Merge branch 'libretro:master' into master

Makefile

@@ -48,6 +48,8 @@ ifeq "$(ASAN)" "1"
 else
 ifeq "$(DEBUG)" "0"
 	CFLAGS += -O3 -DNDEBUG
+else
+	CFLAGS += -O1
 endif
 endif
 	LD = $(CC)

configure

@@ -221,7 +221,7 @@ fi
 #fi
 
 if [ -z "$ARCH" ]; then
-  ARCH=`$CC -dumpmachine | awk -F '-' '{print $1}'`
+  ARCH=`$CC $MFLAGS $CFLAGS -dumpmachine | awk -F '-' '{print $1}'`
 fi
 
 # CPU/ABI stuff first, else compile test may fail

cpu/drc/emit_x86.c

@@ -934,11 +934,12 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI,	// x86-64,i386 common
 
 #define emith_call_reg(r) do { \
 	EMIT_REX_IF(0, 0, r); \
-	EMIT_OP_MODRM(0xff, 3, 2, (r)&7); \
+	EMIT_OP_MODRM64(0xff, 3, 2, r); \
 } while (0)
 
 #define emith_abicall_ctx(offs) do { \
-	EMIT_OP_MODRM(0xff, 2, 2, CONTEXT_REG); \
+	EMIT_REX_IF(0, 0, CONTEXT_REG); \
+	EMIT_OP_MODRM64(0xff, 2, 2, CONTEXT_REG); \
 	EMIT(offs, u32); \
 } while (0)
 
@@ -955,11 +956,12 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI,	// x86-64,i386 common
 
 #define emith_jump_reg(r) do { \
 	EMIT_REX_IF(0, 0, r); \
-	EMIT_OP_MODRM(0xff, 3, 4, (r)&7); \
+	EMIT_OP_MODRM64(0xff, 3, 4, r); \
 } while (0)
 
 #define emith_jump_ctx(offs) do { \
-	EMIT_OP_MODRM(0xff, 2, 4, CONTEXT_REG); \
+	EMIT_REX_IF(0, 0, CONTEXT_REG); \
+	EMIT_OP_MODRM64(0xff, 2, 4, CONTEXT_REG); \
 	EMIT(offs, u32); \
 } while (0)
 
@@ -1209,13 +1211,12 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI,	// x86-64,i386 common
 } while (0)
 
 #define emith_sh2_rcall(a, tab, func, mask) do { \
+	int scale_ = PTR_SCALE <= 2 ? PTR_SCALE : 2; \
 	emith_lsr(mask, a, SH2_READ_SHIFT); \
+	if (PTR_SCALE > scale_) emith_lsl(mask, mask, PTR_SCALE-scale_); \
 	EMIT_XREX_IF(1, tab, tab, mask); \
 	EMIT_OP_MODRM64(0x8d, 0, tab, 4); \
-	EMIT_SIB64(PTR_SCALE, mask, tab); /* lea tab, [tab + mask * {4,8}] */ \
-	EMIT_XREX_IF(1, tab, tab, mask); \
-	EMIT_OP_MODRM64(0x8d, 0, tab, 4); \
-	EMIT_SIB64(PTR_SCALE, mask, tab); /* lea tab, [tab + mask * {4,8}] */ \
+	EMIT_SIB64(scale_+1, mask, tab); /* lea tab, [tab + mask*(2*scale)] */ \
 	EMIT_REX_IF(1, func,  tab); \
 	emith_deref_modrm(0x8b, 0, func, tab); /* mov func, [tab] */ \
 	EMIT_REX_IF(0, mask, tab); \

