phase 4: transformer core kernels

CUDA kernels (csrc/):
- common.cuh: shared warp_reduce_sum/max, block_reduce_sum/max
- normalization/rmsnorm.cu: RMSNorm (F32 + BF16)
- normalization/layernorm.cu: LayerNorm with Welford (F32 + BF16)
- activation/activations.cu: GELU tanh-approx + SiLU (F32 + BF16)
- reduce/softmax.cu: safe softmax, 3-pass (F32 + BF16)
- embedding/embedding.cu: gather lookup (F32 + BF16)
- embedding/rope.cu: RoPE in-place + precomputed cos/sin cache (F32 + BF16)

Rust wrappers (xserv-kernels/src/):
- rmsnorm.rs, layernorm.rs, activation.rs, softmax.rs, embedding.rs, rope.rs
- RopeCache struct with GPU-side precomputation

Tests: 12 new tests (ops_test.rs), all passing with good precision:
- F32: max_err 1e-6 ~ 1e-9
- BF16: max_err 2e-3 ~ 7e-3
Total: 29 kernel tests + 27 prior = 56 tests passing

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/xserv-kernels/src/activation.rs

@@ -0,0 +1,41 @@
+use std::ffi::c_void;
+use xserv_tensor::{DType, Device, Tensor};
+
+unsafe extern "C" {
+    fn launch_gelu_f32(x: *const c_void, out: *mut c_void, n: i32, stream: *mut c_void);
+    fn launch_gelu_bf16(x: *const c_void, out: *mut c_void, n: i32, stream: *mut c_void);
+    fn launch_silu_f32(x: *const c_void, out: *mut c_void, n: i32, stream: *mut c_void);
+    fn launch_silu_bf16(x: *const c_void, out: *mut c_void, n: i32, stream: *mut c_void);
+}
+
+pub fn gelu(x: &Tensor) -> Tensor {
+    assert!(x.is_contiguous());
+    assert!(matches!(x.device(), Device::Cuda(_)));
+    let out = Tensor::zeros(x.shape(), x.dtype(), x.device());
+    let n = x.numel() as i32;
+    unsafe {
+        match x.dtype() {
+            DType::F32 => launch_gelu_f32(x.data_ptr() as _, out.data_ptr() as *mut c_void, n, std::ptr::null_mut()),
+            DType::BF16 => launch_gelu_bf16(x.data_ptr() as _, out.data_ptr() as *mut c_void, n, std::ptr::null_mut()),
+            _ => panic!("unsupported dtype for gelu"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+    out
+}
+
+pub fn silu(x: &Tensor) -> Tensor {
+    assert!(x.is_contiguous());
+    assert!(matches!(x.device(), Device::Cuda(_)));
+    let out = Tensor::zeros(x.shape(), x.dtype(), x.device());
+    let n = x.numel() as i32;
+    unsafe {
+        match x.dtype() {
+            DType::F32 => launch_silu_f32(x.data_ptr() as _, out.data_ptr() as *mut c_void, n, std::ptr::null_mut()),
+            DType::BF16 => launch_silu_bf16(x.data_ptr() as _, out.data_ptr() as *mut c_void, n, std::ptr::null_mut()),
+            _ => panic!("unsupported dtype for silu"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+    out
+}

