add avx compile-time feature

paritytech · Dec 12, 2023 · a0cdeb9 · a0cdeb9
1 parent 69612f1
commit a0cdeb9
Show file tree

Hide file tree

Showing 14 changed files with 90 additions and 42 deletions.
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -24,7 +24,28 @@ jobs:
       - uses: actions-rs/cargo@v1
         with:
           command: check
-          args: --workspace --all-features
+          args: --workspace
+
+  check:
+    name: Check AVX
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions-rs/toolchain@v1
+        with:
+          profile: minimal
+          toolchain: stable
+          override: true
+
+      - name: Rust Cache
+        uses: Swatinem/rust-cache@3cf7f8cc28d1b4e7d01e3783be10a97d55d483c8 # v2.7.1
+
+      - uses: actions-rs/cargo@v1
+        env:
+          RUSTFLAGS: "-C target-cpu=native"
+        with:
+          command: check
+          args: --features avx --workspace
 
   test:
     name: Test
@@ -51,6 +72,33 @@ jobs:
           command: test
           args: --workspace
 
+  test:
+    name: Test AVX
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        os:
+          - ubuntu-latest
+          - macOS-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions-rs/toolchain@v1
+        with:
+          profile: minimal
+          toolchain: stable
+          override: true
+
+      - name: Rust Cache
+        uses: Swatinem/rust-cache@3cf7f8cc28d1b4e7d01e3783be10a97d55d483c8 # v2.7.1
+
+      - name: Test with avx feature
+        uses: actions-rs/cargo@v1
+        env:
+          RUSTFLAGS: "-C target-cpu=native"
+        with:
+          command: test
+          args: --workspace --features avx
+
   test_windows:
     name: Test Windows
     runs-on: windows-latest
@@ -98,4 +146,4 @@ jobs:
       - uses: actions-rs/cargo@v1
         with:
           command: clippy
-          args: --workspace --all-features
+          args: --workspace --all-features -- -D warnings
diff --git a/reed-solomon-novelpoly-fuzzit/src/additive_mpy.rs b/reed-solomon-novelpoly-fuzzit/src/additive_mpy.rs
@@ -14,7 +14,7 @@ struct FieldMpyParams {
 	idx_to_test: usize,
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 impl<'a> Arbitrary<'a> for FieldMpyParams {
 	fn arbitrary(u: &mut Unstructured<'a>) -> arbitrary::Result<Self> {
 		let additive = Additive(u.int_in_range(0..=u16::MAX)?);
@@ -26,14 +26,14 @@ impl<'a> Arbitrary<'a> for FieldMpyParams {
 }
 
 fn main() {
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	run();
 
 	#[cfg(not(target_feature = "avx"))]
 	panic!("Nothing to do for non avx enabled targets")
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 fn run() {
 	// You have full control over the loop but
 	// you're supposed to call `fuzz` ad vitam aeternam

diff --git a/reed-solomon-novelpoly-fuzzit/src/afft.rs b/reed-solomon-novelpoly-fuzzit/src/afft.rs
@@ -26,14 +26,14 @@ impl<'a> Arbitrary<'a> for AfftParams {
 }
 
 fn main() {
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	run();
 
 	#[cfg(not(target_feature = "avx"))]
 	panic!("Nothing to do for non avx enabled targets")
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 fn run() {
 	// You have full control over the loop but
 	// you're supposed to call `fuzz` ad vitam aeternam

diff --git a/reed-solomon-novelpoly-fuzzit/src/inverse_afft.rs b/reed-solomon-novelpoly-fuzzit/src/inverse_afft.rs
@@ -26,14 +26,14 @@ impl<'a> Arbitrary<'a> for InverseAfftParams {
 }
 
 fn main() {
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	run();
 
 	#[cfg(not(target_feature = "avx"))]
 	panic!("Nothing to do for non avx enabled targets")
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 fn run() {
 	// You have full control over the loop but
 	// you're supposed to call `fuzz` ad vitam aeternam

diff --git a/reed-solomon-novelpoly/src/field/f256.rs b/reed-solomon-novelpoly/src/field/f256.rs
@@ -18,5 +18,5 @@ impl std::fmt::Debug for Additive {
 }
 
