Markup languages coprocesing unit

In general, I find that an FPGA or other custom hardware implementation is only justified if there is a strict real-time constraint that can't be met by a general-purpose computer, or there is a need for much higher performance than CPUs can provide and there is enough parallelism in the application to take advantage of a hardware implementation.

It seems to be that parsing a linear language like HTML does not have enough parallelism, and later parallelizable stages like rendering high-quality text and graphics can be done quite nicely by a GPU.  Also, a language like HTML has a lot of complexity that is expensive to implement in hardware but cheap to implement in software.

I'd be interested in your comments about my impressions and what I may have missed.

Hi John

Thanks for your observations and interest.

The purpose of the FPGA is just having a working prototype. If successful the project may be eventually embedded as parts of CPUs.

The focus is very low level and aims to reduce some cycles when analyzing serialized data that may be HTML but can also be just end of string tokens on network communications, praising a script language (or any text file), regular expressions or similar stuff.

The idea is to design/build a comparison and pattern searching unit with multiple simplified ALUs capable of just few kinds of comparisons instead of using the CPU’s ALU to compare and do pattern searching. This unit may be very purpose specific but given the high rate of networking and the extensive use of scripting languages it may be worth it.

Taking the HTML case, and SUPER oversimplifying, to analyze a HTML string a program must look for the tags. This is done looking for a < and then the following characters that must match certain rules and then fin the > and then the correspondent closing tag or /> for single tags. I know this is done at a very high level but at the end the CPU has to push tons of characters into the ALU to compare one by one. Parallelism in separate treads is hard to apply because what you are looking for depends on what you have fond before. I don’t know if there are algorithms to analyze this kind of languages in multiple cores but even if they are so is kind of overkill.

The comparison unit I’m thinking of will allow to see more than one character on the string at a time and there for reduce the times you have to back and forward throw it looking for specific things.

In the HTML to look for tags you can look for < followed only by characters.

A very high level diagram is like follows:

Data will pased from registry 1 to 2 from 2 to 3 and so on on each clock cicle.

Comparators 2 to N-1 can be configured to search anything between a-z or A-Z,
comparator 1 to search for either “  ” or >
and comparator N configured to look only for <

The metacomparator to accept either  all to 1
or all but the first
or all but the last

So in the string depending on N and the data we will have 1 on the “I found what you are looking for interruption” on strings like “<tag ”, <tag> or “ tag>”

The comparators will have some registers to set some comparison values and a command registry to tell for example use comparison values 1 and 2  as well as comparison values 3 and 4 as lookup ranges, or find any of the masks(value) on the comparison value registers.

Hope this clarifies my intentions and calls your attention.

Hi John

Thanks for your observations and interest.

The purpose of the FPGA is just having a working prototype. If successful the project may be eventually embedded as parts of CPUs.

The focus is very low level and aims to reduce some cycles when analyzing serialized data that may be HTML but can also be just end of string tokens on network communications, praising a script language (or any text file), regular expressions or similar stuff.

The idea is to design/build a comparison and pattern searching unit with multiple simplified ALUs capable of just few kinds of comparisons instead of using the CPU’s ALU to compare and do pattern searching. This unit may be very purpose specific but given the high rate of networking and the extensive use of scripting languages it may be worth it.

Taking the HTML case, and SUPER oversimplifying, to analyze a HTML string a program must look for the tags. This is done looking for a < and then the following characters that must match certain rules and then fin the > and then the correspondent closing tag or /> for single tags. I know this is done at a very high level but at the end the CPU has to push tons of characters into the ALU to compare one by one. Parallelism in separate treads is hard to apply because what you are looking for depends on what you have fond before. I don’t know if there are algorithms to analyze this kind of languages in multiple cores but even if they are so is kind of overkill.

The comparison unit I’m thinking of will allow to see more than one character on the string at a time and there for reduce the times you have to back and forward throw it looking for specific things.

In the HTML to look for tags you can look for < followed only by characters.

A very high level diagram is like follows:

Data will pased from registry 1 to 2 from 2 to 3 and so on on each clock cicle.

Comparators 2 to N-1 can be configured to search anything between a-z or A-Z,
comparator 1 to search for either “  ” or >
and comparator N configured to look only for <

The metacomparator to accept either  all to 1
or all but the first
or all but the last

So in the string depending on N and the data we will have 1 on the “I found what you are looking for interruption” on strings like “<tag ”, <tag> or “ tag>”

The comparators will have some registers to set some comparison values and a command registry to tell for example use comparison values 1 and 2  as well as comparison values 3 and 4 as lookup ranges, or find any of the masks(value) on the comparison value registers.

Hope this clarifies my intentions and calls your attention.