gpu: rework dma vs busy timing

maybe should implement actual fifo instead someday
libretro/pcsx_rearmed#809

diff --git a/frontend/plugin.c b/frontend/plugin.c
index 0235463..c400165 100644 (file)
--- a/frontend/plugin.c
+++ b/frontend/plugin.c
@@ -122,7 +122,7 @@ extern void GPUwriteDataMem(uint32_t *, int);
 extern uint32_t GPUreadStatus(void);
 extern uint32_t GPUreadData(void);
 extern void GPUreadDataMem(uint32_t *, int);
-extern long GPUdmaChain(uint32_t *, uint32_t, uint32_t *);
+extern long GPUdmaChain(uint32_t *, uint32_t, uint32_t *, int32_t *);
 extern void GPUupdateLace(void);
 extern long GPUfreeze(uint32_t, void *);
 extern void GPUvBlank(int, int);

diff --git a/libpcsxcore/plugins.h b/libpcsxcore/plugins.h
index d080bae..df8ed87 100644 (file)
--- a/libpcsxcore/plugins.h
+++ b/libpcsxcore/plugins.h
@@ -58,7 +58,7 @@ typedef void (CALLBACK* GPUwriteDataMem)(uint32_t *, int);
 typedef uint32_t (CALLBACK* GPUreadStatus)(void);\r
 typedef uint32_t (CALLBACK* GPUreadData)(void);\r
 typedef void (CALLBACK* GPUreadDataMem)(uint32_t *, int);\r
-typedef long (CALLBACK* GPUdmaChain)(uint32_t *,uint32_t, uint32_t *);\r
+typedef long (CALLBACK* GPUdmaChain)(uint32_t *, uint32_t, uint32_t *, int32_t *);\r
 typedef void (CALLBACK* GPUupdateLace)(void);\r
 typedef void (CALLBACK* GPUmakeSnapshot)(void);\r
 typedef void (CALLBACK* GPUkeypressed)(int);\r

diff --git a/libpcsxcore/psxdma.c b/libpcsxcore/psxdma.c
index 3ec42ed..25ee2f0 100644 (file)
--- a/libpcsxcore/psxdma.c
+++ b/libpcsxcore/psxdma.c
@@ -90,6 +90,7 @@ void psxDma4(u32 madr, u32 bcr, u32 chcr) { // SPU
        DMA_INTERRUPT(4);
 }
 
+#if 0
 // Taken from PEOPS SOFTGPU
 static inline boolean CheckForEndlessLoop(u32 laddr, u32 *lUsedAddr) {
        if (laddr == lUsedAddr[1]) return TRUE;
@@ -130,11 +131,12 @@ static u32 gpuDmaChainSize(u32 addr) {
 
        return size;
 }
+#endif
 
 void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
-       u32 *ptr, madr_next, *madr_next_p, size;
+       u32 *ptr, madr_next, *madr_next_p;
        u32 words, words_left, words_max, words_copy;
-       int do_walking;
+       int cycles_sum, cycles_last_cmd = 0, do_walking;
 
        madr &= ~3;
        switch (chcr) {
@@ -195,18 +197,19 @@ void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
                                do_walking = Config.hacks.gpu_slow_list_walking;
                        madr_next_p = do_walking ? &madr_next : NULL;
 
-                       size = GPU_dmaChain((u32 *)psxM, madr & 0x1fffff, madr_next_p);
-                       if ((int)size <= 0)
-                               size = gpuDmaChainSize(madr);
+                       cycles_sum = GPU_dmaChain((u32 *)psxM, madr & 0x1fffff,
+                                       madr_next_p, &cycles_last_cmd);
 
                        HW_DMA2_MADR = SWAPu32(madr_next);
 
                        // a hack for Judge Dredd which is annoyingly sensitive to timing
                        if (Config.hacks.gpu_timing1024)
-                               size = 1024;
+                               cycles_sum = 1024;
 
-                       psxRegs.gpuIdleAfter = psxRegs.cycle + size + 16;
-                       set_event(PSXINT_GPUDMA, size);
+                       psxRegs.gpuIdleAfter = psxRegs.cycle + cycles_sum + cycles_last_cmd;
+                       set_event(PSXINT_GPUDMA, cycles_sum);
+                       //printf("%u dma2cf: %d,%d %08x\n", psxRegs.cycle, cycles_sum,
+                       //      cycles_last_cmd, HW_DMA2_MADR);
                        return;
 
                default:
@@ -221,11 +224,17 @@ void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
 void gpuInterrupt() {
        if (HW_DMA2_CHCR == SWAP32(0x01000401) && !(HW_DMA2_MADR & SWAP32(0x800000)))
        {
-               u32 size, madr_next = 0xffffff, madr = SWAPu32(HW_DMA2_MADR);
-               size = GPU_dmaChain((u32 *)psxM, madr & 0x1fffff, &madr_next);
+               u32 madr_next = 0xffffff, madr = SWAPu32(HW_DMA2_MADR);
+               int cycles_sum, cycles_last_cmd = 0;
+               cycles_sum = GPU_dmaChain((u32 *)psxM, madr & 0x1fffff,
+                               &madr_next, &cycles_last_cmd);
                HW_DMA2_MADR = SWAPu32(madr_next);
-               psxRegs.gpuIdleAfter = psxRegs.cycle + size + 64;
-               set_event(PSXINT_GPUDMA, size);
+               if ((s32)(psxRegs.gpuIdleAfter - psxRegs.cycle) > 0)
+                       cycles_sum += psxRegs.gpuIdleAfter - psxRegs.cycle;
+               psxRegs.gpuIdleAfter = psxRegs.cycle + cycles_sum + cycles_last_cmd;
+               set_event(PSXINT_GPUDMA, cycles_sum);
+               //printf("%u dma2cn: %d,%d %08x\n", psxRegs.cycle, cycles_sum,
+               //      cycles_last_cmd, HW_DMA2_MADR);
                return;
        }
        if (HW_DMA2_CHCR & SWAP32(0x01000000))

