Skip to content

numr 0.3.0 #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
farhan-syah merged 5 commits into from
Feb 10, 2026
Merged

numr 0.3.0 #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
28fe830
chore: initialize 0.3.0 development branch
farhan-syah Feb 10, 2026
8a0c919
refactor(tensor): extract Shape and Strides into dedicated types
farhan-syah Feb 10, 2026
a0a5e14
refactor(cpu): reorganize FFT and reduce into modular structure
farhan-syah Feb 10, 2026
1e0d396
feat(ops): add composite shape operations via impl_generic
farhan-syah Feb 10, 2026
a6405e2
refactor: add configurable parallelism and improve kernel consistency
farhan-syah Feb 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions src/autograd/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -121,6 +121,7 @@ mod forward;
pub mod ops;

// Reverse-mode exports
pub use crate::tensor::id::TensorId;
pub use backward::{backward, backward_with_graph};
pub use grad_fn::GradFn;
pub use grad_store::GradStore;
Expand Down
4 changes: 2 additions & 2 deletions src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,7 +96,7 @@ pub mod prelude {
// Core types
pub use crate::dtype::DType;
pub use crate::error::{Error, Result};
pub use crate::tensor::{Layout, Tensor};
pub use crate::tensor::{Layout, Shape, Strides, Tensor};

// Runtime traits
pub use crate::runtime::{Device, Runtime, RuntimeClient};
Expand All @@ -116,7 +116,7 @@ pub mod prelude {

// Backend runtimes
#[cfg(feature = "cpu")]
pub use crate::runtime::cpu::{CpuClient, CpuDevice, CpuRuntime};
pub use crate::runtime::cpu::{CpuClient, CpuDevice, CpuRuntime, ParallelismConfig};

#[cfg(feature = "cuda")]
pub use crate::runtime::cuda::{CudaClient, CudaDevice, CudaRuntime};
Expand Down
219 changes: 149 additions & 70 deletions src/ops/cpu/complex.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,17 +29,30 @@ impl ComplexOps<CpuRuntime> for CpuClient {

let a_ptr = a_contig.storage().ptr();
let out_ptr = out.storage().ptr();
let chunk_size = self.chunk_size_hint();

unsafe {
match dtype {
DType::Complex64 => {
kernels::conj_complex64(a_ptr as *const _, out_ptr as *mut _, numel);
}
DType::Complex128 => {
kernels::conj_complex128(a_ptr as *const _, out_ptr as *mut _, numel);
}
_ => unreachable!("conj called on non-complex dtype"),
match dtype {
DType::Complex64 => {
self.install_parallelism(|| unsafe {
kernels::conj_complex64(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
DType::Complex128 => {
self.install_parallelism(|| unsafe {
kernels::conj_complex128(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
_ => unreachable!("conj called on non-complex dtype"),
}

Ok(out)
Expand Down Expand Up @@ -71,17 +84,30 @@ impl ComplexOps<CpuRuntime> for CpuClient {

let a_ptr = a_contig.storage().ptr();
let out_ptr = out.storage().ptr();
let chunk_size = self.chunk_size_hint();

unsafe {
match dtype {
DType::Complex64 => {
kernels::real_complex64(a_ptr as *const _, out_ptr as *mut _, numel);
}
DType::Complex128 => {
kernels::real_complex128(a_ptr as *const _, out_ptr as *mut _, numel);
}
_ => unreachable!("real called on non-complex dtype"),
match dtype {
DType::Complex64 => {
self.install_parallelism(|| unsafe {
kernels::real_complex64(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
DType::Complex128 => {
self.install_parallelism(|| unsafe {
kernels::real_complex128(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
_ => unreachable!("real called on non-complex dtype"),
}

Ok(out)
Expand Down Expand Up @@ -113,17 +139,30 @@ impl ComplexOps<CpuRuntime> for CpuClient {

let a_ptr = a_contig.storage().ptr();
let out_ptr = out.storage().ptr();
let chunk_size = self.chunk_size_hint();

unsafe {
match dtype {
DType::Complex64 => {
kernels::imag_complex64(a_ptr as *const _, out_ptr as *mut _, numel);
}
DType::Complex128 => {
kernels::imag_complex128(a_ptr as *const _, out_ptr as *mut _, numel);
}
_ => unreachable!("imag called on non-complex dtype"),
match dtype {
DType::Complex64 => {
self.install_parallelism(|| unsafe {
kernels::imag_complex64(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
DType::Complex128 => {
self.install_parallelism(|| unsafe {
kernels::imag_complex128(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
_ => unreachable!("imag called on non-complex dtype"),
}

