issue/848: add paged attention prefill for nvidia gpu with test pass

include/infiniop.h

@@ -32,6 +32,7 @@
 #include "infiniop/ops/zeros.h"
 #include "infiniop/tensor_descriptor.h"
 #include "infiniop/ops/paged_attention.h"
+#include "infiniop/ops/paged_attention_prefill.h"
 #include "infiniop/ops/paged_caching.h"
 
 #endif // __INFINIOP_API_H__

include/infiniop/ops/paged_attention_prefill.h

@@ -0,0 +1,79 @@
+#ifndef __INFINIOP_PAGED_ATTENTION_PREFILL_API_H__
+#define __INFINIOP_PAGED_ATTENTION_PREFILL_API_H__
+
+#include "../operator_descriptor.h"
+
+// Define an opaque handle for the Paged Attention Prefill descriptor.
+typedef struct InfiniopDescriptor *infiniopPagedAttentionPrefillDescriptor_t;
+
+/**
+ * @brief Creates a descriptor for the Paged Attention Prefill operation.
+ * * @param handle The handle to the InfiniOP library context.
+ * @param desc_ptr A pointer to store the created descriptor.
+ * @param out_desc Descriptor for the output tensor.
+ * @param q_desc Descriptor for the query tensor.
+ * @param k_cache_desc Descriptor for the global physical key cache.
+ * @param v_cache_desc Descriptor for the global physical value cache.
+ * @param block_tables_desc Descriptor for the block tables mapping logic to physical blocks.
+ * @param cache_lens_desc Descriptor for the total sequence lengths (history + current).
+ * @param seq_lens_desc Descriptor for the current prefill sequence lengths.
+ * @param alibi_slopes_desc Optional descriptor for the ALiBi slopes tensor. Can be NULL.
+ * @param scale The attention scaling factor.
+ * @return infiniStatus_t Status code of the operation.
+ */
+__C __export infiniStatus_t infiniopCreatePagedAttentionPrefillDescriptor(
+    infiniopHandle_t handle,
+    infiniopPagedAttentionPrefillDescriptor_t *desc_ptr,
+    infiniopTensorDescriptor_t out_desc,
+    infiniopTensorDescriptor_t q_desc,
+    infiniopTensorDescriptor_t k_cache_desc,
+    infiniopTensorDescriptor_t v_cache_desc,
+    infiniopTensorDescriptor_t block_tables_desc,
+    infiniopTensorDescriptor_t cache_lens_desc,
+    infiniopTensorDescriptor_t seq_lens_desc,
+    infiniopTensorDescriptor_t alibi_slopes_desc,
+    float scale);
+
+/**
+ * @brief Retrieves the workspace size required for the Paged Attention Prefill operation.
+ */
+__C __export infiniStatus_t infiniopGetPagedAttentionPrefillWorkspaceSize(
+    infiniopPagedAttentionPrefillDescriptor_t desc, size_t *size);
+
+/**
+ * @brief Executes the Paged Attention Prefill operation.
+ * * @param desc The Paged Attention Prefill descriptor.
+ * @param workspace Pointer to the workspace memory.
+ * @param workspace_size The size of the workspace.
+ * @param out Pointer to the output tensor data.
+ * @param q Pointer to the query tensor data.
+ * @param k_cache Pointer to the global key cache data.
+ * @param v_cache Pointer to the global value cache data.
+ * @param block_tables Pointer to the block tables data.
+ * @param cache_lens Pointer to the total sequence lengths data.
+ * @param seq_lens Pointer to the current prefill sequence lengths data.
+ * @param alibi_slopes Pointer to the ALiBi slopes data. Can be NULL.
+ * @param stream The CUDA/device stream for the operation.
+ * @return infiniStatus_t Status code of the operation.
+ */
+__C __export infiniStatus_t infiniopPagedAttentionPrefill(
+    infiniopPagedAttentionPrefillDescriptor_t desc,
+    void *workspace,
+    size_t workspace_size,
+    void *out,
+    const void *q,
+    const void *k_cache,
+    const void *v_cache,
+    const void *block_tables,
+    const void *cache_lens,
+    const void *seq_lens,
+    const void *alibi_slopes,
+    void *stream);
+
+/**
+ * @brief Destroys a Paged Attention Prefill descriptor.
+ */
+__C __export infiniStatus_t infiniopDestroyPagedAttentionPrefillDescriptor(
+    infiniopPagedAttentionPrefillDescriptor_t desc);
+
+#endif // __INFINIOP_PAGED_ATTENTION_PREFILL_API_H__