cpu/sh2/compiler.c

@@ -194,7 +194,7 @@ static char sh2dasm_buff[64];
 		(ulong)(sh2)->r[8], (ulong)(sh2)->r[9], (ulong)(sh2)->r[10], (ulong)(sh2)->r[11], \
 		(ulong)(sh2)->r[12], (ulong)(sh2)->r[13], (ulong)(sh2)->r[14], (ulong)(sh2)->r[15]); \
 	printf("%csh2 pc-ml %08lx %08lx %08lx %08lx %08lx %08lx %08lx %08lx\n", ms, \
-		(ulong)(sh2)->pc, (ulong)(sh2)->ppc, (ulong)(sh2)->pr, (ulong)(sh2)->sr&0xfff, \
+		(ulong)(sh2)->pc, (ulong)(sh2)->ppc, (ulong)(sh2)->pr, (ulong)(sh2)->sr, \
 		(ulong)(sh2)->gbr, (ulong)(sh2)->vbr, (ulong)(sh2)->mach, (ulong)(sh2)->macl); \
 	printf("%csh2 tmp-p  %08x %08x %08x %08x %08x %08lx %08x %08x\n", ms, \
 		(sh2)->drc_tmp, (sh2)->irq_cycles, \
@@ -269,12 +269,14 @@ static void REGPARM(3) *sh2_drc_log_entry(void *block, SH2 *sh2, u32 sr)
         SH2_DUMP(&fsh2, "file");
         SH2_DUMP(sh2, "current");
         SH2_DUMP(&csh2[idx][0], "previous");
+        SH2_DUMP(&csh2[idx][1], "previous");
 	char *ps = (char *)sh2, *pf = (char *)&fsh2;
 	for (idx = 0; idx < offsetof(SH2, read8_map); idx += sizeof(u32))
 		if (*(u32 *)(ps+idx) != *(u32 *)(pf+idx))
 			printf("diff reg %ld\n",(long)idx/sizeof(u32));
         exit(1);
       }
+      memcpy(&csh2[idx][1], &csh2[idx][0], offsetof(SH2, poll_cnt)+4);
       csh2[idx][0] = fsh2;
     }
   }
@@ -4649,6 +4651,7 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
           FLUSH_CYCLES(sr);
           rcache_get_reg_arg(0, GET_Rn(), NULL);
           tmp = emit_memhandler_read(0);
+          sr  = rcache_get_reg(SHR_SR, RC_GR_RMW, NULL);
           emith_clr_t_cond(sr);
           emith_cmp_r_imm(tmp, 0);
           emith_set_t_cond(sr, DCOND_EQ);
@@ -5262,8 +5265,6 @@ static void sh2_generate_utils(void)
   host_arg2reg(arg1, 1);
   host_arg2reg(arg2, 2);
   host_arg2reg(arg3, 3);
-  emith_move_r_r(arg0, arg0); // nop
-  emith_flush();
 
   // sh2_drc_write8(u32 a, u32 d)
   sh2_drc_write8 = (void *)tcache_ptr;
@@ -5645,7 +5646,6 @@ static void sh2_smc_rm_blocks(u32 a, int len, int tcache_id, int free)
   u32 start_lit, end_lit;
   struct block_desc *block;
   int removed = 0, rest;
-  u32 _a = a;
 
   // ignore cache-through
   a &= wtmask;

cpu/sh2/compiler.h

@@ -50,9 +50,10 @@ u16 scan_block(u32 base_pc, int is_slave, u8 *op_flags, u32 *end_pc,
 #elif defined(__powerpc__) || defined(__ppc__) || defined(__PPC__)
 #define	DRC_SR_REG	"r28"
 #define DRC_REG_LL	0	// no ABI for __ILP32__
-#elif defined(__i386__)
-#define	DRC_SR_REG	"edi"
-#define DRC_REG_LL	0	// 32 bit
+//i386 only has 8 registers and reserving one of them causes too much spilling
+//#elif defined(__i386__)
+//#define	DRC_SR_REG	"edi"
+//#define DRC_REG_LL	0	// 32 bit
 #elif defined(__x86_64__)
 #define	DRC_SR_REG	"rbx"
 #define DRC_REG_LL	(__ILP32__ || _WIN32)

pico/draw.c

@@ -566,7 +566,7 @@ static void DrawLayer(int plane_sh, u32 *hcache, int cellskip, int maxcells,
     ts.nametab+=(ts.line>>4)<<shift[width];
 
     DrawStripInterlace(&ts, plane_sh);
-  } else if( pvid->reg[11]&4) {
+  } else if (pvid->reg[11]&4) {
     // shit, we have 2-cell column based vscroll
     // luckily this doesn't happen too often
     ts.line=ymask|(shift[width]<<24); // save some stuff instead of line
@@ -2125,7 +2125,6 @@ void PicoDrawSync(int to, int off, int on)
   {
     int width2 = (est->Pico->video.reg[12]&1) ? 160 : 128;
 
