Improve Z280 DMA Memory Copy

The Z280 interbank memory copy uses DMA and did not properly handle scenarios where the source or destination memory range crossed over the banked/common memory boundary.  It added a bunch of code, but it is fixed now.  This seems to be required by CP/M 3.

Source/HBIOS/hbios.asm

@@ -561,7 +561,7 @@ HBX_MMA	.DB	0			; TEMPORARY STORAGE FOR REG A
 HBX_BNKCPY:
 #IF (MEMMGR == MM_Z280)
 	.DB	$ED,$71			; SC
-	.DW	Z280_BNKCPY		; SC PARAMETER
+	.DW	Z280_BNKCPYX		; SC PARAMETER
 	RET
 #ELSE
   #IF (CPUFAM == CPU_Z280)
@@ -2600,6 +2600,71 @@ HB_WDZ:
 INITSYS3:
 	CALL	PRTSUM			; PRINT UNIT/DEVICE SUMMARY TABLE
 ;
+#IF 0
+	CALL	NEWLINE
+	CALL	NEWLINE
+	CALL	NEWLINE
+
+	; SRC & DEST BELOW BND
+	CALL	NEWLINE
+	LD	HL,$4000
+	LD	DE,$5000
+	LD	BC,$3000
+	CALL	NEWLINE
+	CALL	REGDMP
+	CALL	Z280_BNKCPYX
+
+	; SRC & DEST ABOVE BND
+	CALL	NEWLINE
+	LD	HL,$8000
+	LD	DE,$9000
+	LD	BC,$1000
+	CALL	NEWLINE
+	CALL	REGDMP
+	CALL	Z280_BNKCPYX
+
+	; SRC CROSSOVER
+	CALL	NEWLINE
+	LD	HL,$7000
+	LD	DE,$9000
+	LD	BC,$2000
+	CALL	NEWLINE
+	CALL	REGDMP
+	CALL	Z280_BNKCPYX
+
+	; DEST CROSSOVER
+	CALL	NEWLINE
+	LD	HL,$9000
+	LD	DE,$7000
+	LD	BC,$2000
+	CALL	NEWLINE
+	CALL	REGDMP
+	CALL	Z280_BNKCPYX
+
+	; DOUBLE CROSSOVER
+	CALL	NEWLINE
+	LD	HL,$7800
+	LD	DE,$7000
+	LD	BC,$2000
+	CALL	NEWLINE
+	CALL	REGDMP
+	CALL	Z280_BNKCPYX
+
+	; DOUBLE CROSSOVER SINGLE BYTES
+	CALL	NEWLINE
+	LD	HL,$7FFE
+	LD	DE,$7FFF
+	LD	BC,$0500
+	CALL	NEWLINE
+	CALL	REGDMP
+	CALL	Z280_BNKCPYX
+
+	CALL	NEWLINE
+	CALL	NEWLINE
+	CALL	NEWLINE
+;
+#ENDIF
+;
 INITSYS4:
 ;
 #IF (MEMMGR == MM_Z280)
@@ -5209,8 +5274,115 @@ Z280_BNKSEL_LEN	.EQU	$ - Z280_BNKSEL
 ;
 ; Z280 BANK COPY (CALLED FROM PROXY)
 ;
+; USE Z280 PHYSICAL MEMORY DMA COPY TO PERFORM AN INTERBANK COPY.
+; COPY FROM (HB_SRCBNK):(HL) TO (HB_DSTBNK):(DE) FOR BC BYTES.  BOTH
+; HB_SRCBNK AND HB_DSTBNK MUST BE INITIALIZED PRIOR TO CALLING THIS
+; ROUTINE.
+;
+; ADDRESSES ARE TRANSLATED FROM LOGICAL (Z80) TO PHYSICAL (Z280) TO
+; SETUP THE DMA COPY PARAMETERS.  IF THE SOURCE OR DESTINATION RANGE
+; CROSSES OVER THE BANKED/COMMON BOUNDARY AT $8000, THEN SPECIAL STEPS
+; MUST BE TAKEN BECAUSE THE BANKED AND COMMON AEAS ARE PROBABLY NOT
+; SEQUENTIALLY LOCATED IN PHYSICAL MEMORY.  TWO ENTRY POINTS ARE
+; PROVIDED.  Z280_BNKCPY IS MUCH FASTER, BUT DOES NOT ACCOUNT FOR THE
+; COPY RANGES CROSSING OVER THE BANKED/COMMON BOUNDARY (WORKS GREAT
+; FOR ANY COPY KNOWN TO STAY WITHIN IT'S OWN AREA).  Z280_BNKCPYX
+; WILL HANDLE COPIES WHERE THE SOURCE AND/OR DESTINATION RANGES
+; CROSS OVER THE BANKED/COMMON MEMORY BOUNDARY.  IT DOES THIS BY
+; BREAKING UP THE COPY REQUESTS INTO MULTIPLE REQUESTS THAT ALL FIT
+; WITHIN A SINGLE BANKED/COMMON MEMORY SEGMENT AND CALLING Z280_BNKCPY
+; ITERATIVELY UNTIL THE COPY IS COMPLETE.
+;
+; THERE IS ESSENTIALLY NO PROTECTION FOR CALLING THESE ROUTINES WITH
+; INVALID PARAMETERS.  FOR EXAMPLE, A REQUEST TO COPY $2000 BYTES
+; STARTING AT $F000 EXCEEDS THE SIZE OF THE Z80 MEMORY SPACES AND
+; RESULTS IN UNDEFINED BEHAVIOR.
+;
