TOS RAM test routine ?

[exxos]

I guess if someone fancies looking at the memory test routines, it could be possible to just copy the routine to run from TTram instead. Of course it would have to test a small memory area is valid before copying the routine to it. But if it is switching to 8MHz to run the test then ramping up to 50MHz would go lightning fast.

[Badwolf]

@Badwolf When you get a moment can you please give this one a try. I tried only altering the number on the TTram test..

Here's TOS206.img.

Looks like the inverse video is gone again.

The ST-RAM test finishes early.

The TT-RAM test seems faster, but with the ST-RAM finishing early, I don't know if it's legit.



For comparison, here's the (slow) TP32.img with all the same settings.



BW

[exxos]

For comparison, here's the (slow) TP32.img with all the same settings.

Thanks. I think I will just abandon making the test faster because I don't know currently what PAK TOS did to speed it up.

[Badwolf]

Thanks. I think I will just abandon making the test faster because I don't know currently what PAK TOS did to speed it up.

Well the first one looks to be about twice as fast to me, but I reckon you'd do better just to turn on the instruction cache (although like Doug says, that won't show up on Hatari). The CACR is 0000 during these tests (according to the debugger). That just needs to be changed to 0001. Presumably something sets them to something sensible later after the tests so it shouldn't be a problem.

Try bunging

Code: Select all

        move.l    #$00000001,d0
        D0_CACR

at around line 2001 in startup.S (the line before bsr memtest, wherever that is).

I can see if it's made the change in Hatari, but it'd need testing on real hardware to know if cache has really helped.

BW

EDIT: although, TBF, who really minds if a memory test everyone skips is a bit slow? :P

[exxos]

I can see if it's made the change in Hatari, but it'd need testing on real hardware to know if cache has really helped.

EDIT: although, TBF, who really minds if a memory test everyone skips is a bit slow? :P

Yeah, it is much time messing about just for the sake of someone pressing spacebar.

I don't know why the inverse video comes up. does not seem to be related to anything I changed as I reversed all the changes.

[stephen_usher]

For comparison, here's the (slow) TP32.img with all the same settings.

Thanks. I think I will just abandon making the test faster because I don't know currently what PAK TOS did to speed it up.

PAK/3 TOS 3.06 is approximately the same speed as TOS 3.06 on the TT, so it's not the PAK/3 changes.

[exxos]

PAK/3 TOS 3.06 is approximately the same speed as TOS 3.06 on the TT, so it's not the PAK/3 changes.

Thats a point.. I altered that tosversion thing as well.. That could be what is sitting the inverse video.

EDIT:

Yep this was the cause..

Code: Select all

#if TOSVERSION >= 0x300
        move.l    d0,-(a7)
        movea.l   #memtsp,a0
        bsr       mempstr
        move.l    (a7)+,d0

[exxos]

@frank.lukas - GERMAN ST536 TOS.

Capture.PNG

tos206de.zip

[exxos]

@Badwolf

Oddly it seems to be hellbent on actually disabling them more than anything :shrug:

Code: Select all

/*
 * disable the instruction/data caches
 */
#if (TOSVERSION >= 0x300) | TP_50
        data
zero:
	    dc.l 0
		text

#if TP_50 /* PAK */
		lea     no020(pc),a0
		move.l  a0,(_illinst).w
#endif
        move.l  #$00000808,d0     /* clear & disable instruction/data caches on */
        D0_CACR              /*  68020-60 (even though bit usage differs!) */

        moveq   #0,d0
        D0_VBR           /* initialise VBR to zero */
no020:

EDIT:

hmmm..

So this is actually turned off because TP_50 = 1 in the defs file..

Code: Select all

#if TP_03 /* RAMRESET */
#include "../tospatch/ramres.S"
#else

#if (TOSVERSION >= 0x300) & (TOSVERSION < 0x400) & !TP_50
        btst      #0,(scu_gp1).w
        beq.s     domemchk
#endif

No idea what this is doing ?

