New optimizations for X86 in upcoming GCC 5.0

#1

Post by ingve » Sat, Nov 29, 2014, 10:14 PM UTC

New optimizations for X86 in upcoming GCC 5.0
software.intel.com

Re: New optimizations for X86 in upcoming GCC 5.0

#2

Post by janvdberg » Sat, Nov 29, 2014, 11:27 PM UTC

As this is the number 1 story on HN now: can someone explain what kind of improvements we will see in day to day use from this?

Re: New optimizations for X86 in upcoming GCC 5.0

#3

Post by justincormack » Sat, Nov 29, 2014, 11:27 PM UTC

Here is a more general list of gcc 5 changes https://gcc.gnu.org/gcc-5/changes.html

EDIT was looking at this for __builtin_mul_overflow which apparently are in clang already, for testing overflow of arbitrary types.

Re: New optimizations for X86 in upcoming GCC 5.0

#4

Post by ams6110 » Sat, Nov 29, 2014, 11:36 PM UTC

janvdberg wrote: post #2

As this is the number 1 story on HN now: can someone explain what kind of improvements we will see in day to day use from this?

From the story....

  1. Image conversion (RGB structure to some other)

  2. N-dimentional coordinates. (Normalize array of XYZ points)

  3. Multiplication of vectors by constant matrix

Re: New optimizations for X86 in upcoming GCC 5.0

#5

Post by azraai » Sat, Nov 29, 2014, 11:38 PM UTC

janvdberg wrote: post #2

As this is the number 1 story on HN now: can someone explain what kind of improvements we will see in day to day use from this?

Faster multi dimensional loops. Quicker raw image compression.

That are few things that came across my mind.

Re: New optimizations for X86 in upcoming GCC 5.0

#6

Post by ape4 » Sat, Nov 29, 2014, 11:40 PM UTC

Competition (with clang) is good.

Re: New optimizations for X86 in upcoming GCC 5.0

#7

Post by _wmd » Sat, Nov 29, 2014, 11:41 PM UTC

janvdberg wrote: post #2

As this is the number 1 story on HN now: can someone explain what kind of improvements we will see in day to day use from this?

The article mentions two: unpacking 24 bit values and matrix multiplication, both might be something you'd find in a software video (de)compressor or a computer game

Re: New optimizations for X86 in upcoming GCC 5.0

#8

Post by medgno » Sat, Nov 29, 2014, 11:51 PM UTC

I'm so glad to see auto-vectorization happening more and more often. However, I wonder whether a language that had built-in support for primitive floating-point vector types (e.g., GLSL's vec3, vec4, mat3, mat4) could help the compiler with performing these sorts of optimizations.

Re: New optimizations for X86 in upcoming GCC 5.0

#9

Post by cremno » Sat, Nov 29, 2014, 11:53 PM UTC

justincormack wrote: post #3

Here is a more general list of gcc 5 changes https://gcc.gnu.org/gcc-5/changes.html EDIT was looking at this for __builtin_mul_overflow which apparently are in clang already, for testing overflow of arbitrary types.

Actually Clang doesn't implement the listed built-in functions (yet?). They're type-generic versions.

http://llvm.org/releases/3.5.0/tools/clang/docs/LanguageExte...

http://clang.llvm.org/docs/LanguageExtensions.html#checked-a...

Re: New optimizations for X86 in upcoming GCC 5.0

#10

Post by vinkelhake » Sun, Nov 30, 2014, 12:01 AM UTC

_wmd wrote: post #7

janvdberg wrote: post #2

As this is the number 1 story on HN now: can someone explain what kind of improvements we will see in day to day use from this?

The article mentions two: unpacking 24 bit values and matrix multiplication, both might be something you'd find in a software video (de)compressor or a computer game

Where did you get 24 from? It's about vectorizing loads of 3 elements. The elements don't have to be 8-bit. The second example uses floats.