05259: H6309 opcode muld does not work

ID	Category [?]	Severity [?]	Reproducibility	Date Submitted	Last Update
05259	Core	Major	Always	Aug 9, 2013, 01:50	Aug 12, 2013, 02:30

Tester	Robert Gault	View Status	Public	Platform	MESS (Official Binary)
Assigned To	Bletch	Resolution	Fixed	OS	Windows XP
Status [?]	Resolved			Driver	trs/coco3.cpp
Version	0.149u1	Fixed in Version		Build	Normal
		Fixed in Git Commit		Github Pull Request #

Summary	05259: H6309 opcode muld does not work
Description	At least one opcode for the Hitachi 6309 is not working, muld. There may be others but I have not done testing. This could make NitrOS-9 6309 systems fail to function.
Steps To Reproduce	Run some program that will permit creating assembly language code. It might be possible to do this with the MESS debugger. A simple test program follows. start ldd #$10 muld #$1234 swi The swi is the equivalent of a breakpoint. Check the registers after the muld instruction. They should be regA =0 regB=1 regE=$23 regF=$40 The actual binary code for this program is $CC,$00,$11,$8F,$12,$34,$3f
Additional Information	There are many other addressing modes for the muld opcode and they all fail. ex muld 1,y muld address The instuction should multiply the 16-bit content of regD by a 16-bit number and store the 32-bit result in regQ which consists of regA regB regE regF. Now the one big problem is that this instruction, divd, and divq is that they are twos complement signed operations. That means for large numbers, the results are not usable. Generally one needs to insure that there is no overflow and that the numbers stay positive. Older versions of MESS worked correctly. I believe this should addressed in mess\src\emu\cpu\m6809\ where one of h6309.c h6309.h or h6309.ops, but I'm not sure just what code is used in coco3h emulation.
Github Commit

Flags
Regression Version
Affected Sets / Systems

Attached Files

No.09716 Robert Gault Tester Aug 9, 2013, 11:38 edited on: Aug 9, 2013, 21:23	Here is more data for the instructions LDD #n MULD #m n m regD regW regCC 2 2 0 4 EZ -2 2 $FFFF $FFFC EN -2=$FFFE $7FFF $7FFF $3FFF 1 E $8000 2 $FFFF 0 EN $8000 is a negative number so the answer is a -$10000 which isn't expressed well. $8001 2 $FFFF 2 EN result -$10002 Easier to understand as a negative number
No.09729 Bletch Developer Aug 12, 2013, 02:30	The 6809/6309/Konami CPU cores were rewritten (by me) a few months ago in order to: * Merge them into a single code bases * Support intra-instruction timing (so Dragon Fire could work correctly) * Run faster This MULD regression was almost certainly introduced during that rewrite. I believe that I've fixed it - let me know if there are further problems.