Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

CodeGen: Rewrite dot product lowering using a dedicated IR instruction #1512

Merged
merged 6 commits into from
Nov 9, 2024
Merged

CodeGen: Rewrite dot product lowering using a dedicated IR instruction #1512

merged 6 commits into from
Nov 9, 2024

Commits on Nov 8, 2024

  1. CodeGen: Implement support for vdpps AVX instruction 

    This will help optimize lowering of vector.dot and vector.normalize
    @zeux
    zeux committed Nov 8, 2024
    Configuration menu
    Copy the full SHA
    d2c008c View commit details
    Browse the repository at this point in the history

  2. CodeGen: Implement DOT_VEC IR opcode 

    This exposes vdpps on X64 and allows to compute a 3-wide dot product for
two vectors, returning the result as a number.
    @zeux
    zeux committed Nov 8, 2024
    Configuration menu
    Copy the full SHA
    74a9128 View commit details
    Browse the repository at this point in the history

  3. CodeGen: Use DOT_VEC under a fast flag for lowering vector lib 

    This is useful for vector.dot, vector.magnitude and vector.normalize.
    @zeux
    zeux committed Nov 8, 2024
    Configuration menu
    Copy the full SHA
    cd73807 View commit details
    Browse the repository at this point in the history

  4. CodeGen: Implement a naive version of A64 DOT_VEC 

    This is using existing instructions and scalar adds to have a baseline.
This is still faster than the original implementation of vector. ops.
    @zeux
    zeux committed Nov 8, 2024
    Configuration menu
    Copy the full SHA
    8fc458e View commit details
    Browse the repository at this point in the history

  5. CodeGen: Implement faddp opcode for A64 

    We now support scalar version of faddp opcode which can add the
first two floats of the vector into the first scalar of the destination.
    @zeux
    zeux committed Nov 8, 2024
    Configuration menu
    Copy the full SHA
    6ebac70 View commit details
    Browse the repository at this point in the history

  6. CodeGen: Rewrite DOT_VEC lowering for A64 using faddp 

    This results in about the same performance as a naive version on M2,
but uses fewer registers and is what clang generates for a similar
source.
    @zeux
    zeux committed Nov 8, 2024
    Configuration menu
    Copy the full SHA
    81b691b View commit details
    Browse the repository at this point in the history