X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=plugins%2Fgpulib%2Fgpu.c;h=e3943a251660ee9c5d4da62be3f151f87dd31fd5;hb=c296224f47ceebab4d6fbd071959bff294e80293;hp=22571fa2b5f98e0ca6a9abbbac53b0a0fcda7b34;hpb=8f8ade9c6a3da1dd6d254789d50df6b8c2997850;p=pcsx_rearmed.git

diff --git a/plugins/gpulib/gpu.c b/plugins/gpulib/gpu.c
index 22571fa2..e3943a25 100644
--- a/plugins/gpulib/gpu.c
+++ b/plugins/gpulib/gpu.c
@@ -14,6 +14,7 @@
 #include <stdlib.h> /* for calloc */
 
 #include "gpu.h"
+#include "gpu_timing.h"
 #include "../../libpcsxcore/gpu.h" // meh
 #include "../../frontend/plugin_lib.h"
 
@@ -35,15 +36,15 @@
 
 struct psx_gpu gpu;
 
-static noinline int do_cmd_buffer(uint32_t *data, int count);
+static noinline int do_cmd_buffer(uint32_t *data, int count, int *cpu_cycles);
 static void finish_vram_transfer(int is_read);
 
 static noinline void do_cmd_reset(void)
 {
+  int dummy = 0;
   renderer_sync();
-
   if (unlikely(gpu.cmd_len > 0))
-    do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len);
+    do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy);
   gpu.cmd_len = 0;
 
   if (unlikely(gpu.dma.h > 0))
@@ -78,7 +79,10 @@ static noinline void update_width(void)
   int hres = hres_all[(gpu.status >> 16) & 7];
   int pal = gpu.status & PSX_GPU_STATUS_PAL;
   int sw = gpu.screen.x2 - gpu.screen.x1;
+  int type = gpu.state.screen_centering_type;
   int x = 0, x_auto;
+  if (type == C_AUTO)
+    type = gpu.state.screen_centering_type_default;
   if (sw <= 0)
     /* nothing displayed? */;
   else {
@@ -87,7 +91,7 @@ static noinline void update_width(void)
     x = (x + 1) & ~1;   // blitter limitation
     sw /= hdiv;
     sw = (sw + 2) & ~3; // according to nocash
-    switch (gpu.state.screen_centering_type) {
+    switch (type) {
     case C_INGAME:
       break;
     case C_MANUAL:
@@ -177,8 +181,8 @@ static noinline void decide_frameskip(void)
     gpu.frameskip.active = 0;
 
   if (!gpu.frameskip.active && gpu.frameskip.pending_fill[0] != 0) {
-    int dummy;
-    do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy);
+    int dummy = 0;
+    do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy, &dummy);
     gpu.frameskip.pending_fill[0] = 0;
   }
 }
@@ -219,8 +223,9 @@ static noinline void get_gpu_info(uint32_t data)
   }
 }
 
-// double, for overdraw guard
-#define VRAM_SIZE ((1024 * 512 * 2 * 2) + 4096)
+#ifndef max
+#define max(a, b) (((a) > (b)) ? (a) : (b))
+#endif
 
 //  Minimum 16-byte VRAM alignment needed by gpu_unai's pixel-skipping
 //  renderer/downscaler it uses in high res modes:
@@ -232,6 +237,9 @@ static noinline void get_gpu_info(uint32_t data)
 	#define VRAM_ALIGN 16
 #endif
 
+// double, for overdraw guard + at least 1 page before
+#define VRAM_SIZE ((1024 * 512 * 2 * 2) + max(VRAM_ALIGN, 4096))
+
 // vram ptr received from mmap/malloc/alloc (will deallocate using this)
 static uint16_t *vram_ptr_orig = NULL;
 
@@ -245,9 +253,9 @@ static uint16_t *vram_ptr_orig = NULL;
 static int map_vram(void)
 {
 #if GPULIB_USE_MMAP
-  gpu.vram = vram_ptr_orig = gpu.mmap(VRAM_SIZE + (VRAM_ALIGN-1));
+  gpu.vram = vram_ptr_orig = gpu.mmap(VRAM_SIZE);
 #else
-  gpu.vram = vram_ptr_orig = calloc(VRAM_SIZE + (VRAM_ALIGN-1), 1);
+  gpu.vram = vram_ptr_orig = calloc(VRAM_SIZE, 1);
 #endif
   if (gpu.vram != NULL && gpu.vram != (void *)(intptr_t)-1) {
     // 4kb guard in front
@@ -510,7 +518,7 @@ static void finish_vram_transfer(int is_read)
     gpu.gpu_state_change(PGS_VRAM_TRANSFER_END);
 }
 
-static void do_vram_copy(const uint32_t *params)
+static void do_vram_copy(const uint32_t *params, int *cpu_cycles)
 {
   const uint32_t sx =  LE32TOH(params[0]) & 0x3FF;
   const uint32_t sy = (LE32TOH(params[0]) >> 16) & 0x1FF;
@@ -522,6 +530,7 @@ static void do_vram_copy(const uint32_t *params)
   uint16_t lbuf[128];
   uint32_t x, y;
 
+  *cpu_cycles += gput_copy(w, h);
   if (sx == dx && sy == dy && msb == 0)
     return;
 
