mirror of
https://github.com/BLAKE3-team/BLAKE3
synced 2024-03-28 10:19:52 +01:00
Implement better target detection for NEON
This commit is contained in:
parent
db436a50c2
commit
0a0bb7126e
11
build.rs
11
build.rs
|
@ -44,6 +44,14 @@ fn is_x86_32() -> bool {
|
|||
arch == "i386" || arch == "i586" || arch == "i686"
|
||||
}
|
||||
|
||||
fn is_arm() -> bool {
|
||||
is_armv7() || is_aarch64() || target_components()[0] == "arm"
|
||||
}
|
||||
|
||||
fn is_aarch64() -> bool {
|
||||
target_components()[0] == "aarch64"
|
||||
}
|
||||
|
||||
fn is_armv7() -> bool {
|
||||
target_components()[0] == "armv7"
|
||||
}
|
||||
|
@ -237,7 +245,8 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
|
|||
}
|
||||
}
|
||||
|
||||
if is_neon() {
|
||||
if (is_arm() && is_neon()) || (is_aarch64() && !is_pure()) {
|
||||
println!("cargo:rustc-cfg=blake3_neon");
|
||||
build_neon_c_intrinsics();
|
||||
}
|
||||
|
||||
|
|
|
@ -94,7 +94,7 @@ mod avx2;
|
|||
#[cfg(blake3_avx512_ffi)]
|
||||
#[path = "ffi_avx512.rs"]
|
||||
mod avx512;
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
#[path = "ffi_neon.rs"]
|
||||
mod neon;
|
||||
mod portable;
|
||||
|
|
|
@ -10,7 +10,7 @@ cfg_if::cfg_if! {
|
|||
pub const MAX_SIMD_DEGREE: usize = 8;
|
||||
}
|
||||
}
|
||||
} else if #[cfg(feature = "neon")] {
|
||||
} else if #[cfg(blake3_neon)] {
|
||||
pub const MAX_SIMD_DEGREE: usize = 4;
|
||||
} else {
|
||||
pub const MAX_SIMD_DEGREE: usize = 1;
|
||||
|
@ -30,7 +30,7 @@ cfg_if::cfg_if! {
|
|||
pub const MAX_SIMD_DEGREE_OR_2: usize = 8;
|
||||
}
|
||||
}
|
||||
} else if #[cfg(feature = "neon")] {
|
||||
} else if #[cfg(blake3_neon)] {
|
||||
pub const MAX_SIMD_DEGREE_OR_2: usize = 4;
|
||||
} else {
|
||||
pub const MAX_SIMD_DEGREE_OR_2: usize = 2;
|
||||
|
@ -49,7 +49,7 @@ pub enum Platform {
|
|||
#[cfg(blake3_avx512_ffi)]
|
||||
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
|
||||
AVX512,
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
NEON,
|
||||
}
|
||||
|
||||
|
@ -76,7 +76,7 @@ impl Platform {
|
|||
}
|
||||
// We don't use dynamic feature detection for NEON. If the "neon"
|
||||
// feature is on, NEON is assumed to be supported.
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
{
|
||||
return Platform::NEON;
|
||||
}
|
||||
|
@ -95,7 +95,7 @@ impl Platform {
|
|||
#[cfg(blake3_avx512_ffi)]
|
||||
#[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
|
||||
Platform::AVX512 => 16,
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
Platform::NEON => 4,
|
||||
};
|
||||
debug_assert!(degree <= MAX_SIMD_DEGREE);
|
||||
|
@ -129,7 +129,7 @@ impl Platform {
|
|||
crate::avx512::compress_in_place(cv, block, block_len, counter, flags)
|
||||
},
|
||||
// No NEON compress_in_place() implementation yet.
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
Platform::NEON => portable::compress_in_place(cv, block, block_len, counter, flags),
|
||||
}
|
||||
}
|
||||
|
@ -161,7 +161,7 @@ impl Platform {
|
|||
crate::avx512::compress_xof(cv, block, block_len, counter, flags)
|
||||
},
|
||||
// No NEON compress_xof() implementation yet.
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
Platform::NEON => portable::compress_xof(cv, block, block_len, counter, flags),
|
||||
}
|
||||
}
|
||||
|
@ -256,7 +256,7 @@ impl Platform {
|
|||
)
|
||||
},
|
||||
// Assumed to be safe if the "neon" feature is on.
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
Platform::NEON => unsafe {
|
||||
crate::neon::hash_many(
|
||||
inputs,
|
||||
|
@ -315,7 +315,7 @@ impl Platform {
|
|||
}
|
||||
}
|
||||
|
||||
#[cfg(feature = "neon")]
|
||||
#[cfg(blake3_neon)]
|
||||
pub fn neon() -> Option<Self> {
|
||||
// Assumed to be safe if the "neon" feature is on.
|
||||
Some(Self::NEON)
|
||||
|
|
Loading…
Reference in New Issue