Jump to content

New AGPGart

Slice

By Slice,
in AGPGart

 Share
941 posts in this topic

Recommended Posts

kramer2k
kramer2k
Posted
Posted
OK!

New version with corrected mistakes noted above.

Tnhx to Kramer2k & Aivarz.

 

Not yet on leo...

 

AGP: not supported 00ed

AGP: Found NVIDIA nForce3-250 AGP Controller

AGPGart: start

AGPgart: BridgeMemoryRange [fa800000, fe8fffff]

AGP: Found AGP Bus

AGP: hammerbus DevID=1100 VenID=1002

AGP: hammerbus DevID=1101 VenID=1002

AGP: hammerbus DevID=1102 VenID=1002

AGP: hammerbus DevID=1103 VenID=1002

AGP: hammerbus DevID=ffff VenID=ffff

AGP: hammerbus DevID=ffff VenID=ffff

AGP: hammerbus DevID=ffff VenID=ffff

 

Then it just hangs there, no further booting

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted

Hi all!

2 kramer2k

I probably found what is wrong with you nForce chipset. If you successful with 2.1 then I can make 2.4.x working. And I don't like your gart-width=0x40 setting by NVinject. Are you ever tried natit?

cyclonefr have gart-width=0x20 and he is successful, more or less.

 

2 aivarz. dunno what is wrong with Intel875. Are you successful with any other versions?

Nov 19 22:37:16 - kernel[0]: NVChannel(GL): Graphics channel exception! status = 0xffff info32 = 0x6 = Fifo: Parse Error

Nov 19 22:37:16 - kernel[0]: 0000000b

I have one think about it.

 

2 cyclonefr

You have slow UI may be because of large IOLog outputs. I need it for development but in next version I cancel unnecessory outputs.

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
Hi!

Is this a solution for Ati X1950GT AGP blue screen with cursor effect??

As I see

switch (dev) {

case 0x1a31:

return ("Intel 82845 host to AGP bridge");

case 0x3341:

return ("Intel 82855 host to AGP bridge");

case 0x2531:

return ("Intel 82860 host to AGP bridge");

case 0x2571:

return ("Intel 82865 host to AGP bridge");

case 0x255d:

return ("Intel E7205 host to AGP bridge");

case 0x2579:

return ("Intel 82875P host to AGP bridge");

case 0x2561:

return ("Intel 82845G host to AGP bridge");

case 0x3581:

return ("Intel 82855GM host to AGP bridge");

};

There are no 848 chipset.

Upload your ioreg from the command

ioreg -l -x -w 1024 >ioregGergo.txt

I shall try to add your bus to the driver.

Link to comment
Share on other sites
Aivarz
Aivarz
Posted
Posted
2 aivarz. dunno what is wrong with Intel875. Are you successful with any other versions?

 

Previously I had P4P800E-Deluxe which is 865PE and moved to P4C800 Deluxe just a week ago. On Leopard and Tiger I kept the same version of AGPGart I used on P4P800E-Deluxe - the only version that have worked for me. It is one of the AGPGart Light versions, cant't check it right now. The only thing is that it's not setting up gart at all, just a bus speed and card shows up as AGP instead of PCI. So I dont know if I can call it success. I understand that it's not very grateful to make AGPGart for 875P since there is very few boards on this chipset, but on the other hand it should be similar to 865PE, so there's still little hope for me.

Let me know if you need more detailed info on my hardware and/or AGPGart I'm using wich you think may help you to develop working AGPGart.

 

Keep up a good work!

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted

2 aivarz

As I see at Linux driver and at previous AGPGart

case 0x25708086: /* i865 */ case 0x25788086: /* i875P */ case 0x25608086: /* i845G */ pciwrite8(hostSpace, AGP_INTEL_I845_MCHCFG, (pciread8(hostSpace, AGP_INTEL_I845_MCHCFG) | (1 your chipset supported as 865. Cyclonefr with i865 successfully booted AGPGart245 and have QE/CI.Why not you? nVidia driver not work as AGP?

