Avoid calling non-#[inline(always)] functions from intrinsics.

This avoids the need to force-enable the +neon feature on ARM.

Author: Amanieu

.github/workflows/main.yml

@@ -116,7 +116,6 @@ jobs:
           os: ubuntu-latest
         - target: armv7-unknown-linux-gnueabihf
           os: ubuntu-latest
-          rustflags: -C target-feature=+neon
         - target: mips-unknown-linux-gnu
           os: ubuntu-latest
           norun: true

crates/core_arch/src/aarch64/neon/generated.rs

@@ -6,8 +6,7 @@ mod generated;
 pub use self::generated::*;
 
 use crate::{
-    convert::TryInto, core_arch::simd::*, core_arch::simd_llvm::*, hint::unreachable_unchecked,
-    mem::transmute,
+    core_arch::simd::*, core_arch::simd_llvm::*, hint::unreachable_unchecked, mem::transmute,
 };
 #[cfg(test)]
 use stdarch_test::assert_instr;

crates/core_arch/src/arm_shared/neon/generated.rs

@@ -10,7 +10,7 @@ macro_rules! simd_ty {
 
         #[allow(clippy::use_self)]
         impl $id {
-            #[inline]
+            #[inline(always)]
             pub(crate) const fn new($($elem_name: $elem_ty),*) -> Self {
                 $id($($elem_name),*)
             }
@@ -43,12 +43,12 @@ macro_rules! simd_m_ty {
 
         #[allow(clippy::use_self)]
         impl $id {
-            #[inline]
+            #[inline(always)]
             const fn bool_to_internal(x: bool) -> $ety {
                 [0 as $ety, !(0 as $ety)][x as usize]
             }
 
-            #[inline]
+            #[inline(always)]
             pub(crate) const fn new($($elem_name: bool),*) -> Self {
                 $id($(Self::bool_to_internal($elem_name)),*)
             }

crates/core_arch/src/arm_shared/neon/mod.rs

@@ -5894,7 +5894,7 @@ name = vqshl
 n-suffix
 constn = N
 multi_fn = static_assert_imm-out_bits_exp_len-N
-multi_fn = vqshl-self-noext, a, {vdup-nself-noext, N.try_into().unwrap()}
+multi_fn = vqshl-self-noext, a, {vdup-nself-noext, N as _}
 a = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
 n = 2
 validate 0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60
@@ -5921,7 +5921,7 @@ name = vqshl
 n-suffix
 constn = N
 multi_fn = static_assert_imm-out_bits_exp_len-N
-multi_fn = vqshl-self-noext, a, {vdup-nsigned-noext, N.try_into().unwrap()}
+multi_fn = vqshl-self-noext, a, {vdup-nsigned-noext, N as _}
 a = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
 n = 2
 validate 0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60
@@ -6480,7 +6480,7 @@ name = vrshr
 n-suffix
 constn = N
 multi_fn = static_assert-N-1-bits
-multi_fn = vrshl-self-noext, a, {vdup-nself-noext, (-N).try_into().unwrap()}
+multi_fn = vrshl-self-noext, a, {vdup-nself-noext, (-N) as _}
 a = 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64
 n = 2
 validate 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
@@ -6507,7 +6507,7 @@ name = vrshr
 n-suffix
 constn = N
 multi_fn = static_assert-N-1-bits
-multi_fn = vrshl-self-noext, a, {vdup-nsigned-noext, (-N).try_into().unwrap()}
+multi_fn = vrshl-self-noext, a, {vdup-nsigned-noext, (-N) as _}
 a = 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64
 n = 2
 validate 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
@@ -6804,7 +6804,7 @@ name = vshl
 n-suffix
 constn = N
 multi_fn = static_assert_imm-out_bits_exp_len-N
-multi_fn = simd_shl, a, {vdup-nself-noext, N.try_into().unwrap()}
+multi_fn = simd_shl, a, {vdup-nself-noext, N as _}
 a = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
 n = 2
 validate 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64
@@ -6818,7 +6818,7 @@ name = vshll
 n-suffix
 constn = N
 multi_fn = static_assert-N-0-bits
-multi_fn = simd_shl, {simd_cast, a}, {vdup-nout-noext, N.try_into().unwrap()}
+multi_fn = simd_shl, {simd_cast, a}, {vdup-nout-noext, N as _}
 a = 1, 2, 3, 4, 5, 6, 7, 8
 n = 2
 validate 4, 8, 12, 16, 20, 24, 28, 32
@@ -6851,7 +6851,7 @@ n-suffix
 constn = N
 multi_fn = static_assert-N-1-bits
 multi_fn = fix_right_shift_imm-N-bits
-multi_fn = simd_shr, a, {vdup-nself-noext, n.try_into().unwrap()}
+multi_fn = simd_shr, a, {vdup-nself-noext, n as _}
 a = 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64
 n = 2
 validate 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
@@ -6867,7 +6867,7 @@ name = vshrn_n
 no-q
 constn = N
 multi_fn = static_assert-N-1-halfbits
-multi_fn = simd_cast, {simd_shr, a, {vdup-nself-noext, N.try_into().unwrap()}}
+multi_fn = simd_cast, {simd_shr, a, {vdup-nself-noext, N as _}}
 a = 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64
 n = 2
 validate 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16

