Also run bench functions in release

Author: tgross35

ci/run.sh

@@ -26,13 +26,14 @@ else
     run="cargo test --manifest-path testcrate/Cargo.toml --no-fail-fast --target $target"
     $run
     $run --release
-    $run --benches
     $run --features c
     $run --features c --release
     $run --features no-asm
     $run --features no-asm --release
     $run --features no-f16-f128
     $run --features no-f16-f128 --release
+    $run --benches
+    $run --benches --release
 fi
 
 if [ "${TEST_VERBATIM:-}" = "1" ]; then

testcrate/Cargo.toml

@@ -22,7 +22,7 @@ default-features = false
 features = ["public-test-deps"]
 
 [dev-dependencies]
-criterion = { version = "0.5.1", default-features = false }
+criterion = { version = "0.5.1", default-features = false, features = ["cargo_bench_support"] }
 paste = "1.0.15"
 
 [target.'cfg(all(target_arch = "arm", not(any(target_env = "gnu", target_env = "musl")), target_os = "linux"))'.dev-dependencies]

testcrate/benches/float_extend.rs

@@ -14,11 +14,11 @@ float_bench! {
     sys_available: not(feature = "no-sys-f16"),
     asm: [
         #[cfg(target_arch = "aarch64")] {
-            // FIXME(f16_f128): use `f16` assembly once support is added (rust-lang/rust/#116909)
+            // FIXME(f16_f128): remove `to_bits()` after f16 asm support (rust-lang/rust/#116909)
             let ret: f32;
             asm!(
-                "fcvt    {ret:s}, h0",
-                // a = in(vreg) a,
+                "fcvt    {ret:s}, {a:h}",
+                a = in(vreg) a.to_bits(),
                 ret = lateout(vreg) ret,
                 options(nomem, nostack),
             );

testcrate/benches/float_trunc.rs

@@ -13,16 +13,16 @@ float_bench! {
     sys_available: not(feature = "no-sys-f16"),
     asm: [
         #[cfg(target_arch = "aarch64")] {
-            // FIXME(f16_f128): use `f16` assembly once support is added (rust-lang/rust/#116909)
-            let ret = core::mem::MaybeUninit::uninit();
+            // FIXME(f16_f128): remove `from_bits()` after f16 asm support (rust-lang/rust/#116909)
+            let ret: u16;
             asm!(
-                "fcvt    h0, {a:s}",
+                "fcvt    {ret:h}, {a:s}",
                 a = in(vreg) a,
-                // ret = lateout(vreg) ret,
+                ret = lateout(vreg) ret,
                 options(nomem, nostack),
             );
 
-            ret.assume_init()
+            f16::from_bits(ret)
         };
     ],
 }
@@ -35,16 +35,16 @@ float_bench! {
     sys_available: not(feature = "no-sys-f128"),
     asm: [
         #[cfg(target_arch = "aarch64")] {
-            // FIXME(f16_f128): use `f16` assembly once support is added (rust-lang/rust/#116909)
-            let ret = core::mem::MaybeUninit::uninit();
+            // FIXME(f16_f128): remove `from_bits()` after f16 asm support (rust-lang/rust/#116909)
+            let ret: u16;
             asm!(
-                "fcvt    h0, {a:d}",
+                "fcvt    {ret:h}, {a:d}",
                 a = in(vreg) a,
-                // ret = lateout(vreg) ret,
+                ret = lateout(vreg) ret,
                 options(nomem, nostack),
             );
 
-            ret.assume_init()
+            f16::from_bits(ret)
         };
     ],
 }

testcrate/src/bench.rs

@@ -206,33 +206,30 @@ macro_rules! float_bench {
                 }
             }
 
-            c.bench_function(&format!("{test_name} compiler-builtins"), |b| {
-                b.iter(|| {
-                    for ($($arg),*) in benchvec.iter().copied() {
-                        black_box(crate_fn( $(black_box($arg)),* ));
-                    }
-                })
-            });
+            let mut group = c.benchmark_group(test_name);
+            group.bench_function("compiler-builtins", |b| b.iter(|| {
+                for ($($arg),*) in benchvec.iter().copied() {
+                    black_box(crate_fn( $(black_box($arg)),* ));
+                }
+            }));
 
             #[cfg($sys_available)]
-            c.bench_function(&format!("{test_name} system"), |b| {
-                b.iter(|| {
-                    for ($($arg),*) in benchvec.iter().copied() {
-                        black_box(sys_fn( $(black_box($arg)),* ));
-                    }
-                })
-            });
+            group.bench_function("system", |b| b.iter(|| {
+                for ($($arg),*) in benchvec.iter().copied() {
+                    black_box(sys_fn( $(black_box($arg)),* ));
+                }
+            }));
 
             #[cfg(any( $($asm_meta),* ))]
-            c.bench_function(&format!(
-                "{test_name} assembly ({} {})", std::env::consts::ARCH, std::env::consts::FAMILY
-            ), |b| {
-                b.iter(|| {
-                    for ($($arg),*) in benchvec.iter().copied() {
-                        black_box(asm_fn( $(black_box($arg)),* ));
-                    }
-                })
-            });
+            group.bench_function(&format!(
+                "assembly ({} {})", std::env::consts::ARCH, std::env::consts::FAMILY
+            ), |b| b.iter(|| {
+                for ($($arg),*) in benchvec.iter().copied() {
+                    black_box(asm_fn( $(black_box($arg)),* ));
+                }
+            }));
+
+            group.finish();
         }
     }};