Native Code Performance on Modern CPUs: A Changing Landscape

Download this episode

Download Video

Download captions

Download Captions

Description

Modern CPUs are fast. Really fast. New instructions, wider vector registers, and more powerful CPUs promise faster code. But reasoning about performance is not as it seems on the surface. This talk will dive deep into how advancements in the latest chips force some rethinking of native code performance, from the point of view of a compiler developer. The performance landscape is changing. Come see what that means.

Tag:

C/C++

Day:

3

Level:

400 - Expert

Code:

4-587

Room:

Room 3024

Embed

Format

Available formats for this video:

Actual format may change based on video formats available and browser capability.

    The Discussion

    • User profile image
      ttmahdy

      Will there be a live stream of this session?

    • User profile image
      lesderid

      This wasn't very fun to watch. The speaker seemed a bit disinterested, or even annoyed. Interesting session nevertheless.

    • User profile image
      anonymous

      I'll have to disagree with you, lesderid. I liked his style and he was clearly very interested in the subject. Great talk.

    • User profile image
      mfn

      Awesome talk! Especially the discussion on how much memory access latency impacts performance, even for operations that one would typically consider to be "mostly CPU".

    • User profile image
      Sri

      excellent talk on an excellent topic. Hope to see more and more talks on native runtime on windows!!

    • User profile image
      CoolOppo

      Awesome talk! I really like the detail pertaining to the nitty-gritty stuff, it's awesome how well it was explained with the flowcharts.

    • User profile image
      dzyashu

      Great talk. But there are audio issues starting at about 3:50 in HQ file.

    • User profile image
      deiruch

      Great session (presentation & content). Would love to hear more about .Net native from him. Specifically if life's easier for the optimizer when dealing with IL. Can more/other optimization be performed? I'm asking because pointers (in C/C++) make many optimizations quite difficult.

    • User profile image
      IanBoyd

      I really like Eric's speaking and presentation style. To the point that i hunted down the three other talks he has given.

      I also really like that he really goes into detail to explain something. Something that might have been a throwaway comment is actually backed up by dragging us into the assembly. Other talks he gives up the architecture diagram and shows instructions getting queued into pipelines. With clear diagrams it really helps drive the point home as real.

      He also seems to have a uhh Jeff Goldbum speaking style. So when it's trying to vectorize my code, it's nice to hear that the compiler uhh...finds a way.

    • User profile image
      anon

      His tonality just makes it sound like he really wants to be somewhere else most of the time. I like the actual choice of words and pacing though.

    • User profile image
      Fernando Pelliccioni

      Excelent talk! Thanks!

    • User profile image
      Nick

      Eric is a great speaker! I like his style, too.
      Unlike many other talks, i found this one really informative.

    • User profile image
      sean

      Great talk! Really enjoyed this.

    • User profile image
      fenbf

      At first sight it seems the presentation is slow and does not provide much of information. But actually it is the opposite: after watching it a lot of stuff is left in my head (a)

      My summary of the talk is here. But in general, the most important thing to remember: "Highly efficient code is actually memory efficient code". 

    Comments closed

    Comments have been closed since this content was published more than 30 days ago, but if you'd like to continue the conversation, please create a new thread in our Forums, or Contact Us and let us know.