gpu_neon: psx_gpu: try to update texture cache instead of invalidating

..mostly just in one case though, invalidate as usual otherwise

6 files changed, 85 insertions, 18 deletions

diff --git a/plugins/gpu_neon/gpu.c b/plugins/gpu_neon/gpu.c
index d6151cd..c31d9a1 100644
--- a/plugins/gpu_neon/gpu.c
+++ b/plugins/gpu_neon/gpu.c
@@ -1,5 +1,5 @@
 /*
- * (C) Gražvydas "notaz" Ignotas, 2011
+ * (C) Gražvydas "notaz" Ignotas, 2011-2012
  *
  * This work is licensed under the terms of any of these licenses
  * (at your option):
@@ -27,13 +27,16 @@
 struct psx_gpu gpu __attribute__((aligned(2048)));
 
 static noinline int do_cmd_buffer(uint32_t *data, int count);
+static void finish_vram_transfer(int is_read);
 
 static noinline void do_cmd_reset(void)
 {
   if (unlikely(gpu.cmd_len > 0))
     do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len);
-
   gpu.cmd_len = 0;
+
+  if (unlikely(gpu.dma.h > 0))
+    finish_vram_transfer(gpu.dma_start.is_read);
   gpu.dma.h = 0;
 }
 
@@ -269,12 +272,16 @@ static int do_vram_io(uint32_t *data, int count, int is_read)
     do_vram_line(x, y, sdata, w, is_read);
   }
 
-  if (h > 0 && count > 0) {
-    y &= 511;
-    do_vram_line(x, y, sdata, count, is_read);
-    o = count;
-    count = 0;
+  if (h > 0) {
+    if (count > 0) {
+      y &= 511;
+      do_vram_line(x, y, sdata, count, is_read);
+      o = count;
+      count = 0;
+    }
   }
+  else
+    finish_vram_transfer(is_read);
   gpu.dma.y = y;
   gpu.dma.h = h;
   gpu.dma.offset = o;
@@ -292,6 +299,8 @@ static void start_vram_transfer(uint32_t pos_word, uint32_t size_word, int is_re
   gpu.dma.w = ((size_word - 1) & 0x3ff) + 1;
   gpu.dma.h = (((size_word >> 16) - 1) & 0x1ff) + 1;
   gpu.dma.offset = 0;
+  gpu.dma.is_read = is_read;
+  gpu.dma_start = gpu.dma;
 
   renderer_flush_queues();
   if (is_read) {
@@ -300,14 +309,20 @@ static void start_vram_transfer(uint32_t pos_word, uint32_t size_word, int is_re
     memcpy(&gpu.gp0, VRAM_MEM_XY(gpu.dma.x, gpu.dma.y), 4);
     gpu.state.last_vram_read_frame = *gpu.state.frame_count;
   }
-  else {
-    renderer_invalidate_caches(gpu.dma.x, gpu.dma.y, gpu.dma.w, gpu.dma.h);
-  }
 
   log_io("start_vram_transfer %c (%d, %d) %dx%d\n", is_read ? 'r' : 'w',
     gpu.dma.x, gpu.dma.y, gpu.dma.w, gpu.dma.h);
 }
 
+static void finish_vram_transfer(int is_read)
+{
+  if (is_read)
+    gpu.status.img = 0;
+  else
+    renderer_update_caches(gpu.dma_start.x, gpu.dma_start.y,
+                           gpu.dma_start.w, gpu.dma_start.h);
+}
+
 static noinline int do_cmd_buffer(uint32_t *data, int count)
 {
   int len, cmd, start, pos;
@@ -541,7 +556,6 @@ long GPUfreeze(uint32_t type, struct GPUFreeze *freeze)
       freeze->ulStatus = gpu.status.reg;
       break;
     case 0: // load
-      renderer_invalidate_caches(0, 0, 1024, 512);
       memcpy(gpu.vram, freeze->psxVRam, sizeof(gpu.vram));
       memcpy(gpu.regs, freeze->ulControl, sizeof(gpu.regs));
       memcpy(gpu.ex_regs, freeze->ulControl + 0xe0, sizeof(gpu.ex_regs));
@@ -551,6 +565,7 @@ long GPUfreeze(uint32_t type, struct GPUFreeze *freeze)
         GPUwriteStatus((i << 24) | (gpu.regs[i] ^ 1));
       }
       renderer_sync_ecmds(gpu.ex_regs);
+      renderer_update_caches(0, 0, 1024, 512);
       break;
   }
 
diff --git a/plugins/gpu_neon/gpu.h b/plugins/gpu_neon/gpu.h
index 36fdef8..d9ad416 100644
--- a/plugins/gpu_neon/gpu.h
+++ b/plugins/gpu_neon/gpu.h
@@ -58,8 +58,8 @@ struct psx_gpu {
   } screen;
   struct {
     int x, y, w, h;
-    int offset;
-  } dma;
+    short int offset, is_read;
+  } dma, dma_start;
   int cmd_len;
   uint32_t zero;
   struct {
@@ -97,7 +97,7 @@ struct rearmed_cbs;
 
