diff options
Diffstat (limited to 'candle-core/src/quantized/k_quants.rs')
| -rw-r--r-- | candle-core/src/quantized/k_quants.rs | 12 |
1 files changed, 12 insertions, 0 deletions
diff --git a/candle-core/src/quantized/k_quants.rs b/candle-core/src/quantized/k_quants.rs index 7f14600b..e7404529 100644 --- a/candle-core/src/quantized/k_quants.rs +++ b/candle-core/src/quantized/k_quants.rs @@ -671,7 +671,11 @@ impl GgmlType for BlockQ2K { const BLCK_SIZE: usize = QK_K; type VecDotType = BlockQ8K; + #[allow(unreachable_code)] fn vec_dot(n: usize, xs: &[Self], ys: &[Self::VecDotType]) -> Result<f32> { + #[cfg(target_feature = "avx")] + return super::avx::vec_dot_q2k_q8k(n, xs, ys); + if n % QK_K != 0 { crate::bail!("vec_dot_q2k_q8k: {n} is not divisible by {QK_K}") } @@ -834,7 +838,11 @@ impl GgmlType for BlockQ3K { const BLCK_SIZE: usize = QK_K; type VecDotType = BlockQ8K; + #[allow(unreachable_code)] fn vec_dot(n: usize, xs: &[Self], ys: &[Self::VecDotType]) -> Result<f32> { + #[cfg(target_feature = "avx")] + return super::avx::vec_dot_q3k_q8k(n, xs, ys); + if n % QK_K != 0 { crate::bail!("vec_dot_q3k_q8k: {n} is not divisible by {QK_K}") } @@ -1288,7 +1296,11 @@ impl GgmlType for BlockQ5K { const BLCK_SIZE: usize = QK_K; type VecDotType = BlockQ8K; + #[allow(unreachable_code)] fn vec_dot(n: usize, xs: &[Self], ys: &[Self::VecDotType]) -> Result<f32> { + #[cfg(target_feature = "avx")] + return super::avx::vec_dot_q5k_q8k(n, xs, ys); + if n % QK_K != 0 { crate::bail!("vec_dot_q5k_q8k: {n} is not divisible by {QK_K}") } |