psx_gpu: start handling vram loads/moves for enhancement

1 files changed, 53 insertions, 3 deletions

diff --git a/plugins/gpu_neon/psx_gpu_if.c b/plugins/gpu_neon/psx_gpu_if.c
index 250aa8c..ca76fe2 100644
--- a/plugins/gpu_neon/psx_gpu_if.c
+++ b/plugins/gpu_neon/psx_gpu_if.c
@@ -9,6 +9,7 @@
  */
 
 #include <stdio.h>
+#include <sys/mman.h>
 
 extern const unsigned char cmd_lengths[256];
 #define command_lengths cmd_lengths
@@ -39,6 +40,8 @@ int do_cmd_list(uint32_t *list, int count, int *last_cmd)
   return ret;
 }
 
+#define ENHANCEMENT_BUF_SIZE (1024 * 1024 * 2 * 4 + 4096)
+
 int renderer_init(void)
 {
   initialize_psx_gpu(&egpu, gpu.vram);
@@ -48,9 +51,12 @@ int renderer_init(void)
     // currently we use 4x 1024*1024 buffers instead of single 2048*1024
     // to be able to reuse 1024-width code better (triangle setup,
     // dithering phase, lines).
-    gpu.enhancement_bufer = malloc(1024 * 1024 * 2 * 4);
-    if (gpu.enhancement_bufer == NULL)
+    gpu.enhancement_bufer = mmap(NULL, ENHANCEMENT_BUF_SIZE,
+      PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+    if (gpu.enhancement_bufer == MAP_FAILED) {
       printf("OOM for enhancement buffer\n");
+      gpu.enhancement_bufer = NULL;
+    }
   }
   egpu.enhancement_buf_ptr = gpu.enhancement_bufer;
 
@@ -59,11 +65,44 @@ int renderer_init(void)
 
 void renderer_finish(void)
 {
-  free(gpu.enhancement_bufer);
+  if (gpu.enhancement_bufer != NULL)
+    munmap(gpu.enhancement_bufer, ENHANCEMENT_BUF_SIZE);
   gpu.enhancement_bufer = NULL;
   egpu.enhancement_buf_ptr = NULL;
 }
 
+static __attribute__((noinline)) void
+sync_enhancement_buffers(int x, int y, int w, int h)
+{
+  int xt = egpu.enhancement_x_threshold;
+  u16 *src, *dst;
+  int wb, i;
+
+  w += x & 7;
+  x &= ~7;
+  w = (w + 7) & ~7;
+  if (y + h > 512)
+    h = 512 - y;
+
+  for (i = 0; i < 4 && w > 0; i++) {
+    if (x < 512) {
+      wb = w;
+      if (x + w > 512)
+        wb = 512 - x;
+      src = gpu.vram + xt * i + y * 1024 + x;
+      dst = egpu.enhancement_buf_ptr +
+        (1024*1024 + xt * 2) * i + (y * 1024 + x) * 2;
+      scale2x_tiles8(dst, src, wb / 8, h);
+    }
+
+    x -= xt;
+    if (x < 0) {
+      w += x;
+      x = 0;
+    }
+  }
+}
+
 void renderer_sync_ecmds(uint32_t *ecmds)
 {
   gpu_parse(&egpu, ecmds + 1, 6 * 4, NULL);
@@ -72,6 +111,8 @@ void renderer_sync_ecmds(uint32_t *ecmds)
 void renderer_update_caches(int x, int y, int w, int h)
 {
   update_texture_cache_region(&egpu, x, y, x + w - 1, y + h - 1);
+  if (gpu.state.enhancement_active && !gpu.status.rgb24)
+    sync_enhancement_buffers(x, y, w, h);
 }
 
 void renderer_flush_queues(void)
@@ -90,6 +131,7 @@ void renderer_set_interlace(int enable, int is_odd)
 
 void renderer_notify_res_change(void)
 {
+  // note: must keep it multiple of 8
   egpu.enhancement_x_threshold = gpu.screen.hres;
 }
 
@@ -97,5 +139,13 @@ void renderer_notify_res_change(void)
 
 void renderer_set_config(const struct rearmed_cbs *cbs)
 {
+  static int enhancement_was_on;
+
   disable_main_render = cbs->gpu_neon.enhancement_no_main;
+  if (egpu.enhancement_buf_ptr != NULL && cbs->gpu_neon.enhancement_enable
+      && !enhancement_was_on)
+  {
+    sync_enhancement_buffers(0, 0, 1024, 512);
+  }
+  enhancement_was_on = cbs->gpu_neon.enhancement_enable;
 }