X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fgpu_neon%2Fpsx_gpu_if.c;h=ad017614b527758764a22177704643556063f358;hp=ca76fe2498af08e9b80b1cee4548e82cadb99eff;hb=HEAD;hpb=50f9355a2338111d940ed408f52fe1defe4df23e

diff --git a/plugins/gpu_neon/psx_gpu_if.c b/plugins/gpu_neon/psx_gpu_if.c
index ca76fe24..84fa9322 100644
--- a/plugins/gpu_neon/psx_gpu_if.c
+++ b/plugins/gpu_neon/psx_gpu_if.c
@@ -9,110 +9,177 @@
  */
 
 #include <stdio.h>
+#include <assert.h>
 #include <sys/mman.h>
 
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
+#ifndef min
+#define min(a, b) ((a) < (b) ? (a) : (b))
+#endif
+#ifndef max
+#define max(a, b) ((a) > (b) ? (a) : (b))
+#endif
+
 extern const unsigned char cmd_lengths[256];
 #define command_lengths cmd_lengths
 
 static unsigned int *ex_regs;
+static int initialized;
 
 #define PCSX
 #define SET_Ex(r, v) \
   ex_regs[r] = v
 
+static __attribute__((noinline)) void
+sync_enhancement_buffers(int x, int y, int w, int h);
+
+#include "../gpulib/gpu.h"
 #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)));
 
-int do_cmd_list(uint32_t *list, int count, int *last_cmd)
+int do_cmd_list(uint32_t *list, int count,
+ int *cycles_sum, int *cycles_last, int *last_cmd)
 {
   int ret;
 
+#if defined(__arm__) && defined(NEON_BUILD) && !defined(SIMD_BUILD)
+  // the asm doesn't bother to save callee-save vector regs, so do it here
+  __asm__ __volatile__("":::"q4","q5","q6","q7");
+#endif
+
   if (gpu.state.enhancement_active)
-    ret = gpu_parse_enhanced(&egpu, list, count * 4, (u32 *)last_cmd);
+    ret = gpu_parse_enhanced(&egpu, list, count * 4,
+            cycles_sum, cycles_last, (u32 *)last_cmd);
   else
-    ret = gpu_parse(&egpu, list, count * 4, (u32 *)last_cmd);
+    ret = gpu_parse(&egpu, list, count * 4,
+            cycles_sum, cycles_last, (u32 *)last_cmd);
+
+#if defined(__arm__) && defined(NEON_BUILD) && !defined(SIMD_BUILD)
+  __asm__ __volatile__("":::"q4","q5","q6","q7");
+#endif
 
   ex_regs[1] &= ~0x1ff;
   ex_regs[1] |= egpu.texture_settings & 0x1ff;
   return ret;
 }
 
-#define ENHANCEMENT_BUF_SIZE (1024 * 1024 * 2 * 4 + 4096)
+#define ENHANCEMENT_BUF_SIZE (1024 * 1024 * 2 * 4 + 4096 * 2)
 
-int renderer_init(void)
+static void *get_enhancement_bufer(int *x, int *y, int *w, int *h,
+ int *vram_h)
 {
-  initialize_psx_gpu(&egpu, gpu.vram);
-  ex_regs = gpu.ex_regs;
+  uint16_t *ret = select_enhancement_buf_ptr(&egpu, *x, *y);
+  if (ret == NULL)
+    return NULL;
+
+  *x *= 2;
+  *y *= 2;
+  *w = *w * 2;
+  *h = *h * 2;
+  *vram_h = 1024;
+  return ret;
+}
 
-  if (gpu.enhancement_bufer == NULL) {
-    // 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 = 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;
-    }
+static void map_enhancement_buffer(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).
+  egpu.enhancement_buf_ptr = gpu.mmap(ENHANCEMENT_BUF_SIZE);
+  if (egpu.enhancement_buf_ptr == NULL) {
+    fprintf(stderr, "failed to map enhancement buffer\n");
+    gpu.get_enhancement_bufer = NULL;
+  }
+  else {
+    egpu.enhancement_buf_ptr += 4096 / 2;
+    gpu.get_enhancement_bufer = get_enhancement_bufer;
   }
-  egpu.enhancement_buf_ptr = gpu.enhancement_bufer;
+}
 
+int renderer_init(void)
+{
+  if (gpu.vram != NULL) {
+    initialize_psx_gpu(&egpu, gpu.vram);
+    initialized = 1;
+  }
+
+  if (gpu.mmap != NULL && egpu.enhancement_buf_ptr == NULL)
+    map_enhancement_buffer();
+
+  ex_regs = gpu.ex_regs;
   return 0;
 }
 
 void renderer_finish(void)
 {
-  if (gpu.enhancement_bufer != NULL)
-    munmap(gpu.enhancement_bufer, ENHANCEMENT_BUF_SIZE);
-  gpu.enhancement_bufer = NULL;
+  if (egpu.enhancement_buf_ptr != NULL) {
+    egpu.enhancement_buf_ptr -= 4096 / 2;
+    gpu.munmap(egpu.enhancement_buf_ptr, ENHANCEMENT_BUF_SIZE);
+  }
   egpu.enhancement_buf_ptr = NULL;
+  egpu.enhancement_current_buf_ptr = NULL;
+  initialized = 0;
 }
 
