gpulib: eliminate list scan-ahead

9 files changed, 198 insertions, 104 deletions

diff --git a/plugins/dfxvideo/gpulib_if.c b/plugins/dfxvideo/gpulib_if.c
index 82bc38d..12aa0a3 100644
--- a/plugins/dfxvideo/gpulib_if.c
+++ b/plugins/dfxvideo/gpulib_if.c
@@ -296,16 +296,27 @@ int renderer_init(void)
 
 extern const unsigned char cmd_lengths[256];
 
-void do_cmd_list(unsigned int *list, int list_len)
+int do_cmd_list(unsigned int *list, int list_len, int *last_cmd)
 {
-  unsigned int cmd, len;
-
+  unsigned int cmd = 0, len;
+  unsigned int *list_start = list;
   unsigned int *list_end = list + list_len;
 
   for (; list < list_end; list += 1 + len)
   {
     cmd = *list >> 24;
     len = cmd_lengths[cmd];
+    if (list + 1 + len > list_end) {
+      cmd = -1;
+      break;
+    }
+
+#ifndef TEST
+    if (cmd == 0xa0 || cmd == 0xc0)
+      break; // image i/o, forward to upper layer
+    else if ((cmd & 0xf8) == 0xe0)
+      gpu.ex_regs[cmd & 7] = list[0];
+#endif
 
     primTableJ[cmd]((void *)list);
 
@@ -313,8 +324,8 @@ void do_cmd_list(unsigned int *list, int list_len)
     {
       case 0x48 ... 0x4F:
       {
-        u32 num_vertexes = 1;
-        u32 *list_position = &(list[2]);
+        u32 num_vertexes = 2;
+        u32 *list_position = &(list[3]);
 
         while(1)
         {
@@ -325,16 +336,14 @@ void do_cmd_list(unsigned int *list, int list_len)
           num_vertexes++;
         }
 
-        if(num_vertexes > 2)
-          len += (num_vertexes - 2);
-
+        len += (num_vertexes - 2);
         break;
       }
 
       case 0x58 ... 0x5F:
       {
-        u32 num_vertexes = 1;
-        u32 *list_position = &(list[2]);
+        u32 num_vertexes = 2;
+        u32 *list_position = &(list[4]);
 
         while(1)
         {
@@ -345,9 +354,7 @@ void do_cmd_list(unsigned int *list, int list_len)
           num_vertexes++;
         }
 
-        if(num_vertexes > 2)
-          len += (num_vertexes - 2) * 2;
-
+        len += (num_vertexes - 2) * 2;
         break;
       }
 
@@ -365,6 +372,12 @@ void do_cmd_list(unsigned int *list, int list_len)
 #endif
     }
   }
+
+  gpu.ex_regs[1] &= ~0x1ff;
+  gpu.ex_regs[1] |= lGPUstatusRet & 0x1ff;
+
+  *last_cmd = cmd;
+  return list - list_start;
 }
 
 void renderer_sync_ecmds(uint32_t *ecmds)
diff --git a/plugins/gpu-gles/gpulib_if.c b/plugins/gpu-gles/gpulib_if.c
index ce32aad..c669b63 100644
--- a/plugins/gpu-gles/gpulib_if.c
+++ b/plugins/gpu-gles/gpulib_if.c
@@ -502,16 +502,28 @@ int renderer_init(void)
 
 extern const unsigned char cmd_lengths[256];
 
-void do_cmd_list(unsigned int *list, int list_len)
+// XXX: mostly dupe code from soft peops
+int do_cmd_list(unsigned int *list, int list_len, int *last_cmd)
 {
   unsigned int cmd, len;
-
+  unsigned int *list_start = list;
   unsigned int *list_end = list + list_len;
 
   for (; list < list_end; list += 1 + len)
   {
     cmd = *list >> 24;
     len = cmd_lengths[cmd];
+    if (list + 1 + len > list_end) {
+      cmd = -1;
+      break;
+    }
+
+#ifndef TEST
+    if (cmd == 0xa0 || cmd == 0xc0)
+      break; // image i/o, forward to upper layer
+    else if ((cmd & 0xf8) == 0xe0)
+      gpu.ex_regs[cmd & 7] = list[0];
+#endif
 
     primTableJ[cmd]((void *)list);
 