diff --git a/plugins/dfxvideo/gpulib_if.c b/plugins/dfxvideo/gpulib_if.c
index ed8d9f2..8a3f2f9 100644 (file)
--- a/plugins/dfxvideo/gpulib_if.c
+++ b/plugins/dfxvideo/gpulib_if.c
@@ -322,12 +322,13 @@ void renderer_notify_scanout_change(int x, int y)
 #include "../gpulib/gpu_timing.h"
 extern const unsigned char cmd_lengths[256];
 
-int do_cmd_list(uint32_t *list, int list_len, int *cpu_cycles_out, int *last_cmd)
+int do_cmd_list(uint32_t *list, int list_len,
+ int *cycles_sum_out, int *cycles_last, int *last_cmd)
 {
+  int cpu_cycles_sum = 0, cpu_cycles = *cycles_last;
   unsigned int cmd = 0, len;
   uint32_t *list_start = list;
   uint32_t *list_end = list + list_len;
-  u32 cpu_cycles = 0;
 
   for (; list < list_end; list += 1 + len)
   {
@@ -357,7 +358,7 @@ int do_cmd_list(uint32_t *list, int list_len, int *cpu_cycles_out, int *last_cmd
 
         while(1)
         {
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           if(list_position >= list_end) {
             cmd = -1;
@@ -382,7 +383,7 @@ int do_cmd_list(uint32_t *list, int list_len, int *cpu_cycles_out, int *last_cmd
 
         while(1)
         {
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           if(list_position >= list_end) {
             cmd = -1;
@@ -403,8 +404,8 @@ int do_cmd_list(uint32_t *list, int list_len, int *cpu_cycles_out, int *last_cmd
 #ifdef TEST
       case 0xA0:          //  sys -> vid
       {
-        u32 load_width = LE2HOST32(slist[4]);
-        u32 load_height = LE2HOST32(slist[5]);
+        u32 load_width = LE2HOST16(slist[4]);
+        u32 load_height = LE2HOST16(slist[5]);
         u32 load_size = load_width * load_height;
 
         len += load_size / 2;
@@ -414,32 +415,32 @@ int do_cmd_list(uint32_t *list, int list_len, int *cpu_cycles_out, int *last_cmd
 
       // timing
       case 0x02:
-        cpu_cycles += gput_fill(LE2HOST32(slist[4]) & 0x3ff,
-            LE2HOST32(slist[5]) & 0x1ff);
+        gput_sum(cpu_cycles_sum, cpu_cycles,
+            gput_fill(LE2HOST16(slist[4]) & 0x3ff, LE2HOST16(slist[5]) & 0x1ff));
         break;
-      case 0x20 ... 0x23: cpu_cycles += gput_poly_base();    break;
-      case 0x24 ... 0x27: cpu_cycles += gput_poly_base_t();  break;
-      case 0x28 ... 0x2B: cpu_cycles += gput_quad_base();    break;
-      case 0x2C ... 0x2F: cpu_cycles += gput_quad_base_t();  break;
-      case 0x30 ... 0x33: cpu_cycles += gput_poly_base_g();  break;
-      case 0x34 ... 0x37: cpu_cycles += gput_poly_base_gt(); break;
-      case 0x38 ... 0x3B: cpu_cycles += gput_quad_base_g();  break;
-      case 0x3C ... 0x3F: cpu_cycles += gput_quad_base_gt(); break;
-      case 0x40 ... 0x47: cpu_cycles += gput_line(0);        break;
-      case 0x50 ... 0x57: cpu_cycles += gput_line(0);        break;
+      case 0x20 ... 0x23: gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base());    break;
+      case 0x24 ... 0x27: gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_t());  break;
+      case 0x28 ... 0x2B: gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base());    break;
+      case 0x2C ... 0x2F: gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_t());  break;
+      case 0x30 ... 0x33: gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_g());  break;
+      case 0x34 ... 0x37: gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_gt()); break;
+      case 0x38 ... 0x3B: gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_g());  break;
+      case 0x3C ... 0x3F: gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_gt()); break;
+      case 0x40 ... 0x47: gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));        break;
+      case 0x50 ... 0x57: gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));        break;
       case 0x60 ... 0x63:
-        cpu_cycles += gput_sprite(LE2HOST32(slist[4]) & 0x3ff,
-            LE2HOST32(slist[5]) & 0x1ff);
+        gput_sum(cpu_cycles_sum, cpu_cycles,
+            gput_sprite(LE2HOST16(slist[4]) & 0x3ff, LE2HOST16(slist[5]) & 0x1ff));
         break;
       case 0x64 ... 0x67:
-        cpu_cycles += gput_sprite(LE2HOST32(slist[6]) & 0x3ff,
-            LE2HOST32(slist[7]) & 0x1ff);
+        gput_sum(cpu_cycles_sum, cpu_cycles,
+            gput_sprite(LE2HOST16(slist[6]) & 0x3ff, LE2HOST16(slist[7]) & 0x1ff));
         break;
-      case 0x68 ... 0x6B: cpu_cycles += gput_sprite(1, 1);   break;
+      case 0x68 ... 0x6B: gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(1, 1));   break;
       case 0x70 ... 0x73:
-      case 0x74 ... 0x77: cpu_cycles += gput_sprite(8, 8);   break;
+      case 0x74 ... 0x77: gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(8, 8));   break;
       case 0x78 ... 0x7B:
-      case 0x7C ... 0x7F: cpu_cycles += gput_sprite(16, 16); break;
+      case 0x7C ... 0x7F: gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(16, 16)); break;
     }
   }
 
