diff --git a/Cargo.toml b/Cargo.toml
index 21df1cc..eee2d1f 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -12,7 +12,7 @@ keywords = ["fft", "dft", "discrete", "fourier", "no_std"]
 categories = ["algorithms", "science"]
 
 [dependencies]
-monarch-derive = "=0.4.0"
+monarch-derive = { path = "crates/monarch-derive" }
 num-complex = "0.4"
 num-traits = { version = "0.2", default-features = false }
 
diff --git a/assets/linear_comparison.png b/assets/linear_comparison.png
index 4073289..a9b1bee 100644
Binary files a/assets/linear_comparison.png and b/assets/linear_comparison.png differ
diff --git a/assets/log_comparison.png b/assets/log_comparison.png
index aa75915..ed4f630 100644
Binary files a/assets/log_comparison.png and b/assets/log_comparison.png differ
diff --git a/crates/benches/generate_benchmark_plot.py b/crates/benches/generate_benchmark_plot.py
index a2ba893..41369fa 100644
--- a/crates/benches/generate_benchmark_plot.py
+++ b/crates/benches/generate_benchmark_plot.py
@@ -38,4 +38,6 @@
 plt.xlabel('FFT Size')
 # plt.yscale('log')
 plt.legend(loc='upper left')
+plt.gcf().set_size_inches(10, 5)
+# plt.savefig('../../linear_comparison.png')
 plt.show()
diff --git a/crates/monarch-derive/src/lib.rs b/crates/monarch-derive/src/lib.rs
index e47c802..4c8627f 100644
--- a/crates/monarch-derive/src/lib.rs
+++ b/crates/monarch-derive/src/lib.rs
@@ -114,7 +114,7 @@ pub fn generate_switch(_input: TokenStream) -> TokenStream {
         /// let input: Vec<_> = (0..8).map(|i| Complex::new(i as f32, 0.0)).collect();
         /// let output = fft::<8, _, _>(input);
         /// ```
-        #[inline]
+        #[inline(always)]
         pub fn fft<const N: usize, T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; N] {
             let x_in = input.as_ref();
             assert_eq!(x_in.len(), N);
@@ -135,7 +135,7 @@ pub fn generate_switch(_input: TokenStream) -> TokenStream {
         /// let input: Vec<_> = (0..8).map(|i| Complex::new(i as f32, 0.0)).collect();
         /// let output = ifft::<8, _, _>(input);
         /// ```
-        #[inline]
+        #[inline(always)]
         pub fn ifft<const N: usize, T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; N] {
             let x_in = input.as_ref();
             assert_eq!(x_in.len(), N);
@@ -187,7 +187,7 @@ pub fn generate_powers_of_two(_input: TokenStream) -> TokenStream {
 
         quote! {
             #[doc = concat!("Inner FFT")]
-            #[inline]
+            #[inline(always)]
             pub fn #func<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; #s] {
                 let n = #s;
                 let x = input.as_ref();
@@ -214,7 +214,7 @@ pub fn generate_powers_of_two(_input: TokenStream) -> TokenStream {
 
     let expanded = quote! {
         
-        #[inline]
+        #[inline(always)]
         pub fn fft1<T: Float, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 1] {
             let n = 1;
             let x = input.as_ref();
@@ -281,7 +281,7 @@ pub fn generate_coprimes(_input: TokenStream) -> TokenStream {
 
         quote! {
             #[doc = concat!("Inner FFT")]
-            #[inline]
+            #[inline(always)]
             pub fn #func<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; #s] {
                 let n = #s;
                 let x = input.as_ref();
@@ -366,7 +366,7 @@ pub fn generate_mixed_radix(_input: TokenStream) -> TokenStream {
 
         quote! {
             #[doc = concat!("Inner FFT")]
-            #[inline]
+            #[inline(always)]
             pub fn #func<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; #s] {
                 let n = #s;
                 let x = input.as_ref();
@@ -548,7 +548,7 @@ pub fn generate_primes(_input: TokenStream) -> TokenStream {
 
         quote! {
             #[doc = concat!("Inner FFT")]
-            #[inline]
+            #[inline(always)]
             pub fn #func<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; #s] {
                 let n = #s;
                 let x = input.as_ref();
@@ -603,7 +603,7 @@ pub fn generate_iffts(_input: TokenStream) -> TokenStream {
 
         quote! {
             #[doc = concat!("Inner iFFT")]
-            #[inline]
+            #[inline(always)]
             pub fn #func<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; #n] {
                 let x = input.as_ref();
                 assert_eq!(x.len(), #n);
@@ -621,7 +621,7 @@ pub fn generate_iffts(_input: TokenStream) -> TokenStream {
 
     let expanded = quote! {
         #[doc = concat!("Inner iFFT")]
-        #[inline]
+        #[inline(always)]
         pub fn ifft1<T: Float, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 1] {
             let n = 1;
             let x = input.as_ref();
diff --git a/src/lib.rs b/src/lib.rs
index 5af4997..466cafb 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -73,7 +73,7 @@ fn _compute_twiddle<T: Float + FloatConst>(index: usize, fft_len: usize) -> Comp
 }
 
 #[doc = concat!("Inner FFT")]
-#[inline]
+#[inline(always)]
 pub fn fft3<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 3] {
     let n = 3;
     let x = input.as_ref();
@@ -99,7 +99,7 @@ pub fn fft3<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex
 }
 
 #[doc = concat!("Inner FFT")]
-#[inline]
+#[inline(always)]
 pub fn fft9<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 9] {
     let n = 9;
     let x = input.as_ref();
@@ -132,7 +132,7 @@ pub fn fft9<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex
 }
 
 #[doc = concat!("Inner FFT")]
-#[inline]
+#[inline(always)]
 pub fn fft18<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 18] {
     let n = 18;
     let x = input.as_ref();
@@ -202,7 +202,7 @@ pub fn fft18<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Comple
 }
 
 #[doc = concat!("Inner FFT")]
-#[inline]
+#[inline(always)]
 pub fn fft27<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 27] {
     let n = 27;
     let x = input.as_ref();
@@ -279,7 +279,7 @@ pub fn fft27<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Comple
 }
 
 #[doc = concat!("Inner FFT")]
-#[inline]
+#[inline(always)]
 pub fn fft125<T: Float + FloatConst, A: AsRef<[Complex<T>]>>(input: A) -> [Complex<T>; 125] {
     let n = 125;
     let x = input.as_ref();