Neoscrypt lookup-gap

[OhGodAPet]

I'll provide the kernel source - only valid values are undefined and defined - when defined, the lookup gap should be 2. Macro should be LOOKUP_GAP.

[ystarnaud]

Ok I'll put that in the next update I push.

[Zuikkis]

How's this going? :) I'm interested in trying new things..

[ystarnaud]

Oh sorry this has been done along with a few other issues brought up before. I've been trying to implement a new feature but it's not been working as it should. I was hoping to have it debugged and fixed so I could push my code up but at this point i'll just have to take it out of my source and clean out the debug stuff I added. I'll probably push an update later tonight or tomorrow (when thanksgiving is over lol)

[ystarnaud]

Added with 4ba8a68043a14f9f3d12c6ae3838cc802ddadd65 (master branch).

[Zuikkis]

Hmm, I can't get any decent speed from neoscrypt with this version. 155kh/s on 280x, was 330kh/s before. Perhaps you pushed wrong kernel? There is no reference to LOOKUP_GAP in the kernel source.

Also in miner.h there is a small bug, line 1494:

define switch_pools(p) __switch_pools(p, TRUE)

"TRUE" should be lowercase "true", does not compile under linux otherwise..

[troky]

Something is wrong in new neoscrypt kernel. 130kh/s, was 310kh/s on 290.

[ystarnaud]

Make sure you have no conflicting TC set. Tc overrides intensity/xI and rI.

On Wednesday, December 3, 2014, troky notifications@github.com wrote:

Something is wrong in new neoscrypt kernel. 130kh/s, was 310kh/s on 290.

— Reply to this email directly or view it on GitHub https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65500263.

[ystarnaud]

Also the lookup gap setting was something wolf was working on its not in the current kernel but I added it for future updates.

On Wednesday, December 3, 2014, Yann St.Arnaud ystarnaud@gmail.com wrote:

Make sure you have no conflicting TC set. Tc overrides intensity/xI and rI.

On Wednesday, December 3, 2014, troky <notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com');> wrote:

Something is wrong in new neoscrypt kernel. 130kh/s, was 310kh/s on 290.

— Reply to this email directly or view it on GitHub https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65500263 .

[Zuikkis]

Nah, you have uploaded the wrong kernel. The old .bin works with same speed as before, you just have to rename the file to add "lg" to the file name.. On 4 Dec 2014 00:32, "ystarnaud" notifications@github.com wrote:

Also the lookup gap setting was something wolf was working on its not in the current kernel but I added it for future updates.

On Wednesday, December 3, 2014, Yann St.Arnaud ystarnaud@gmail.com wrote:

Make sure you have no conflicting TC set. Tc overrides intensity/xI and rI.

On Wednesday, December 3, 2014, troky <notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com');> wrote:

Something is wrong in new neoscrypt kernel. 130kh/s, was 310kh/s on 290.

— Reply to this email directly or view it on GitHub < https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65500263> .

— Reply to this email directly or view it on GitHub https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65503848.

[Zuikkis]

Also, you can copy the neoscrypt.cl from the previous build, to get full speed.

If you look at the cl, it's obvious that this is the first unoptimized version.. block swap is done with temp variables instead of xor, copy uses int instead of int4, etc.

I'm a bit disappointed if you don't push the lookup gap version. I had some ideas for optimization, but was waiting to see what wolf has done so I don't do duplicate work.

At least the current salsa code seems awfully slow when compared to regular scrypt.. On 4 Dec 2014 00:36, "Teemu Suikki" zuikkis@gmail.com wrote:

Nah, you have uploaded the wrong kernel. The old .bin works with same speed as before, you just have to rename the file to add "lg" to the file name.. On 4 Dec 2014 00:32, "ystarnaud" notifications@github.com wrote:

Also the lookup gap setting was something wolf was working on its not in the current kernel but I added it for future updates.

On Wednesday, December 3, 2014, Yann St.Arnaud ystarnaud@gmail.com wrote:

Make sure you have no conflicting TC set. Tc overrides intensity/xI and rI.

On Wednesday, December 3, 2014, troky <notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com');> wrote:

Something is wrong in new neoscrypt kernel. 130kh/s, was 310kh/s on 290.

— Reply to this email directly or view it on GitHub < https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65500263> .

— Reply to this email directly or view it on GitHub https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65503848 .

[ystarnaud]

Ok I'll copy the kernel code again. I didn't think I used the wrong cl but it's true that when I tested I only replaced the miner program. Sorry about that.

On Wednesday, December 3, 2014, Zuikkis notifications@github.com wrote:

Also, you can copy the neoscrypt.cl from the previous build, to get full speed.

If you look at the cl, it's obvious that this is the first unoptimized version.. block swap is done with temp variables instead of xor, copy uses int instead of int4, etc.

I'm a bit disappointed if you don't push the lookup gap version. I had some ideas for optimization, but was waiting to see what wolf has done so I don't do duplicate work.

At least the current salsa code seems awfully slow when compared to regular scrypt.. On 4 Dec 2014 00:36, "Teemu Suikki" <zuikkis@gmail.com javascript:_e(%7B%7D,'cvml','zuikkis@gmail.com');> wrote:

Nah, you have uploaded the wrong kernel. The old .bin works with same speed as before, you just have to rename the file to add "lg" to the file name.. On 4 Dec 2014 00:32, "ystarnaud" <notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com');> wrote:

Also the lookup gap setting was something wolf was working on its not in the current kernel but I added it for future updates.

On Wednesday, December 3, 2014, Yann St.Arnaud <ystarnaud@gmail.com javascript:_e(%7B%7D,'cvml','ystarnaud@gmail.com');> wrote:

Make sure you have no conflicting TC set. Tc overrides intensity/xI and rI.

On Wednesday, December 3, 2014, troky <notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com'); <javascript:_e(%7B%7D,'cvml','notifications@github.com javascript:_e(%7B%7D,'cvml','notifications@github.com');');>> wrote:

Something is wrong in new neoscrypt kernel. 130kh/s, was 310kh/s on 290.

— Reply to this email directly or view it on GitHub < https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65500263>

.

— Reply to this email directly or view it on GitHub < https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65503848> .

— Reply to this email directly or view it on GitHub https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65509058.

[ystarnaud]

387bb28952140ff5ec419852e70fc59d5fba996e should have the correct cl. I wiped my bins, kernel files on my test rig and copied everything over from my build directory. I was able to get 326kh on my 7970 with the same old config file.

[Zuikkis]

Confirmed, works fine again.

2014-12-04 2:25 GMT+02:00 ystarnaud notifications@github.com:

387bb28 https://github.com/sgminer-dev/sgminer/commit/387bb28952140ff5ec419852e70fc59d5fba996e should have the correct cl. I wiped my bins, kernel files on my test rig and copied everything over from my build directory. I was able to get 326kh on my 7970 with the same old config file.

— Reply to this email directly or view it on GitHub https://github.com/sgminer-dev/sgminer/issues/400#issuecomment-65517765.

Teemu Suikki http://www.ksportfinland.com http://www.z-power.fi/

[OhGodAPet]

I totally forgot to close this shit.