src/infiniop/ops/paged_attention_prefill/cuda/kernel.cuh

@@ -0,0 +1,161 @@
+#ifndef __PAGED_ATTENTION_PREFILL_KERNEL_CUH__
+#define __PAGED_ATTENTION_PREFILL_KERNEL_CUH__
+
+#include <cuda_fp16.h>
+#include <float.h>
+#include <math.h>
+
+namespace op::paged_attention_prefill::cuda {
+
+// =============================================================
+//  Correctness prioritized.
+// =============================================================
+
+template <typename Tdata, typename Tcompute>
+__global__ void pagedAttentionPrefillKernel(
+    Tdata *out_,
+    const Tdata *q_,
+    const Tdata *k_cache_,
+    const Tdata *v_cache_,
+    const int64_t *block_tables_,
+    const int64_t *cache_lens_,
+    const int64_t *seq_lens_,
+    const float *alibi_slopes_,
+    const size_t num_heads,
+    const size_t num_kv_heads,
+    const float scale,
+    const size_t max_num_blocks_per_seq,
+    const size_t block_size,
+    const size_t max_new_len,
+    const ptrdiff_t q_stride,
+    const ptrdiff_t kv_block_stride,
+    const ptrdiff_t kv_head_stride,
+    const ptrdiff_t o_stride,
+    const size_t head_size_const) {
+
+    // --- 1. Coordinate Setup ---
+    // Grid: (num_heads, max_new_len, num_seqs)
+    // Block: (HEAD_SIZE, 1, 1)
+    const int seq_idx = blockIdx.z;
+    const int q_token_idx = blockIdx.y;
+    const int head_idx = blockIdx.x;
+    const int dim_idx = threadIdx.x;
+
+    // Check boundary
+    const int64_t cur_new_len = seq_lens_[seq_idx];
+    if (q_token_idx >= cur_new_len) {
+        return;
+    }
+
+    // Safety check for head size
+    if (dim_idx >= head_size_const) {
+        return;
+    }
+
+    // Dimensions
+    const int64_t total_seq_len = cache_lens_[seq_idx];
+    const int64_t history_len = total_seq_len - cur_new_len;
+    const int64_t global_token_idx = history_len + q_token_idx;
+
+    const size_t num_queries_per_kv = num_heads / num_kv_heads;
+    const size_t kv_head_idx = head_idx / num_queries_per_kv;
+
+    const int64_t *block_table = block_tables_ + seq_idx * max_num_blocks_per_seq;
+
+    const Tdata *q_ptr_base = q_ + seq_idx * q_stride + q_token_idx * (num_heads * head_size_const) + head_idx * head_size_const;
+
+    Tdata *out_ptr = out_ + seq_idx * o_stride + q_token_idx * (num_heads * head_size_const) + head_idx * head_size_const;
+
+    const float alibi_slope = (alibi_slopes_ == nullptr) ? 0.0f : alibi_slopes_[head_idx];
+
+    // --- Pass 1: Find Global Max ---
+    Tcompute max_score = -FLT_MAX;
+
+    for (int t = 0; t < total_seq_len; ++t) {
+        if (t > global_token_idx) {
+            break;
+        }
+
+        const int64_t b_idx = t / block_size;
+        const int64_t t_off = t % block_size;
+        const int64_t physical_block = block_table[b_idx];
