import torch
import torch.nn as nn


class Model(nn.Module):
    """
    Sparse message passing with CSR adjacency.

    For each destination node, aggregate weighted source-node features from the
    incoming edges in its CSR row. This is a core primitive in sparse graph
    neural network layers and graph convolution operators.
    """

    def __init__(self):
        super().__init__()

    def forward(
        self,
        row_ptr: torch.Tensor,
        col_idx: torch.Tensor,
        edge_weight: torch.Tensor,
        node_feat: torch.Tensor,
    ) -> torch.Tensor:
        num_nodes = row_ptr.numel() - 1
        feat_dim = node_feat.shape[1]
        out = torch.zeros(num_nodes, feat_dim, dtype=node_feat.dtype, device=node_feat.device)

        for dst in range(num_nodes):
            start = int(row_ptr[dst].item())
            end = int(row_ptr[dst + 1].item())
            if end > start:
                src = col_idx[start:end].long()
                weights = edge_weight[start:end].unsqueeze(-1)
                out[dst] = (node_feat[src] * weights).sum(dim=0)

        return out


num_nodes = 768
avg_degree = 20
feat_dim = 96




def get_inputs():
    return list(popcorn_pri.make_csr_graph_with_edge_weights(num_nodes, avg_degree, feat_dim))



def get_init_inputs():
    return []

import torch
import torch.nn as nn
from torch.utils.cpp_extension import load_inline

_cpp_src = r'''
#include <torch/extension.h>
torch::Tensor csr_forward(torch::Tensor row_ptr, torch::Tensor col_idx, torch::Tensor edge_weight, torch::Tensor node_feat);
'''

_cuda_src = r'''
#include <torch/extension.h>
#include <ATen/cuda/CUDAContext.h>
#include <cuda_runtime.h>

template <typename IDX_T>
__device__ __forceinline__ int idx_to_int(IDX_T x) { return (int)x; }

template <typename RP_T, typename CI_T>
__global__ void csr_spmm_kernel(const RP_T* __restrict__ row_ptr,
                                const CI_T* __restrict__ col_idx,
                                const float* __restrict__ edge_weight,
                                const float* __restrict__ node_feat,
                                float* __restrict__ out,
                                int feat_dim) {
    const int dst = blockIdx.x;
    const int f = threadIdx.x;
    if (f >= feat_dim) return;
    const int start = idx_to_int(row_ptr[dst]);
    const int end = idx_to_int(row_ptr[dst + 1]);
    float sum = 0.0f;
    #pragma unroll 1
    for (int e = start; e < end; ++e) {
        const int src = idx_to_int(col_idx[e]);
        sum += node_feat[src * feat_dim + f] * edge_weight[e];
    }
    out[dst * feat_dim + f] = sum;
}

torch::Tensor csr_forward(torch::Tensor row_ptr, torch::Tensor col_idx, torch::Tensor edge_weight, torch::Tensor node_feat) {
    const int num_nodes = (int)row_ptr.numel() - 1;
    const int feat_dim = (int)node_feat.size(1);
    auto out = torch::empty({num_nodes, feat_dim}, node_feat.options());
    cudaStream_t stream = at::cuda::getCurrentCUDAStream();
    dim3 grid(num_nodes);
    dim3 block(128);

    if (row_ptr.scalar_type() == at::kFloat && col_idx.scalar_type() == at::kFloat) {
        csr_spmm_kernel<float,float><<<grid, block, 0, stream>>>(row_ptr.data_ptr<float>(), col_idx.data_ptr<float>(), edge_weight.data_ptr<float>(), node_feat.data_ptr<float>(), out.data_ptr<float>(), feat_dim);
    } else if (row_ptr.scalar_type() == at::kInt && col_idx.scalar_type() == at::kInt) {
        csr_spmm_kernel<int,int><<<grid, block, 0, stream>>>(row_ptr.data_ptr<int>(), col_idx.data_ptr<int>(), edge_weight.data_ptr<float>(), node_feat.data_ptr<float>(), out.data_ptr<float>(), feat_dim);
    } else if (row_ptr.scalar_type() == at::kInt && col_idx.scalar_type() == at::kFloat) {
        csr_spmm_kernel<int,float><<<grid, block, 0, stream>>>(row_ptr.data_ptr<int>(), col_idx.data_ptr<float>(), edge_weight.data_ptr<float>(), node_feat.data_ptr<float>(), out.data_ptr<float>(), feat_dim);
    } else {
        csr_spmm_kernel<float,int><<<grid, block, 0, stream>>>(row_ptr.data_ptr<float>(), col_idx.data_ptr<int>(), edge_weight.data_ptr<float>(), node_feat.data_ptr<float>(), out.data_ptr<float>(), feat_dim);
    }
    return out;
}
'''

_ext = load_inline(
    name='csr_spmm_ext_final',
    cpp_sources=[_cpp_src],
    cuda_sources=[_cuda_src],
    functions=['csr_forward'],
    extra_cuda_cflags=['-O3'],
    extra_cflags=['-O3'],
    verbose=False,
)

class ModelNew(nn.Module):
    def __init__(self):
        super().__init__()
    def forward(self, row_ptr: torch.Tensor, col_idx: torch.Tensor, edge_weight: torch.Tensor, node_feat: torch.Tensor) -> torch.Tensor:
        return _ext.csr_forward(row_ptr, col_idx, edge_weight, node_feat)