crates/core_arch/src/simd.rs

@@ -1304,7 +1304,7 @@ fn gen_aarch64(
                 };
                 format!(
                     r#"{}
-    {}{}({}, {} as i64, a.cast())"#,
+    {}{}({}, {} as i64, a as _)"#,
                     multi_calls,
                     ext_c,
                     current_fn,
@@ -1327,7 +1327,7 @@ fn gen_aarch64(
                 }
             }
         } else if link_aarch64.is_some() && matches!(fn_type, Fntype::Store) {
-            let cast = if is_vstx(&name) { ".cast()" } else { "" };
+            let cast = if is_vstx(&name) { " as _" } else { "" };
             match type_sub_len(in_t[1]) {
                 1 => format!(r#"{}{}(b, a{})"#, ext_c, current_fn, cast),
                 2 => format!(r#"{}{}(b.0, b.1, a{})"#, ext_c, current_fn, cast),
@@ -1336,7 +1336,7 @@ fn gen_aarch64(
                 _ => panic!("unsupported type: {}", in_t[1]),
             }
         } else if link_aarch64.is_some() && is_vldx(&name) {
-            format!(r#"{}{}(a.cast())"#, ext_c, current_fn,)
+            format!(r#"{}{}(a as _)"#, ext_c, current_fn,)
         } else {
             let trans: [&str; 2] = if link_t[3] != out_t {
                 ["transmute(", ")"]
@@ -1553,7 +1553,7 @@ fn gen_store_test(
         let a: [{}; {}] = {};
         let e: [{}; {}] = {};
         let mut r: [{}; {}] = [0{}; {}];
-        {}{}(r.as_mut_ptr(), core::ptr::read_unaligned(a[1..].as_ptr().cast()));
+        {}{}(r.as_mut_ptr(), core::ptr::read_unaligned(a[1..].as_ptr() as _));
         assert_eq!(r, e);
 "#,
             type_to_native_type(in_t[1]),
@@ -2196,7 +2196,7 @@ fn gen_arm(
                         _ => "",
                     };
                     format!(
-                        "{}(a.cast(), {}, {}, {})",
+                        "{}(a as _, {}, {}, {})",
                         current_fn,
                         subs,
                         constn.as_deref().unwrap(),
@@ -2235,7 +2235,7 @@ fn gen_arm(
             } else if matches!(fn_type, Fntype::Store) {
                 let (cast, size) = if is_vstx(&name) {
                     (
-                        ".cast()",
+                        " as _",
                         format!(", {}", type_bits(&type_to_sub_type(in_t[1])) / 8),
                     )
                 } else {
@@ -2276,7 +2276,7 @@ fn gen_arm(
                             _ => "",
                         };
                         format!(
-                            "{}({}, {} as i64, a.cast())",
+                            "{}({}, {} as i64, a as _)",
                             current_fn,
                             subs,
                             constn.as_deref().unwrap()
@@ -2307,7 +2307,7 @@ fn gen_arm(
                         _ => String::new(),
                     }
                 } else if matches!(fn_type, Fntype::Store) {
-                    let cast = if is_vstx(&name) { ".cast()" } else { "" };
+                    let cast = if is_vstx(&name) { " as _" } else { "" };
                     match type_sub_len(in_t[1]) {
                         1 => format!("{}(b, a{})", current_fn, cast),
                         2 => format!("{}(b.0, b.1, a{})", current_fn, cast),
@@ -2316,7 +2316,7 @@ fn gen_arm(
                         _ => String::new(),
                     }
                 } else if link_aarch64.is_some() && is_vldx(&name) {
-                    format!("{}(a.cast())", current_fn)
+                    format!("{}(a as _)", current_fn)
                 } else {
                     String::new()
                 };