diff --git a/libc/arch-x86_64/string/sse2-memmove-slm.S b/libc/arch-x86_64/string/sse2-memmove-slm.S
index ee8440e..0dbffad 100644
--- a/libc/arch-x86_64/string/sse2-memmove-slm.S
+++ b/libc/arch-x86_64/string/sse2-memmove-slm.S
@@ -99,7 +99,7 @@
 /* Check whether we should copy backward or forward.  */
 	cmp	%rsi, %rdi
 	je	L(mm_return)
-	ja	L(mm_len_0_or_more_backward)
+	jg	L(mm_len_0_or_more_backward)
 
 /* Now do checks for lengths. We do [0..16], [0..32], [0..64], [0..128]
 	separately.  */
@@ -107,7 +107,7 @@
 	jbe	L(mm_len_0_16_bytes_forward)
 
 	cmp	$32, %rdx
-	jg	L(mm_len_32_or_more_forward)
+	ja	L(mm_len_32_or_more_forward)
 
 /* Copy [0..32] and return.  */
 	movdqu	(%rsi), %xmm0
@@ -118,7 +118,7 @@
 
 L(mm_len_32_or_more_forward):
 	cmp	$64, %rdx
-	jg	L(mm_len_64_or_more_forward)
+	ja	L(mm_len_64_or_more_forward)
 
 /* Copy [0..64] and return.  */
 	movdqu	(%rsi), %xmm0
@@ -133,7 +133,7 @@
 
 L(mm_len_64_or_more_forward):
 	cmp	$128, %rdx
-	jg	L(mm_len_128_or_more_forward)
+	ja	L(mm_len_128_or_more_forward)
 
 /* Copy [0..128] and return.  */
 	movdqu	(%rsi), %xmm0
@@ -155,13 +155,6 @@
 	jmp	L(mm_return)
 
 L(mm_len_128_or_more_forward):
-
-	cmp	$SHARED_CACHE_SIZE_HALF, %rdx
-	jae	L(mm_large_page_forward)
-
-	mov	%rsi, %r8  // copy src to r8
-	mov	%rdi, %r9  // copy dst to r9
-
 /* Aligning the address of destination.  */
 /*  save first unaligned 64 bytes */
 	movdqu	(%rsi), %xmm0
@@ -169,56 +162,57 @@
 	movdqu	32(%rsi), %xmm2
 	movdqu	48(%rsi), %xmm3
 
-	lea	64(%r9), %rdi
-	and	$-64, %rdi  /* rdi now aligned to next 64 byte boundary */
+	lea	64(%rdi), %r8
+	and	$-64, %r8  /* r8 now aligned to next 64 byte boundary */
+	sub	%rdi, %rsi /* rsi = src - dst = diff */
 
-	sub	%r9, %rsi /* rsi = src - dst = diff */
+	movdqu	(%r8, %rsi), %xmm4
+	movdqu	16(%r8, %rsi), %xmm5
+	movdqu	32(%r8, %rsi), %xmm6
+	movdqu	48(%r8, %rsi), %xmm7
 
-	movdqu	(%rdi, %rsi), %xmm4
-	movdqu	16(%rdi, %rsi), %xmm5
-	movdqu	32(%rdi, %rsi), %xmm6
-	movdqu	48(%rdi, %rsi), %xmm7
+	movdqu	%xmm0, (%rdi)
+	movdqu	%xmm1, 16(%rdi)
+	movdqu	%xmm2, 32(%rdi)
+	movdqu	%xmm3, 48(%rdi)
+	movdqa	%xmm4, (%r8)
+	movaps	%xmm5, 16(%r8)
+	movaps	%xmm6, 32(%r8)
+	movaps	%xmm7, 48(%r8)
+	add	$64, %r8
 
