DEVCON2: Dr. Greg Colvin's video commentary makes the Ethereum Virtual Machine scream!

Reporter: Lydia

Dr. Greg Colvin is a member of the ANSI/ISO C++ standards committee. At the meeting, he used pictures and sounds to vividly describe the technology that can make the Ethereum virtual machine scream.

This is how he explained the whole thing:

Here is a video where the Jeep sedan is moving very slowly.

I also don't want to make architectural changes to the machine, just take things out of the way that kill performance, if architectural changes require infinitely changing algorithms. Going from 256 bytes to 64 bytes will be faster.

It's comparable to the speed of the Dodge.

There are several methods:

Properly protect change

Small testable changes

No structural changes

Specific changes

Operation: From limited 256 bytes to 512 bytes

Gasoline computing: from 256 bytes to 64 bytes

Provides the necessary gas and memory for each operation

This is today's speed, it can't be faster at the loop level, there are still some number generators here, the Rc5 generator is not used much anymore, and needs constant values, pushing one at a time, so a loaded constant is created, which is only one byte.

To check whether go to is a valid address, use the loading time instead of the running time, and use jumpvi instead. RC5 does this very well.

There is a racing car here, and on the right is the interpreter. From here we need to do different things, and we need to make some improvements and challenges. We are currently using a 256-byte arithmetic library, a 56-byte library, and a 64-byte library for the GMP library to see if we can do better. If everything works properly, it should be enough. We will select some operation codes to make better use of the interpretation library. If it is 64 bytes, it can greatly improve performance. Now there are many programs in Silicon Valley that use SIMD. Now the EVM is a stack machine, and now we need to use a registered machine. We hope to have faster algorithms, so we can move from stack code to algorithm code. Now it may be a 256-byte register, and it also uses a single byte, so the code does not need to be too long.