X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fgpu_neon%2Fgpu.c;h=a18e05ffd0d23fd2868c16701e6321420fa2f79b;hp=990fc520d86977fcecdc2c1c75000d476b2597ed;hb=b78edec75aea5e9015e11dd71f7736d7e92b347b;hpb=096ec49b4bae537b45c4b7178af6ad97c6bfabb2

diff --git a/plugins/gpu_neon/gpu.c b/plugins/gpu_neon/gpu.c
index 990fc520..a18e05ff 100644
--- a/plugins/gpu_neon/gpu.c
+++ b/plugins/gpu_neon/gpu.c
@@ -24,7 +24,7 @@
 //#define log_anomaly gpu_log
 #define log_anomaly(...)
 
-struct psx_gpu gpu __attribute__((aligned(64)));
+struct psx_gpu gpu __attribute__((aligned(2048)));
 
 static noinline void do_reset(void)
 {
@@ -59,14 +59,32 @@ static noinline void update_height(void)
 
 static noinline void decide_frameskip(void)
 {
-  gpu.frameskip.frame_ready = !gpu.frameskip.active;
+  if (gpu.frameskip.active)
+    gpu.frameskip.cnt++;
+  else {
+    gpu.frameskip.cnt = 0;
+    gpu.frameskip.frame_ready = 1;
+  }
 
-  if (!gpu.frameskip.active && (*gpu.frameskip.advice || gpu.frameskip.set == 1))
+  if (!gpu.frameskip.active && *gpu.frameskip.advice)
+    gpu.frameskip.active = 1;
+  else if (gpu.frameskip.set > 0 && gpu.frameskip.cnt < gpu.frameskip.set)
     gpu.frameskip.active = 1;
   else
     gpu.frameskip.active = 0;
 }
 
+static noinline void decide_frameskip_allow(uint32_t cmd_e3)
+{
+  // no frameskip if it decides to draw to display area,
+  // but not for interlace since it'll most likely always do that
+  uint32_t x = cmd_e3 & 0x3ff;
+  uint32_t y = (cmd_e3 >> 10) & 0x3ff;
+  gpu.frameskip.allow = gpu.status.interlace ||
+    (uint32_t)(x - gpu.screen.x) >= (uint32_t)gpu.screen.w ||
+    (uint32_t)(y - gpu.screen.y) >= (uint32_t)gpu.screen.h;
+}
+
 static noinline void get_gpu_info(uint32_t data)
 {
   switch (data & 0x0f) {
@@ -132,9 +150,12 @@ void GPUwriteStatus(uint32_t data)
     case 0x05:
       gpu.screen.x = data & 0x3ff;
       gpu.screen.y = (data >> 10) & 0x3ff;
-      if (gpu.frameskip.set && gpu.frameskip.last_flip_frame != *gpu.state.frame_count) {
-        decide_frameskip();
-        gpu.frameskip.last_flip_frame = *gpu.state.frame_count;
+      if (gpu.frameskip.set) {
+        decide_frameskip_allow(gpu.ex_regs[3]);
+        if (gpu.frameskip.last_flip_frame != *gpu.state.frame_count) {
+          decide_frameskip();
+          gpu.frameskip.last_flip_frame = *gpu.state.frame_count;
+        }
       }
       break;
     case 0x06:
@@ -249,10 +270,15 @@ static void start_vram_transfer(uint32_t pos_word, uint32_t size_word, int is_re
   gpu.dma.h = size_word >> 16;
   gpu.dma.offset = 0;
 
-  if (is_read)
+  renderer_flush_queues();
+  if (is_read) {
     gpu.status.img = 1;
-  else
+    // XXX: wrong for width 1
+    memcpy(&gpu.gp0, VRAM_MEM_XY(gpu.dma.x, gpu.dma.y), 4);
+  }
+  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);
@@ -294,15 +320,8 @@ static int check_cmd(uint32_t *data, int count)
         gpu.ex_regs[1] |= list[5] & 0x1ff;
       }
       else if (cmd == 0xe3)
-      {
-        // no frameskip if it decides to draw to display area,
-        // but not for interlace since it'll most likely always do that
-        uint32_t x = list[0] & 0x3ff;
-        uint32_t y = (list[0] >> 10) & 0x3ff;
-        gpu.frameskip.allow = gpu.status.interlace ||
-          (uint32_t)(x - gpu.screen.x) >= (uint32_t)gpu.screen.w ||
-          (uint32_t)(y - gpu.screen.y) >= (uint32_t)gpu.screen.h;
-      }
+        decide_frameskip_allow(list[0]);
+
       if (2 <= cmd && cmd < 0xc0)
         vram_dirty = 1;
       else if ((cmd & 0xf8) == 0xe0)
@@ -343,7 +362,7 @@ static int check_cmd(uint32_t *data, int count)
   return count - pos;
 }
 
-static void flush_cmd_buffer(void)
+void flush_cmd_buffer(void)
 {
   int left = check_cmd(gpu.cmd_buffer, gpu.cmd_len);
   if (left > 0)
@@ -378,7 +397,7 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
   uint32_t addr, *list;
   uint32_t *llist_entry = NULL;
   int len, left, count;
-  long dma_words = 0;
+  long cpu_cycles = 0;
 
   if (unlikely(gpu.cmd_len > 0))
     flush_cmd_buffer();
@@ -386,7 +405,7 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
   // ff7 sends it's main list twice, detect this
   if (*gpu.state.frame_count == gpu.state.last_list.frame &&
       *gpu.state.hcnt - gpu.state.last_list.hcnt <= 1 &&
-       gpu.state.last_list.words > 1024)
+       gpu.state.last_list.cycles > 2048)
   {
     llist_entry = rambase + (gpu.state.last_list.addr & 0x1fffff) / 4;
     *llist_entry |= 0x800000;
@@ -399,7 +418,9 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
     list = rambase + (addr & 0x1fffff) / 4;
     len = list[0] >> 24;
     addr = list[0] & 0xffffff;
-    dma_words += 1 + len;
+    cpu_cycles += 10;
+    if (len > 0)
+      cpu_cycles += 5 + len;
 
     log_io(".chain %08x #%d\n", (list - rambase) * 4, len);
 
@@ -430,10 +451,10 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
 
   gpu.state.last_list.frame = *gpu.state.frame_count;
   gpu.state.last_list.hcnt = *gpu.state.hcnt;
-  gpu.state.last_list.words = dma_words;
+  gpu.state.last_list.cycles = cpu_cycles;
   gpu.state.last_list.addr = start_addr;
 
-  return dma_words;
+  return cpu_cycles;
 }
 
 void GPUreadDataMem(uint32_t *mem, int count)
@@ -449,15 +470,17 @@ void GPUreadDataMem(uint32_t *mem, int count)
 
 uint32_t GPUreadData(void)
 {
-  log_io("gpu_read\n");
+  uint32_t ret;
 
   if (unlikely(gpu.cmd_len > 0))
     flush_cmd_buffer();
 
+  ret = gpu.gp0;
   if (gpu.dma.h)
-    do_vram_io(&gpu.gp0, 1, 1);
+    do_vram_io(&ret, 1, 1);
 
-  return gpu.gp0;
+  log_io("gpu_read %08x\n", ret);
+  return ret;
 }
 
 uint32_t GPUreadStatus(void)