core, testpico fixes

pico/memory.c

@@ -1039,10 +1039,24 @@ static void m68k_mem_setup(void)
 static int get_scanline(int is_from_z80)
 {
   if (is_from_z80) {
-    int mclk_z80 = (z80_cyclesLeft<0 ? Pico.t.z80c_aim : z80_cyclesDone()) * 15;
+    // ugh... compute by dividing cycles since frame start by cycles per line
-    int mclk_line = Pico.t.z80_scanline * 488 * 7;
+    // need some fractional resolution here, else there may be an extra line
-    while (mclk_z80 - mclk_line >= 488 * 7)
+    int cycles_line = cycles_68k_to_z80(488 << 8); // cycles per line, as Q8
-      Pico.t.z80_scanline++, mclk_line += 488 * 7;
+    int cycles_z80 = (z80_cyclesLeft<0 ? Pico.t.z80c_aim:z80_cyclesDone())<<8;
+    int cycles = cycles_line * Pico.t.z80_scanline;
+    // approximation by multiplying with inverse
+    if (cycles_z80 - cycles >= 2*cycles_line) {
+      // compute 1/cycles_line, storing the result to avoid future dividing
+      static int cycles_line_o, cycles_line_i;
+      if (cycles_line_o != cycles_line)
+        { cycles_line_o = cycles_line, cycles_line_i = (1<<22) / cycles_line; }
+      // compute lines = diff/cycles_line = diff*(1/cycles_line)
+      int lines = ((cycles_z80 - cycles) * cycles_line_i) >> 22;
+      Pico.t.z80_scanline += lines, cycles += cycles_line * lines;
+    }
+    // handle any rounding leftover
+    while (cycles_z80 - cycles >= cycles_line)
+      Pico.t.z80_scanline ++, cycles += cycles_line;
     return Pico.t.z80_scanline;
   }
 
@@ -1342,7 +1356,8 @@ void PicoWrite16_32x(u32 a, u32 d) {}
 static unsigned char z80_md_vdp_read(unsigned short a)
 {
   if ((a & 0xff00) == 0x7f00) {
-    z80_subCLeft(3);
+    static int f; f = (f&0xff) + 0x8c; // 0.6
+    z80_subCLeft(2+(f>>8)); // 3.3 per kabuto, but notaz' test implies 2.6 ?!?
     Pico.t.z80_buscycles += 7;
 
     switch (a & 0x0d)
@@ -1368,9 +1383,10 @@ static unsigned char z80_md_bank_read(unsigned short a)
   unsigned char ret;
 
   // account for 68K bus access on both CPUs.
+  static int f; f = (f&0xff) + 0x4c; // 0.3
+  z80_subCLeft(3+(f>>8)); // 3.3 per kabuto
   // don't use SekCyclesBurn(7) 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
-  z80_subCLeft(3);
   Pico.t.z80_buscycles += 7;
 
   addr68k = Pico.m.z80_bank68k << 15;
@@ -1413,9 +1429,10 @@ static void z80_md_bank_write(unsigned int a, unsigned char data)
   unsigned int addr68k;
 
   // account for 68K bus access on both CPUs.
+  static int f; f = (f&0xff) + 0x4c; // 0.3
+  z80_subCLeft(3+(f>>8)); // 3.3 per kabuto
   // don't use SekCyclesBurn(7) 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
-  z80_subCLeft(3);
   Pico.t.z80_buscycles += 7;
 
   addr68k = Pico.m.z80_bank68k << 15;

pico/pico_int.h

@@ -205,7 +205,7 @@ extern struct DrZ80 drZ80;
   (Pico.t.z80c_aim - z80_cyclesLeft)
 
 // one line has 488 68K cycles and 228 Z80 cycles, 228/488*8192=3827
-#define cycles_68k_to_z80(x) ((x) * 3857 >> 13)
+#define cycles_68k_to_z80(x) ((x) * 3847 >> 13)
 
 // ----------------------- SH2 CPU -----------------------
 

pico/videoport.c

@@ -17,7 +17,8 @@ enum { clkdiv = 2 };    // CPU clock granularity: one of 1,2,4,8
 
 // VDP Slot timing, taken from http://gendev.spritesmind.net/
 //     forum/viewtopic.php?f=22&t=851&sid=d5701a71396ee7f700c74fb7cd85cb09
-// Thank you very much for the great work, Nemesis!
+//     http://plutiedev.com/mirror/kabuto-hardware-notes
+// Thank you very much for the great work, Nemesis, Kabuto!
 
 // Slot clock is sysclock/20 for h32 and sysclock/16 for h40.
 // One scanline is 63.7us/63.5us (h32/h40) long which is 488.6/487.4 68k cycles.
@@ -30,13 +31,15 @@ enum { slcpu = 488 };
 // is generated also depends on the video mode.
 enum { hint32 = 0x85, gapstart32 = 0x94, gapend32 = 0xe9};
 enum { hint40 = 0xa5, gapstart40 = 0xb7, gapend40 = 0xe5};
+// XXX Kabuto says gapend40 is 0xe4, but then a line would've 211 slots, while
+// it's 210 in all other sources I looked at?
 
 // The horizontal sync period (HBLANK) is 30/37 slots (h32/h40):
 // h32: 4 slots front porch (1.49us), 13 HSYNC (4.84us), 13 back porch (4.84us)
 // h40: 5 slots front porch (1.49us), 16 HSYNC (4.77us), 16 back porch (4.77us)
-// HBLANK starts in slot 0x93/0xb4, according to Nemesis' measurements.
+// HBLANK starts in slot 0x93/0xb3 and ends after slot 0x05 (from Kabuto's doc)
 enum { hboff32 = 0x93-hint32, hblen32 = 0xf8-(gapend32-gapstart32)-hint32};//30
-enum { hboff40 = 0xb4-hint40, hblen40 = 0xf8-(gapend40-gapstart40)-hint40};//37
+enum { hboff40 = 0xb3-hint40, hblen40 = 0xf8-(gapend40-gapstart40)-hint40};//37
 
 // number of slots in a scanline
 #define slots32	(0x100-(gapend32-gapstart32)) // 171
@@ -1030,7 +1033,8 @@ static u32 VideoSr(const struct PicoVideo *pv)
 {
   unsigned int hp = pv->reg[12]&1 ? hboff40*488/slots40 : hboff32*488/slots32;
   unsigned int hl = pv->reg[12]&1 ? hblen40*488/slots40 : hblen32*488/slots32;
-  unsigned int c = SekCyclesDone() - Pico.t.m68c_line_start;
+  // XXX -2 is to please notaz' testpico, but why is this?
+  unsigned int c = SekCyclesDone()-2 - Pico.t.m68c_line_start;
   u32 d;
 
   PicoVideoFIFOSync(c);