psx_gpu: start handling vram loads/moves for enhancement

4 files changed, 109 insertions, 31 deletions

diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu_arm_neon.S b/plugins/gpu_neon/psx_gpu/psx_gpu_arm_neon.S
index 3239412..87a14f6 100644
--- a/plugins/gpu_neon/psx_gpu/psx_gpu_arm_neon.S
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu_arm_neon.S
@@ -5410,3 +5410,38 @@ function(update_texture_8bpp_cache_slice)
   vpop { q0 - q3 }
   ldmia sp!, { r4 - r11, pc }
 
+
+/* void scale2x_tiles8(void *dst, const void *src, int w8, int h) */
+function(scale2x_tiles8)
+  push { r4, r14 }
+
+  mov r4, r1
+  add r12, r0, #1024*2
+  mov r14, r2
+
+0:
+  vld1.u16 { q0 }, [ r1, :128 ]!
+  vld1.u16 { q2 }, [ r1, :128 ]!
+  vmov q1, q0
+  vmov q3, q2
+  vzip.16 q0, q1
+  vzip.16 q2, q3
+  subs r14, #2
+  vst1.u16 { q0, q1 }, [ r0, :128 ]!
+  vst1.u16 { q0, q1 }, [ r12, :128 ]!
+  blt 1f
+  vst1.u16 { q2, q3 }, [ r0, :128 ]!
+  vst1.u16 { q2, q3 }, [ r12, :128 ]!
+  bgt 0b
+1:
+  subs r3, #1
+  mov r14, r2
+  add r0, #1024*2*2
+  add r4, #1024*2
+  sub r0, r2, lsl #4+1
+  mov r1, r4
+  add r12, r0, #1024*2
+  bgt 0b
+  nop
+
+  pop { r4, pc }
diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c b/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
index 86a816e..af82d7e 100644
--- a/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
@@ -798,6 +798,8 @@ breakloop:
 #define shift_triangle_area() \
   psx_gpu->triangle_area *= 4
 
+extern void scale2x_tiles8(void *dst, const void *src, int w8, int h);
+
 static int disable_main_render;
 
 static void do_triangle_enhanced(psx_gpu_struct *psx_gpu,
@@ -1292,37 +1294,28 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size, u32 *last_c
       }
   
       case 0x80:          //  vid -> vid
-        render_block_move(psx_gpu, list_s16[2] & 0x3FF, list_s16[3] & 0x1FF,
-         list_s16[4] & 0x3FF, list_s16[5] & 0x1FF,
-         ((list_s16[6] - 1) & 0x3FF) + 1, ((list_s16[7] - 1) & 0x1FF) + 1);
+      {
+        u32 sx = list_s16[2] & 0x3FF;
+        u32 sy = list_s16[3] & 0x1FF;
+        u32 dx = list_s16[4] & 0x3FF;
+        u32 dy = list_s16[5] & 0x1FF;
+        u32 w = ((list_s16[6] - 1) & 0x3FF) + 1;
+        u32 h = ((list_s16[7] - 1) & 0x1FF) + 1;
+
+        render_block_move(psx_gpu, sx, sy, dx, dy, w, h);
+        if (dy + h > 512)
+          h = 512 - dy;
+        sx = sx & ~7; // FIXME?
+        dx = dx * 2 & ~7;
+        dy *= 2;
+        scale2x_tiles8(psx_gpu->enhancement_buf_ptr + dy * 1024 + dx,
+          psx_gpu->vram_ptr + sy * 1024 + sx, w / 8, h);
         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;
-        u32 load_y = list_s16[3] & 0x1FF;
-        u32 load_width = list_s16[4] & 0x3FF;
-        u32 load_height = list_s16[5] & 0x1FF;
-        u32 load_size = load_width * load_height;
-  
-        command_length += load_size / 2;
-
-        if(load_size & 1)
-          command_length++;
-
-        render_block_copy(psx_gpu, (u16 *)&(list_s16[6]), load_x, load_y,
-         load_width, load_height, load_width);
-        break;
-      }
-
-      case 0xC0:          //  vid -> sys
-        break;
-#endif
 
       case 0xE1:
         set_texture(psx_gpu, list[0]);
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;
 }
diff --git a/plugins/gpulib/vout_pl.c b/plugins/gpulib/vout_pl.c
index cbd8034..5c74914 100644
--- a/plugins/gpulib/vout_pl.c
+++ b/plugins/gpulib/vout_pl.c
@@ -77,8 +77,8 @@ static void blit(void)
       (x + 8) / stride * 1024 * 1024;
     x *= 2;
     y *= 2;
-    w *= 2;
-    h *= 2;
+    w = (w - 2) * 2;
+    h = (h * 2) - 1;
     stride *= 2;
     vram_mask = 1024 * 1024 - 1;
   }