diff options
| author | Laurent Mazare <laurent.mazare@gmail.com> | 2023-08-25 10:14:49 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2023-08-25 10:14:49 +0100 |
| commit | afc10a3232b218dcdb8c3b0989f1066940ea992b (patch) | |
| tree | 0648bca15ca71d1e78c61ae79afa2d039447b247 /candle-core/src/quantized/k_quants.rs | |
| parent | d728e646c20e773498b859fe41f4109f86320ca6 (diff) | |
| download | candle-afc10a3232b218dcdb8c3b0989f1066940ea992b.tar.gz candle-afc10a3232b218dcdb8c3b0989f1066940ea992b.tar.bz2 candle-afc10a3232b218dcdb8c3b0989f1066940ea992b.zip | |
AVX version for the q8-0 multiplications. (#598)
Diffstat (limited to 'candle-core/src/quantized/k_quants.rs')
| -rw-r--r-- | candle-core/src/quantized/k_quants.rs | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/candle-core/src/quantized/k_quants.rs b/candle-core/src/quantized/k_quants.rs index 36efe2f2..02022480 100644 --- a/candle-core/src/quantized/k_quants.rs +++ b/candle-core/src/quantized/k_quants.rs @@ -421,7 +421,11 @@ impl GgmlType for BlockQ8_0 { Ok(()) } + #[allow(unreachable_code)] fn vec_dot(n: usize, xs: &[Self], ys: &[Self::VecDotType]) -> Result<f32> { + #[cfg(target_feature = "avx")] + return super::avx::vec_dot_q8_0_q8_0(n, xs, ys); + let qk = QK8_0; if n % QK8_0 != 0 { crate::bail!("vec_dot_q8_0_q8_0: {n} is not divisible by {qk}") |