32x: improve 'simple' scheduling, works for 'interesting' games

git-svn-id: file:///home/notaz/opt/svn/PicoDrive@798 be3aeb3a-fb24-0410-a615-afba39da0efa

pico/32x/32x.c

@@ -99,28 +99,20 @@ static void p32x_start_blank(void)
   p32x_poll_event(3, 1);
 }
 
-// FIXME..
+static __inline void run_m68k(int cyc)
-static __inline void SekRunM68k(int cyc)
 {
-  int cyc_do;
-  SekCycleAim += cyc;
   if (Pico32x.emu_flags & P32XF_68KPOLL) {
-    SekCycleCnt = SekCycleAim;
+    SekCycleCnt += cyc;
     return;
   }
-  if ((cyc_do = SekCycleAim - SekCycleCnt) <= 0)
+#if defined(EMU_C68K)
-    return;
+  PicoCpuCM68k.cycles = cyc;
-#if defined(EMU_CORE_DEBUG)
-  // this means we do run-compare
-  SekCycleCnt+=CM_compareRun(cyc_do, 0);
-#elif defined(EMU_C68K)
-  PicoCpuCM68k.cycles=cyc_do;
   CycloneRun(&PicoCpuCM68k);
-  SekCycleCnt+=cyc_do-PicoCpuCM68k.cycles;
+  SekCycleCnt += cyc - PicoCpuCM68k.cycles;
 #elif defined(EMU_M68K)
-  SekCycleCnt+=m68k_execute(cyc_do);
+  SekCycleCnt += m68k_execute(cyc);
 #elif defined(EMU_F68K)
-  SekCycleCnt+=fm68k_emulate(cyc_do+1, 0, 0);
+  SekCycleCnt += fm68k_emulate(cyc+1, 0, 0);
 #endif
 }
 
@@ -131,18 +123,28 @@ static __inline void SekRunM68k(int cyc)
 
 #define PICO_32X
 #define CPUS_RUN_SIMPLE(m68k_cycles,s68k_cycles) \
-  SekRunM68k(m68k_cycles); \
+{ \
-  if (!(Pico32x.emu_flags & (P32XF_MSH2POLL|P32XF_MSH2VPOLL))) \
+  int slice; \
-    sh2_execute(&msh2, CYCLES_M68K2SH2(m68k_cycles)); \
+  SekCycleAim += m68k_cycles; \
-  if (!(Pico32x.emu_flags & (P32XF_SSH2POLL|P32XF_SSH2VPOLL))) \
+  while (SekCycleCnt < SekCycleAim) { \
-    sh2_execute(&ssh2, CYCLES_M68K2SH2(m68k_cycles))
+    slice = SekCycleCnt; \
+    run_m68k(SekCycleAim - SekCycleCnt); \
+    slice = SekCycleCnt - slice; /* real count from 68k */ \
+    if (SekCycleCnt < SekCycleAim) \
+      elprintf(EL_32X, "slice %d", slice); \
+    if (!(Pico32x.emu_flags & (P32XF_SSH2POLL|P32XF_SSH2VPOLL))) \
+      sh2_execute(&ssh2, CYCLES_M68K2SH2(slice)); \
+    if (!(Pico32x.emu_flags & (P32XF_MSH2POLL|P32XF_MSH2VPOLL))) \
+      sh2_execute(&msh2, CYCLES_M68K2SH2(slice)); \
+  } \
+}
 
 #define STEP_68K 24
 #define CPUS_RUN_LOCKSTEP(m68k_cycles,s68k_cycles) \
 { \
   int i; \
   for (i = 0; i <= (m68k_cycles) - STEP_68K; i += STEP_68K) { \
-    SekRunM68k(STEP_68K); \
+    run_m68k(STEP_68K); \
     if (!(Pico32x.emu_flags & (P32XF_MSH2POLL|P32XF_MSH2VPOLL))) \
       sh2_execute(&msh2, CYCLES_M68K2SH2(STEP_68K)); \
     if (!(Pico32x.emu_flags & (P32XF_SSH2POLL|P32XF_SSH2VPOLL))) \
@@ -150,15 +152,15 @@ static __inline void SekRunM68k(int cyc)
   } \
   /* last step */ \
   i = (m68k_cycles) - i; \
-  SekRunM68k(i); \
+  run_m68k(i); \
   if (!(Pico32x.emu_flags & (P32XF_MSH2POLL|P32XF_MSH2VPOLL))) \
     sh2_execute(&msh2, CYCLES_M68K2SH2(i)); \
   if (!(Pico32x.emu_flags & (P32XF_SSH2POLL|P32XF_SSH2VPOLL))) \
     sh2_execute(&ssh2, CYCLES_M68K2SH2(i)); \
 }
 
