68020+ scale factor behaviour?

roondar · 30 October 2020, 16:08

This is an offshoot of the setting a pixel thread. There was a tiny bit of discussion about compilers and how they selected words vs longwords for operations and chb correctly pointed out that the compiler can't always know whether a result will fit in a word or not and thus could in some circumstances decide to use longwords even if the variable was set to being a word to make sure the result was correct.

This all had to do with using a register as an index (something like move.w d0,0(a0,d1.l)). Made me wonder how the 68020 deals with this if you use the built-in scaling facility.

In other words, I wonder what would happen if I do this:

Code:

move.w #32767,d0
move.w 0(a0,d0.w*2),d1

I guess that same would apply to d0 being 2^31, but that is a bit more of an edge case, IMHO.

a/b · 30 October 2020, 16:55

d0.w is sign-extended to d0.l (000+ behavior), then scaled.

roondar · 30 October 2020, 17:05

Thanks!
I had just finished writing a tiny test program to see for myself and that is indeed what it does

30 October 2020, 16:08	#1
roondar Registered User Join Date: Jul 2015 Location: The Netherlands Posts: 3,409	68020+ scale factor behaviour? This is an offshoot of the setting a pixel thread. There was a tiny bit of discussion about compilers and how they selected words vs longwords for operations and chb correctly pointed out that the compiler can't always know whether a result will fit in a word or not and thus could in some circumstances decide to use longwords even if the variable was set to being a word to make sure the result was correct. This all had to do with using a register as an index (something like move.w d0,0(a0,d1.l)). Made me wonder how the 68020 deals with this if you use the built-in scaling facility. In other words, I wonder what would happen if I do this: Code: move.w #32767,d0 move.w 0(a0,d0.w*2),d1 I guess that same would apply to d0 being 2^31, but that is a bit more of an edge case, IMHO.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Relations between Factor 5 and Graftgold?	mc68060	Retrogaming General Discussion	43	19 September 2018 19:25
68030 vs 68020 line-f behaviour	jotd	Coders. Asm / Hardware	7	16 July 2017 20:29
Factor 5	hit	News	40	19 March 2017 15:44
Factor 5 dates	Predseda	HOL data problems	2	12 January 2016 12:57
What Chips Are What Form Factor	Loedown	support.Hardware	1	17 November 2009 13:17

30 October 2020, 16:55	#2
a/b Registered User Join Date: Jun 2016 Location: europe Posts: 1,039	d0.w is sign-extended to d0.l (000+ behavior), then scaled.

30 October 2020, 17:05	#3
roondar Registered User Join Date: Jul 2015 Location: The Netherlands Posts: 3,409	Thanks! I had just finished writing a tiny test program to see for myself and that is indeed what it does

Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)