc <= 0) lcdc_trans();
+	cpu->lcdc -= cnt;
+	if (cpu->lcdc <= 0) lcdc_trans();
 }
 
 /* cnt - time to emulate, expressed in 2MHz units */
 inline void sound_advance(int cnt)
 {
-	cpu.snd += cnt;
+	cpu->snd += cnt;
 }
 
 /* cnt - time to emulate, expressed in 2MHz units */
 void cpu_timers(int cnt)
 {
-	div_advance(cnt << cpu.speed);
-	timer_advance(cnt << cpu.speed);
+	div_advance(cnt << cpu->speed);
+	timer_advance(cnt << cpu->speed);
 	lcdc_advance(cnt);
 	sound_advance(cnt);
 }
@@ -356,16 +350,16 @@ int cpu_idle(int max)
 	int cnt, unit;
 
 
-	if (!(cpu.halt && IME)) return 0;
+	if (!(cpu->halt && IME)) return 0;
 	if (R_IF & R_IE)
 	{
-		cpu.halt = 0;
+		cpu->halt = 0;
 		return 0;
 	}
 
 	/* Make sure we don't miss lcdc status events! */
-	if ((R_IE & (IF_VBLANK | IF_STAT)) && (max > cpu.lcdc))
-		max = cpu.lcdc;
+	if ((R_IE & (IF_VBLANK | IF_STAT)) && (max > cpu->lcdc))
+		max = cpu->lcdc;
 
 	/* If timer interrupt cannot happen, this is very simple! */
 	if (!((R_IE & IF_TIMER) && (R_TAC & 0x04)))
@@ -376,7 +370,7 @@ int cpu_idle(int max)
 
 	/* Figure out when the next timer interrupt will happen */
 	unit = ((-R_TAC) & 3) << 1;
-	cnt = (511 - cpu.tim + (1<<unit)) >> unit;
+	cnt = (511 - cpu->tim + (1<<unit)) >> unit;
 	cnt += (255 - R_TIMA) << (9 - unit);
 
 	if (max < cnt)
@@ -909,15 +903,15 @@ int cpu_emulate(int cycles)
 		PC++;
 		if (R_KEY1 & 1)
 		{
-			cpu.speed = cpu.speed ^ 1;
-			R_KEY1 = (R_KEY1 & 0x7E) | (cpu.speed << 7);
+			cpu->speed = cpu->speed ^ 1;
+			R_KEY1 = (R_KEY1 & 0x7E) | (cpu->speed << 7);
 			break;
 		}
 		/* NOTE - we do not implement dmg STOP whatsoever */
 		break;
 
 	case 0x76: /* HALT */
-		cpu.halt = 1;
+		cpu->halt = 1;
 		break;
 
 	case 0xCB: /* CB prefix */
@@ -956,7 +950,7 @@ int cpu_emulate(int cycles)
 	clen <<= 1;
 	div_advance(clen);
 	timer_advance(clen);
-	clen >>= cpu.speed;
+	clen >>= cpu->speed;
 	lcdc_advance(clen);
 	sound_advance(clen);