X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fgpulib%2Fgpu.c;h=337e27a7ad98182c5450a989e5876bf8f5038bdc;hp=ea172376fdb0719069656f616c156fd8b59e4b42;hb=c4c66b22ae607aa8dc5e6983f7ab030ead01faae;hpb=3d47ef173590f99d22b33e2c6261e068b2f66871

diff --git a/plugins/gpulib/gpu.c b/plugins/gpulib/gpu.c
index ea172376..337e27a7 100644
--- a/plugins/gpulib/gpu.c
+++ b/plugins/gpulib/gpu.c
@@ -24,7 +24,7 @@
 //#define log_anomaly gpu_log
 #define log_anomaly(...)
 
-struct psx_gpu gpu __attribute__((aligned(2048)));
+struct psx_gpu gpu;
 
 static noinline int do_cmd_buffer(uint32_t *data, int count);
 static void finish_vram_transfer(int is_read);
@@ -68,10 +68,11 @@ static noinline void update_width(void)
 
 static noinline void update_height(void)
 {
+  // TODO: emulate this properly..
   int sh = gpu.screen.y2 - gpu.screen.y1;
   if (gpu.status.dheight)
     sh *= 2;
-  if (sh <= 0)
+  if (sh <= 0 || sh > gpu.screen.vres)
     sh = gpu.screen.vres;
 
   gpu.screen.h = sh;
@@ -92,6 +93,12 @@ static noinline void decide_frameskip(void)
     gpu.frameskip.active = 1;
   else
     gpu.frameskip.active = 0;
+
+  if (!gpu.frameskip.active && gpu.frameskip.pending_fill[0] != 0) {
+    int dummy;
+    do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy);
+    gpu.frameskip.pending_fill[0] = 0;
+  }
 }
 
 static noinline int decide_frameskip_allow(uint32_t cmd_e3)
@@ -127,6 +134,22 @@ static noinline void get_gpu_info(uint32_t data)
   }
 }
 
+// double, for overdraw guard
+#define VRAM_SIZE (1024 * 512 * 2 * 2)
+
+static int map_vram(void)
+{
+  gpu.vram = gpu.mmap(VRAM_SIZE);
+  if (gpu.vram != NULL) {
+    gpu.vram += 4096 / 2;
+    return 0;
+  }
+  else {
+    fprintf(stderr, "could not map vram, expect crashes\n");
+    return -1;
+  }
+}
+
 long GPUinit(void)
 {
   int ret;
@@ -139,12 +162,26 @@ long GPUinit(void)
   gpu.cmd_len = 0;
   do_reset();
 
+  if (gpu.mmap != NULL) {
+    if (map_vram() != 0)
+      ret = -1;
+  }
   return ret;
 }
 
 long GPUshutdown(void)
 {
-  return vout_finish();
+  long ret;
+
+  renderer_finish();
+  ret = vout_finish();
+  if (gpu.vram != NULL) {
+    gpu.vram -= 4096 / 2;
+    gpu.munmap(gpu.vram, VRAM_SIZE);
+  }
+  gpu.vram = NULL;
+
+  return ret;
 }
 
 void GPUwriteStatus(uint32_t data)
@@ -176,7 +213,7 @@ void GPUwriteStatus(uint32_t data)
       break;
     case 0x05:
       gpu.screen.x = data & 0x3ff;
-      gpu.screen.y = (data >> 10) & 0x3ff;
+      gpu.screen.y = (data >> 10) & 0x1ff;
       if (gpu.frameskip.set) {
         decide_frameskip_allow(gpu.ex_regs[3]);
         if (gpu.frameskip.last_flip_frame != *gpu.state.frame_count) {
@@ -201,6 +238,7 @@ void GPUwriteStatus(uint32_t data)
       gpu.screen.vres = vres[(gpu.status.reg >> 19) & 3];
       update_width();
       update_height();
+      renderer_notify_res_change();
       break;
     default:
       if ((cmd & 0xf0) == 0x10)
@@ -330,41 +368,62 @@ static void finish_vram_transfer(int is_read)
 
 static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd)
 {
-  int cmd = 0, pos = 0, len, dummy;
+  int cmd = 0, pos = 0, len, dummy, v;
   int skip = 1;
 
