GCC generates Buggy Code II.

Hehe... sorry - but since "we" used gcc 6809 quite a lot in the last 2 years... some things have come up: main.c

In the example code (compiled with options: -O2 -quiet -fverbose-asm -W -Wall -Wextra -Wconversion -fomit-frame-pointer -mint8 -msoft-reg-count=0 -std=gnu99 -fno-time-report ) (I have produced similar buggy code with other options, but it is difficult to reproduce)

GCC produces buggy code. When calculating "16bit values" / "the address of a pointer" it mixes "d" and "x" registers so thorowly, that the resulting d register contains the wrong value.

The example seems rather exotic - the bug occured in "real life" but the example is a bit far fetched, since in order to produce the bug, it seems gcc has to be "stressed".

Generated code (somehwere in the middle, my source at about line 90):

	addb	_tmp+9	; , <variable>.count1
	stb	2,s	; , D.1214
	stb	_tmp+9	; , <variable>.count1
	sex		;extendqihi2: R:b -> R:d	; ,
!	tfr	d,x	; , D.1214
!	exg	d,x	; , D.1217
!	addd	,s	; ,
!	exg	d,x	; , D.1217
	std	_tmp+3	;  D.1217, <variable>.count3
	ldb	2,s	; , D.1214
	sex		;extendqihi2: R:b -> R:d	; ,

The "!" code is wrong. After the sex the value is to be calculated in "D". The tfr "doubles" the value in D to the X register, the following exg is "nonsense", than the addd actually calculates the correct value in D, the following exg is not only nonsense but defenitly a bug, since the register D ist reset with its former value from register X.

following would be Ok:

	addb	_tmp+9	; , <variable>.count1
	stb	2,s	; , D.1214
	stb	_tmp+9	; , <variable>.count1
	sex		;extendqihi2: R:b -> R:d	; ,
!	addd	,s	; ,
!	tfr	d,x	; , D.1214
	std	_tmp+3	;  D.1217, <variable>.count3
	ldb	2,s	; , D.1214
	sex		;extendqihi2: R:b -> R:d	; ,

This would result in the correct value in D and in "X", when it is needed. (I have seen same code generated, where this section needs to calculate "X" instead of "D", so X should probably be kept).