crates/xserv-kernels/src/embedding.rs

@@ -0,0 +1,51 @@
+use std::ffi::c_void;
+use xserv_cuda::GpuBuffer;
+use xserv_tensor::{DType, Device, Tensor};
+
+unsafe extern "C" {
+    fn launch_embedding_f32(table: *const c_void, token_ids: *const c_void, out: *mut c_void,
+                            num_tokens: i32, hidden_size: i32, stream: *mut c_void);
+    fn launch_embedding_bf16(table: *const c_void, token_ids: *const c_void, out: *mut c_void,
+                             num_tokens: i32, hidden_size: i32, stream: *mut c_void);
+}
+
+/// Embedding lookup: table[token_ids[i]] for each i.
+/// table: [vocab_size, hidden_size], token_ids: [num_tokens] (i32 on CPU)
+pub fn embedding(table: &Tensor, token_ids: &[u32]) -> Tensor {
+    assert_eq!(table.ndim(), 2);
+    assert!(table.is_contiguous());
+    assert!(matches!(table.device(), Device::Cuda(_)));
+
+    let hidden_size = table.shape()[1];
+    let num_tokens = token_ids.len();
+
+    // Upload token_ids to GPU
+    let ids_bytes = unsafe {
+        std::slice::from_raw_parts(
+            token_ids.as_ptr() as *const u8,
+            num_tokens * std::mem::size_of::<u32>(),
+        )
+    };
+    let mut ids_gpu = GpuBuffer::alloc(ids_bytes.len()).expect("alloc token_ids");
+    ids_gpu.copy_from_host(ids_bytes).unwrap();
+
+    let out = Tensor::zeros(&[num_tokens, hidden_size], table.dtype(), table.device());
+
+    unsafe {
+        match table.dtype() {
+            DType::F32 => launch_embedding_f32(
+                table.data_ptr() as _, ids_gpu.as_ptr() as _,
+                out.data_ptr() as *mut c_void,
+                num_tokens as i32, hidden_size as i32, std::ptr::null_mut(),
+            ),
+            DType::BF16 => launch_embedding_bf16(
+                table.data_ptr() as _, ids_gpu.as_ptr() as _,
+                out.data_ptr() as *mut c_void,
+                num_tokens as i32, hidden_size as i32, std::ptr::null_mut(),
+            ),
+            _ => panic!("unsupported dtype for embedding"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+    out
+}

crates/xserv-kernels/src/layernorm.rs

@@ -0,0 +1,39 @@
+use std::ffi::c_void;
+use xserv_tensor::{DType, Device, Tensor};
+
+unsafe extern "C" {
+    fn launch_layernorm_f32(x: *const c_void, gamma: *const c_void, beta: *const c_void,
+                            out: *mut c_void, rows: i32, hidden_size: i32, eps: f32, stream: *mut c_void);
+    fn launch_layernorm_bf16(x: *const c_void, gamma: *const c_void, beta: *const c_void,
+                             out: *mut c_void, rows: i32, hidden_size: i32, eps: f32, stream: *mut c_void);
+}
+
+pub fn layernorm(x: &Tensor, gamma: &Tensor, beta: &Tensor, eps: f32) -> Tensor {
+    assert!(x.ndim() >= 1);
+    assert!(x.is_contiguous() && gamma.is_contiguous() && beta.is_contiguous());
+    assert!(matches!(x.device(), Device::Cuda(_)));
+    let hidden_size = *x.shape().last().unwrap();
+    assert_eq!(gamma.shape(), &[hidden_size]);
+    assert_eq!(beta.shape(), &[hidden_size]);
+
+    let rows = x.numel() / hidden_size;
+    let out = Tensor::zeros(x.shape(), x.dtype(), x.device());
+
+    unsafe {
+        match x.dtype() {
+            DType::F32 => launch_layernorm_f32(
+                x.data_ptr() as _, gamma.data_ptr() as _, beta.data_ptr() as _,
+                out.data_ptr() as *mut c_void,
+                rows as i32, hidden_size as i32, eps, std::ptr::null_mut(),
+            ),
+            DType::BF16 => launch_layernorm_bf16(
+                x.data_ptr() as _, gamma.data_ptr() as _, beta.data_ptr() as _,
+                out.data_ptr() as *mut c_void,
+                rows as i32, hidden_size as i32, eps, std::ptr::null_mut(),
+            ),
+            _ => panic!("unsupported dtype for layernorm"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+    out
+}

crates/xserv-kernels/src/lib.rs

@@ -1,3 +1,15 @@
+pub mod activation;
+pub mod embedding;
 pub mod gemm;
+pub mod layernorm;
+pub mod rmsnorm;
+pub mod rope;
+pub mod softmax;
 
