huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-16 18:48:51 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add a matmul cuda example.

Browse Source
This commit is contained in:
laurent
2023-06-22 19:44:26 +01:00
parent 0671b8c369
commit aebffcfc13
2 changed files with 11 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
examples/cuda_basics.rs
View File

@ -1,5 +1,5 @@
use anyhow::Result;
use candle::{DType, Device, Tensor};
use candle::{Device, Tensor};
fn main() -> Result<()> {
let device = Device::new_cuda(0)?;
@ -9,7 +9,10 @@ fn main() -> Result<()> {
let z = (y + x * 3.)?;
println!("{:?}", z.to_vec1::<f32>()?);
println!("{:?}", z.sqrt()?.to_vec1::<f32>()?);
let x = Tensor::ones((3, 2), DType::F32, &device)?;
println!("{:?}", x.to_vec2::<f32>()?);
let x = Tensor::new(&[[11f32, 22.], [33., 44.], [55., 66.], [77., 78.]], &device)?;
let y = Tensor::new(&[[1f32, 2., 3.], [4., 5., 6.]], &device)?;
println!("{:?}", y.to_vec2::<f32>()?);
let z = x.matmul(&y)?;
println!("{:?}", z.to_vec2::<f32>()?);
Ok(())
}

9
src/cuda_backend.rs
View File

@ -159,14 +159,15 @@ fn gemm_config<T>(
) -> StridedBatchedConfig<T> {
// https://docs.nvidia.com/cuda/cublas/index.html#cublas-t-gemm
use cudarc::cublas::sys::cublasOperation_t;
println!("{:?} {:?} {:?}", lhs_stride, rhs_stride, (b, m, n, k));
let gemm = GemmConfig {
alpha,
beta,
m: m as i32,
n: n as i32,
k: k as i32,
lda: lhs_stride[lhs_stride.len() - 2] as i32,
ldb: rhs_stride[rhs_stride.len() - 2] as i32,
lda: m as i32,
ldb: k as i32,
ldc: m as i32,
transa: cublasOperation_t::CUBLAS_OP_N,
transb: cublasOperation_t::CUBLAS_OP_N,
@ -174,8 +175,8 @@ fn gemm_config<T>(
StridedBatchedConfig {
batch_size: b as i32,
gemm,
stride_a: lhs_stride[0] as i64,
stride_b: rhs_stride[0] as i64,
stride_a: (m * k) as i64,
stride_b: (n * k) as i64,
stride_c: (m * n * k) as i64,
}
}