@@ -447,7 +448,8 @@ breakloop:
   gpu.ex_regs[1] &= ~0x1ff;
   gpu.ex_regs[1] |= lGPUstatusRet & 0x1ff;
 
-  *cpu_cycles_out += cpu_cycles;
+  *cycles_sum_out += cpu_cycles_sum;
+  *cycles_last = cpu_cycles;
   *last_cmd = cmd;
   return list - list_start;
 }

diff --git a/plugins/gpu-gles/gpulib_if.c b/plugins/gpu-gles/gpulib_if.c
index 561b7cf..ab95c64 100644 (file)
--- a/plugins/gpu-gles/gpulib_if.c
+++ b/plugins/gpu-gles/gpulib_if.c
@@ -521,7 +521,8 @@ void renderer_notify_scanout_change(int x, int y)
 extern const unsigned char cmd_lengths[256];
 
 // XXX: mostly dupe code from soft peops
-int do_cmd_list(unsigned int *list, int list_len, int *cycles, int *last_cmd)
+int do_cmd_list(uint32_t *list, int list_len,
+ int *cycles_sum_out, int *cycles_last, int *last_cmd)
 {
   unsigned int cmd, len;
   unsigned int *list_start = list;

diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu.h b/plugins/gpu_neon/psx_gpu/psx_gpu.h
index 764c2e7..88f3df5 100644 (file)
--- a/plugins/gpu_neon/psx_gpu/psx_gpu.h
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu.h
@@ -256,7 +256,7 @@ void flush_render_block_buffer(psx_gpu_struct *psx_gpu);
 
 void initialize_psx_gpu(psx_gpu_struct *psx_gpu, u16 *vram);
 u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
- s32 *cpu_cycles, u32 *last_command);
+ s32 *cpu_cycles_sum_out, s32 *cpu_cycles_last, u32 *last_command);
 
 void triangle_benchmark(psx_gpu_struct *psx_gpu);
 

diff --git a/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c b/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
index af26fa3..d401522 100644 (file)
--- a/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
+++ b/plugins/gpu_neon/psx_gpu/psx_gpu_parse.c
@@ -252,10 +252,11 @@ static void do_fill(psx_gpu_struct *psx_gpu, u32 x, u32 y,
 #endif
 
 u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
- s32 *cpu_cycles_out, u32 *last_command)
+ s32 *cpu_cycles_sum_out, s32 *cpu_cycles_last, u32 *last_command)
 {
   vertex_struct vertexes[4] __attribute__((aligned(16))) = {};
-  u32 current_command = 0, command_length, cpu_cycles = 0;
+  u32 current_command = 0, command_length;
+  u32 cpu_cycles_sum = 0, cpu_cycles = *cpu_cycles_last;
 
   u32 *list_start = list;
   u32 *list_end = list + (size / 4);
@@ -284,7 +285,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         u32 color = list[0] & 0xFFFFFF;
 
         do_fill(psx_gpu, x, y, width, height, color);
-        cpu_cycles += gput_fill(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_fill(width, height));
         break;
       }
 
@@ -297,7 +298,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy(2, 6);
           
         render_triangle(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base());
         break;
       }
   
@@ -312,7 +313,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_uv(2, 10);
   
         render_triangle(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base_t();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_t());
         break;
       }
   
@@ -327,7 +328,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   
         render_triangle(psx_gpu, vertexes, current_command);
         render_triangle(psx_gpu, &(vertexes[1]), current_command);
-        cpu_cycles += gput_quad_base();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base());
         break;
       }
   
@@ -344,7 +345,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   
         render_triangle(psx_gpu, vertexes, current_command);
         render_triangle(psx_gpu, &(vertexes[1]), current_command);
-        cpu_cycles += gput_quad_base_t();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_t());
         break;
       }
   
@@ -355,7 +356,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_rgb(2, 8);
   
         render_triangle(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base_g();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_g());
         break;
       }
   
@@ -369,7 +370,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_uv_rgb(2, 12);
 
         render_triangle(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base_gt();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_gt());
         break;
       }
   
