How is --mine working on ARM?

[Marc Karasek]

I have not been abel to figure out how this is working on ARM (--mine option).

The void generateDatasetInitCode() {} does not copy any dataset_init() code. So when this is run via
jit.getDatasetInitFunc()(cache, (uint8_t*)&datasetItem, 0, 1);
It crashes?? since no code is there?

Caveat I do not have an ARM platform to test on. I am porting RandomX over to another processor and was using ARM as the template. It crashes at this spot due to no dataset_init() code being present and the generateDatasetInitCode() doing nothing.

Under x86 I see that the generate() function copies code over, etc..

[SChernykh]

I wrote the ARM code and it's how it works there. generateDatasetInitCode() is empty because the code is already there and doesn't need to be generated, see JitCompilerA64::getDatasetInitFunc(). The naming of functions is probably not good because actual dataset init code is generated in generateSuperscalarHash.

[Marc Karasek]

I see that when the jit is created this copies the code over..
Yeah not too clear as to what is going on ..

[SChernykh]

The whole ARM code is in jit_compiler_a64_static.S and generateSuperscalarHash just overwrites the part that starts at randomx_calc_dataset_item_aarch64 IIRC

[Marc Karasek]

Question on the randomx_calc_dataset_item_aarch64() function # rl[0] = (itemNumber + 1) * superscalarMul0; madd x0, x2, x12, x12 Do you expect the above to add 1 to the itemnumber and then mul it by superscalarMul? According to ARM spec that is not what this does? result = UInt(operand3) + (UInt(operand1) * UInt(operand2));

…

On Wed, Feb 24, 2021 at 6:44 PM SChernykh ***@***.***> wrote: The whole ARM code is in jit_compiler_a64_static.S and generateSuperscalarHash just overwrites the part that starts at randomx_calc_dataset_item_aarch64 IIRC — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AD72ZMYHHZOHRDEMNNCGIBLTAWFNLANCNFSM4YFM4PIA> .

-- /**************** Marc Karasek ASM T648 Herder of Cats ****************/

[SChernykh]

madd x0, x2, x12, x12 calculates itemNumber * superscalarMul0 + superscalarMul0 which is the same.

[Marc Karasek]

that is not the same as (item +1) * superscalarMul0 which is the comment in the code.. :-)

…

On Fri, Feb 26, 2021 at 2:04 PM SChernykh ***@***.***> wrote: madd x0, x2, x12, x12 calculates itemNumber * superscalarMul0 + superscalarMul0 which is the same. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AD72ZMYFUURYTPC5I4K36V3TA7WFRANCNFSM4YFM4PIA> .

-- /**************** Marc Karasek ASM T648 Herder of Cats ****************/

[SChernykh]

It is the same. I don't remember, but I think it's 5th grade math. (a+1)*b = a*b + b

[Marc Karasek]

I have one question on the generateSuperscalarHash() In the other generate program you used the IntRegMap[] to grab the registers for the opcodes. In generateSuperscalarHash() you use the instr.src / instr.dst directly? These should be offsets (which were used to access the IntRegMap[] array). How does generateSuperscalarHash() figure out what registers it can use? Marc

…

On Fri, Feb 26, 2021 at 3:32 PM SChernykh ***@***.***> wrote: It is the same. I don't remember, but I think it's 5th grade math. (a+1)*b = a*b + b — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AD72ZM76I6MFCMDTLTNCBLDTBAANPANCNFSM4YFM4PIA> .

-- /**************** Marc Karasek ASM T648 Herder of Cats ****************/

[SChernykh]

SuperscalarHash uses registers x0-x7 unlike the main RandomX loop.

[Marc Karasek]

ok that makes sense now.. I have gotten it to work on the new SOC but the hashes are wrong.. Trying to debug to see what is wrong..

…

On Wed, Mar 3, 2021 at 2:40 PM SChernykh ***@***.***> wrote: SuperscalarHash uses registers x0-x7 unlike the main RandomX loop. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#204 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AD72ZM6XNM7T7FLSO7F3O43TB2GDFANCNFSM4YFM4PIA> .

-- /**************** Marc Karasek ASM T648 Herder of Cats ****************/