@@ -519,8 +531,8 @@ void do_cmd_list(unsigned int *list, int list_len)
     {
       case 0x48 ... 0x4F:
       {
-        uint32_t num_vertexes = 1;
-        uint32_t *list_position = &(list[2]);
+        uint32_t num_vertexes = 2;
+        uint32_t *list_position = &(list[3]);
 
         while(1)
         {
@@ -531,16 +543,14 @@ void do_cmd_list(unsigned int *list, int list_len)
           num_vertexes++;
         }
 
-        if(num_vertexes > 2)
-          len += (num_vertexes - 2);
-
+        len += (num_vertexes - 2);
         break;
       }
 
       case 0x58 ... 0x5F:
       {
-        uint32_t num_vertexes = 1;
-        uint32_t *list_position = &(list[2]);
+        uint32_t num_vertexes = 2;
+        uint32_t *list_position = &(list[4]);
 
         while(1)
         {
@@ -551,9 +561,7 @@ void do_cmd_list(unsigned int *list, int list_len)
           num_vertexes++;
         }
 
-        if(num_vertexes > 2)
-          len += (num_vertexes - 2) * 2;
-
+        len += (num_vertexes - 2) * 2;
         break;
       }
 
@@ -571,6 +579,12 @@ void do_cmd_list(unsigned int *list, int list_len)
 #endif
     }
   }
+
+  gpu.ex_regs[1] &= ~0x1ff;
+  gpu.ex_regs[1] |= lGPUstatusRet & 0x1ff;
+
+  *last_cmd = cmd;
+  return list - list_start;
 }
 
 void renderer_sync_ecmds(uint32_t *ecmds)
diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu.h b/plugins/gpu_neon/psx_gpu/psx_gpu.h
index 4605c39..53a8717 100644
--- a/plugins/gpu_neon/psx_gpu/psx_gpu.h
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu.h
@@ -231,7 +231,7 @@ u32 texture_region_mask(s32 x1, s32 y1, s32 x2, s32 y2);
 void flush_render_block_buffer(psx_gpu_struct *psx_gpu);
 
 void initialize_psx_gpu(psx_gpu_struct *psx_gpu, u16 *vram);
-void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size);
+u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size, u32 *last_command);
 
 void triangle_benchmark(psx_gpu_struct *psx_gpu);
 
diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu_main.c b/plugins/gpu_neon/psx_gpu/psx_gpu_main.c
index 6c17b0a..8ca3ad0 100644
--- a/plugins/gpu_neon/psx_gpu/psx_gpu_main.c
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu_main.c
@@ -192,7 +192,7 @@ int main(int argc, char *argv[])
   init_counter();
 #endif
 
-  gpu_parse(psx_gpu, list, size);
+  gpu_parse(psx_gpu, list, size, NULL);
   flush_render_block_buffer(psx_gpu);
 
   clear_stats();
@@ -201,7 +201,7 @@ int main(int argc, char *argv[])
   u32 cycles = get_counter();
 #endif
 
-  gpu_parse(psx_gpu, list, size);
+  gpu_parse(psx_gpu, list, size, NULL);
   flush_render_block_buffer(psx_gpu);
 
 #ifdef NEON_BUILD
diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c b/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
index ac35631..3fc040d 100644
--- a/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
@@ -199,12 +199,17 @@ void set_triangle_color(psx_gpu_struct *psx_gpu, u32 triangle_color)
   get_vertex_data_xy(vertex_number, offset16);                                 \
   set_vertex_color_constant(vertex_number, color)                              \
 