@@ -382,7 +383,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   
         render_triangle(psx_gpu, vertexes, current_command);
         render_triangle(psx_gpu, &(vertexes[1]), current_command);
-        cpu_cycles += gput_quad_base_g();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_g());
         break;
       }
   
@@ -398,7 +399,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   
         render_triangle(psx_gpu, vertexes, current_command);
         render_triangle(psx_gpu, &(vertexes[1]), current_command);
-        cpu_cycles += gput_quad_base_gt();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_gt());
         break;
       }
   
@@ -410,7 +411,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         vertexes[1].y = list_s16[5] + psx_gpu->offset_y;
 
         render_line(psx_gpu, vertexes, current_command, list[0], 0);
-        cpu_cycles += gput_line(0);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
         break;
       }
   
@@ -432,7 +433,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
           vertexes[1].y = (xy >> 16) + psx_gpu->offset_y;
 
           render_line(psx_gpu, vertexes, current_command, list[0], 0);
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           list_position++;
           num_vertexes++;
@@ -467,7 +468,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         vertexes[1].y = list_s16[7] + psx_gpu->offset_y;
 
         render_line(psx_gpu, vertexes, current_command, 0, 0);
-        cpu_cycles += gput_line(0);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
         break;
       }
  
@@ -498,7 +499,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
           vertexes[1].y = (xy >> 16) + psx_gpu->offset_y;
 
           render_line(psx_gpu, vertexes, current_command, 0, 0);
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           list_position += 2;
           num_vertexes++;
@@ -527,7 +528,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, 0, 0, &width, &height,
            current_command, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -543,7 +544,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, uv & 0xFF, (uv >> 8) & 0xFF,
            &width, &height, current_command, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -555,7 +556,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, 0, 0, &width, &height,
            current_command, list[0]);
-        cpu_cycles += gput_sprite(1, 1);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(1, 1));
         break;
       }
   
@@ -567,7 +568,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, 0, 0, &width, &height,
            current_command, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -582,7 +583,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, uv & 0xFF, (uv >> 8) & 0xFF,
            &width, &height, current_command, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -594,7 +595,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, 0, 0, &width, &height,
            current_command, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -609,7 +610,7 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         render_sprite(psx_gpu, x, y, uv & 0xFF, (uv >> 8) & 0xFF,
            &width, &height, current_command, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -789,7 +790,8 @@ u32 gpu_parse(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   }
 
 breakloop:
-  *cpu_cycles_out += cpu_cycles;
+  *cpu_cycles_sum_out += cpu_cycles_sum;
+  *cpu_cycles_last = cpu_cycles;
   *last_command = current_command;
   return list - list_start;
 }
@@ -1202,10 +1204,11 @@ static void do_sprite_enhanced(psx_gpu_struct *psx_gpu, int x, int y,
 #endif
 
 u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
- s32 *cpu_cycles_out, u32 *last_command)
+ s32 *cpu_cycles_sum_out, s32 *cpu_cycles_last, u32 *last_command)
 {
   vertex_struct vertexes[4] __attribute__((aligned(16))) = {};
-  u32 current_command = 0, command_length, cpu_cycles = 0;
+  u32 current_command = 0, command_length;
+  u32 cpu_cycles_sum = 0, cpu_cycles = *cpu_cycles_last;
 
   u32 *list_start = list;
   u32 *list_end = list + (size / 4);
@@ -1244,7 +1247,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         x &= ~0xF;
         width = ((width + 0xF) & ~0xF);
-        cpu_cycles += gput_fill(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_fill(width, height));
         if (width == 0 || height == 0)
           break;
 
@@ -1275,7 +1278,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy(2, 6);
 
         do_triangle_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base());
         break;
       }
   
@@ -1290,7 +1293,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_uv(2, 10);
   
         do_triangle_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base_t();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_t());
         break;
       }
   
@@ -1304,7 +1307,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy(3, 8);
 
         do_quad_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_quad_base();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base());
         break;
       }
   
@@ -1321,7 +1324,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   
         uv_hack(vertexes, 4);
         do_quad_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_quad_base_t();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_t());
         break;
       }
   
@@ -1332,7 +1335,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_rgb(2, 8);
   
         do_triangle_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base_g();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_g());
         break;
       }
   
@@ -1346,7 +1349,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_uv_rgb(2, 12);
 
         do_triangle_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_poly_base_gt();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_gt());
         break;
       }
   
@@ -1358,7 +1361,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         get_vertex_data_xy_rgb(3, 12);
   
         do_quad_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_quad_base_g();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_g());
         break;
       }
   
@@ -1374,7 +1377,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         uv_hack(vertexes, 4);
         do_quad_enhanced(psx_gpu, vertexes, current_command);
-        cpu_cycles += gput_quad_base_gt();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_gt());
         break;
       }
   
@@ -1388,7 +1391,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         render_line(psx_gpu, vertexes, current_command, list[0], 0);
         if (enhancement_enable(psx_gpu))
           render_line(psx_gpu, vertexes, current_command, list[0], 1);
-        cpu_cycles += gput_line(0);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
         break;
       }
   
