Skip to main content

ComputeBackend

yule_gpu

Trait ComputeBackend 

Source 
pub trait ComputeBackend: Send + Sync {

Show 19 methods    // Required methods
    fn name(&self) -> &str;
    fn device_info(&self) -> DeviceInfo;
    fn allocate(&self, size_bytes: usize) -> Result<BufferHandle>;
    fn free(&self, handle: BufferHandle) -> Result<()>;
    fn matmul(
        &self,
        a: &BufferHandle,
        b: &BufferHandle,
        out: &BufferHandle,
        m: u32,
        n: u32,
        k: u32,
    ) -> Result<()>;
    fn softmax(
        &self,
        input: &BufferHandle,
        output: &BufferHandle,
        size: u32,
    ) -> Result<()>;
    fn rms_norm(
        &self,
        input: &BufferHandle,
        weight: &BufferHandle,
        output: &BufferHandle,
        size: u32,
        eps: f32,
    ) -> Result<()>;
    fn rope(
        &self,
        q: &BufferHandle,
        k: &BufferHandle,
        pos: u32,
        head_dim: u32,
        freq_base: f32,
        n_heads_q: u32,
        n_heads_k: u32,
    ) -> Result<()>;
    fn silu(
        &self,
        input: &BufferHandle,
        output: &BufferHandle,
        size: u32,
    ) -> Result<()>;
    fn element_mul(
        &self,
        a: &BufferHandle,
        b: &BufferHandle,
        output: &BufferHandle,
        size: u32,
    ) -> Result<()>;
    fn add(
        &self,
        a: &BufferHandle,
        b: &BufferHandle,
        output: &BufferHandle,
        size: u32,
    ) -> Result<()>;
    fn copy_to_device(&self, data: &[u8], handle: &BufferHandle) -> Result<()>;
    fn copy_from_device(
        &self,
        handle: &BufferHandle,
        data: &mut [u8],
    ) -> Result<()>;
    fn copy_buffer(
        &self,
        src: &BufferHandle,
        dst: &BufferHandle,
        size: usize,
    ) -> Result<()>;
    fn copy_buffer_offset(
        &self,
        src: &BufferHandle,
        dst: &BufferHandle,
        src_offset: usize,
        dst_offset: usize,
        size: usize,
    ) -> Result<()>;
    fn synchronize(&self) -> Result<()>;

    // Provided methods
    fn attn_score(
        &self,
        _q: &BufferHandle,
        _k_cache: &BufferHandle,
        _scores: &BufferHandle,
        _head_dim: u32,
        _seq_len: u32,
        _head_offset: u32,
        _kv_offset: u32,
        _kv_stride: u32,
    ) -> Result<()> { ... }
    fn attn_value(
        &self,
        _weights: &BufferHandle,
        _v_cache: &BufferHandle,
        _output: &BufferHandle,
        _head_dim: u32,
        _seq_len: u32,
        _kv_offset: u32,
        _kv_stride: u32,
        _out_offset: u32,
    ) -> Result<()> { ... }
    fn quantized_matmul(
        &self,
        _weights: &BufferHandle,
        _input: &BufferHandle,
        _output: &BufferHandle,
        _n_rows: u32,
        _n_cols: u32,
        _dtype: DType,
    ) -> Result<()> { ... }

}

Required Methods§

Source

fn name(&self) -> &str

Source

fn device_info(&self) -> DeviceInfo

Source

fn allocate(&self, size_bytes: usize) -> Result<BufferHandle>

Source

fn free(&self, handle: BufferHandle) -> Result<()>

Source

fn matmul( &self, a: &BufferHandle, b: &BufferHandle, out: &BufferHandle, m: u32, n: u32, k: u32, ) -> Result<()>

Source

fn softmax( &self, input: &BufferHandle, output: &BufferHandle, size: u32, ) -> Result<()>

Source

fn rms_norm( &self, input: &BufferHandle, weight: &BufferHandle, output: &BufferHandle, size: u32, eps: f32, ) -> Result<()>

Source

fn rope( &self, q: &BufferHandle, k: &BufferHandle, pos: u32, head_dim: u32, freq_base: f32, n_heads_q: u32, n_heads_k: u32, ) -> Result<()>

Source

fn silu( &self, input: &BufferHandle, output: &BufferHandle, size: u32, ) -> Result<()>

Source

fn element_mul( &self, a: &BufferHandle, b: &BufferHandle, output: &BufferHandle, size: u32, ) -> Result<()>

Source

fn add( &self, a: &BufferHandle, b: &BufferHandle, output: &BufferHandle, size: u32, ) -> Result<()>

Source

fn copy_to_device(&self, data: &[u8], handle: &BufferHandle) -> Result<()>

Source

fn copy_from_device(&self, handle: &BufferHandle, data: &mut [u8]) -> Result<()>

Source

fn copy_buffer( &self, src: &BufferHandle, dst: &BufferHandle, size: usize, ) -> Result<()>

Source

fn copy_buffer_offset( &self, src: &BufferHandle, dst: &BufferHandle, src_offset: usize, dst_offset: usize, size: usize, ) -> Result<()>

Source

fn synchronize(&self) -> Result<()>

Provided Methods§

Source

fn attn_score( &self, _q: &BufferHandle, _k_cache: &BufferHandle, _scores: &BufferHandle, _head_dim: u32, _seq_len: u32, _head_offset: u32, _kv_offset: u32, _kv_stride: u32, ) -> Result<()>

Compute attention scores: scores[pos] = Q[head_offset..] · K_cache[pos*kv_stride+kv_offset..]

Source

fn attn_value( &self, _weights: &BufferHandle, _v_cache: &BufferHandle, _output: &BufferHandle, _head_dim: u32, _seq_len: u32, _kv_offset: u32, _kv_stride: u32, _out_offset: u32, ) -> Result<()>

Compute weighted value aggregation: out[out_offset+d] = sum_pos(weights[pos] * V[pos*kv_stride+kv_offset+d])

Source

fn quantized_matmul( &self, _weights: &BufferHandle, _input: &BufferHandle, _output: &BufferHandle, _n_rows: u32, _n_cols: u32, _dtype: DType, ) -> Result<()>

Fused dequantize + matrix-vector multiply for quantized weights. GPU backends override this for fused VRAM kernels. Default impl falls back to regular matmul (assumes pre-dequantized data).

Implementors§