 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;
-    }
+  int i, right = x + w, bottom = y + h;
+  const u16 *src = gpu.vram;
+  // use these because the scanout struct may hold reduced w, h
+  // due to intersection stuff, see the update_enhancement_buf_scanouts() mess
+  int s_w = max(gpu.screen.hres, gpu.screen.w);
+  int s_h = gpu.screen.vres;
+  if (gpu.screen.y < 0)
+    s_h -= gpu.screen.y;
+  s_w = min(s_w, 512);
+  for (i = 0; i < ARRAY_SIZE(egpu.enhancement_scanouts); i++) {
+    const struct psx_gpu_scanout *s = &egpu.enhancement_scanouts[i];
+    u16 *dst = select_enhancement_buf_by_index(&egpu, i);
+    int x1, x2, y1, y2;
+    if (s->w == 0) continue;
+    if (s->x >= right) continue;
+    if (s->x + s_w <= x) continue;
+    if (s->y >= bottom) continue;
+    if (s->y + s_h <= y) continue;
+    x1 = max(x, s->x);
+    x2 = min(right, s->x + s_w);
+    y1 = max(y, s->y);
+    y2 = min(bottom, s->y + s_h);
+    // 16-byte align for the asm version
+    x2 += x1 & 7;
+    x1 &= ~7;
+    scale2x_tiles8(dst + y1 * 1024*2 + x1 * 2,
+        src + y1 * 1024 + x1, (x2 - x1 + 7) / 8u, y2 - y1);
   }
 }
 
 void renderer_sync_ecmds(uint32_t *ecmds)
 {
-  gpu_parse(&egpu, ecmds + 1, 6 * 4, NULL);
+  s32 dummy0 = 0;
+  u32 dummy1 = 0;
+  gpu_parse(&egpu, ecmds + 1, 6 * 4, &dummy0, &dummy0, &dummy1);
 }
 
-void renderer_update_caches(int x, int y, int w, int h)
+void renderer_update_caches(int x, int y, int w, int h, int state_changed)
 {
   update_texture_cache_region(&egpu, x, y, x + w - 1, y + h - 1);
-  if (gpu.state.enhancement_active && !gpu.status.rgb24)
+
+  if (gpu.state.enhancement_active) {
+    if (state_changed) {
+      int vres = gpu.screen.vres;
+      if (gpu.screen.y < 0)
+        vres -= gpu.screen.y;
+      memset(egpu.enhancement_scanouts, 0, sizeof(egpu.enhancement_scanouts));
+      egpu.enhancement_scanout_eselect = 0;
+      update_enhancement_buf_scanouts(&egpu,
+        gpu.screen.src_x, gpu.screen.src_y, gpu.screen.hres, vres);
+      return;
+    }
     sync_enhancement_buffers(x, y, w, h);
+  }
 }
 
 void renderer_flush_queues(void)
@@ -131,21 +198,37 @@ 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;
+  renderer_notify_scanout_change(gpu.screen.src_x, gpu.screen.src_y);
+}
+
+void renderer_notify_scanout_change(int x, int y)
+{
+  int vres = gpu.screen.vres;
+  if (!gpu.state.enhancement_active || !egpu.enhancement_buf_ptr)
+    return;
+
+  if (gpu.screen.y < 0)
+    vres -= gpu.screen.y;
+  update_enhancement_buf_scanouts(&egpu, x, y, gpu.screen.hres, vres);
 }
 
 #include "../../frontend/plugin_lib.h"
 
 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);
+  if (!initialized) {
+    initialize_psx_gpu(&egpu, gpu.vram);
+    initialized = 1;
+  }
+  if (cbs->pl_set_gpu_caps)
+    cbs->pl_set_gpu_caps(GPU_CAP_SUPPORTS_2X);
+
+  egpu.hack_disable_main = cbs->gpu_neon.enhancement_no_main;
+  egpu.hack_texture_adj = cbs->gpu_neon.enhancement_tex_adj;
+  if (gpu.state.enhancement_enable) {
+    if (gpu.mmap != NULL && egpu.enhancement_buf_ptr == NULL)
+      map_enhancement_buffer();
   }
-  enhancement_was_on = cbs->gpu_neon.enhancement_enable;
 }
+
+// vim:ts=2:sw=2:expandtab