@@ -1413,7 +1416,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
           render_line(psx_gpu, vertexes, current_command, list[0], 0);
           if (enhancement_enable(psx_gpu))
             render_line(psx_gpu, vertexes, current_command, list[0], 1);
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           list_position++;
           num_vertexes++;
@@ -1450,7 +1453,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
         render_line(psx_gpu, vertexes, current_command, 0, 0);
         if (enhancement_enable(psx_gpu))
           render_line(psx_gpu, vertexes, current_command, 0, 1);
-        cpu_cycles += gput_line(0);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
         break;
       }
  
@@ -1484,7 +1487,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
           render_line(psx_gpu, vertexes, current_command, 0, 0);
           if (enhancement_enable(psx_gpu))
             render_line(psx_gpu, vertexes, current_command, 0, 1);
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           list_position += 2;
           num_vertexes++;
@@ -1516,7 +1519,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + width))
           do_sprite_enhanced(psx_gpu, x, y, 0, 0, width, height, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -1536,7 +1539,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + width))
           do_sprite_enhanced(psx_gpu, x, y, u, v, width, height, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -1551,7 +1554,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + 1))
           do_sprite_enhanced(psx_gpu, x, y, 0, 0, width, height, list[0]);
-        cpu_cycles += gput_sprite(1, 1);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(1, 1));
         break;
       }
   
@@ -1566,7 +1569,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + 8))
           do_sprite_enhanced(psx_gpu, x, y, 0, 0, width, height, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -1585,7 +1588,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + 8))
           do_sprite_enhanced(psx_gpu, x, y, u, v, width, height, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -1600,7 +1603,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + 16))
           do_sprite_enhanced(psx_gpu, x, y, 0, 0, width, height, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
   
@@ -1619,7 +1622,7 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
 
         if (check_enhanced_range(psx_gpu, x, x + 16))
           do_sprite_enhanced(psx_gpu, x, y, u, v, width, height, list[0]);
-        cpu_cycles += gput_sprite(width, height);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(width, height));
         break;
       }
 
@@ -1773,7 +1776,8 @@ u32 gpu_parse_enhanced(psx_gpu_struct *psx_gpu, u32 *list, u32 size,
   enhancement_disable();
 
 breakloop:
-  *cpu_cycles_out += cpu_cycles;
+  *cpu_cycles_sum_out += cpu_cycles_sum;
+  *cpu_cycles_last = cpu_cycles;
   *last_command = current_command;
   return list - list_start;
 }

diff --git a/plugins/gpu_neon/psx_gpu_if.c b/plugins/gpu_neon/psx_gpu_if.c
index 3f45b0e..313f1f0 100644 (file)
--- a/plugins/gpu_neon/psx_gpu_if.c
+++ b/plugins/gpu_neon/psx_gpu_if.c
@@ -38,7 +38,8 @@ sync_enhancement_buffers(int x, int y, int w, int h);
 
 static psx_gpu_struct egpu __attribute__((aligned(256)));
 
-int do_cmd_list(uint32_t *list, int count, int *cycles, int *last_cmd)
+int do_cmd_list(uint32_t *list, int count,
+ int *cycles_sum, int *cycles_last, int *last_cmd)
 {
   int ret;
 
@@ -48,9 +49,11 @@ int do_cmd_list(uint32_t *list, int count, int *cycles, int *last_cmd)
 #endif
 
   if (gpu.state.enhancement_active)
-    ret = gpu_parse_enhanced(&egpu, list, count * 4, cycles, (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, cycles, (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");
@@ -156,7 +159,7 @@ void renderer_sync_ecmds(uint32_t *ecmds)
 {
   s32 dummy0 = 0;
   u32 dummy1 = 0;
-  gpu_parse(&egpu, ecmds + 1, 6 * 4, &dummy0, &dummy1);
+  gpu_parse(&egpu, ecmds + 1, 6 * 4, &dummy0, &dummy0, &dummy1);
 }
 
 void renderer_update_caches(int x, int y, int w, int h, int state_changed)

diff --git a/plugins/gpu_unai/gpulib_if.cpp b/plugins/gpu_unai/gpulib_if.cpp
index 45c73a7..6816e2b 100644 (file)
--- a/plugins/gpu_unai/gpulib_if.cpp
+++ b/plugins/gpu_unai/gpulib_if.cpp
@@ -393,13 +393,14 @@ static void gpuGP0Cmd_0xEx(gpu_unai_t &gpu_unai, u32 cmd_word)
 #include "../gpulib/gpu_timing.h"
 extern const unsigned char cmd_lengths[256];
 
-int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
+int do_cmd_list(u32 *list_, int list_len,
+ int *cycles_sum_out, int *cycles_last, int *last_cmd)
 {
+  int cpu_cycles_sum = 0, cpu_cycles = *cycles_last;
   u32 cmd = 0, len, i;
-  le32_t *list = (le32_t *)_list;
+  le32_t *list = (le32_t *)list_;
   le32_t *list_start = list;
   le32_t *list_end = list + list_len;
-  u32 cpu_cycles = 0;
 