-    // technically, VDP starts active display output at slot 12
     if (unlikely(on|off) && (off >= width2 ||
           // hack for timing inaccuracy, if on/off near borders
           (off && off <= 24) || (on < width2 && on >= width2-24)))
@@ -2178,7 +2177,7 @@ void PicoDrawBgcDMA(u16 *base, u32 source, u32 mask, int dlen, int sl)
   BgcDMAoffs = 0;
 
   // handle slot offset in 1st line
-  if (sl-12 > 0) // active display output only starts at slot 12
+  if (sl-12 > 0)
     BgcDMAoffs = 2*(sl-12);
   else if (sl < 0) { // DMA starts before active display
     BgcDMAsrc += 2*-sl;

pico/draw_arm.S

@@ -461,7 +461,6 @@ DrawLayer:
     subs    r10,r10, #0x01000000
     bmi     .dsloop_exit
 
-.dsloop_enter:
     and     r7, r5, r8
     add     r7, lr, r7, lsl #1 @ Pico.vram+((tilex&ts->xmask) as halfwords)
     ldrh    r7, [r7, r12]      @ r7=code (int, but from unsigned, no sign extend)
@@ -1385,6 +1384,9 @@ DrawWindow:
     and     r10, r10, #7
     mov     r10, r10, lsl #1      @ r10=ty
 
+    ldr     r6, [r11, #OFS_EST_rendstatus]
+    ldr     lr, [r11, #OFS_EST_PicoMem_vram]
+
     mov     r12, r12, lsl #10
 
     tst     r4, #1                @ 40 cell mode?
@@ -1394,12 +1396,6 @@ DrawWindow:
     addeq   r12, r12, r5, lsl #6  @ nametab
     add     r12, r12, r0, lsl #2  @ +starttile
 
-    ldr     lr, [r11, #OFS_EST_PicoMem_vram]
-    ldr     r6, [r11, #OFS_EST_rendstatus]
-
-    @ fetch the first code now
-    ldrh    r7, [lr, r12]
-
     ands    r6, r6, #PDRAW_WND_DIFF_PRIO
     cmpeq   r2, #1                @ prio && !(rendstatus & WND_DIFF_PRIO)?
     ldmeqfd sp!, {r4-r11,pc}      @ yes, assume that whole window uses same priority
@@ -1412,20 +1408,18 @@ DrawWindow:
     @ cache some stuff to avoid mem access
     ldr     r11, [r11, #OFS_EST_HighCol]
     mov     r8, r8, lsl #1        @ cells
-    add     r11,r11,#8
     mvn     r9, #0                @ r9=prevcode=-1
-    add     r1, r11, r0, lsl #4   @ r1=pdest
+    add     r1, r11, r0, lsl #4   @ r1=pdest=HighCol+starttile (+8 added in loop)
     mov     r0, #0xf
-    b       .dwloop_enter
 
     @ r4,r5 are scratch in this loop
 .dwloop:
     add     r1, r1, #8
 .dwloop_nor1:
-    add     r12, r12, #2    @ halfwords
     ldrh    r7, [lr, r12]   @ r7=code (int, but from unsigned, no sign extend)
+    add     r12, r12, #2    @ halfwords
     subs    r8, r8, #1
-    beq     .dwloop_end     @ done
+    bmi     .dwloop_end     @ done
 
     eor     r5, r6, r7, lsr #15
     tst     r5, #1
@@ -1435,7 +1429,6 @@ DrawWindow:
     cmp     r7, r9
     beq     .dw_samecode    @ we know stuff about this tile already
 
-.dwloop_enter:
     mov     r9, r7          @ remember code
 
     movs    r2, r9, lsl #20 @ if (code&0x1000)

pico/media.c

@@ -350,7 +350,7 @@ enum media_type_e PicoLoadMedia(const char *filename,
   // simple test for GG. Do this here since m.hardware is nulled in Insert
   if ((PicoIn.AHW & PAHW_SMS) && !PicoIn.hwSelect) {
     const char *ext = NULL;
-    if (rom_file && rom_file->ext && (*rom_file->ext != '\0')) {
+    if (rom_file && (*rom_file->ext != '\0')) {
       ext = rom_file->ext;
     }
     else if ((ext = strrchr(filename, '.'))) {

pico/memory.c

@@ -565,7 +565,6 @@ void NOINLINE ctl_write_z80reset(u32 d)
         pprof_end_sub(m68k);
       }
       Pico.t.z80_busdelay &= 0xff; // also resets bus request
-      Pico.video.status &= ~PVS_Z80WAIT;
       YM2612ResetChip();
       timers_reset();
     }
@@ -1174,7 +1173,7 @@ static int ym2612_write_local(u32 a, u32 d, int is_from_z80)
 