+
+        const Tdata *k_vec = k_cache_ + physical_block * kv_block_stride + kv_head_idx * kv_head_stride + t_off * head_size_const;
+
+        Tcompute score = 0.0f;
+        for (int d = 0; d < head_size_const; ++d) {
+            score += static_cast<Tcompute>(q_ptr_base[d]) * static_cast<Tcompute>(k_vec[d]);
+        }
+        score *= scale;
+
+        if (alibi_slope != 0.0f) {
+            score += alibi_slope * (t - total_seq_len + 1);
+        }
+
+        if (score > max_score) {
+            max_score = score;
+        }
+    }
+
+    // --- Pass 2: Calculate Denominator (Sum Exp) ---
+    Tcompute sum_exp = 0.0f;
+
+    for (int t = 0; t < total_seq_len; ++t) {
+        if (t > global_token_idx) {
+            break;
+        }
+
+        const int64_t b_idx = t / block_size;
+        const int64_t t_off = t % block_size;
+        const int64_t physical_block = block_table[b_idx];
+        const Tdata *k_vec = k_cache_ + physical_block * kv_block_stride + kv_head_idx * kv_head_stride + t_off * head_size_const;
+
+        Tcompute score = 0.0f;
+        for (int d = 0; d < head_size_const; ++d) {
+            score += static_cast<Tcompute>(q_ptr_base[d]) * static_cast<Tcompute>(k_vec[d]);
+        }
+        score *= scale;
+        if (alibi_slope != 0.0f) {
+            score += alibi_slope * (t - total_seq_len + 1);
+        }
+
+        sum_exp += expf(score - max_score);
+    }
+
+    // --- Pass 3: Calculate Weighted Sum (V) ---
+    Tcompute acc = 0.0f;
+    Tcompute inv_sum = 1.0f / (sum_exp + 1e-6f);
+
+    for (int t = 0; t < total_seq_len; ++t) {
+        if (t > global_token_idx) {
+            break;
+        }
+
+        const int64_t b_idx = t / block_size;
+        const int64_t t_off = t % block_size;
+        const int64_t physical_block = block_table[b_idx];
+
+        // Re-compute Score
+        const Tdata *k_vec = k_cache_ + physical_block * kv_block_stride + kv_head_idx * kv_head_stride + t_off * head_size_const;
+        Tcompute score = 0.0f;
+        for (int d = 0; d < head_size_const; ++d) {
+            score += static_cast<Tcompute>(q_ptr_base[d]) * static_cast<Tcompute>(k_vec[d]);
+        }
+        score *= scale;
+        if (alibi_slope != 0.0f) {
+            score += alibi_slope * (t - total_seq_len + 1);
+        }
+
+        Tcompute prob = expf(score - max_score) * inv_sum;
+
+        const Tdata *v_vec = v_cache_ + physical_block * kv_block_stride + kv_head_idx * kv_head_stride + t_off * head_size_const;
+
+        acc += prob * static_cast<Tcompute>(v_vec[dim_idx]);
+    }
+
+    out_ptr[dim_idx] = static_cast<Tdata>(acc);
+}
+
+} // namespace op::paged_attention_prefill::cuda
+
+#endif // __PAGED_ATTENTION_PREFILL_KERNEL_CUH__