 #[cfg(table_bootstrap_complete)]
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub use faster8::f256::*;
diff --git a/reed-solomon-novelpoly/src/field/f2e16.rs b/reed-solomon-novelpoly/src/field/f2e16.rs
@@ -35,5 +35,5 @@ impl std::fmt::Debug for Additive {
 }
 
 #[cfg(table_bootstrap_complete)]
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub use crate::field::faster8::f2e16::*;
diff --git a/reed-solomon-novelpoly/src/field/faster8/mod.rs b/reed-solomon-novelpoly/src/field/faster8/mod.rs
@@ -1,4 +1,4 @@
-#![cfg(target_feature = "avx")]
+#![cfg(all(target_feature = "avx", feature = "avx"))]
 
 #[cfg(feature = "f256")]
 pub mod f256;

diff --git a/reed-solomon-novelpoly/src/field/inc_afft.rs b/reed-solomon-novelpoly/src/field/inc_afft.rs
@@ -94,7 +94,7 @@ pub fn inverse_afft(data: &mut [Additive], size: usize, index: usize) {
 	unsafe { &AFFT }.inverse_afft(data,size,index)
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub fn inverse_afft_faster8(data: &mut [Additive], size: usize, index: usize) {
 	unsafe { &AFFT }.inverse_afft_faster8(data,size,index)
 }
@@ -104,7 +104,7 @@ pub fn afft(data: &mut [Additive], size: usize, index: usize) {
 	unsafe { &AFFT }.afft(data,size,index)
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 /// Additive FFT in the "novel polynomial basis"
 pub fn afft_faster8(data: &mut [Additive], size: usize, index: usize) {
 	unsafe { &AFFT }.afft_faster8(data,size,index)
@@ -114,7 +114,7 @@ pub fn afft_faster8(data: &mut [Additive], size: usize, index: usize) {
 impl AdditiveFFT {
 
 	/// `data[i + depart_no] ^= data[i];`
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	#[inline(always)]
 	fn butterfly_down(data: &mut [Additive], i_8x: usize, depart_no_8x: usize) {
 		let rhs = Additive8x::load(&data[(i_8x * Additive8x::LANE) .. ][.. Additive8x::LANE]);
@@ -125,7 +125,7 @@ impl AdditiveFFT {
 	}
 
 	// `data[i] ^= data[i + depart_no].mul(skew)`;
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	#[inline(always)]
 	fn butterfly_up(data: &mut [Additive], i_8x: usize, depart_no_8x: usize, skew: Multiplier) {
 		let rhs = Additive8x::load(&data[((i_8x + depart_no_8x) * Additive8x::LANE) .. ][.. Additive8x::LANE]).mul(skew);
@@ -218,7 +218,7 @@ impl AdditiveFFT {
 	}
 
 	/// Inverse additive FFT in the "novel polynomial basis", but do 8 at once using available vector units
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	pub fn inverse_afft_faster8(&self, data: &mut [Additive], size: usize, index: usize) {
 		let mut depart_no = 1_usize;
 		while depart_no < Additive8x::LANE {
@@ -335,7 +335,7 @@ impl AdditiveFFT {
 	/// Additive FFT in the "novel polynomial basis", but do 8 at once using available vector units
 	///
 	/// `size` is the count of the individual additive field elements, so 8x larger than `data.len()`.
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	pub fn afft_faster8(&self, data: &mut [Additive], size: usize, index: usize) {
 		let mut depart_no = size >> 1_usize;
 		while depart_no >= Additive8x::LANE {
@@ -490,19 +490,19 @@ pub mod test_utils {
 		Vec::from_iter(rng.sample_iter::<Elt, _>(dist).take(size).map(Additive))
 	}
 
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	pub fn gen_faster8_from_plain(data: impl AsRef<[Additive]>) -> Vec<Additive> {
 		let data = data.as_ref();
 		data.to_vec()
 	}
 