       // the busy flag in the YM2612 status is actually a 32 cycle timer
       // (89.6 Z80 cycles), triggered by any write to the data port.
-      Pico.t.ym2612_busy = (cycles + 90) << 8; // Q8 for convenience
+      Pico.t.ym2612_busy = (cycles << 8) + YMBUSY_ZCYCLES; // Q8 for convenience
 
       switch (addr)
       {
@@ -1284,9 +1283,11 @@ static u32 ym2612_read_local_68k(void)
 void ym2612_pack_state(void)
 {
   // timers are saved as tick counts, in 16.16 int format
-  int tac, tat = 0, tbc, tbt = 0;
+  int tac, tat = 0, tbc, tbt = 0, busy = 0;
   tac = 1024 - ym2612.OPN.ST.TA;
   tbc = 256  - ym2612.OPN.ST.TB;
+  if (Pico.t.ym2612_busy > 0)
+    busy = cycles_z80_to_68k(Pico.t.ym2612_busy);
   if (Pico.t.timer_a_next_oflow != TIMER_NO_OFLOW)
     tat = (int)((double)(Pico.t.timer_a_step - Pico.t.timer_a_next_oflow)
           / (double)Pico.t.timer_a_step * tac * 65536);
@@ -1301,12 +1302,12 @@ void ym2612_pack_state(void)
     YM2612PicoStateSave2_940(tat, tbt);
   else
 #endif
-    YM2612PicoStateSave2(tat, tbt);
+    YM2612PicoStateSave2(tat, tbt, busy);
 }
 
 void ym2612_unpack_state(void)
 {
-  int i, ret, tac, tat, tbc, tbt;
+  int i, ret, tac, tat, tbc, tbt, busy = 0;
   YM2612PicoStateLoad();
 
   // feed all the registers and update internal state
@@ -1336,12 +1337,13 @@ void ym2612_unpack_state(void)
     ret = YM2612PicoStateLoad2_940(&tat, &tbt);
   else
 #endif
-    ret = YM2612PicoStateLoad2(&tat, &tbt);
+    ret = YM2612PicoStateLoad2(&tat, &tbt, &busy);
   if (ret != 0) {
     elprintf(EL_STATUS, "old ym2612 state");
     return; // no saved timers
   }
 
+  Pico.t.ym2612_busy = cycles_68k_to_z80(busy);
   tac = (1024 - ym2612.OPN.ST.TA) << 16;
   tbc = (256  - ym2612.OPN.ST.TB) << 16;
   if (ym2612.OPN.ST.mode & 1)
@@ -1370,11 +1372,6 @@ void PicoWrite16_32x(u32 a, u32 d) {}
 static void access_68k_bus(int delay) // bus delay as Q8
 {
   // TODO: if the 68K is in DMA wait, Z80 has to wait until DMA ends
-  if (Pico.video.status & (PVS_CPUWR|PVS_CPURD)) {
-    z80_subCLeft(z80_cyclesLeft); // rather rough on both condition and action
-    // TODO the next line will cause audio lag in Overdrive 2 demo?
-    //Pico.video.status |= PVS_Z80WAIT;
-  }
 
   // 68k bus access delay for z80. The fractional part needs to be accumulated
   // until an additional cycle is full. That is then added to the integer part.
@@ -1383,7 +1380,7 @@ static void access_68k_bus(int delay) // bus delay as Q8
   Pico.t.z80_busdelay &= 0xff; // leftover cycle fraction
   // don't use SekCyclesBurn() here since the Z80 doesn't run in cycle lock to
   // the 68K. Count the stolen cycles to be accounted later in the 68k CPU runs
-  Pico.t.z80_buscycles += 8;
+  Pico.t.z80_buscycles += 8; // TODO <=8.4 for Rick 2, but >=8.9 for misc_test
 }
 
 static unsigned char z80_md_vdp_read(unsigned short a)

pico/pico.c

@@ -239,6 +239,7 @@ void PicoLoopPrepare(void)
     Pico.t.vcnt_wrap = 0xEB;
     Pico.t.vcnt_adj = 6;
   }
+  PicoVideoFIFOMode(Pico.video.reg[1]&0x40, Pico.video.reg[12]&1);
 