-pub use gemm::{GemmBackend, matmul};
+pub use activation::{gelu, silu};
+pub use embedding::embedding;
+pub use gemm::{matmul, GemmBackend};
+pub use layernorm::layernorm;
+pub use rmsnorm::rmsnorm;
+pub use rope::{rope_inplace, RopeCache};
+pub use softmax::softmax;

crates/xserv-kernels/src/rmsnorm.rs

@@ -0,0 +1,37 @@
+use std::ffi::c_void;
+use xserv_tensor::{DType, Device, Tensor};
+
+unsafe extern "C" {
+    fn launch_rmsnorm_f32(x: *const c_void, gamma: *const c_void, out: *mut c_void,
+                          rows: i32, hidden_size: i32, eps: f32, stream: *mut c_void);
+    fn launch_rmsnorm_bf16(x: *const c_void, gamma: *const c_void, out: *mut c_void,
+                           rows: i32, hidden_size: i32, eps: f32, stream: *mut c_void);
+}
+
+pub fn rmsnorm(x: &Tensor, gamma: &Tensor, eps: f32) -> Tensor {
+    assert!(x.ndim() >= 1);
+    assert!(x.is_contiguous() && gamma.is_contiguous());
+    assert!(matches!(x.device(), Device::Cuda(_)));
+    let hidden_size = *x.shape().last().unwrap();
+    assert_eq!(gamma.shape(), &[hidden_size]);
+    assert_eq!(x.dtype(), gamma.dtype());
+
+    let rows = x.numel() / hidden_size;
+    let out = Tensor::zeros(x.shape(), x.dtype(), x.device());
+
+    unsafe {
+        match x.dtype() {
+            DType::F32 => launch_rmsnorm_f32(
+                x.data_ptr() as _, gamma.data_ptr() as _, out.data_ptr() as *mut c_void,
+                rows as i32, hidden_size as i32, eps, std::ptr::null_mut(),
+            ),
+            DType::BF16 => launch_rmsnorm_bf16(
+                x.data_ptr() as _, gamma.data_ptr() as _, out.data_ptr() as *mut c_void,
+                rows as i32, hidden_size as i32, eps, std::ptr::null_mut(),
+            ),
+            _ => panic!("unsupported dtype for rmsnorm"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+    out
+}