   //TODO: set ilace_mask when resolution changes instead of every time,
   // eliminate #ifdef below.
@@ -432,8 +433,8 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
     {
       case 0x02:
         gpuClearImage(packet);
-        cpu_cycles += gput_fill(le16_to_s16(packet.U2[4]) & 0x3ff,
-           le16_to_s16(packet.U2[5]) & 0x1ff);
+        gput_sum(cpu_cycles_sum, cpu_cycles,
+           gput_fill(le16_to_s16(packet.U2[4]) & 0x3ff, le16_to_s16(packet.U2[5]) & 0x1ff));
         break;
 
       case 0x20:
@@ -446,7 +447,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.Masking | Blending | gpu_unai.PixelMSB
         ];
         gpuDrawPolyF(packet, driver, false);
-        cpu_cycles += gput_poly_base();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base());
       } break;
 
       case 0x24:
@@ -471,7 +472,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
 
         PP driver = gpuPolySpanDrivers[driver_idx];
         gpuDrawPolyFT(packet, driver, false);
-        cpu_cycles += gput_poly_base_t();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_t());
       } break;
 
       case 0x28:
@@ -484,7 +485,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.Masking | Blending | gpu_unai.PixelMSB
         ];
         gpuDrawPolyF(packet, driver, true); // is_quad = true
-        cpu_cycles += gput_quad_base();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base());
       } break;
 
       case 0x2C:
@@ -509,7 +510,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
 
         PP driver = gpuPolySpanDrivers[driver_idx];
         gpuDrawPolyFT(packet, driver, true); // is_quad = true
-        cpu_cycles += gput_quad_base_t();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_t());
       } break;
 
       case 0x30:
@@ -527,7 +528,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.Masking | Blending | 129 | gpu_unai.PixelMSB
         ];
         gpuDrawPolyG(packet, driver, false);
-        cpu_cycles += gput_poly_base_g();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_g());
       } break;
 
       case 0x34:
@@ -543,7 +544,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.Masking | Blending | ((Lighting)?129:0) | gpu_unai.PixelMSB
         ];
         gpuDrawPolyGT(packet, driver, false);
-        cpu_cycles += gput_poly_base_gt();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_poly_base_gt());
       } break;
 
       case 0x38:
@@ -558,7 +559,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.Masking | Blending | 129 | gpu_unai.PixelMSB
         ];
         gpuDrawPolyG(packet, driver, true); // is_quad = true
-        cpu_cycles += gput_quad_base_g();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_g());
       } break;
 
       case 0x3C:
@@ -574,7 +575,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.Masking | Blending | ((Lighting)?129:0) | gpu_unai.PixelMSB
         ];
         gpuDrawPolyGT(packet, driver, true); // is_quad = true
-        cpu_cycles += gput_quad_base_gt();
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_quad_base_gt());
       } break;
 
       case 0x40:
@@ -585,7 +586,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
         u32 driver_idx = (Blending_Mode | gpu_unai.Masking | Blending | (gpu_unai.PixelMSB>>3)) >> 1;
         PSD driver = gpuPixelSpanDrivers[driver_idx];
         gpuDrawLineF(packet, driver);
-        cpu_cycles += gput_line(0);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
       } break;
 
       case 0x48 ... 0x4F: { // Monochrome line strip
@@ -602,7 +603,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.PacketBuffer.U4[1] = gpu_unai.PacketBuffer.U4[2];
           gpu_unai.PacketBuffer.U4[2] = *list_position++;
           gpuDrawLineF(packet, driver);
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           num_vertexes++;
           if(list_position >= list_end) {
@@ -626,7 +627,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
         driver_idx |= (1 << 5);
         PSD driver = gpuPixelSpanDrivers[driver_idx];
         gpuDrawLineG(packet, driver);
-        cpu_cycles += gput_line(0);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
       } break;
 
       case 0x58 ... 0x5F: { // Gouraud-shaded line strip
@@ -647,7 +648,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           gpu_unai.PacketBuffer.U4[2] = *list_position++;
           gpu_unai.PacketBuffer.U4[3] = *list_position++;
           gpuDrawLineG(packet, driver);
-          cpu_cycles += gput_line(0);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_line(0));
 
           num_vertexes++;
           if(list_position >= list_end) {
@@ -668,7 +669,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
         PT driver = gpuTileSpanDrivers[(Blending_Mode | gpu_unai.Masking | Blending | (gpu_unai.PixelMSB>>3)) >> 1];
         s32 w = 0, h = 0;
         gpuDrawT(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(w, h);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
       } break;
 
       case 0x64:
@@ -697,7 +698,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           driver_idx |= Lighting;
         PS driver = gpuSpriteSpanDrivers[driver_idx];
         gpuDrawS(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(w, h);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
       } break;
 
       case 0x68:
@@ -708,7 +709,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
         PT driver = gpuTileSpanDrivers[(Blending_Mode | gpu_unai.Masking | Blending | (gpu_unai.PixelMSB>>3)) >> 1];
         s32 w = 0, h = 0;
         gpuDrawT(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(1, 1);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(1, 1));
       } break;
 
       case 0x70:
@@ -719,7 +720,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
         PT driver = gpuTileSpanDrivers[(Blending_Mode | gpu_unai.Masking | Blending | (gpu_unai.PixelMSB>>3)) >> 1];
         s32 w = 0, h = 0;
         gpuDrawT(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(w, h);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
       } break;
 
       case 0x74:
@@ -738,7 +739,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           driver_idx |= Lighting;
         PS driver = gpuSpriteSpanDrivers[driver_idx];
         gpuDrawS(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(w, h);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
       } break;
 