As I have a free time I made some changes in the driver

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
sounds great same speed as 2.4.3 :( opengl is fast as rock but not under games weirdo.

Quartz=73 and UI=37 : poor

 

here is a little ioreg to light ur candle :)

 

worst is with GLexntesion viewer tester : the reference for opengl test i do a poor 500 with/without AGPGART :) still point zero

What value do you expect?

My Core2Duo 2.5 GHz + nVidia7300 PCIE give me (both overclocked)

System Info

Xbench Version 1.3

System Version 10.4.9 (8P2137)

Physical RAM 1024 MB

Model ACPI

Drive Type ST3200820AS

CPU Test 130.78

GCD Loop 300.45 15.84 Mops/sec

Floating Point Basic 145.28 3.45 Gflop/sec

vecLib FFT 101.95 3.36 Gflop/sec

Floating Point Library 94.65 16.48 Mops/sec

Thread Test 255.67

Computation 233.92 4.74 Mops/sec, 4 threads

Lock Contention 281.87 12.13 Mlocks/sec, 4 threads

Memory Test 167.39

System 171.59

Allocate 137.03 503.22 Kalloc/sec

Fill 206.37 10033.99 MB/sec

Copy 187.28 3868.14 MB/sec

Stream 163.38

Copy 148.73 3071.91 MB/sec

Scale 150.70 3113.45 MB/sec

Add 180.61 3847.40 MB/sec

Triad 179.02 3829.61 MB/sec

Quartz Graphics Test 188.11

Line 153.21 10.20 Klines/sec [50% alpha]

Rectangle 198.32 59.21 Krects/sec [50% alpha]

Circle 187.83 15.31 Kcircles/sec [50% alpha]

Bezier 171.03 4.31 Kbeziers/sec [50% alpha]

Text 260.42 16.29 Kchars/sec

OpenGL Graphics Test 184.13

Spinning Squares 184.13 233.58 frames/sec

User Interface Test 488.48

Elements 488.48 2.24 Krefresh/sec

And OpenGL extension viewer

Picture_1.png

Picture_2.png

Link to comment
Share on other sites
Aivarz
Aivarz
Posted
Posted

2.4.6 on Leopard

 

with NVinject 0.1.3

 

Nov 20 19:02:39 - kernel[0]: NVChannel(GL): Graphics channel exception! status = 0xffff info32 = 0x6 = Fifo: Parse Error

Nov 20 19:02:39 - kernel[0]: 0000000b

 

with Titan

 

Nov 20 19:14:00 hole kernel[0]: NVChannel(GL): Graphics channel exception! status = 0xffff info32 = 0x6 = Fifo: Parse Error

Nov 20 19:14:00 hole kernel[0]: 0000000b

Nov 20 19:14:00 hole kernel[0]: NVChannel(GL): Graphics channel exception! status = 0xffff info32 = 0x3 = Fifo: Unknown Method Error

Nov 20 19:14:00 hole kernel[0]: 0000000b

Nov 20 19:14:00: --- last message repeated 1 time ---

 

At the same time I can get results like this with AGPGart light and QuartzGL enabled.

 

Quartz Graphics Test 129.17

Line 129.62 8.63 Klines/sec [50% alpha]

Rectangle 126.70 37.83 Krects/sec [50% alpha]

Circle 160.50 13.08 Kcircles/sec [50% alpha]

Bezier 74.71 1.88 Kbeziers/sec [50% alpha]

Text 286.97 17.95 Kchars/sec

OpenGL Graphics Test 110.79

Spinning Squares 110.79 140.54 frames/sec

User Interface Test 115.46

Elements 115.46 529.92 refresh/sec

 

post-55901-1195583593_thumb.png

 

I don't know how it works.

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted

AGPlight don't create gart table. In this case graphic driver create own gart table as PCIE. But I don't understand how it may work with AGP bus controlled by own chip. Or may be Leo already support Intel875?