+#ifndef SET_Ex
+#define SET_Ex(r, v)
+#endif
+
 vertex_struct vertexes[4] __attribute__((aligned(32)));
 
-void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
+u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size, u32 *last_command)
 {
-  u32 current_command, command_length;
-  
+  u32 current_command = 0, command_length;
+
+  u32 *list_start = list;
   u32 *list_end = list + (size / 4);
 
   for(; list < list_end; list += 1 + command_length)
@@ -212,6 +217,10 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
   	s16 *list_s16 = (void *)list;
   	current_command = *list >> 24;
   	command_length = command_lengths[current_command];
+  	if (list + 1 + command_length > list_end) {
+  	  current_command = (u32)-1;
+  	  break;
+  	}
 
   	switch(current_command)
   	{
@@ -590,7 +599,12 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
         render_block_move(psx_gpu, list_s16[2] & 0x3FF, list_s16[3] & 0x1FF,
          list_s16[4] & 0x3FF, list_s16[5] & 0x1FF, list_s16[6], list_s16[7]);
   			break;
-  
+ 
+#ifdef PCSX
+  		case 0xA0:          //  sys -> vid
+  		case 0xC0:          //  vid -> sys
+  			goto breakloop;
+#else
   		case 0xA0:          //  sys -> vid
       {
         u32 load_x = list_s16[2] & 0x3FF;
@@ -608,10 +622,11 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
          load_width, load_height, load_width);
   			break;
       }
-  
+
   		case 0xC0:          //  vid -> sys
   			break;
-  
+#endif
+
   		case 0xE1:
         set_texture(psx_gpu, list[0] & 0x1FF);
 
@@ -621,6 +636,7 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
           psx_gpu->render_state_base &= ~RENDER_STATE_DITHER;
 
         psx_gpu->display_area_draw_enable = (list[0] >> 10) & 0x1;
+  			SET_Ex(1, list[0]);
   			break;
   
   		case 0xE2:
@@ -653,6 +669,7 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
 
           update_texture_ptr(psx_gpu);
         }
+        SET_Ex(2, list[0]);
         break;
   		}
   
@@ -666,6 +683,7 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
          psx_gpu->viewport_start_y, psx_gpu->viewport_end_x,
          psx_gpu->viewport_end_y);
 #endif
+  			SET_Ex(3, list[0]);
   			break;
   
   		case 0xE4:
@@ -678,6 +696,7 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
          psx_gpu->viewport_start_y, psx_gpu->viewport_end_x,
          psx_gpu->viewport_end_y);
 #endif
+  			SET_Ex(4, list[0]);
   			break;
   
   		case 0xE5:
@@ -687,6 +706,7 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
         psx_gpu->offset_x = offset_x >> 21;
         psx_gpu->offset_y = offset_y >> 21; 
   
+  			SET_Ex(5, list[0]);
   			break;
   		}
 
@@ -706,6 +726,7 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
           psx_gpu->mask_msb = mask_msb;
         }
 
+  			SET_Ex(6, list[0]);
   			break;
       }
   
@@ -713,5 +734,12 @@ void gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size)
   			break;
   	}
   }
+
+#ifdef PCSX
+breakloop:
+#endif
+  if (last_command != NULL)
+    *last_command = current_command;
+  return list - list_start;
 }
 
diff --git a/plugins/gpu_neon/psx_gpu_if.c b/plugins/gpu_neon/psx_gpu_if.c
index 3ff6e48..ff31c27 100644
--- a/plugins/gpu_neon/psx_gpu_if.c
+++ b/plugins/gpu_neon/psx_gpu_if.c
@@ -13,26 +13,37 @@
 extern const unsigned char cmd_lengths[256];
 #define command_lengths cmd_lengths
 