 int  renderer_init(void);
 void renderer_sync_ecmds(uint32_t * ecmds);
-void renderer_invalidate_caches(int x, int y, int w, int h);
+void renderer_update_caches(int x, int y, int w, int h);
 void renderer_flush_queues(void);
 void renderer_set_interlace(int enable, int is_odd);
 void renderer_set_config(const struct rearmed_cbs *config);
diff --git a/plugins/gpu_neon/peops_if.c b/plugins/gpu_neon/peops_if.c
index 2e967f2..191fce1 100644
--- a/plugins/gpu_neon/peops_if.c
+++ b/plugins/gpu_neon/peops_if.c
@@ -377,7 +377,7 @@ void renderer_sync_ecmds(uint32_t *ecmds)
   cmdSTP((unsigned char *)&ecmds[6]);
 }
 
-void renderer_invalidate_caches(int x, int y, int w, int h)
+void renderer_update_caches(int x, int y, int w, int h)
 {
 }
 
diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu.c b/plugins/gpu_neon/psx_gpu/psx_gpu.c
index 9e84ace..cf6e2be 100644
--- a/plugins/gpu_neon/psx_gpu/psx_gpu.c
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu.c
@@ -244,6 +244,58 @@ u32 invalidate_texture_cache_region_viewport(psx_gpu_struct *psx_gpu, u32 x1,
   return mask;
 }
 
+void update_texture_cache_region(psx_gpu_struct *psx_gpu, u32 x1, u32 y1,
+ u32 x2, u32 y2)
+{
+  u32 mask = texture_region_mask(x1, y1, x2, y2);
+  u32 texture_page;
+  u8 *texture_page_ptr;
+  u16 *vram_ptr;
+  u32 texel_block;
+  u32 sub_x, sub_y;
+
+  psx_gpu->dirty_textures_8bpp_mask |= mask;
+  psx_gpu->dirty_textures_8bpp_alternate_mask |= mask;
+
+  if ((psx_gpu->dirty_textures_4bpp_mask & mask) == 0 &&
+      (x1 & 3) == 0 && (y1 & 15) == 0 && x2 - x1 < 4 && y2 - y1 < 16)
+  {
+    texture_page = ((x1 / 64) & 15) + (y1 / 256) * 16;
+    texture_page_ptr = psx_gpu->texture_4bpp_cache[texture_page];
+    texture_page_ptr += (x1 / 4 & 15) * 16*16 + (y1 / 16 & 15) * 16*16*16;
+    vram_ptr = psx_gpu->vram_ptr + x1 + y1 * 1024;
+    sub_x = 4;
+    sub_y = 16;
+
+    while(sub_y)
+    {
+      while(sub_x)
+      {
+        texel_block = *vram_ptr;
+
+        texture_page_ptr[0] = texel_block & 0xF;
+        texture_page_ptr[1] = (texel_block >> 4) & 0xF;
+        texture_page_ptr[2] = (texel_block >> 8) & 0xF;
+        texture_page_ptr[3] = texel_block >> 12;
+        
+        vram_ptr++;
+        texture_page_ptr += 4;
+
+        sub_x--;          
+      }
+
+      vram_ptr -= 4;
+      sub_x = 4;
+
+      sub_y--;
+      vram_ptr += 1024;
+    }
+  }
+  else
+  {
+    psx_gpu->dirty_textures_4bpp_mask |= mask;
+  }
+}
 
 void update_texture_8bpp_cache_slice(psx_gpu_struct *psx_gpu,
  u32 texture_page);
diff --git a/plugins/gpu_neon/psx_gpu_if.c b/plugins/gpu_neon/psx_gpu_if.c
index e8bbe59..1c462e1 100644
--- a/plugins/gpu_neon/psx_gpu_if.c
+++ b/plugins/gpu_neon/psx_gpu_if.c
@@ -35,9 +35,9 @@ void renderer_sync_ecmds(uint32_t *ecmds)
   gpu_parse(&egpu, ecmds + 1, 6 * 4);
 }
 
-void renderer_invalidate_caches(int x, int y, int w, int h)
+void renderer_update_caches(int x, int y, int w, int h)
 {
-  invalidate_texture_cache_region(&egpu, x, y, x + w - 1, y + h - 1);
+  update_texture_cache_region(&egpu, x, y, x + w - 1, y + h - 1);
 }
 
 void renderer_flush_queues(void)
diff --git a/plugins/gpu_neon/unai_if.cpp b/plugins/gpu_neon/unai_if.cpp
index 072993a..8917f2a 100644
--- a/plugins/gpu_neon/unai_if.cpp
+++ b/plugins/gpu_neon/unai_if.cpp
@@ -264,7 +264,7 @@ void renderer_sync_ecmds(uint32_t *ecmds)
   unai_do_prim(0xe6, &ecmds[6], 1);
 }
 
-void renderer_invalidate_caches(int x, int y, int w, int h)
+void renderer_update_caches(int x, int y, int w, int h)
 {
 }