As I know Cyclonefr change some graphics kexts by tiger's versions. May be tiger's driver better support AGP and Leo PCIE only?

 

 

NEXT

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
Hi,

Stop booting with blank screen.

that's my output.

 

2.4.6 on Tiger

Many reports with the same problem

"AAPL,gart-width" = 0x40

nVidia driver don't work with AGP gart table that has width=0x20, i.e. 32bits

I think about AGPGart64 :) but what about 32 bit AGP bus?

Link to comment
Share on other sites
kramer2k
kramer2k
Posted
Posted
2..4.6 tested CEQO but same perf exactly as 2.4.5

opengl:150

UI:37

Quart 76

 

First things first...

A question that really hasn't been asked yet...

Is this complied for Tiger or Leo? Or can we use it for both?

 

Secondly, 246 still doesn't work with Leo using NVinject, NvidiaEFI 1.0.2, or Natitdual

All hang at my previously posted error

 

I know I'm not a coder, but what changed from 2.1 to 2.2 to make it not work on nF3? Anyway that we can implement that on the current version? And why the heck won't 2.1 work on Leo?

 

K

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted
First things first...

A question that really hasn't been asked yet...

Is this complied for Tiger or Leo? Or can we use it for both?

 

Secondly, 246 still doesn't work with Leo using NVinject, NvidiaEFI 1.0.2, or Natitdual

All hang at my previously posted error

 

I know I'm not a coder, but what changed from 2.1 to 2.2 to make it not work on nF3? Anyway that we can implement that on the current version? And why the heck won't 2.1 work on Leo?

 

K

I am a coder.

Here is the natit version that do not make gart-width=0x40.

Version 2.2 and later has a different classes structure then 2.1. It is not a problem, but during rewrite joblo makes some mistakes that I repeat. Now I am catching bugs.

Second thing is that 2.1 use very few gart table entries so it don't crashes using its.

Try this natit and AGPGart246. May be I found one more mistake for nForce.

Compare natit.info.plist with your existing, may be you need to add some value.

 

NEXT

Natit.kext.zip

Link to comment
Share on other sites
GERGÖ
GERGÖ
Posted
Posted
As I see

 

There are no 848 chipset.

Upload your ioreg from the command

ioreg -l -x -w 1024 >ioregGergo.txt

I shall try to add your bus to the driver.

Hi! Thank u for your help.My dev. id is listed there: 848 is 865 like and has the same 2571 id. I'll try it and report back.THX :unsure:

Link to comment
Share on other sites
Slice
Slice
Posted
  • Author
Posted

O yes!

I found great difference 2.1 and 2.2

2.1

	while( (phys = mem->getPhysicalSegment64(offset, &len)))
{
	offset += len;
	len = (len + PAGE_MASK) & ~PAGE_MASK;
	while( len > 0) 
	{
		BUG_ON(phys & 0xffffff0000000ffcULL);
		pte = (phys & 0x000000ff00000000ULL) >> 28;
		pte |=(phys & 0x00000000fffff000ULL);
		pte |= GPTE_VALID | GPTE_COHERENT;
		OSWriteLittleInt32( agpInfo->gartArray, agpOffset, pte);
		agpOffset += 4;

2.2 

	while( (phys = mem->getPhysicalSegment(offset, &len)))
{
	offset += len;
	len = (len + PAGE_MASK) & ~PAGE_MASK;
	while( len > 0) 
	{
		phys |= addrMask;
		OSWriteLittleInt32(gartArray, gartOffset, phys);

This is a difference in 32bit and 64bit gart. But in 2.1 agpOffset +=4 - the same 4 bytes :unsure:

Other change - "nvidia_set_aperture" that is not used by 2.1 but used by 2.2 :blink: Is it bad setting?

Without good advice I can only propose you test different combination to find a good one.

Link to comment
Share on other sites
 Share
Go to topic listing
×
×
  • Create New...