   Pico.m.dirtyPal = 1;
   rendstatus_old = -1;

pico/pico_cmn.c

@@ -125,12 +125,6 @@ static void do_timing_hacks_start(struct PicoVideo *pv)
   int cycles = PicoVideoFIFOHint();
 
   SekCyclesBurn(cycles); // prolong cpu HOLD if necessary
-  if (pv->status & PVS_Z80WAIT) {
-    Pico.t.z80c_cnt += cycles_68k_to_z80(cycles);
-    if (!(pv->status & (PVS_CPUWR|PVS_CPURD)))
-      pv->status &= ~PVS_Z80WAIT;
-  }
-
   // XXX how to handle Z80 bus cycle stealing during DMA correctly?
   if ((Pico.t.z80_buscycles -= cycles) < 0)
     Pico.t.z80_buscycles = 0;

pico/pico_int.h

@@ -209,6 +209,7 @@ extern struct DrZ80 drZ80;
 
 // 68k clock = OSC/7, z80 clock = OSC/15, 68k:z80 ratio = 7/15 = 3822.9/8192
 #define cycles_68k_to_z80(x) ((x) * 3823 >> 13)
+#define cycles_z80_to_68k(x) ((x) * 8777 >> 12)
 
 // ----------------------- SH2 CPU -----------------------
 
@@ -299,7 +300,6 @@ extern SH2 sh2s[2];
 #define PVS_DMAFILL   (1 << 20) // DMA fill is waiting for fill data
 #define PVS_DMABG     (1 << 21) // background DMA operation is running
 #define PVS_FIFORUN   (1 << 22) // FIFO is processing
-#define PVS_Z80WAIT   (1 << 23) // Z80 blocked by VDP DMA
 
 struct PicoVideo
 {
@@ -898,10 +898,12 @@ void ym2612_unpack_state(void);
 
 #define TIMER_NO_OFLOW 0x70000000
 
-// tA =    72 * (1024 - TA) / M, with M = mclock/2
-#define TIMER_A_TICK_ZCYCLES cycles_68k_to_z80(256LL*   72*2) // Q8
-// tB = 16*72 * ( 256 - TB) / M
-#define TIMER_B_TICK_ZCYCLES cycles_68k_to_z80(256LL*16*72*2) // Q8
+// tA =    24*3 * (1024 - TA) / M, with M = mclock/2
+#define TIMER_A_TICK_ZCYCLES cycles_68k_to_z80(256LL*   24*3*2) // Q8
+// tB = 16*24*3 * ( 256 - TB) / M
+#define TIMER_B_TICK_ZCYCLES cycles_68k_to_z80(256LL*16*24*3*2) // Q8
+// busy =  32*3 / M
+#define YMBUSY_ZCYCLES       cycles_68k_to_z80(256LL*   32*3*2) // Q8
 
 #define timers_cycle(ticks) \
   if (Pico.t.ym2612_busy > 0) \

pico/sek.c

@@ -221,7 +221,8 @@ PICO_INTERNAL void SekPackCpu(unsigned char *cpu, int is_sub)
     *(u32 *)(cpu+0x50) = SekCycleCntS68k;
     *(s16 *)(cpu+0x4e) = SekCycleCntS68k - SekCycleAimS68k;
   } else {
-    *(u32 *)(cpu+0x50) = Pico.t.m68c_cnt;
+    *(u32 *)(cpu+0x50) = Pico.t.m68c_cnt + Pico.t.z80_buscycles +
+                          ((Pico.t.refresh_delay + (1<<14)/2) >> 14);
     *(s16 *)(cpu+0x4e) = Pico.t.m68c_cnt - Pico.t.m68c_aim;
   }
 }
@@ -266,6 +267,8 @@ PICO_INTERNAL void SekUnpackCpu(const unsigned char *cpu, int is_sub)
   } else {
     Pico.t.m68c_cnt = *(u32 *)(cpu+0x50);
     Pico.t.m68c_aim = Pico.t.m68c_cnt - *(s16 *)(cpu+0x4e);
+    Pico.t.z80_buscycles = 0;
+    Pico.t.refresh_delay = 0;
   }
 }