Make JR ignore page boundaries; relative jumps are now twenty times as fast.

3 files changed, 29 insertions, 22 deletions

diff --git a/README.markdown b/README.markdown
index 2c39862..7f1c6b5 100644
--- a/README.markdown
+++ b/README.markdown
@@ -44,7 +44,7 @@ instruction with a fixed 42 cycle overhead:
 
 (Using techniques borrowed from
 [Tezxas](http://tezxas.ticalc.org/technica.htm) I will be able to get
-this to 30 cycles.)
+this to 26 cycles.)
 
 From there, an instruction will take anywhere from 0 to, well, lots of
 additional cycles.  Generally, however, it will take under 50, for 92
@@ -66,3 +66,11 @@ expected.
 * [More z80 instruction set reference](http://nemesis.lonestar.org/computers/tandy/software/apps/m4/qd/opcodes.html)
 * [Details on flags and other side effects](http://www.gaby.de/z80/z80code.htm)
 
+Incompatibilities with Real Hardware
+------------------------------------
+
+1. PC-relative accesses (reads, writes, jumps) across page boundaries
+   may not work.  This is apparently a rare enough occurence that
+   ignoring it is reasonable.
+
+2. 
diff --git a/instruction-timing.gnumeric b/instruction-timing.gnumeric
index 0965839..58a6f30 100644
--- a/instruction-timing.gnumeric
+++ b/instruction-timing.gnumeric
diff --git a/opcodes.asm.m4 b/opcodes.asm.m4
index 051d33e..fe92245 100644
--- a/opcodes.asm.m4
+++ b/opcodes.asm.m4
@@ -344,18 +344,15 @@ OPCODE(0f,«
 	;;  if B not zero
 	;; No flags
 	;; 24 bytes
-	;; take: 22+4+ 8+8+4+300+8+94+4+22 = 474
-	;; skip: 22+4+10+               22 = 58
+	;; take: 22+4+ 8+8+4+8+22 = 76
+	;; skip: 22+4+10+      22 = 58
 OPCODE(10,«
 	LOHI	ebc
 	subq.b	#1,ebc
 	beq.s	local(end)	; slooooow
 	FETCHBI	d1
-	move.l	epc,a0
-	bsr	underef
-	add.w	d1,d0		; ??? Can I avoid underef/deref cycle?
-	bsr	deref
-	move.l	a0,epc
+	ext.w	d1
+	add.w	d1,epc
 local(end):
 	HILO	ebc
 	»,,,32)
@@ -430,14 +427,18 @@ OPCODE(17,«
 	;; PC <- immed.b
 	;; Branch relative by a signed immediate byte
 	;; No flags
+	;; 20 cycles
+
+	;; XXX
+	;; Yes, I can avoid the underef/deref cycle.  To do so, put a
+	;; sled of emulator trap instructions on either side of each
+	;; 16k page.  When that trap is executed, undo the shortcut
+	;; and redo it by the book.
+
 OPCODE(18,«
 	clr.w	d1
 	FETCHBI	d1
-	move.l	epc,a0
-	bsr	underef
-	add.w	d0,d1		; ??? Can I avoid underef/deref cycle?
-	bsr	deref
-	move.l	a0,epc
+	add.w	d1,epc
 	»)
 				;nok
 
@@ -505,8 +506,8 @@ OPCODE(1f,«
 	;;  PC <- PC+immed.b
 	;; No flags
 	;; 10 bytes
-	;; take: 40+10+422(=JR immed.b) = 472
-	;; skip: 40+12+12               =  64
+	;; take: 40+10+20(=JR immed.b) = 70
+	;; skip: 40+12+12              = 64
 OPCODE(20,«
 	bsr	f_norm_z
 	;; if the emulated Z flag is set, this will be clear
@@ -573,20 +574,20 @@ OPCODE(26,«
 	;; DAA
 	;; Decrement, adjust accum
 	;; http://www.z80.info/z80syntx.htm#DAA
-	;; Flags: oh lord they're fucked up
-	;; XXX DO THIS
+	;; Flags: ummm, go find a manual.
+	;; XXX
 	;; ? cycles
 OPCODE(27,«
 	F_PAR	eaf
 	»)
 				;nok
 
-	;; JR Z,immed.b
+	;; JR	Z,immed.b
 	;; If zero
 	;;  PC <- PC+immed.b
 	;; SPEED can be made faster
 	;; No flags
-	;; ~472 cycles
+	;; ~130 cycles
 OPCODE(28,«
 	bsr	f_norm_z
 	bne	emu_op_18
@@ -747,9 +748,7 @@ OPCODE(38,«
 OPCODE(39,«
 	move.l	esp,a0
 	bsr	underef
-	F_ADD_W	ehl,d0		; ??? Can I avoid underef/deref cycle?
-	bsr	deref
-	move.l	a0,esp
+	F_ADD_W	d0,ehl
 	»)
 				;nok