(WIP) Add no_std support

.travis.yml

@@ -5,7 +5,7 @@ dist: trusty
 
 matrix:
   include:
-    - rust: 1.28.0
+    - rust: 1.36.0
       env:
         TARGET=x86_64-unknown-linux-gnu
     - rust: stable

Cargo.toml

@@ -32,6 +32,11 @@ rawpointer = "0.2"
 bencher = "0.1.2"
 itertools = "0.7.11"
 
+[features]
+default = ["std"]
+
+std = []
+
 [profile.release]
 [profile.bench]
 

benches/benchmarks.rs

@@ -158,7 +158,7 @@ gemm_layout!{layout_f64_032, dgemm,
 }
 
 
-use std::ops::{Add, Mul};
+use core::ops::{Add, Mul};
 
 trait Z {
     fn zero() -> Self;

blas-bench/benches/benchmarks.rs

@@ -7,7 +7,7 @@ pub use matrixmultiply::dgemm;
 extern crate bencher;
 extern crate blas;
 
-use std::os::raw::c_int;
+use core::os::raw::c_int;
 
 
 #[allow(non_camel_case_types)]

examples/usegemm.rs

@@ -5,6 +5,7 @@
 //
 // Jump down to the next place where it says EXAMPLE.
 
+extern crate core;
 extern crate itertools;
 extern crate matrixmultiply;
 
@@ -16,7 +17,7 @@ use itertools::{
     enumerate,
     repeat_n,
 };
-use std::fmt::{Display, Debug};
+use core::fmt::{Display, Debug};
 
 trait Float : Copy + Display + Debug + PartialEq {
     fn zero() -> Self;

src/aligned_alloc.rs

@@ -1,12 +1,11 @@
 
-use std::alloc;
-use std::alloc::{Layout, handle_alloc_error};
-use std::{mem, cmp};
+use alloc::alloc::{self, Layout, handle_alloc_error};
+use core::{mem, cmp};
 
 #[cfg(test)]
-use std::ops::{Deref, DerefMut};
+use core::ops::{Deref, DerefMut};
 #[cfg(test)]
-use std::slice;
+use core::slice;
 
 pub(crate) struct Alloc<T> { ptr: *mut T, len: usize, align: usize }
 

src/dgemm_kernel.rs

@@ -12,9 +12,9 @@ use kernel::{U4, U8};
 use archparam;
 
 #[cfg(target_arch="x86")]
-use std::arch::x86::*;
+use core::arch::x86::*;
 #[cfg(target_arch="x86_64")]
-use std::arch::x86_64::*;
+use core::arch::x86_64::*;
 #[cfg(any(target_arch="x86", target_arch="x86_64"))]
 use x86::{FusedMulAdd, AvxMulAdd, DMultiplyAdd};
 
@@ -812,6 +812,7 @@ unsafe fn at(ptr: *const T, i: usize) -> T {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use alloc::vec;
     use aligned_alloc::Alloc;
 
     fn aligned_alloc<T>(elt: T, n: usize) -> Alloc<T> where T: Copy
@@ -864,6 +865,7 @@ mod tests {
     mod test_arch_kernels {
         use super::test_a_kernel;
         use super::super::*;
+        use std::println;
         macro_rules! test_arch_kernels_x86 {
             ($($feature_name:tt, $name:ident, $kernel_ty:ty),*) => {
                 $(

src/gemm.rs

@@ -6,9 +6,9 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-use std::cmp::min;
-use std::mem::size_of;
-use std::ptr::copy_nonoverlapping;
+use core::cmp::min;
+use core::mem::size_of;
+use core::ptr::copy_nonoverlapping;
 
 use aligned_alloc::Alloc;
 

src/kernel.rs

@@ -6,7 +6,7 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-use std::ops::{AddAssign, MulAssign};
+use core::ops::{AddAssign, MulAssign};
 
 /// General matrix multiply kernel
 pub trait GemmKernel {

src/lib.rs

@@ -58,13 +58,38 @@
 //!   - `avx`
 //!   - `sse2`
 //!
+//! ## Features
+//!
+//! This crate can be used without the standard library (`#![no_std]`) by
+//! disabling the default `std` feature. To do so, use this in your
+//! `Cargo.toml`:
+//!
+//! ```toml
+//! matrixmultiply = { version = "0.2", default-features = false }
+//! ```
+//!
+//! Runtime CPU feature detection is available only when `std` is enabled.
+//! Without the `std` feature, the crate uses special CPU features only if they
+//! are enabled at compile time. (To enable CPU features at compile time, pass
+//! the relevant
+//! [`target-cpu`](https://doc.rust-lang.org/rustc/codegen-options/index.html#target-cpu)
+//! or
+//! [`target-feature`](https://doc.rust-lang.org/rustc/codegen-options/index.html#target-feature)
+//! option to `rustc`.)
+//!
 //! ## Other Notes
 //!
 //! The functions in this crate are thread safe, as long as the destination
 //! matrix is distinct.
 
