docs.rs failed to build ferrotorch-gpu-0.5.1
Please check the build logs for more information.
See Builds for ideas on how to fix a failed build, or Metadata for how to configure docs.rs builds.
If you believe this is docs.rs' fault, open an issue.

Visit the last successful build: ferrotorch-gpu-0.4.2

ferrotorch-gpu

CUDA GPU backend for ferrotorch — BLAS, cuSPARSE, cuFFT, FlashAttention-2, bf16 mixed-precision, and memory safety.

What it provides

GpuDevice -- CUDA device management and context handling
Data transfer -- cpu_to_gpu, gpu_to_cpu, tensor_to_gpu, tensor_to_cpu, alloc_zeros
GpuTensor -- GPU-resident tensor wrapper with cuda() and cuda_default() helpers
CUDA buffer -- CudaBuffer for raw GPU memory, CudaAllocator for memory management
BLAS -- gpu_matmul_f32, gpu_matmul_f64 via cuBLAS, gpu_bmm_* batched
bf16 mixed-precision -- gpu_matmul_bf16, gpu_bmm_bf16 via cuBLAS, plus 8 elementwise/reduction/activation kernels (gpu_add_bf16, gpu_mul_bf16, gpu_silu_bf16, gpu_relu_bf16, gpu_softmax_bf16, gpu_rmsnorm_bf16, gpu_rope_half_bf16, gpu_scale_bf16) and supporting utilities for LLM bf16 pipelines
cuSOLVER -- gpu_lu_solve, gpu_cholesky, gpu_eigh, gpu_eig, gpu_lstsq via cuSOLVER
cuFFT -- axes-aware FFT via cufftPlanMany: gpu_rfft_r2c_*, gpu_irfft_c2r_*, gpu_hfft_*, gpu_ihfft_*, gpu_fftn_axes_c2c_* (n-d, arbitrary axis selection), gpu_fftn2d_c2c_*, gpu_fftn3d_c2c_*
cuSPARSE -- gpu_spmm_csr_* SpMM, gpu_sparse_to_dense_csr_* / gpu_dense_to_sparse_csr_* CSR conversions, gpu_csc_to_dense_*, gpu_csr_to_csc_*, gpu_coo_to_csr_*, gpu_csr_to_coo_* format conversions
cuSPARSELt -- gpu_sparse_matmul_24 for 2:4 structured sparse matrix multiply (NVIDIA Ampere and later)
FlashAttention-2 -- gpu_flash_attention_f32 / gpu_flash_attention_f64 forward-pass PTX kernel
Convolution -- gpu_conv2d_f32
Element-wise kernels -- gpu_add, gpu_sub, gpu_mul, gpu_neg, gpu_relu, plus 50+ PTX kernels for reductions, scans, masked ops, scatter/gather, strided copies
Graph capture -- GpuGraphPool for CUDA-graph stream capture and replay
Memory guard -- MemoryGuard, MemoryWatchdog, MemoryStats, OomPolicy for GPU memory management and pressure monitoring

Feature flags

Feature	Default	Description
`cuda`	yes	Links against CUDA driver API via cudarc

Quick start

use ferrotorch_gpu::{GpuDevice, cpu_to_gpu, gpu_to_cpu};

let device = GpuDevice::new(0)?;
let host_data = vec![1.0f32, 2.0, 3.0];
let gpu_buf = cpu_to_gpu(&host_data, &device)?;
let back = gpu_to_cpu(&gpu_buf, &device)?;
assert_eq!(back, host_data);

Part of ferrotorch

This crate is one component of the ferrotorch workspace. See the workspace README for full documentation.

License