From 1fffefbe582f348864161b2b96b117ef62ce7b8d Mon Sep 17 00:00:00 2001
From: Olivier Giniaux <oginiaux@gmail.com>
Date: Sun, 17 Nov 2024 11:19:20 +0100
Subject: [PATCH] Inline 8x16 compression

---
 src/gxhash/mod.rs          | 13 +++++++------
 src/gxhash/platform/arm.rs |  2 +-
 src/gxhash/platform/x86.rs |  4 ++--
 3 files changed, 10 insertions(+), 9 deletions(-)

diff --git a/src/gxhash/mod.rs b/src/gxhash/mod.rs
index 3ef2604..984633e 100644
--- a/src/gxhash/mod.rs
+++ b/src/gxhash/mod.rs
@@ -65,11 +65,6 @@ macro_rules! load_unaligned {
 
 pub(crate) use load_unaligned;
 
-#[cfg(target_arch = "arm")]
-use core::arch::arm::*;
-#[cfg(target_arch = "aarch64")]
-use core::arch::aarch64::*;
-
 #[inline(always)]
 pub(crate) unsafe fn gxhash(input: &[u8], seed: State) -> State {
     return finalize(gxhash_no_finish(input, seed));
@@ -94,7 +89,13 @@ pub(crate) unsafe fn gxhash_no_finish(input: &[u8], seed: State) -> State {
             if len < 16 {
                 break 'p1;
             } else if len < 128 {
-                // Possibly we can have something here
+                // Turbo mode, but is it worth the extra code size?
+                // It's not even great quality-wise, and v0 mix v1 may be a source of DOS collision attack
+                // if len >= 32 {
+                //     load_unaligned!(ptr, v0, v1);
+                //     state = aes_encrypt(state, aes_encrypt(v0, v1));
+                //     whole_vector_count -= 2;
+                // }
                 break 'p2;
             }
 
diff --git a/src/gxhash/platform/arm.rs b/src/gxhash/platform/arm.rs
index 1621983..309c800 100644
--- a/src/gxhash/platform/arm.rs
+++ b/src/gxhash/platform/arm.rs
@@ -77,7 +77,7 @@ pub unsafe fn ld(array: *const u32) -> State {
     vreinterpretq_s8_u32(vld1q_u32(array))
 }
 
-#[inline(never)]
+#[inline(always)]
 pub unsafe fn compress_8(mut ptr: *const State, whole_vector_count: usize, hash_vector: State, len: usize) -> (State, *const State, usize) {
 
     let end_address = ptr.add((whole_vector_count / 8) * 8) as usize;
diff --git a/src/gxhash/platform/x86.rs b/src/gxhash/platform/x86.rs
index a4023bf..8c63e24 100644
--- a/src/gxhash/platform/x86.rs
+++ b/src/gxhash/platform/x86.rs
@@ -69,7 +69,7 @@ pub unsafe fn ld(array: *const u32) -> State {
 }
 
 #[cfg(not(feature = "hybrid"))]
-#[inline(never)]
+#[inline(always)]
 pub unsafe fn compress_8(mut ptr: *const State, whole_vector_count: usize, hash_vector: State, len: usize) -> State {
 
     let end_address = ptr.add((whole_vector_count / 8) * 8) as usize;
@@ -113,7 +113,7 @@ pub unsafe fn compress_8(mut ptr: *const State, whole_vector_count: usize, hash_
 }
 
 #[cfg(feature = "hybrid")]
-#[inline(never)]
+#[inline(always)]
 pub unsafe fn compress_8(ptr: *const State, end_address: usize, hash_vector: State, len: usize) -> State {
     macro_rules! load_unaligned_x2 {
         ($ptr:ident, $($var:ident),+) => {