huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-19 19:58:35 +00:00
Code Issues Packages Projects Releases Wiki Activity

Support the Accelerate BLAS on macOS. (#325)

Browse Source 
* Add the accelerate feature.

* Ffi tweaks.
This commit is contained in:
Laurent Mazare
2023-08-05 17:25:24 +01:00
committed by GitHub
parent 0b175fcbbd
commit b278834267
12 changed files with 241 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

111
candle-core/src/accelerate.rs Normal file
View File

@ -0,0 +1,111 @@
#![allow(dead_code)]
use libc::{c_char, c_double, c_float, c_int};
mod ffi {
use super::*;
extern "C" {
// It would be nice to be able to switch to the NEWLAPACK version of the function but this
// seems to trigger some link error. Available function names can be seen here:
// /Library/Developer/CommandLineTools/SDKs/MacOSX13.3.sdk/System/Library/Frameworks/Accelerate.framework/Versions/A/Accelerate.tbd
#[link_name = "sgemm_"]
pub fn sgemm_ffi(
transa: *const c_char,
transb: *const c_char,
m: *const c_int,
n: *const c_int,
k: *const c_int,
alpha: *const c_float,
a: *const c_float,
lda: *const c_int,
b: *const c_float,
ldb: *const c_int,
beta: *const c_float,
c: *mut c_float,
ldc: *const c_int,
);
#[link_name = "dgemm_"]
pub fn dgemm_ffi(
transa: *const c_char,
transb: *const c_char,
m: *const c_int,
n: *const c_int,
k: *const c_int,
alpha: *const c_double,
a: *const c_double,
lda: *const c_int,
b: *const c_double,
ldb: *const c_int,
beta: *const c_double,
c: *mut c_double,
ldc: *const c_int,
);
}
}
#[allow(clippy::too_many_arguments)]
#[inline]
pub unsafe fn sgemm(
transa: u8,
transb: u8,
m: i32,
n: i32,
k: i32,
alpha: f32,
a: &[f32],
lda: i32,
b: &[f32],
ldb: i32,
beta: f32,
c: &mut [f32],
ldc: i32,
) {
ffi::sgemm_ffi(
&(transa as c_char),
&(transb as c_char),
&m,
&n,
&k,
&alpha,
a.as_ptr(),
&lda,
b.as_ptr(),
&ldb,
&beta,
c.as_mut_ptr(),
&ldc,
)
}
#[allow(clippy::too_many_arguments)]
#[inline]
pub unsafe fn dgemm(
transa: u8,
transb: u8,
m: i32,
n: i32,
k: i32,
alpha: f64,
a: &[f64],
lda: i32,
b: &[f64],
ldb: i32,
beta: f64,
c: &mut [f64],
ldc: i32,
) {
ffi::dgemm_ffi(
&(transa as c_char),
&(transb as c_char),
&m,
&n,
&k,
&alpha,
a.as_ptr(),
&lda,
b.as_ptr(),
&ldb,
&beta,
c.as_mut_ptr(),
&ldc,
)
}