remove simd_memcpy with official memcpy

Signed-off-by: Duc Tri Nguyen <[email protected]>

mlkem/asm/clean/rej_uniform_asm.s

@@ -1,78 +1,13 @@
 // SPDX-License-Identifier: Apache-2.0
-
-/*************************************************
-* Name:        simd_memcpy
-*
-* Function: void simd_mempcy(void *dst, void *src, size_t length);
-* Description: optimized memcpy with SIMD instructions for small data
-*
-* Arguments:   - void *dst:          pointer to destination buffer
-*              - void *src:          pointer to source buffer
-*              - size_t len:         length of src in bytes
-**************************************************/
-.align 4
-.global simd_memcpy
-.global _simd_memcpy
-simd_memcpy:
-_simd_memcpy:
-    /* Input registers */
-    dst                         .req x0
-    src                         .req x1
-    copylen                     .req x2
-
-    /* Temporary registers, no need to preserve */
-    data                        .req w12
-    aligned_addr                .req x13
-    aligned_len                 .req x14
-    unaligned_len               .req x15
-
-    cmp     copylen, #0                     // If copylen is less than or equal to 0
-    b.le    end                             // then we exit the function
-    cmp     copylen, #32                    // Check if size is less than 32 bytes
-    b.lo    small_copy                      // Branch to small copy if size is less than 32
-
-    // Align destination address
-    mov     aligned_addr, dst
-    and     aligned_addr, aligned_addr, #15
-    cbz     aligned_addr, aligned_copy      // If already aligned, jump to aligned copy
-
-    mov     unaligned_len, #16
-    sub     unaligned_len, unaligned_len, aligned_addr
-    cmp     copylen, unaligned_len
-    b.lo    small_copy                      // If size is smaller than alignment padding, use small copy
-    sub     copylen, copylen, unaligned_len
-
-    small_align:
-    ldrb    data, [src], #1
-    strb    data, [dst], #1
-    subs    unaligned_len, unaligned_len, #1
-    b.ne    small_align
-
-aligned_copy:
-    // Main copy loop using 128-bit Neon registers
-    lsr     aligned_len, copylen, #5        // Calculate the number of 32-byte chunks
-    cbz     aligned_len, tail_copy          // If no chunks, jump to tail copy
-
-    copy_loop:
-    ldp     q0, q1, [src], #32
-    stp     q0, q1, [dst], #32
-    subs    aligned_len, aligned_len, #1
-    b.ne    copy_loop
-
-tail_copy:
-    // Copy any remaining bytes less than 32 bytes
-    and     copylen, copylen, #31
-    cbz     copylen, end
-
-small_copy:
-    ldrb    data, [src], #1
-    strb    data, [dst], #1
-    subs    copylen, copylen, #1
-    b.ne    small_copy
-
-end:
-    ret
-
+#ifdef __APPLE__
+    .macro MEMCPY dst, src, len
+    bl _memcpy
+    .endm
+#else
+    .macro MEMCPY dst, src, len
+    bl memcpy
+    .endm
+#endif
 /*************************************************
 * Name:        rej_uniform_asm
 *
@@ -362,7 +297,8 @@ _rej_uniform_asm:
 
         ubfiz   x2, min, #1, #32
         mov     x1, sp_copy
-        bl      simd_memcpy
+
+        MEMCPY  x0, x1, x2
 
     return: