From b13e54dddb034f880110d1d9828202fe14228961 Mon Sep 17 00:00:00 2001
From: zymon <s@zymon.org>
Date: Mon, 28 Aug 2023 15:56:36 +0200
Subject: [PATCH] SIMD instruction added

---
 Project.toml | 7 +++----
 src/STFT.jl  | 5 +++--
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/Project.toml b/Project.toml
index 161c00c..e5fcece 100644
--- a/Project.toml
+++ b/Project.toml
@@ -1,13 +1,12 @@
 name = "STFT"
 uuid = "58bb99bf-048b-48b7-93e7-1cbf3ee61509"
-authors = ["Szymon M. Woźniak"]
-version = "1.0.1"
+authors = ["Szymon M. Woźniak <s@zymon.org>"]
+version = "1.1.0"
 
 [deps]
 FFTW = "7a1cc6ca-52ef-59f5-83cd-3a7055c09341"
-
+LoopVectorization = "bdcacae8-1622-11e9-2a5c-532679323890"
 
 [compat]
-
 FFTW = "1.4"
 julia = "1.6"
diff --git a/src/STFT.jl b/src/STFT.jl
index 778070d..32c78a1 100644
--- a/src/STFT.jl
+++ b/src/STFT.jl
@@ -1,5 +1,6 @@
 module STFT
 
+using LoopVectorization
 using FFTW
 
 export stft, istft
@@ -106,7 +107,7 @@ function analysis(
     N = N < W ? W : N   # DFT size
     sc = zeros(T, N, S) # Allocate container for signal segments
 
-    for s ∈ 1:S, n ∈ 1:W # Slice the signal
+    @turbo for s ∈ 1:S, n ∈ 1:W # Slice the signal
         sc[n, s] = w[n] * x[(s-1)*H+n]
     end
     _fft(sc, 1) # Convert segments to frequency-domain
@@ -229,7 +230,7 @@ function synthesis(
 
     xs = irfft(X, N, 1) # Convert segments to time-domain
 
-    for s ∈ 1:S
+    @turbo for s ∈ 1:S
         ss = (s-1)*H # Segment start
         for n = 1:W
             xn[ss+n] += xs[n, s] * w[n]