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	pub fn gen_faster8<R: Rng + SeedableRng<Seed = [u8; 32]>>(size: usize) -> Vec<Additive> {
 		let data = gen_plain::<R>(size);
 		gen_faster8_from_plain(data)
 	}
 
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	pub fn assert_plain_eq_faster8(plain: impl AsRef<[Additive]>, faster8: impl AsRef<[Additive]>) {
 		let plain = plain.as_ref();
 		let faster8 = faster8.as_ref();
@@ -515,13 +515,13 @@ pub mod test_utils {
 #[cfg(test)]
 mod afft_tests {
 
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	mod simd {
 		use super::super::*;
 		use super::super::test_utils::*;
 		use rand::rngs::SmallRng;
 
-		#[cfg(target_feature = "avx")]
+		#[cfg(all(target_feature = "avx", feature = "avx"))]
 		#[test]
 		fn afft_output_plain_eq_faster8_size_16() {
 			let index = 0;
@@ -540,7 +540,7 @@ mod afft_tests {
 			assert_plain_eq_faster8(data_plain, data_faster8);
 		}
 
-		#[cfg(target_feature = "avx")]
+		#[cfg(all(target_feature = "avx", feature = "avx"))]
 		#[test]
 		fn afft_output_plain_eq_faster8_size_32() {
 			let index = 0;
@@ -559,7 +559,7 @@ mod afft_tests {
 			assert_plain_eq_faster8(data_plain, data_faster8);
 		}
 
-		#[cfg(target_feature = "avx")]
+		#[cfg(all(target_feature = "avx", feature = "avx"))]
 		#[test]
 		fn afft_output_plain_eq_faster8_impulse_data() {
 			let index = 0;
@@ -583,7 +583,7 @@ mod afft_tests {
 			assert_plain_eq_faster8(data_plain, data_faster8);
 		}
 
-		#[cfg(target_feature = "avx")]
+		#[cfg(all(target_feature = "avx", feature = "avx"))]
 		#[test]
 		fn inverse_afft_output_plain_eq_faster8_size_8() {
 			let index = 0;
@@ -604,7 +604,7 @@ mod afft_tests {
 			assert_plain_eq_faster8(data_plain, data_faster8);
 		}
 
-		#[cfg(target_feature = "avx")]
+		#[cfg(all(target_feature = "avx", feature = "avx"))]
 		#[test]
 		fn inverse_afft_output_plain_eq_faster8_size_32() {
 			let index = 0;

diff --git a/reed-solomon-novelpoly/src/field/inc_encode.rs b/reed-solomon-novelpoly/src/field/inc_encode.rs
@@ -1,6 +1,6 @@
 #[inline(always)]
 pub fn encode_low(data: &[Additive], k: usize, codeword: &mut [Additive], n: usize) {
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	if k >= 16 && k % 8 == 0 && n % 8 == 0 && (n-k) % 8 == 0 {
 		encode_low_faster8(data, k, codeword, n);		
 	} else {
@@ -51,7 +51,7 @@ pub fn encode_low_plain(data: &[Additive], k: usize, codeword: &mut [Additive],
 }
 
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub fn encode_low_faster8(data: &[Additive], k: usize, codeword: &mut [Additive], n: usize) {
 	assert!(k + k <= n);
 	assert_eq!(codeword.len(), n);
@@ -102,7 +102,7 @@ pub fn encode_low_faster8(data: &[Additive], k: usize, codeword: &mut [Additive]
 //Encoding alg for k/n>0.5: parity is a power of two.
 #[inline(always)]
 pub fn encode_high(data: &[Additive], k: usize, parity: &mut [Additive], mem: &mut [Additive], n: usize) {
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	if (n-k) % Additive8x::LANE == 0 && n % Additive8x::LANE == 0 && k % Additive8x::LANE == 0 {
 		encode_high_faster8(data, k, parity, mem, n);
 	} else {
@@ -135,12 +135,12 @@ pub fn encode_high_plain(data: &[Additive], k: usize, parity: &mut [Additive], m
 	afft(parity, t, 0);
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub fn encode_high_faster8_adapter(data: &[Additive], k: usize, parity: &mut [Additive], mem: &mut [Additive], n: usize) {
 	encode_high_faster8(&data, k, parity, mem, n);
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub fn encode_high_faster8(data: &[Additive], k: usize, parity: &mut [Additive], mem: &mut [Additive], n: usize) {
 	let t: usize = n - k;
 	assert!(t >= 8);
@@ -164,7 +164,7 @@ pub fn encode_high_faster8(data: &[Additive], k: usize, parity: &mut [Additive],
 }
 