Code: Select all

                lea     reset(PC),A6
                bra     memchk

/*   ---------------------------------------- */

newkres:        move    #$2700,SR       /* disable interrupts */

                clr.l   (_memvalid).w   /* make memory configuration invalid */

                movea.l 4.w,A0
                lea     (_buserror).w,A1
                moveq   #(($0100-$0008)/4)-1,D0 /* clear $8-$100 */
newkres1:       move.l  A0,(A1)+
                dbra    D0,newkres1
                bra     _main

resethandler:   cmpa.l  #resetvec,A6
                beq.s   noreset

                lea     _main(pc),a6

crc:            lea     _os_entry(PC),A3     /* start address */
                lea     rom_crc_table,A1
                moveq   #numbanks-1,D7

crc0:           move.l  #banksize,D2    /* bytes to check per loop */
                moveq   #0,D0           /* set starting values */
                moveq   #0,D5
                movea.l A3,A0

crc1:           move.w  (A0),D3
                addq.l  #numbanks,A0

                move.w  D5,D6
                lsl.w   #8,D5
                lsr.w   #8,D6
                eor.b   D3,D6
                add.w   D6,D6
                move.w  0(A1,D6.w),D4
                eor.w   D4,D5

                move.w  D0,D1
                lsl.w   #8,D0
                eor.w   D3,D1
                lsr.w   #8,D1
                add.w   D1,D1
                move.w  0(A1,D1.w),D4
                eor.w   D4,D0

                subq.l  #1,D2           /* already done? */
                bne.s   crc1            /* no => */

                move.w  (A0),D1
                move.b  D1,D6
                addq.l  #numbanks,A0
                move.b  (A0)+,D1
                cmp.w   D1,D0
                bne.s   oldkeyresetj

                lsl.w   #8,D6
                move.b  (A0),D6
                cmp.w   D6,D5
oldkeyresetj:   bne     coldboot

                addq.l  #2,A3           /* 2 passes */
                subq.l  #2,D7           /* again? */
                bpl.s   crc0            /* yes */

                jmp     (A6)

/* ---------------------------------------- */

reset:          beq     go_on

                lea     checkmem(PC),A6
                bra.s   crc

checkmem:       lea     ($00007FFE).w,SP

                movea.l (_buserror).w,A4          /* save buserror vector */
                move.l  #coldboot,(_buserror).w /* set new vector */
                lea     _os_entry(PC),A2     /* set end address */
                move.l  #$00020000,D7   /* distance of probes */
                movea.l D7,A0           /* set start address */
                move.w  #$FB55,D3       /* value to add */

loop1:          movea.l A0,A1
                move.w  D0,D2
                moveq   #$2A,D1
loop2:          move.w  D2,-(A1)
                add.w   D3,D2
                dbra    D1,loop2
                movea.l A0,A1
                moveq   #$2A,D1
loop3:          eor.w   D0,-(A1)
                bne.s   oldkeyresetj
                add.w   D3,D0
                dbra    D1,loop3
                adda.l  D7,A0
                cmpa.l  A2,A0
                bls.s   loop1
                movea.l A2,A0
                move.l  A4,(_buserror).w

                movem.l (A1),D0-D7/A1/A3-A6 /* get zeroes */
clrmem:         movem.l D0-D7/A1/A3-A6,-(A0) /* clear page */
                movem.l D0-D7/A1/A3-A6,-(A0)
                movem.l D0-D7/A1/A3-A6,-(A0)
                movem.l D0-D7/A1/A3-A6,-(A0)
                movem.l D0-D7/A1/A4-A6,-(A0)
                cmpa.w  #$0100,A0       /* clear memory to $100 */
                bhi.s   clrmem

                move.l  A2,(_phystop).w  /* highest address as phystop */

                move.l  #$752019F3,(_memvalid).w
                move.l  #$237698AA,(_memval2).w
                move.l  #$5555AAAA,(_memval3).w
                clr.l   (_ramtop).w
                move.l  #$1357BD13,(_ramvalid).w

                move.b  (memctrl).w,(memconf).w

go_on:          move.l  #$31415926,(resvalid).w
                move.l  #resethandler,(resvector).w

Looks like a totally new bank of routines for testing ROM CRC and RAM as its got the same $20000 value in there.

Looks to be triggered from TP_03

Code: Select all

TP_03: RAMRESET: Modified reset routine.
 * Will keep RAM-TOS resident even after Hardreset (via keyboard)

Code: Select all

#
# Set this to 1 to build a RAM version relocated to 0x380000.
# Only the PL version uses this so far.
# Note that the PL version is not an official TOS, it is
# an image found on the net on several sites, and contains
# some patches for the BIOS.
#
RAMVERSION=0

[frank.lukas]

Many thanks ...

Fastram bar only ...

PAK030/50Mhz 64MB = 24s
PAK020/40Mhz 64MB = 35s
TF536 /50Mhz 64MB = 4m40s

... the ST Ram test bar is faster on the PAK Atari ST than on the TF536 Atari ST.


Atari TT 32Mhz 256MB = 2m20s

... the PAK machines are already faster than the Atari TT.

2m20s for 256MB / 4 = 35s (This is the same as the PAK020 with 40Mhz and 64MB)