+static unsigned int *ex_regs;
+
+#define PCSX
+#define SET_Ex(r, v) \
+  ex_regs[r] = v
+
 #include "psx_gpu/psx_gpu.c"
 #include "psx_gpu/psx_gpu_parse.c"
 #include "../gpulib/gpu.h"
 
 static psx_gpu_struct egpu __attribute__((aligned(256)));
 
-void do_cmd_list(uint32_t *list, int count)
+int do_cmd_list(uint32_t *list, int count, int *last_cmd)
 {
-  gpu_parse(&egpu, list, count * 4);
+  int ret = gpu_parse(&egpu, list, count * 4, (u32 *)last_cmd);
+
+  ex_regs[1] &= ~0x1ff;
+  ex_regs[1] |= egpu.texture_settings & 0x1ff;
+  return ret;
 }
 
 int renderer_init(void)
 {
   initialize_psx_gpu(&egpu, gpu.vram);
+  ex_regs = gpu.ex_regs;
   return 0;
 }
 
 void renderer_sync_ecmds(uint32_t *ecmds)
 {
-  gpu_parse(&egpu, ecmds + 1, 6 * 4);
+  gpu_parse(&egpu, ecmds + 1, 6 * 4, NULL);
 }
 
 void renderer_update_caches(int x, int y, int w, int h)
diff --git a/plugins/gpu_unai/gpulib_if.cpp b/plugins/gpu_unai/gpulib_if.cpp
index 22fe31e..4994ef4 100644
--- a/plugins/gpu_unai/gpulib_if.cpp
+++ b/plugins/gpu_unai/gpulib_if.cpp
@@ -162,9 +162,10 @@ int renderer_init(void)
 
 extern const unsigned char cmd_lengths[256];
 
-void do_cmd_list(unsigned int *list, int list_len)
+int do_cmd_list(unsigned int *list, int list_len, int *last_cmd)
 {
-  unsigned int cmd, len;
+  unsigned int cmd = 0, len;
+  unsigned int *list_start = list;
   unsigned int *list_end = list + list_len;
 
   linesInterlace = force_interlace;
@@ -176,6 +177,17 @@ void do_cmd_list(unsigned int *list, int list_len)
   {
     cmd = *list >> 24;
     len = cmd_lengths[cmd];
+    if (list + 1 + len > list_end) {
+      cmd = -1;
+      break;
+    }
+
+#ifndef TEST
+    if (cmd == 0xa0 || cmd == 0xc0)
+      break; // image i/o, forward to upper layer
+    else if ((cmd & 0xf8) == 0xe0)
+      gpu.ex_regs[cmd & 7] = list[0];
+#endif
 
     switch(cmd)
     {
@@ -190,19 +202,16 @@ void do_cmd_list(unsigned int *list, int list_len)
 
         while(1)
         {
-          if((*list_position & 0xf000f000) == 0x50005000 || list_position >= list_end)
-            break;
-
           PacketBuffer.U4[1] = PacketBuffer.U4[2];
           PacketBuffer.U4[2] = *list_position++;
           gpuDrawLF(gpuPixelDrivers [ (Blending_Mode | Masking | Blending | (PixelMSB>>3)) >> 1]);
 
           num_vertexes++;
+          if((*list_position & 0xf000f000) == 0x50005000 || list_position >= list_end)
+            break;
         }
 
-        if(num_vertexes > 2)
-          len += (num_vertexes - 2);
-
+        len += (num_vertexes - 2);
         break;
       }
 
@@ -217,9 +226,6 @@ void do_cmd_list(unsigned int *list, int list_len)
 
         while(1)
         {
-          if((*list_position & 0xf000f000) == 0x50005000 || list_position >= list_end)
-            break;
-
           PacketBuffer.U4[0] = PacketBuffer.U4[2];
           PacketBuffer.U4[1] = PacketBuffer.U4[3];
           PacketBuffer.U4[2] = *list_position++;
@@ -227,11 +233,11 @@ void do_cmd_list(unsigned int *list, int list_len)
           gpuDrawLG(gpuPixelDrivers [ (Blending_Mode | Masking | Blending | (PixelMSB>>3)) >> 1]);
 
           num_vertexes++;
+          if((*list_position & 0xf000f000) == 0x50005000 || list_position >= list_end)
+            break;
         }
 
