vasm movem optimization issue?

dalton · 20 September 2016, 14:36

I get this issue when I compile some code with movem optimaztions enabled. The situation seems to be that a branch transcending the movems gets the wrong destination address after the optimization has been implemented.

This is my code:

Code:

		bra		.foo

		movem.l	(a0),d2/d3
		movem.l	(a1),d4/d5
.foo:
		nop
		nop
		nop
		nop
		nop

After compiling:

Code:

	BRA.S	LAB_039E
	MOVE.L	(A0),D2
	MOVE.L	4(A0),D3
	MOVE.L	(A1),D4
	MOVE.L	4(A1),D5
	NOP
	NOP
	NOP
	NOP
LAB_039E:
	NOP

As you can see, the branch has changed its adress by four rows.

Version info:

vasm 1.7e (c) in 2002-2016 Volker Barthelmann
vasm M68k/CPU32/ColdFire cpu backend 2.1 (c) 2002-2015 Frank Wille
vasm motorola syntax module 3.9a (c) 2002-2016 Frank Wille
vasm test output module 1.0 (c) 2002 Volker Barthelmann

phx · 22 September 2016, 12:06

Wow! Yes, thanks a lot! This showed me three different bugs with MOVEM, which I fixed now.

Optimizing/translating to 68020+ addressing modes didn't work so well for MOVEM. For example MOVEM (40000,A0) was not automatically translated into the appropriate 020 addressing mode, but you had to explicitely write (40000.L,A0) instead.
Optimizing MOVEM was not attempted at all, when a register list symbol was used in one operand and a label in the other.
Finally your MOVEM-optimization with two registers for 020+ CPUs, enabled by -opt-movem and -opt-speed.

The patch is too big to post it here. Please use tomorrow's source snapshot.

dalton · 23 September 2016, 14:02

Thanks for investigating so thoroughly!

20 September 2016, 14:36	#1
dalton tulou Join Date: Jun 2006 Location: Gothenburg / Sweden Posts: 88	vasm movem optimization issue? I get this issue when I compile some code with movem optimaztions enabled. The situation seems to be that a branch transcending the movems gets the wrong destination address after the optimization has been implemented. This is my code: Code: bra .foo movem.l (a0),d2/d3 movem.l (a1),d4/d5 .foo: nop nop nop nop nop After compiling: Code: BRA.S LAB_039E MOVE.L (A0),D2 MOVE.L 4(A0),D3 MOVE.L (A1),D4 MOVE.L 4(A1),D5 NOP NOP NOP NOP LAB_039E: NOP As you can see, the branch has changed its adress by four rows. Version info: vasm 1.7e (c) in 2002-2016 Volker Barthelmann vasm M68k/CPU32/ColdFire cpu backend 2.1 (c) 2002-2015 Frank Wille vasm motorola syntax module 3.9a (c) 2002-2016 Frank Wille vasm test output module 1.0 (c) 2002 Volker Barthelmann

22 September 2016, 12:06	#2
phx Natteravn Join Date: Nov 2009 Location: Herford / Germany Posts: 2,510	Wow! Yes, thanks a lot! This showed me three different bugs with MOVEM, which I fixed now. Optimizing/translating to 68020+ addressing modes didn't work so well for MOVEM. For example MOVEM (40000,A0) was not automatically translated into the appropriate 020 addressing mode, but you had to explicitely write (40000.L,A0) instead. Optimizing MOVEM was not attempted at all, when a register list symbol was used in one operand and a label in the other. Finally your MOVEM-optimization with two registers for 020+ CPUs, enabled by -opt-movem and -opt-speed. The patch is too big to post it here. Please use tomorrow's source snapshot.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
3D Graphics: possible optimization?	sandruzzo	Coders. General	3	26 February 2016 08:01
Loop optimization + cycle counts	losso	Coders. Asm / Hardware	8	05 November 2013 11:50
ARM Assembler Optimization	finkel	Coders. General	10	01 December 2010 11:56
Movem Pickout Amiga fun issue 06	AliasXZ	support.Games	0	12 August 2009 00:32
Something like Movem	MarlboroMan	Looking for a game name ?	2	10 October 2004 12:47

23 September 2016, 14:02	#3
dalton tulou Join Date: Jun 2006 Location: Gothenburg / Sweden Posts: 88	Thanks for investigating so thoroughly!

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