Ok(out)
Expand All @@ -133,6 +172,7 @@ impl ComplexOps<CpuRuntime> for CpuClient {
let dtype = a.dtype();
let shape = a.shape();
let numel = a.numel();
let chunk_size = self.chunk_size_hint();

let a_contig = ensure_contiguous(a);

Expand All @@ -148,19 +188,31 @@ impl ComplexOps<CpuRuntime> for CpuClient {
let a_ptr = a_contig.storage().ptr();
let out_ptr = out.storage().ptr();

unsafe {
match dtype {
DType::F32 => {
kernels::angle_real_f32(a_ptr as *const _, out_ptr as *mut _, numel);
}
DType::F64 => {
kernels::angle_real_f64(a_ptr as *const _, out_ptr as *mut _, numel);
}
_ => {
// For integer types, angle doesn't make mathematical sense
// Return zeros
return Ok(Tensor::<CpuRuntime>::zeros(shape, dtype, &self.device));
}
match dtype {
DType::F32 => {
self.install_parallelism(|| unsafe {
kernels::angle_real_f32(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
DType::F64 => {
self.install_parallelism(|| unsafe {
kernels::angle_real_f64(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
_ => {
// For integer types, angle doesn't make mathematical sense
// Return zeros
return Ok(Tensor::<CpuRuntime>::zeros(shape, dtype, &self.device));
}
}
return Ok(out);
Expand All @@ -181,16 +233,28 @@ impl ComplexOps<CpuRuntime> for CpuClient {
let a_ptr = a_contig.storage().ptr();
let out_ptr = out.storage().ptr();

unsafe {
match dtype {
DType::Complex64 => {
kernels::angle_complex64(a_ptr as *const _, out_ptr as *mut _, numel);
}
DType::Complex128 => {
kernels::angle_complex128(a_ptr as *const _, out_ptr as *mut _, numel);
}
_ => unreachable!("angle called on non-complex dtype"),
match dtype {
DType::Complex64 => {
self.install_parallelism(|| unsafe {
kernels::angle_complex64(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
DType::Complex128 => {
self.install_parallelism(|| unsafe {
kernels::angle_complex128(
a_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
});
}
_ => unreachable!("angle called on non-complex dtype"),
}

Ok(out)
Expand Down Expand Up @@ -226,27 +290,32 @@ impl ComplexOps<CpuRuntime> for CpuClient {
let real_ptr = real_contig.storage().ptr();
let imag_ptr = imag_contig.storage().ptr();
let out_ptr = out.storage().ptr();
let chunk_size = self.chunk_size_hint();

unsafe {
match input_dtype {
DType::F32 => {
match input_dtype {
DType::F32 => {
self.install_parallelism(|| unsafe {
kernels::from_real_imag_f32(
real_ptr as *const _,
imag_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
}
DType::F64 => {
});
}
DType::F64 => {
self.install_parallelism(|| unsafe {
kernels::from_real_imag_f64(
real_ptr as *const _,
imag_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
}
_ => unreachable!("validated above"),
});
}
_ => unreachable!("validated above"),
}

Ok(out)
Expand Down Expand Up @@ -275,27 +344,32 @@ impl ComplexOps<CpuRuntime> for CpuClient {
let complex_ptr = complex_contig.storage().ptr();
let real_ptr = real_contig.storage().ptr();
let out_ptr = out.storage().ptr();
let chunk_size = self.chunk_size_hint();

unsafe {
match dtype {
DType::Complex64 => {
match dtype {
DType::Complex64 => {
self.install_parallelism(|| unsafe {
kernels::complex64_mul_real(
complex_ptr as *const _,
real_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
}
DType::Complex128 => {
});
}
DType::Complex128 => {
self.install_parallelism(|| unsafe {
kernels::complex128_mul_real(
complex_ptr as *const _,
real_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
}
_ => unreachable!("validated above"),
});
}
_ => unreachable!("validated above"),
}

Ok(out)
Expand Down Expand Up @@ -324,27 +398,32 @@ impl ComplexOps<CpuRuntime> for CpuClient {
let complex_ptr = complex_contig.storage().ptr();
let real_ptr = real_contig.storage().ptr();
let out_ptr = out.storage().ptr();
let chunk_size = self.chunk_size_hint();

unsafe {
match dtype {
DType::Complex64 => {
match dtype {
DType::Complex64 => {
self.install_parallelism(|| unsafe {
kernels::complex64_div_real(
complex_ptr as *const _,
real_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
}
DType::Complex128 => {
});
}
DType::Complex128 => {
self.install_parallelism(|| unsafe {
kernels::complex128_div_real(
complex_ptr as *const _,
real_ptr as *const _,
out_ptr as *mut _,
numel,
chunk_size,
);
}
_ => unreachable!("validated above"),
});
}
_ => unreachable!("validated above"),
}

Ok(out)
Expand Down
Loading