https://github.com/Rust-for-Linux/linux/pull/356#issuecomment-859621440

By Bjorn3:

With opt-level=0 you can't do any realistic benchmarks. The compiler won't optimize away any zero-cost abstractions. For example for item in vec.iter() { ... } can be faster than while i < vec.len() { ...; i+=1; } with optimizations due to no bounds checking, but without optimizations it is likely much slower.

black_box calls didn't help as you likely passed a function item and not function pointer through the black box. This means that it is still known which function is called based on the type of the function item. Taking fn() -> Result<V, E> instead of F: Fn() -> Result<V, E> fixes this issue. Replacing the rt = match ... with rt += match ... inside use_result is also necessary to prevent optimizing away the match.

Cargo.toml

@@ -5,14 +5,5 @@ edition = "2018"
 
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
-[profile.dev]
-opt-level = 0
-
-[profile.release]
-opt-level = 0
-
-[profile.bench]
-opt-level = 0
-
 
 [dependencies]

src/main.rs

@@ -1,8 +1,10 @@
-#![feature(test)]
+#![feature(test, bench_black_box)]
 
-use core::num::{NonZeroI16, NonZeroI32};
 extern crate test;
 
+use std::hint::black_box;
+use std::num::{NonZeroI16, NonZeroI32};
+
 #[derive(Clone, Copy, PartialEq, Eq)]
 pub struct ErrorNzi32(NonZeroI32);
 pub struct ErrorNzi16(NonZeroI16);
@@ -67,13 +69,12 @@ mod bench {
         }
     }
 
-    fn use_result<F, V, E>(n: i32, f: F) -> (i32, Result<V, E>)
-    where
-        F: Fn() -> Result<V, E>,
+    #[inline(never)]
+    fn use_result<V, E>(n: i32, f: fn() -> Result<V, E>) -> (i32, Result<V, E>)
     {
         let mut rt :i32 = 0;
         for _ in 0..n {
-            rt = match f() {
+            rt += match f() {
                 Ok(_) => 0,
                 Err(_) => -1,
             };
@@ -84,62 +85,62 @@ mod bench {
 
     #[bench]
     fn bench_nzi32_100(b: &mut Bencher) {
-        b.iter(|| use_result(100, return_nzi32));
+        b.iter(|| use_result(100, black_box(return_nzi32)));
     }
 
     #[bench]
     fn bench_nzi32_10000(b: &mut Bencher) {
-        b.iter(|| use_result(10000, return_nzi32));
+        b.iter(|| use_result(10000, black_box(return_nzi32)));
     }
 
     #[bench]
     fn bench_nzi32_1000000(b: &mut Bencher) {
-        b.iter(|| use_result(1000000, return_nzi32));
+        b.iter(|| use_result(1000000, black_box(return_nzi32)));
     }
 
     #[bench]
     fn bench_nzi16_100(b: &mut Bencher) {
-        b.iter(|| use_result(100, return_nzi16));        
+        b.iter(|| use_result(100, black_box(return_nzi16)));
     }
 
     #[bench]
     fn bench_nzi16_10000(b: &mut Bencher) {
-        b.iter(|| use_result(10000, return_nzi16));        
+        b.iter(|| use_result(10000, black_box(return_nzi16)));
     }
 
     #[bench]
     fn bench_nzi16_1000000(b: &mut Bencher) {
-        b.iter(|| use_result(1000000, return_nzi16));        
+        b.iter(|| use_result(1000000, black_box(return_nzi16)));
     }
 
     #[bench]
     fn bench_i32_100(b: &mut Bencher) {
-        b.iter(|| use_result(100, return_i32)); 
+        b.iter(|| use_result(100, black_box(return_i32)));
     }
 
     #[bench]
     fn bench_i32_10000(b: &mut Bencher) {
-        b.iter(|| use_result(10000, return_i32)); 
+        b.iter(|| use_result(10000, black_box(return_i32)));
     }
 
     #[bench]
     fn bench_i32_1000000(b: &mut Bencher) {
-        b.iter(|| use_result(1000000, return_i32)); 
+        b.iter(|| use_result(1000000, black_box(return_i32)));
     }    
 
     #[bench]
     fn bench_i16_100(b: &mut Bencher) {
-        b.iter(|| use_result(100, return_i16)); 
+        b.iter(|| use_result(100, black_box(return_i16)));
     }
 
     #[bench]
     fn bench_i16_10000(b: &mut Bencher) {
-        b.iter(|| use_result(10000, return_i16)); 
+        b.iter(|| use_result(10000, black_box(return_i16)));
     }
 
     #[bench]
     fn bench_i16_1000000(b: &mut Bencher) {
-        b.iter(|| use_result(1000000, return_i16)); 
+        b.iter(|| use_result(1000000, black_box(return_i16)));
     }    
 }