src/infiniop/ops/paged_attention_prefill/info.h

@@ -0,0 +1,105 @@
+#ifndef __PAGED_ATTENTION_PREFILL_INFO_H__
+#define __PAGED_ATTENTION_PREFILL_INFO_H__
+
+#include "../../../utils.h"
+#include "../../tensor.h"
+#include <iostream>
+#include <optional>
+#include <vector>
+
+namespace op::paged_attention_prefill {
+
+class PagedAttentionPrefillInfo {
+    PagedAttentionPrefillInfo() = default;
+
+public:
+    // --- Data Types and Scale ---
+    infiniDtype_t dtype;
+    float scale;
+
+    // --- Shape Dimensions ---
+    size_t num_seqs;
+    size_t num_heads;
+    size_t num_kv_heads;
+    size_t head_size;
+    size_t block_size;
+    size_t max_num_blocks_per_seq;
+    size_t max_new_len;
+
+    // --- Strides for Memory Layout ---
+    ptrdiff_t q_stride;
+    ptrdiff_t kv_block_stride;
+    ptrdiff_t kv_head_stride;
+    ptrdiff_t o_stride;
+
+    static utils::Result<PagedAttentionPrefillInfo> create(
+        infiniopTensorDescriptor_t out_desc,
+        infiniopTensorDescriptor_t q_desc,
+        infiniopTensorDescriptor_t k_cache_desc,
+        infiniopTensorDescriptor_t v_cache_desc,
+        infiniopTensorDescriptor_t block_tables_desc,
+        infiniopTensorDescriptor_t cache_lens_desc, // Total lengths
+        infiniopTensorDescriptor_t seq_lens_desc,   // New lengths (Prefill length)
+        const std::optional<infiniopTensorDescriptor_t> &alibi_slopes_desc,
+        float scale) {
+
+        auto dtype = q_desc->dtype();
+        CHECK_DTYPE(dtype, INFINI_DTYPE_F16, INFINI_DTYPE_BF16, INFINI_DTYPE_F32);
+        if (out_desc->dtype() != dtype || k_cache_desc->dtype() != dtype || v_cache_desc->dtype() != dtype) {
+            return INFINI_STATUS_BAD_TENSOR_DTYPE;
+        }
+
+        if (q_desc->ndim() < 3 || k_cache_desc->ndim() < 4 || v_cache_desc->ndim() < 4 || block_tables_desc->ndim() != 2 || cache_lens_desc->ndim() != 1 || seq_lens_desc->ndim() != 1) {
+            return INFINI_STATUS_BAD_TENSOR_SHAPE;
+        }
+
+        if (block_tables_desc->dtype() != INFINI_DTYPE_I64 || cache_lens_desc->dtype() != INFINI_DTYPE_I64 || seq_lens_desc->dtype() != INFINI_DTYPE_I64) {
+            return INFINI_STATUS_BAD_TENSOR_DTYPE;
+        }
+
+        // --- Extract shape dimensions ---
+        // Assuming Q shape: [num_seqs, max_new_len, num_heads, head_size]
+        auto q_shape = q_desc->shape();
+        auto k_cache_shape = k_cache_desc->shape();
+
+        size_t num_seqs = q_shape[0];
+        size_t max_new_len = q_shape[1];
+        size_t num_heads = q_shape[2];
+        size_t head_size = q_shape[3];
+
+        if (head_size != 128) {
+            std::cerr << "[Error] PagedAttentionPrefill now only supports head_size = 128, but got "
+                      << head_size << "." << std::endl;
+            return INFINI_STATUS_BAD_TENSOR_SHAPE;
+        }
+
+        size_t num_kv_heads = k_cache_shape[1];
+        size_t block_size = v_cache_desc->shape()[2];
+        size_t max_num_blocks_per_seq = block_tables_desc->shape()[1];
+
+        // --- Extract strides ---
+        ptrdiff_t q_stride = q_desc->stride(0); // Stride between sequences in Q
+        ptrdiff_t kv_block_stride = k_cache_desc->stride(0);
+        ptrdiff_t kv_head_stride = k_cache_desc->stride(1);
+        ptrdiff_t o_stride = out_desc->stride(0);
+
+        return utils::Result<PagedAttentionPrefillInfo>(PagedAttentionPrefillInfo{
+            dtype,
+            scale,
+            num_seqs,
+            num_heads,
+            num_kv_heads,
+            head_size,
+            block_size,
+            max_num_blocks_per_seq,
+            max_new_len,
+            q_stride,
+            kv_block_stride,
+            kv_head_stride,
+            o_stride});
+    }
+};
+
+} // namespace op::paged_attention_prefill
+
+#endif // __PAGED_ATTENTION_PREFILL_INFO_H__