       case 0x78:
@@ -749,7 +750,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
         PT driver = gpuTileSpanDrivers[(Blending_Mode | gpu_unai.Masking | Blending | (gpu_unai.PixelMSB>>3)) >> 1];
         s32 w = 0, h = 0;
         gpuDrawT(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(w, h);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
       } break;
 
       case 0x7C:
@@ -760,7 +761,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           s32 w = 0, h = 0;
           gpuSetCLUT(le32_to_u32(gpu_unai.PacketBuffer.U4[2]) >> 16);
           gpuDrawS16(packet, &w, &h);
-          cpu_cycles += gput_sprite(w, h);
+          gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
           break;
         }
         // fallthrough
@@ -778,7 +779,7 @@ int do_cmd_list(u32 *_list, int list_len, int *cpu_cycles_out, int *last_cmd)
           driver_idx |= Lighting;
         PS driver = gpuSpriteSpanDrivers[driver_idx];
         gpuDrawS(packet, driver, &w, &h);
-        cpu_cycles += gput_sprite(w, h);
+        gput_sum(cpu_cycles_sum, cpu_cycles, gput_sprite(w, h));
       } break;
 
 #ifdef TEST
@@ -815,7 +816,8 @@ breakloop:
   gpu.ex_regs[1] &= ~0x1ff;
   gpu.ex_regs[1] |= gpu_unai.GPU_GP1 & 0x1ff;
 
-  *cpu_cycles_out += cpu_cycles;
+  *cycles_sum_out += cpu_cycles_sum;
+  *cycles_last = cpu_cycles;
   *last_cmd = cmd;
   return list - list_start;
 }
@@ -823,7 +825,7 @@ breakloop:
 void renderer_sync_ecmds(u32 *ecmds)
 {
   int dummy;
-  do_cmd_list(&ecmds[1], 6, &dummy, &dummy);
+  do_cmd_list(&ecmds[1], 6, &dummy, &dummy, &dummy);
 }
 
 void renderer_update_caches(int x, int y, int w, int h, int state_changed)

diff --git a/plugins/gpulib/gpu.c b/plugins/gpulib/gpu.c
index e3943a2..fdaf388 100644 (file)
--- a/plugins/gpulib/gpu.c
+++ b/plugins/gpulib/gpu.c
@@ -36,7 +36,8 @@
 
 struct psx_gpu gpu;
 
-static noinline int do_cmd_buffer(uint32_t *data, int count, int *cpu_cycles);
+static noinline int do_cmd_buffer(uint32_t *data, int count,
+    int *cycles_sum, int *cycles_last);
 static void finish_vram_transfer(int is_read);
 
 static noinline void do_cmd_reset(void)
@@ -44,7 +45,7 @@ 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, &dummy);
+    do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy, &dummy);
   gpu.cmd_len = 0;
 
   if (unlikely(gpu.dma.h > 0))
@@ -182,7 +183,7 @@ static noinline void decide_frameskip(void)
 
   if (!gpu.frameskip.active && gpu.frameskip.pending_fill[0] != 0) {
     int dummy = 0;
-    do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy, &dummy);
+    do_cmd_list(gpu.frameskip.pending_fill, 3, &dummy, &dummy, &dummy);
     gpu.frameskip.pending_fill[0] = 0;
   }
 }
@@ -580,7 +581,7 @@ 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, &dummy);
+          do_cmd_list(list, 3, &dummy, &dummy, &dummy);
         else
           memcpy(gpu.frameskip.pending_fill, list, 3 * 4);
         break;
@@ -630,7 +631,8 @@ 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, int *cpu_cycles)
+static noinline int do_cmd_buffer(uint32_t *data, int count,
+    int *cycles_sum, int *cycles_last)
 {
   int cmd, pos;
   uint32_t old_e3 = gpu.ex_regs[3];
@@ -664,7 +666,9 @@ static noinline int do_cmd_buffer(uint32_t *data, int count, int *cpu_cycles)
         cmd = -1; // incomplete cmd, can't consume yet
         break;
       }
-      do_vram_copy(data + pos + 1, cpu_cycles);
+      *cycles_sum += *cycles_last;
+      *cycles_last = 0;
+      do_vram_copy(data + pos + 1, cycles_last);
       vram_dirty = 1;
       pos += 4;
       continue;
@@ -679,7 +683,7 @@ static noinline int do_cmd_buffer(uint32_t *data, int count, int *cpu_cycles)
     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, cpu_cycles, &cmd);
+      pos += do_cmd_list(data + pos, count - pos, cycles_sum, cycles_last, &cmd);
       vram_dirty = 1;
     }
 
@@ -703,7 +707,7 @@ static noinline int do_cmd_buffer(uint32_t *data, int count, int *cpu_cycles)
 static noinline void flush_cmd_buffer(void)
 {
   int dummy = 0, left;
-  left = do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy);
+  left = do_cmd_buffer(gpu.cmd_buffer, gpu.cmd_len, &dummy, &dummy);
   if (left > 0)
     memmove(gpu.cmd_buffer, gpu.cmd_buffer + gpu.cmd_len - left, left * 4);
   if (left != gpu.cmd_len) {
@@ -722,7 +726,7 @@ void GPUwriteDataMem(uint32_t *mem, int count)
   if (unlikely(gpu.cmd_len > 0))
     flush_cmd_buffer();
 