+; THE COPY IS ALWAYS DONE FROM START TO END.  IF THE SOURCE AND
+; DESTINATION RANGES OVERLAP, THEN YOU MUST TAKE THIS INTO ACCOUNT.
+;
+; THE ROUTINE FUNCTIONS LOGICALLY LIKE THE Z80 LDIR INSTRUCTION.  ON
+; RETURN THE SOURCE (HL) AND DESTINATION (DE) REGISTERS WILL BE LEFT
+; POINTING TO THE NEXT BYTE THAT WOULD BE COPIED IF THE COPY ROUTINE
+; CONTINUED.  BC WILL BE 0.  AF IS UNDEFINED.
+;
 #IF (MEMMGR == MM_Z280)
 ;
+; ADJUST THE LENGTH OF THE COPY SUCH THAT BOTH THE SOURCE AND
+; DESTINATION RANGES DO NOT CROSS OVER THE BANKED/COMMON MEMORY
+; BOUNDARY.  CALL Z280_BNKCPY TO DO AS MANY ITERATIONS AS NEEDED TO
+; COMPLETE THE COPY.
+;
+;
+Z280_BNKCPYX:
+	LD	(Z280_BNKCPY_LEN),BC	; SAVE LENGTH
+;
+	CALL	Z280_BNKCPY_XOVER	; ADJUST FOR XOVER AS NEEDED
+	EX	DE,HL			; SWAP SOURCE/DEST
+	CALL	Z280_BNKCPY_XOVER	; ADJUST FOR XOVER AS NEEDED
+	EX	DE,HL			; SWAP BACK
+;
+	; DO THE WORK, SAVE THE LEN OF THIS ITERATION
+	PUSH	BC			; SAVE ITER LENGTH
+	CALL	Z280_BNKCPY		; DO THE WORK
+;
+	;;; *DEBUG* SIMULATE CALL TO Z280_BNKCPY
+	;;CALL	NEWLINE
+	;;CALL	REGDMP			; *DEBUG*
+	;;ADD	HL,BC			; INCREMENT SRC ADR BY COUNT
+	;;EX	DE,HL			; SWAP
+	;;ADD	HL,BC			; INCREMENT DST ADR BY COUNT
+	;;EX	DE,HL			; SWAP BACK
+	;;LD	BC,0			; COUNT IS NOW ZERO
+	;;; END *DEBUG*
+;
+	POP	BC			; RECOVER ITER LENGTH
+;
+	; ACCUNT FOR WORK ACTUALLY PERFORMED
+	PUSH	HL			; SAVE SOURCE ADR
+	LD	HL,(Z280_BNKCPY_LEN)	; GET PENDING LENGTH
+	OR	A			; CLEAR CARRY
+	SBC	HL,BC			; SUBTRACT WHAT WE DID
+	PUSH	HL			; MOVE NEW PENDING LEN
+	POP	BC			; TO BC
+	POP	HL			; RECOVER SOURCE ADR
+;
+	; SEE IF WE NEED TO ITERATE
+	LD	A,B			; IS LENGTH
+	OR	C			; ... NOW ZERO?
+	RET	Z			; IF SO, ALL DONE
+	JR	Z280_BNKCPYX		; ELSE ITERATE UNTIL DONE
+;
+Z280_BNKCPY_LEN	.DW	0		; TEMP STORAGE FOR BC
+;
+Z280_BNKCPY_XOVER:
+	; DETECT XOVER IN RANGE AND ADJUST COPY LEN IF SO
+	; HL=START, BC=LEN
+	; BC IS REDUCED AS NEEDED TO AVOID XOVER
+	BIT	7,H			; START ABOVE 32K?
+	RET	NZ			; YES, NO XOVER
+	PUSH	HL			; SAVE START ADR
+	ADD	HL,BC			; ADD COPY LEN
+	DEC	HL			; CONVERT TO "LAST" BYTE OF RANGE
+	BIT	7,H			; ABOVE 32K?
+	POP	HL			; RESTORE HL
+	RET	Z			; IF NOT, NO XOVER
+;
+	; START IS BELOW 32K, END IS OVER 32K, XOVER IN SOURCE!
+	; REDUCE LENGTH TO AVOID IT
+	; COMPUTE (32K - START) FOR NEW LEN
+	PUSH	DE			; SAVE DEST (DE)
+	PUSH	HL			; SAVE START (HL)
+	LD	DE,$8000
+	EX	DE,HL			; DE=START, HL=32K
+	OR	A			; CLEAR CARRY
+	SBC	HL,DE			; HL = NEW LEN
+	PUSH	HL			; MOVE NEW LEN
+	POP	BC			; ... TO BC
+	POP	HL			; RECOVER START
+	POP	DE			; RECOVER DEST
+	RET				; RETURN
+;
 Z280_BNKCPY:
 	; Z280 MEMORY TO MEMORY DMA
 	; USE FLOW THROUGH MODE
@@ -5324,6 +5496,7 @@ Z2DMAADR2:
 	INC	C			; BUMP TO NEXT REG
 ;
 	RET
+;
 #ENDIF
 ;
 ; Z280 SYSCALL VECTOR ENTRY POINT.  TAKES STACK PARAMETER AS A BRANCH

Source/ver.inc

@@ -2,4 +2,4 @@
 #DEFINE	RMN	1
 #DEFINE	RUP	1
 #DEFINE	RTP	0
-#DEFINE BIOSVER	"3.1.1-pre.155"
+#DEFINE BIOSVER	"3.1.1-pre.156"

Source/ver.lib

@@ -3,5 +3,5 @@ rmn	equ	1
 rup	equ	1
 rtp	equ	0
 biosver	macro
-	db	"3.1.1-pre.155"
+	db	"3.1.1-pre.156"
 	endm