 pub fn encode_sub(bytes: &[u8], n: usize, k: usize) -> Result<Vec<Additive>> {
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	if (k % Additive8x::LANE) == 0 && (k >> 1) >= Additive8x::LANE {
 		encode_sub_faster8(bytes, n, k)
 	} else {
@@ -221,7 +221,7 @@ pub fn encode_sub_plain(bytes: &[u8], n: usize, k: usize) -> Result<Vec<Additive
 
 
 /// Bytes shall only contain payload data
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub fn encode_sub_faster8(bytes: &[u8], n: usize, k: usize) -> Result<Vec<Additive>> {
 	assert!(is_power_of_2(n), "Algorithm only works for 2^i sizes for N");
 	assert!(is_power_of_2(k), "Algorithm only works for 2^i sizes for K");
@@ -266,7 +266,7 @@ pub fn encode_sub_faster8(bytes: &[u8], n: usize, k: usize) -> Result<Vec<Additi
 	Ok(codeword)
 }
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 #[cfg(test)]
 mod tests_plain_vs_faster8 {
 	use super::*;
@@ -289,7 +289,7 @@ mod tests_plain_vs_faster8 {
 	}
 
 
-	#[cfg(target_feature = "avx")]
+	#[cfg(all(target_feature = "avx", feature = "avx"))]
 	#[test]
 	fn encode_sub_output_plain_eq_faster8() {
 		let n = 64;

diff --git a/reed-solomon-novelpoly/src/field/inc_log_mul.rs b/reed-solomon-novelpoly/src/field/inc_log_mul.rs
@@ -113,7 +113,7 @@ pub fn walsh_plain(data: &mut [Multiplier], size: usize) {
 
 
 #[cfg(table_bootstrap_complete)]
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub fn walsh_faster8(data: &mut [Multiplier], size: usize) {
 	const LANE: usize = 8;
 
@@ -244,7 +244,7 @@ fn cantor_basis() {
 
 
 #[cfg(table_bootstrap_complete)]
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 #[test]
 fn walsh_output_plain_eq_faster8() {
 	use reed_solomon_tester::*;

diff --git a/reed-solomon-novelpoly/src/field/mod.rs b/reed-solomon-novelpoly/src/field/mod.rs
@@ -9,5 +9,5 @@ pub mod f256;
 
 pub mod f2e16;
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub mod faster8;
diff --git a/reed-solomon-novelpoly/src/lib.rs b/reed-solomon-novelpoly/src/lib.rs
@@ -12,7 +12,7 @@ pub mod field;
 pub use self::field::f256;
 pub use self::field::f2e16;
 
-#[cfg(target_feature = "avx")]
+#[cfg(all(target_feature = "avx", feature = "avx"))]
 pub use self::field::faster8;
 
 mod novel_poly_basis;

diff --git a/reed-solomon-novelpoly/src/novel_poly_basis/mod.rs b/reed-solomon-novelpoly/src/novel_poly_basis/mod.rs
@@ -62,7 +62,7 @@ impl CodeParams {
 
 	/// Check if this could use the `faster8` code path, possibly utilizing `avx` SIMD instructions
 	pub fn is_faster8(&self) -> bool {
-		#[cfg(target_feature = "avx")]
+		#[cfg(all(target_feature = "avx", feature = "avx"))]
 		{
 			self.k >= (Additive8x::LANE << 1) && self.n % Additive8x::LANE == 0
 		}