Use sleef cinz functions for platform bit-wise consistency

greatest-ape · Feb 13, 2021 · 13a91a7 · 13a91a7
1 parent 0bb7485
commit 13a91a7
Show file tree

Hide file tree

Showing 2 changed files with 18 additions and 18 deletions.
diff --git a/octasine/benches/process.rs b/octasine/benches/process.rs
@@ -15,41 +15,41 @@ use octasine::gen::AudioGen;
 /// --- Benchmarking OctaSine process_f32 variant: fallback (std) ---
 /// Total number of samples:      12800000
 /// Equivalent to audio duration: 290.24942 seconds
-/// Processing time in total:     27099 milliseconds
-/// Processing time per sample:   2117.1865 nanoseconds
-/// Estimated CPU use:            9.3364525%
+/// Processing time in total:     28821 milliseconds
+/// Processing time per sample:   2251.696 nanoseconds
+/// Estimated CPU use:            9.929736%
 /// Output hash (first 16 bytes): ad 0d 1d 04 5e 38 95 7f 
 ///                               6b dc 96 33 94 3e 15 72 
 /// 
 /// --- Benchmarking OctaSine process_f32 variant: fallback (sleef) ---
 /// Total number of samples:      12800000
 /// Equivalent to audio duration: 290.24942 seconds
-/// Processing time in total:     26339 milliseconds
-/// Processing time per sample:   2057.746 nanoseconds
-/// Estimated CPU use:            9.074609%
+/// Processing time in total:     21378 milliseconds
+/// Processing time per sample:   1670.2142 nanoseconds
+/// Estimated CPU use:            7.3653893%
 /// Output hash (first 16 bytes): ac fd ce 1e a2 7b 79 e1 
 ///                               75 06 b6 94 fe be c9 5f 
-/// Speed compared to std fallback:  1.0288862x
+/// Speed compared to std fallback:  1.348148x
 /// 
 /// --- Benchmarking OctaSine process_f32 variant: sse2 ---
 /// Total number of samples:      12800000
 /// Equivalent to audio duration: 290.24942 seconds
-/// Processing time in total:     17407 milliseconds
-/// Processing time per sample:   1359.9438 nanoseconds
-/// Estimated CPU use:            5.997256%
+/// Processing time in total:     18493 milliseconds
+/// Processing time per sample:   1444.8228 nanoseconds
+/// Estimated CPU use:            6.371417%
 /// Output hash (first 16 bytes): ac fd ce 1e a2 7b 79 e1 
 ///                               75 06 b6 94 fe be c9 5f 
-/// Speed compared to std fallback:  1.5568191x
+/// Speed compared to std fallback:  1.5584583x
 /// 
 /// --- Benchmarking OctaSine process_f32 variant: avx ---
 /// Total number of samples:      12800000
 /// Equivalent to audio duration: 290.24942 seconds
-/// Processing time in total:     10654 milliseconds
-/// Processing time per sample:   832.36273 nanoseconds
-/// Estimated CPU use:            3.6706362%
+/// Processing time in total:     12484 milliseconds
+/// Processing time per sample:   975.3484 nanoseconds
+/// Estimated CPU use:            4.3011284%
 /// Output hash (first 16 bytes): ac fd ce 1e a2 7b 79 e1 
 ///                               75 06 b6 94 fe be c9 5f 
-/// Speed compared to std fallback:  2.5435865x
+/// Speed compared to std fallback:  2.3086069x
 /// ```
 fn main(){
     #[allow(unused_variables)]

diff --git a/octasine/src/gen/mod.rs b/octasine/src/gen/mod.rs
@@ -77,7 +77,7 @@ pub struct FallbackSineSleef;
 impl FallbackSine for FallbackSineSleef {
     fn sin([a1, a2]: [f64; 2]) -> [f64; 2] {
         unsafe {
-            [sleef_sys::Sleef_sin_u35(a1), sleef_sys::Sleef_sin_u35(a2)]
+            [sleef_sys::Sleef_cinz_sind1_u35purec(a1), sleef_sys::Sleef_cinz_sind1_u35purec(a2)]
         }
     }
 }
@@ -192,7 +192,7 @@ impl Simd for Sse2 {
     }
     #[target_feature(enable = "sse2")]
     unsafe fn pd_fast_sin(a: __m128d) -> __m128d {
-        sleef_sys::Sleef_sind2_u35sse2(a)
+        sleef_sys::Sleef_cinz_sind2_u35sse2(a)
     }
     #[target_feature(enable = "sse2")]
     unsafe fn pd_pairwise_horizontal_sum(a: __m128d) -> __m128d {
@@ -257,7 +257,7 @@ impl Simd for Avx {
     }
     #[target_feature(enable = "avx")]
     unsafe fn pd_fast_sin(a: __m256d) -> __m256d {
-        sleef_sys::Sleef_sind4_u35avx(a)
+        sleef_sys::Sleef_cinz_sind4_u35avx(a)
     }
     #[target_feature(enable = "avx")]
     unsafe fn pd_pairwise_horizontal_sum(a: __m256d) -> __m256d {