+  gpu.frameskip.pending_fill[0] = 0;
+
   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;
+    switch (cmd) {
+      case 0x02:
+        if ((list[2] & 0x3ff) > gpu.screen.w || ((list[2] >> 16) & 0x1ff) > gpu.screen.h)
+          // clearing something large, don't skip
+          do_cmd_list(list, 3, &dummy);
+        else
+          memcpy(gpu.frameskip.pending_fill, list, 3 * 4);
+        break;
+      case 0x24 ... 0x27:
+      case 0x2c ... 0x2f:
+      case 0x34 ... 0x37:
+      case 0x3c ... 0x3f:
+        gpu.ex_regs[1] &= ~0x1ff;
+        gpu.ex_regs[1] |= list[4 + ((cmd >> 4) & 1)] & 0x1ff;
+        break;
+      case 0x48 ... 0x4F:
+        for (v = 3; pos + v < count; v++)
+        {
+          if ((list[v] & 0xf000f000) == 0x50005000)
+            break;
+        }
+        len += v - 3;
+        break;
+      case 0x58 ... 0x5F:
+        for (v = 4; pos + v < count; v += 2)
+        {
+          if ((list[v] & 0xf000f000) == 0x50005000)
+            break;
+        }
+        len += v - 4;
+        break;
+      default:
+        if (cmd == 0xe3)
+          skip = decide_frameskip_allow(list[0]);
+        if ((cmd & 0xf8) == 0xe0)
+          gpu.ex_regs[cmd & 7] = list[0];
+        break;
     }
-    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)
+    if (0xa0 <= cmd && cmd <= 0xdf)
       break; // image i/o
+
     pos += len;
   }
 
@@ -390,9 +449,9 @@ static noinline int do_cmd_buffer(uint32_t *data, int count)
     }
 
     cmd = data[pos] >> 24;
-    if (cmd == 0xa0 || cmd == 0xc0) {
+    if (0xa0 <= cmd && cmd <= 0xdf) {
       // consume vram write/read cmd
-      start_vram_transfer(data[pos + 1], data[pos + 2], cmd == 0xc0);
+      start_vram_transfer(data[pos + 1], data[pos + 2], (cmd & 0xe0) == 0xc0);
       pos += 3;
       continue;
     }
@@ -571,13 +630,13 @@ long GPUfreeze(uint32_t type, struct GPUFreeze *freeze)
     case 1: // save
       if (gpu.cmd_len > 0)
         flush_cmd_buffer();
-      memcpy(freeze->psxVRam, gpu.vram, sizeof(gpu.vram));
+      memcpy(freeze->psxVRam, gpu.vram, 1024 * 512 * 2);
       memcpy(freeze->ulControl, gpu.regs, sizeof(gpu.regs));
       memcpy(freeze->ulControl + 0xe0, gpu.ex_regs, sizeof(gpu.ex_regs));
       freeze->ulStatus = gpu.status.reg;
       break;
     case 0: // load
-      memcpy(gpu.vram, freeze->psxVRam, sizeof(gpu.vram));
+      memcpy(gpu.vram, freeze->psxVRam, 1024 * 512 * 2);
       memcpy(gpu.regs, freeze->ulControl, sizeof(gpu.regs));
       memcpy(gpu.ex_regs, freeze->ulControl + 0xe0, sizeof(gpu.ex_regs));
       gpu.status.reg = freeze->ulStatus;
@@ -600,7 +659,16 @@ void GPUupdateLace(void)
     flush_cmd_buffer();
   renderer_flush_queues();
 
-  if (gpu.status.blanking || !gpu.state.fb_dirty)
+  if (gpu.status.blanking) {
+    if (!gpu.state.blanked) {
+      vout_blank();
+      gpu.state.blanked = 1;
+      gpu.state.fb_dirty = 1;
+    }
+    return;
+  }
+
+  if (!gpu.state.fb_dirty)
     return;
 
   if (gpu.frameskip.set) {
@@ -614,6 +682,7 @@ void GPUupdateLace(void)
 
   vout_update();
   gpu.state.fb_dirty = 0;
+  gpu.state.blanked = 0;
 }
 
 void GPUvBlank(int is_vblank, int lcf)
@@ -648,6 +717,14 @@ void GPUrearmedCallbacks(const struct rearmed_cbs *cbs)
   gpu.state.hcnt = cbs->gpu_hcnt;
   gpu.state.frame_count = cbs->gpu_frame_count;
   gpu.state.allow_interlace = cbs->gpu_neon.allow_interlace;
+  gpu.state.enhancement_enable = cbs->gpu_neon.enhancement_enable;
+
+  gpu.mmap = cbs->mmap;
+  gpu.munmap = cbs->munmap;
+
+  // delayed vram mmap
+  if (gpu.vram == NULL)
+    map_vram();
 
   if (cbs->pl_vout_set_raw_vram)
     cbs->pl_vout_set_raw_vram(gpu.vram);