Optimizing .NET for Modern Hardware

Sign in to queue

Description

Charles Torre sits down with Habib Heydarian and Shawn Farkas to discuss .NET optimization for modern hardware.

Day:

3

Code:

9-014

Room:

C9 Stage

Embed

Download

The Discussion

  • User profile image
    carlospined​ag

    Why C# native is faster than C# CLR-bytecode?
    The fact is that CLR JIT compiler must generate native code for a superscalar processor, but CLR is a stack machine.
    What about the conversion from stack machine to RISC-CISC machine?
    This issue is also a research topic for Java compilation, because JVM and CLR are stack machines.

  • User profile image
    carlospined​ag

    C# JIT compiler generate code for take advantage of Instruction Level Parallelism of modern processors?

  • User profile image
    carlospined​ag

    What about a SPEC benchmark for C#?

    I know exists this benchmarks for Java, but I think C# compiler researchers need this benchmarks in order to test their optimizations.

  • User profile image
    carlospined​ag

    Is the superscalar processor the best platform for execute .NET?
    (considering that .NET CLR is a stack machine and a superscalar processor is a machine based in registers)

    Do you remember the idea of Java processors (Sun's specification named "picoJava")?

    Why not a .NET hardware processor, where native code is the same bytecode?

    What thinks Microsoft about this idea?

    Imagine Azure datacenters with big Symmetric Multiprocessor servers with hundred of .NET processors.

  • User profile image
    gpt

    @carlospinedag:obviously there are no mass deployment of picoJava around. why would the idea work with .NET this time? What do you think is different nowadays that will make it work?

Add Your 2 Cents