-        if(num_vertexes > 2)
-          len += (num_vertexes - 2) * 2;
-
+        len += (num_vertexes - 2) * 2;
         break;
       }
 
@@ -252,6 +258,12 @@ void do_cmd_list(unsigned int *list, int list_len)
         break;
     }
   }
+
+  gpu.ex_regs[1] &= ~0x1ff;
+  gpu.ex_regs[1] |= GPU_GP1 & 0x1ff;
+
+  *last_cmd = cmd;
+  return list - list_start;
 }
 
 void renderer_sync_ecmds(uint32_t *ecmds)
diff --git a/plugins/gpulib/gpu.c b/plugins/gpulib/gpu.c
index 99b8eda..df0099c 100644
--- a/plugins/gpulib/gpu.c
+++ b/plugins/gpulib/gpu.c
@@ -94,7 +94,7 @@ static noinline void decide_frameskip(void)
     gpu.frameskip.active = 0;
 }
 
-static noinline void decide_frameskip_allow(uint32_t cmd_e3)
+static noinline int decide_frameskip_allow(uint32_t cmd_e3)
 {
   // no frameskip if it decides to draw to display area,
   // but not for interlace since it'll most likely always do that
@@ -103,6 +103,7 @@ static noinline void decide_frameskip_allow(uint32_t cmd_e3)
   gpu.frameskip.allow = gpu.status.interlace ||
     (uint32_t)(x - gpu.screen.x) >= (uint32_t)gpu.screen.w ||
     (uint32_t)(y - gpu.screen.y) >= (uint32_t)gpu.screen.h;
+  return gpu.frameskip.allow;
 }
 
 static noinline void get_gpu_info(uint32_t data)
@@ -327,70 +328,84 @@ static void finish_vram_transfer(int is_read)
                            gpu.dma_start.w, gpu.dma_start.h);
 }
 
+static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd)
+{
+  int cmd = 0, pos = 0, len, dummy;
+  int skip = 1;
+
+  while (pos < count && skip) {
+    uint32_t *list = data + pos;
+    cmd = list[0] >> 24;
+    len = 1 + cmd_lengths[cmd];
+
+    if (cmd == 0x02) {
+      if ((list[2] & 0x3ff) > gpu.screen.w || ((list[2] >> 16) & 0x1ff) > gpu.screen.h)
+        // clearing something large, don't skip
+        do_cmd_list(data + pos, 3, &dummy);
+    }
+    else if ((cmd & 0xf4) == 0x24) {
+      // flat textured prim
+      gpu.ex_regs[1] &= ~0x1ff;
+      gpu.ex_regs[1] |= list[4] & 0x1ff;
+    }
+    else if ((cmd & 0xf4) == 0x34) {
+      // shaded textured prim
+      gpu.ex_regs[1] &= ~0x1ff;
+      gpu.ex_regs[1] |= list[5] & 0x1ff;
+    }
+    else if (cmd == 0xe3)
+      skip = decide_frameskip_allow(list[0]);
+
+    if ((cmd & 0xf8) == 0xe0)
+      gpu.ex_regs[cmd & 7] = list[0];
+
+    if (pos + len > count) {
+      cmd = -1;
+      break; // incomplete cmd
+    }
+    if (cmd == 0xa0 || cmd == 0xc0)
+      break; // image i/o
+    pos += len;
+  }
+
+  renderer_sync_ecmds(gpu.ex_regs);
+  *last_cmd = cmd;
+  return pos;
+}
+
 static noinline int do_cmd_buffer(uint32_t *data, int count)
 {
-  int len, cmd, start, pos;
+  int cmd, pos;
+  uint32_t old_e3 = gpu.ex_regs[3];
   int vram_dirty = 0;
 