 #![doc(html_root_url = "https://docs.rs/matrixmultiply/0.2/")]
 
+#![no_std]
+#[cfg(feature = "std")]
+extern crate std;
+
+extern crate alloc;
+
 extern crate rawpointer;
 
 #[macro_use] mod debugmacros;

src/sgemm_kernel.rs

@@ -13,9 +13,9 @@ use archparam;
 
 
 #[cfg(target_arch="x86")]
-use std::arch::x86::*;
+use core::arch::x86::*;
 #[cfg(target_arch="x86_64")]
-use std::arch::x86_64::*;
+use core::arch::x86_64::*;
 #[cfg(any(target_arch="x86", target_arch="x86_64"))]
 use x86::{FusedMulAdd, AvxMulAdd, SMultiplyAdd};
 
@@ -499,6 +499,7 @@ unsafe fn at(ptr: *const T, i: usize) -> T {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use alloc::vec;
     use aligned_alloc::Alloc;
 
     fn aligned_alloc<T>(elt: T, n: usize) -> Alloc<T> where T: Copy
@@ -551,6 +552,7 @@ mod tests {
     mod test_arch_kernels {
         use super::test_a_kernel;
         use super::super::*;
+        use std::println;
         macro_rules! test_arch_kernels_x86 {
             ($($feature_name:tt, $name:ident, $kernel_ty:ty),*) => {
                 $(

src/util.rs

@@ -6,7 +6,7 @@
 // option. This file may not be copied, modified, or distributed
 // except according to those terms.
 
-use std::cmp::min;
+use core::cmp::min;
 
 pub struct RangeChunk { i: usize, n: usize, chunk: usize }
 

src/x86/macros.rs

@@ -10,11 +10,23 @@ macro_rules! compile_env_matches_or_is_empty {
 }
 
 macro_rules! is_x86_feature_detected_ {
-    ($name:tt) => {
-        // for testing purposes, we can make sure only one specific feature
-        // is enabled by setting MMTEST_FEATURE=featurename (all others
-        // disabled). This does not force it to be detected, it must also be.
-        compile_env_matches_or_is_empty!("MMTEST_FEATURE", $name) && is_x86_feature_detected!($name)
-    }
+    ($name:tt) => {{
+        #[cfg(feature="std")]
+        {
+            // For testing purposes, we can make sure only one specific feature
+            // is enabled by setting MMTEST_FEATURE=featurename (all others
+            // disabled). This does not force it to be detected, it must also be.
+             compile_env_matches_or_is_empty!("MMTEST_FEATURE", $name) && std::is_x86_feature_detected!($name)
+        }
+        #[cfg(not(feature="std"))]
+        {
+            // For testing purposes, we can make sure only one specific feature
+            // is enabled by setting MMTEST_FEATURE=featurename (all others
+            // disabled). This does not force it to be detected, it must also
+            // be. In the `no_std` case, the `is_86_feature_detected` macro is
+            // not available, so we have to fall back to checking whether the
+            // feature is enabled at compile-time.
+            compile_env_matches_or_is_empty!("MMTEST_FEATURE", $name) && cfg!(target_feature=$name)
+        }
+    }};
 }
-

src/x86/mod.rs

@@ -1,8 +1,8 @@
 
 #[cfg(target_arch="x86")]
-use std::arch::x86::*;
+use core::arch::x86::*;
 #[cfg(target_arch="x86_64")]
-use std::arch::x86_64::*;
+use core::arch::x86_64::*;
 
 #[macro_use]
 mod macros;

tests/sgemm.rs

@@ -1,3 +1,4 @@
+extern crate core;
 extern crate itertools;
 extern crate matrixmultiply;
 
@@ -9,7 +10,7 @@ use itertools::{
     enumerate,
     repeat_n,
 };
-use std::fmt::{Display, Debug};
+use core::fmt::{Display, Debug};
 
 trait Float : Copy + Display + Debug + PartialEq {
     fn zero() -> Self;