-//#define CPUS_RUN CPUS_RUN_SIMPLE
+#define CPUS_RUN CPUS_RUN_SIMPLE
-#define CPUS_RUN CPUS_RUN_LOCKSTEP
+//#define CPUS_RUN CPUS_RUN_LOCKSTEP
 
 #include "../pico_cmn.c"
 

pico/32x/memory.c

@@ -1,7 +1,7 @@
 #include "../pico_int.h"
 #include "../memory.h"
 
-#if 1
+#if 0
 #undef ash2_end_run
 #undef SekEndRun
 #define ash2_end_run(x)
@@ -159,7 +159,7 @@ static u32 p32x_reg_read16(u32 a)
     return sh2_comm_faker(a);
 #else
   if ((a & 0x30) == 0x20 && p32x_poll_detect(&m68k_poll, a, SekCyclesDoneT(), 0)) {
-    SekEndRun(16);
+    SekEndTimeslice(16);
   }
 #endif
 
@@ -220,9 +220,11 @@ static void p32x_reg_write8(u32 a, u32 d)
   if ((a & 0x30) == 0x20) {
     u8 *r8 = (u8 *)r;
     r8[a ^ 1] = d;
-    if (p32x_poll_undetect(&sh2_poll[0], 0) || p32x_poll_undetect(&sh2_poll[1], 0))
+    p32x_poll_undetect(&sh2_poll[0], 0);
-      // if some SH2 is busy waiting, it needs to see the result ASAP
+    p32x_poll_undetect(&sh2_poll[1], 0);
-      SekEndRun(16);
+    // if some SH2 is busy waiting, it needs to see the result ASAP
+    if (SekCyclesLeftNoMCD > 32)
+      SekEndRun(32);
     return;
   }
 }
@@ -265,9 +267,11 @@ static void p32x_reg_write16(u32 a, u32 d)
   // comm port
   else if ((a & 0x30) == 0x20 && r[a / 2] != d) {
     r[a / 2] = d;
-    if (p32x_poll_undetect(&sh2_poll[0], 0) || p32x_poll_undetect(&sh2_poll[1], 0))
+    p32x_poll_undetect(&sh2_poll[0], 0);
-      // if some SH2 is busy waiting, it needs to see the result ASAP
+    p32x_poll_undetect(&sh2_poll[1], 0);
-      SekEndRun(16);
+    // same as for w8
+    if (SekCyclesLeftNoMCD > 32)
+      SekEndRun(32);
     return;
   }
   // PWM

pico/pico_cmn.c

@@ -5,6 +5,7 @@
 #define CYCLES_M68K_VINT_LAG  68
 #define CYCLES_M68K_ASD      148
 #define CYCLES_S68K_LINE     795
+#define CYCLES_S68K_VINT_LAG 111
 #define CYCLES_S68K_ASD      241
 
 // pad delay (for 6 button pads)
@@ -178,7 +179,7 @@ static int PicoFrameHints(void)
   // there must be a delay after vblank bit is set and irq is asserted (Mazin Saga)
   // also delay between F bit (bit 7) is set in SR and IRQ happens (Ex-Mutants)
   // also delay between last H-int and V-int (Golden Axe 3)
-  SekRunM68k(CYCLES_M68K_VINT_LAG);
+  CPUS_RUN(CYCLES_M68K_VINT_LAG, CYCLES_S68K_VINT_LAG);
 
   if (pv->reg[1]&0x20) {
     elprintf(EL_INTS, "vint: @ %06x [%i]", SekPc, SekCycleCnt);
@@ -201,7 +202,7 @@ static int PicoFrameHints(void)
   // Run scanline:
   if (Pico.m.dma_xfers) SekCyclesBurn(CheckDMA());
   CPUS_RUN(CYCLES_M68K_LINE - CYCLES_M68K_VINT_LAG - CYCLES_M68K_ASD,
-    CYCLES_S68K_LINE - CYCLES_S68K_ASD);
+    CYCLES_S68K_LINE - CYCLES_S68K_VINT_LAG - CYCLES_S68K_ASD);
 
 #ifdef PICO_CD
   update_chips();

pico/pico_int.h

@@ -231,7 +231,7 @@ typedef void (z80_write_f)(unsigned int a, unsigned char data);
 #include "cpu/sh2mame/sh2.h"
 
 SH2 msh2, ssh2;
-#define ash2_end_run(after) sh2_icount = after
+#define ash2_end_run(after) if (sh2_icount > (after)) sh2_icount = after
 #define ash2_cycles_done() (sh2->cycles_aim - sh2_icount)
 
 #define sh2_pc(c)     (c) ? ssh2.ppc : msh2.ppc