@@ -557,7 +566,7 @@ static void do_vram_copy(const uint32_t *params)
 
 static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd)
 {
-  int cmd = 0, pos = 0, len, dummy, v;
+  int cmd = 0, pos = 0, len, dummy = 0, v;
   int skip = 1;
 
   gpu.frameskip.pending_fill[0] = 0;
@@ -571,7 +580,7 @@ static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd)
       case 0x02:
         if ((LE32TOH(list[2]) & 0x3ff) > gpu.screen.w || ((LE32TOH(list[2]) >> 16) & 0x1ff) > gpu.screen.h)
           // clearing something large, don't skip
-          do_cmd_list(list, 3, &dummy);
+          do_cmd_list(list, 3, &dummy, &dummy);
         else
           memcpy(gpu.frameskip.pending_fill, list, 3 * 4);
         break;
@@ -621,7 +630,7 @@ static noinline int do_cmd_list_skip(uint32_t *data, int count, int *last_cmd)
   return pos;
 }
 
-static noinline int do_cmd_buffer(uint32_t *data, int count)
+static noinline int do_cmd_buffer(uint32_t *data, int count, int *cpu_cycles)
 {
   int cmd, pos;
   uint32_t old_e3 = gpu.ex_regs[3];
@@ -655,17 +664,22 @@ static noinline int do_cmd_buffer(uint32_t *data, int count)
         cmd = -1; // incomplete cmd, can't consume yet
         break;
       }
-      do_vram_copy(data + pos + 1);
+      do_vram_copy(data + pos + 1, cpu_cycles);
       vram_dirty = 1;
       pos += 4;
       continue;
     }
+    else if (cmd == 0x1f) {
+      log_anomaly("irq1?\n");
+      pos++;
+      continue;
+    }
 
     // 0xex cmds might affect frameskip.allow, so pass to do_cmd_list_skip
     if (gpu.frameskip.active && (gpu.frameskip.allow || ((LE32TOH(data[pos]) >> 24) & 0xf0) == 0xe0))
       pos += do_cmd_list_skip(data + pos, count - pos, &cmd);
     else {
-      pos += do_cmd_list(data + pos, count - pos, &cmd);
+      pos += do_cmd_list(data + pos, count - pos, cpu_cycles, &cmd);
       vram_dirty = 1;
     }
 
@@ -688,7 +702,8 @@ static noinline int do_cmd_buffer(uint32_t *data, int count)
 
 static noinline void flush_cmd_buffer(void)
 {
-  int left = do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len);
+  int dummy = 0, left;
+  left = do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy);
   if (left > 0)
     memmove(gpu.cmd_buffer, gpu.cmd_buffer + gpu.cmd_len - left, left * 4);
   if (left != gpu.cmd_len) {
@@ -700,14 +715,14 @@ static noinline void flush_cmd_buffer(void)
 
 void GPUwriteDataMem(uint32_t *mem, int count)
 {
-  int left;
+  int dummy = 0, left;
 
   log_io("gpu_dma_write %p %d\n", mem, count);
 
   if (unlikely(gpu.cmd_len > 0))
     flush_cmd_buffer();
 
-  left = do_cmd_buffer(mem, count);
+  left = do_cmd_buffer(mem, count, &dummy);
   if (left)
     log_anomaly("GPUwriteDataMem: discarded %d/%d words\n", left, count);
 }
@@ -724,7 +739,7 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr
 {
   uint32_t addr, *list, ld_addr = 0;
   int len, left, count;
-  long cpu_cycles = 0;
+  int cpu_cycles = 0;
 
   preload(rambase + (start_addr & 0x1fffff) / 4);
 
@@ -744,8 +759,8 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr
     if (len > 0)
       cpu_cycles += 5 + len;
 
-    log_io(".chain %08lx #%d+%d\n",
-      (long)(list - rambase) * 4, len, gpu.cmd_len);
+    log_io(".chain %08lx #%d+%d %u\n",
+      (long)(list - rambase) * 4, len, gpu.cmd_len, cpu_cycles);
     if (unlikely(gpu.cmd_len > 0)) {
       if (gpu.cmd_len + len > ARRAY_SIZE(gpu.cmd_buffer)) {
         log_anomaly("cmd_buffer overflow, likely garbage commands\n");
@@ -758,7 +773,7 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr
     }
 
     if (len) {
-      left = do_cmd_buffer(list + 1, len);
+      left = do_cmd_buffer(list + 1, len, &cpu_cycles);
       if (left) {
         memcpy(gpu.cmd_buffer, list + 1 + len - left, left * 4);
         gpu.cmd_len = left;
@@ -968,6 +983,7 @@ void GPUrearmedCallbacks(const struct rearmed_cbs *cbs)
   gpu.state.frame_count = cbs->gpu_frame_count;
   gpu.state.allow_interlace = cbs->gpu_neon.allow_interlace;
   gpu.state.enhancement_enable = cbs->gpu_neon.enhancement_enable;
+  gpu.state.screen_centering_type_default = cbs->screen_centering_type_default;
   if (gpu.state.screen_centering_type != cbs->screen_centering_type
       || gpu.state.screen_centering_x != cbs->screen_centering_x
       || gpu.state.screen_centering_y != cbs->screen_centering_y) {