crates/xserv-kernels/src/rope.rs

@@ -0,0 +1,85 @@
+use std::ffi::c_void;
+use xserv_cuda::GpuBuffer;
+use xserv_tensor::{DType, Device, Tensor};
+
+unsafe extern "C" {
+    fn launch_rope_f32(x: *mut c_void, cos_cache: *const c_void, sin_cache: *const c_void,
+                       positions: *const c_void, num_tokens: i32, num_heads: i32,
+                       head_dim: i32, stream: *mut c_void);
+    fn launch_rope_bf16(x: *mut c_void, cos_cache: *const c_void, sin_cache: *const c_void,
+                        positions: *const c_void, num_tokens: i32, num_heads: i32,
+                        head_dim: i32, stream: *mut c_void);
+    fn launch_compute_rope_cache(cos_cache: *mut c_void, sin_cache: *mut c_void,
+                                 max_seq_len: i32, half_dim: i32, theta: f32,
+                                 stream: *mut c_void);
+}
+
+pub struct RopeCache {
+    pub cos: GpuBuffer,
+    pub sin: GpuBuffer,
+    pub max_seq_len: usize,
+    pub half_dim: usize,
+}
+
+impl RopeCache {
+    pub fn new(max_seq_len: usize, head_dim: usize, theta: f32) -> Self {
+        let half_dim = head_dim / 2;
+        let nbytes = max_seq_len * half_dim * std::mem::size_of::<f32>();
+        let mut cos = GpuBuffer::alloc(nbytes).expect("alloc cos_cache");
+        let mut sin = GpuBuffer::alloc(nbytes).expect("alloc sin_cache");
+
+        unsafe {
+            launch_compute_rope_cache(
+                cos.as_mut_ptr() as _, sin.as_mut_ptr() as _,
+                max_seq_len as i32, half_dim as i32, theta, std::ptr::null_mut(),
+            );
+        }
+        xserv_cuda::device::synchronize().unwrap();
+
+        Self { cos, sin, max_seq_len, half_dim }
+    }
+}
+
+/// Apply RoPE in-place to x.
+/// x: [num_tokens, num_heads, head_dim] on GPU
+/// positions: [num_tokens] (u32 on CPU, will be uploaded)
+pub fn rope_inplace(x: &Tensor, cache: &RopeCache, positions: &[u32]) {
+    assert_eq!(x.ndim(), 3);
+    assert!(x.is_contiguous());
+    assert!(matches!(x.device(), Device::Cuda(_)));
+    let num_tokens = x.shape()[0];
+    let num_heads = x.shape()[1];
+    let head_dim = x.shape()[2];
+    assert_eq!(head_dim / 2, cache.half_dim);
+    assert_eq!(positions.len(), num_tokens);
+
+    let pos_bytes = unsafe {
+        std::slice::from_raw_parts(
+            positions.as_ptr() as *const u8,
+            num_tokens * std::mem::size_of::<u32>(),
+        )
+    };
+    let mut pos_gpu = GpuBuffer::alloc(pos_bytes.len()).expect("alloc positions");
+    pos_gpu.copy_from_host(pos_bytes).unwrap();
+
+    unsafe {
+        match x.dtype() {
+            DType::F32 => launch_rope_f32(
+                x.data_ptr() as *mut c_void,
+                cache.cos.as_ptr() as _, cache.sin.as_ptr() as _,
+                pos_gpu.as_ptr() as _,
+                num_tokens as i32, num_heads as i32, head_dim as i32,
+                std::ptr::null_mut(),
+            ),
+            DType::BF16 => launch_rope_bf16(
+                x.data_ptr() as *mut c_void,
+                cache.cos.as_ptr() as _, cache.sin.as_ptr() as _,
+                pos_gpu.as_ptr() as _,
+                num_tokens as i32, num_heads as i32, head_dim as i32,
+                std::ptr::null_mut(),
+            ),
+            _ => panic!("unsupported dtype for rope"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+}

crates/xserv-kernels/src/softmax.rs

@@ -0,0 +1,34 @@
+use std::ffi::c_void;
+use xserv_tensor::{DType, Device, Tensor};
+
+unsafe extern "C" {
+    fn launch_softmax_f32(x: *const c_void, out: *mut c_void, rows: i32, cols: i32, stream: *mut c_void);
+    fn launch_softmax_bf16(x: *const c_void, out: *mut c_void, rows: i32, cols: i32, stream: *mut c_void);
+}
+
+/// Softmax along the last dimension.
+pub fn softmax(x: &Tensor) -> Tensor {
+    assert!(x.ndim() >= 1);
+    assert!(x.is_contiguous());
+    assert!(matches!(x.device(), Device::Cuda(_)));
+
+    let cols = *x.shape().last().unwrap();
+    let rows = x.numel() / cols;
+    let out = Tensor::zeros(x.shape(), x.dtype(), x.device());
+
+    unsafe {
+        match x.dtype() {
+            DType::F32 => launch_softmax_f32(
+                x.data_ptr() as _, out.data_ptr() as *mut c_void,
+                rows as i32, cols as i32, std::ptr::null_mut(),
+            ),
+            DType::BF16 => launch_softmax_bf16(
+                x.data_ptr() as _, out.data_ptr() as *mut c_void,
+                rows as i32, cols as i32, std::ptr::null_mut(),
+            ),
+            _ => panic!("unsupported dtype for softmax"),
+        }
+    }
+    xserv_cuda::device::synchronize().unwrap();
+    out
+}