-  left = do_cmd_buffer(mem, count, &dummy);
+  left = do_cmd_buffer(mem, count, &dummy, &dummy);
   if (left)
     log_anomaly("GPUwriteDataMem: discarded %d/%d words\n", left, count);
 }
@@ -735,11 +739,13 @@ void GPUwriteData(uint32_t data)
     flush_cmd_buffer();
 }
 
-long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr)
+long GPUdmaChain(uint32_t *rambase, uint32_t start_addr,
+  uint32_t *progress_addr, int32_t *cycles_last_cmd)
 {
   uint32_t addr, *list, ld_addr = 0;
   int len, left, count;
-  int cpu_cycles = 0;
+  int cpu_cycles_sum = 0;
+  int cpu_cycles_last = 0;
 
   preload(rambase + (start_addr & 0x1fffff) / 4);
 
@@ -755,12 +761,12 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr
     addr = LE32TOH(list[0]) & 0xffffff;
     preload(rambase + (addr & 0x1fffff) / 4);
 
-    cpu_cycles += 10;
+    cpu_cycles_sum += 10;
     if (len > 0)
-      cpu_cycles += 5 + len;
+      cpu_cycles_sum += 5 + len;
 
-    log_io(".chain %08lx #%d+%d %u\n",
-      (long)(list - rambase) * 4, len, gpu.cmd_len, cpu_cycles);
+    log_io(".chain %08lx #%d+%d %u+%u\n",
+      (long)(list - rambase) * 4, len, gpu.cmd_len, cpu_cycles_sum, cpu_cycles_last);
     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");
@@ -773,7 +779,7 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr
     }
 
     if (len) {
-      left = do_cmd_buffer(list + 1, len, &cpu_cycles);
+      left = do_cmd_buffer(list + 1, len, &cpu_cycles_sum, &cpu_cycles_last);
       if (left) {
         memcpy(gpu.cmd_buffer, list + 1 + len - left, left * 4);
         gpu.cmd_len = left;
@@ -810,12 +816,14 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr, uint32_t *progress_addr
     }
   }
 
+  //printf(" -> %d %d\n", cpu_cycles_sum, cpu_cycles_last);
   gpu.state.last_list.frame = *gpu.state.frame_count;
   gpu.state.last_list.hcnt = *gpu.state.hcnt;
-  gpu.state.last_list.cycles = cpu_cycles;
+  gpu.state.last_list.cycles = cpu_cycles_sum + cpu_cycles_last;
   gpu.state.last_list.addr = start_addr;
 
-  return cpu_cycles;
+  *cycles_last_cmd = cpu_cycles_last;
+  return cpu_cycles_sum;
 }
 
 void GPUreadDataMem(uint32_t *mem, int count)

diff --git a/plugins/gpulib/gpu.h b/plugins/gpulib/gpu.h
index 201b5a4..7625c41 100644 (file)
--- a/plugins/gpulib/gpu.h
+++ b/plugins/gpulib/gpu.h
@@ -122,7 +122,8 @@ extern struct psx_gpu gpu;
 
 extern const unsigned char cmd_lengths[256];
 
-int do_cmd_list(uint32_t *list, int count, int *cycles, int *last_cmd);
+int do_cmd_list(uint32_t *list, int count,
+       int *cycles_sum, int *cycles_last, int *last_cmd);
 
 struct rearmed_cbs;
 
@@ -150,7 +151,8 @@ struct GPUFreeze;
 long GPUinit(void);
 long GPUshutdown(void);
 void GPUwriteDataMem(uint32_t *mem, int count);
-long GPUdmaChain(uint32_t *rambase, uint32_t addr, uint32_t *progress_addr);
+long GPUdmaChain(uint32_t *rambase, uint32_t addr,
+               uint32_t *progress_addr, int32_t *cycles_last_cmd);
 void GPUwriteData(uint32_t data);
 void GPUreadDataMem(uint32_t *mem, int count);
 uint32_t GPUreadData(void);

diff --git a/plugins/gpulib/gpu_timing.h b/plugins/gpulib/gpu_timing.h
index 363e608..9991fd8 100644 (file)
--- a/plugins/gpulib/gpu_timing.h
+++ b/plugins/gpulib/gpu_timing.h
@@ -1,6 +1,6 @@
 
 // very conservative and wrong
-#define gput_fill(w, h)     (23 + (4 + (w) / 32u) * (h))
+#define gput_fill(w, h)     (23 + (4 + (w) / 16u) * (h))
 #define gput_copy(w, h)     ((w) * (h))
 #define gput_poly_base()    (23)
 #define gput_poly_base_t()  (gput_poly_base() + 90)
@@ -13,3 +13,7 @@
 #define gput_line(k)        (8 + (k))
 #define gput_sprite(w, h)   (8 + ((w) / 2u) * (h))
 
+// sort of a workaround for lack of proper fifo emulation
+#define gput_sum(sum, cnt, new_cycles) do { \
+  sum += cnt; cnt = new_cycles; \
+} while (0)