   // process buffer
-  for (start = pos = 0; pos < count; )
+  for (pos = 0; pos < count; )
   {
-    cmd = -1;
-    len = 0;
-
-    if (gpu.dma.h) {
+    if (gpu.dma.h && !gpu.dma_start.is_read) { // XXX: need to verify
+      vram_dirty = 1;
       pos += do_vram_io(data + pos, count - pos, 0);
       if (pos == count)
         break;
-      start = pos;
-    }
-
-    // do look-ahead pass to detect SR changes and VRAM i/o
-    while (pos < count) {
-      uint32_t *list = data + pos;
-      cmd = list[0] >> 24;
-      len = 1 + cmd_lengths[cmd];
-
-      //printf("  %3d: %02x %d\n", pos, cmd, len);
-      if ((cmd & 0xf4) == 0x24) {
-        // flat textured prim
-        gpu.ex_regs[1] &= ~0x1ff;
-        gpu.ex_regs[1] |= list[4] & 0x1ff;
-      }
-      else if ((cmd & 0xf4) == 0x34) {
-        // shaded textured prim
-        gpu.ex_regs[1] &= ~0x1ff;
-        gpu.ex_regs[1] |= list[5] & 0x1ff;
-      }
-      else if (cmd == 0xe3)
-        decide_frameskip_allow(list[0]);
-
-      if (2 <= cmd && cmd < 0xc0)
-        vram_dirty = 1;
-      else if ((cmd & 0xf8) == 0xe0)
-        gpu.ex_regs[cmd & 7] = list[0];
-
-      if (pos + len > count) {
-        cmd = -1;
-        break; // incomplete cmd
-      }
-      if (cmd == 0xa0 || cmd == 0xc0)
-        break; // image i/o
-      pos += len;
-    }
-
-    if (pos - start > 0) {
-      if (!gpu.frameskip.active || !gpu.frameskip.allow)
-        do_cmd_list(data + start, pos - start);
-      start = pos;
     }
 
+    cmd = data[pos] >> 24;
     if (cmd == 0xa0 || cmd == 0xc0) {
       // consume vram write/read cmd
       start_vram_transfer(data[pos + 1], data[pos + 2], cmd == 0xc0);
-      pos += len;
+      pos += 3;
+      continue;
     }
-    else if (cmd == -1)
+
+    if (gpu.frameskip.active && gpu.frameskip.allow)
+      pos += do_cmd_list_skip(data + pos, count - pos, &cmd);
+    else {
+      pos += do_cmd_list(data + pos, count - pos, &cmd);
+      vram_dirty = 1;
+    }
+
+    if (cmd == -1)
+      // incomplete cmd
       break;
   }
 
@@ -398,10 +413,11 @@ static noinline int do_cmd_buffer(uint32_t *data, int count)
   gpu.status.reg |= gpu.ex_regs[1] & 0x7ff;
   gpu.status.reg |= (gpu.ex_regs[6] & 3) << 11;
 
-  if (gpu.frameskip.active)
-    renderer_sync_ecmds(gpu.ex_regs);
   gpu.state.fb_dirty |= vram_dirty;
 
+  if (old_e3 != gpu.ex_regs[3])
+    decide_frameskip_allow(gpu.ex_regs[3]);
+
   return count - pos;
 }
 
diff --git a/plugins/gpulib/gpu.h b/plugins/gpulib/gpu.h
index d9ad416..11bfe46 100644
--- a/plugins/gpulib/gpu.h
+++ b/plugins/gpulib/gpu.h
@@ -91,7 +91,7 @@ extern struct psx_gpu gpu;
 
 extern const unsigned char cmd_lengths[256];
 
-void do_cmd_list(uint32_t *list, int count);
+int do_cmd_list(uint32_t *list, int count, int *last_cmd);
 
 struct rearmed_cbs;