-	movdqu	%xmm0, (%r9)
-	movdqu	%xmm1, 16(%r9)
-	movdqu	%xmm2, 32(%r9)
-	movdqu	%xmm3, 48(%r9)
-	movdqa	%xmm4, (%rdi)
-	movdqa	%xmm5, 16(%rdi)
-	movdqa	%xmm6, 32(%rdi)
-	movdqa	%xmm7, 48(%rdi)
-	add	$64, %rdi
-
-	lea	(%r9, %rdx), %rbx
+	lea	(%rdi, %rdx), %rbx
 	and	$-64, %rbx
-
-	cmp	%rdi, %rbx
+	cmp	%r8, %rbx
 	jbe	L(mm_copy_remaining_forward)
 
+	cmp	$SHARED_CACHE_SIZE_HALF, %rdx
+	jae	L(mm_large_page_loop_forward)
+
 	.p2align 4
 L(mm_main_loop_forward):
 
-	prefetcht0 128(%rdi, %rsi)
+	prefetcht0 128(%r8, %rsi)
 
-	movdqu	(%rdi, %rsi), %xmm0
-	movdqu	16(%rdi, %rsi), %xmm1
-	movdqu	32(%rdi, %rsi), %xmm2
-	movdqu	48(%rdi, %rsi), %xmm3
-	movdqa	%xmm0, (%rdi)
-	movdqa	%xmm1, 16(%rdi)
-	movdqa	%xmm2, 32(%rdi)
-	movdqa	%xmm3, 48(%rdi)
-	lea	64(%rdi), %rdi
-	cmp	%rdi, %rbx
+	movdqu	(%r8, %rsi), %xmm0
+	movdqu	16(%r8, %rsi), %xmm1
+	movdqu	32(%r8, %rsi), %xmm2
+	movdqu	48(%r8, %rsi), %xmm3
+	movdqa	%xmm0, (%r8)
+	movaps	%xmm1, 16(%r8)
+	movaps	%xmm2, 32(%r8)
+	movaps	%xmm3, 48(%r8)
+	lea	64(%r8), %r8
+	cmp	%r8, %rbx
 	ja	L(mm_main_loop_forward)
 
 L(mm_copy_remaining_forward):
-	add	%r9, %rdx
-	sub	%rdi, %rdx
+	add	%rdi, %rdx
+	sub	%r8, %rdx
 /* We copied all up till %rdi position in the dst.
 	In %rdx now is how many bytes are left to copy.
 	Now we need to advance %r8. */
-	lea	(%rdi, %rsi), %r8
+	lea	(%r8, %rsi), %r9
 
 L(mm_remaining_0_64_bytes_forward):
 	cmp	$32, %rdx
@@ -237,49 +231,49 @@
 	cmpb	$2, %dl
 	.p2align 4,,1
 	ja	L(mm_remaining_3_4_bytes_forward)
-	movzbl	-1(%r8,%rdx), %esi
-	movzbl	(%r8), %ebx
-	movb	%sil, -1(%rdi,%rdx)
-	movb	%bl, (%rdi)
+	movzbl	-1(%r9,%rdx), %esi
+	movzbl	(%r9), %ebx
+	movb	%sil, -1(%r8,%rdx)
+	movb	%bl, (%r8)
 	jmp	L(mm_return)
 
 L(mm_remaining_33_64_bytes_forward):
-	movdqu	(%r8), %xmm0
-	movdqu	16(%r8), %xmm1
-	movdqu	-32(%r8, %rdx), %xmm2
-	movdqu	-16(%r8, %rdx), %xmm3
-	movdqu	%xmm0, (%rdi)
-	movdqu	%xmm1, 16(%rdi)
-	movdqu	%xmm2, -32(%rdi, %rdx)
-	movdqu	%xmm3, -16(%rdi, %rdx)
+	movdqu	(%r9), %xmm0
+	movdqu	16(%r9), %xmm1
+	movdqu	-32(%r9, %rdx), %xmm2
+	movdqu	-16(%r9, %rdx), %xmm3
+	movdqu	%xmm0, (%r8)
+	movdqu	%xmm1, 16(%r8)
+	movdqu	%xmm2, -32(%r8, %rdx)
+	movdqu	%xmm3, -16(%r8, %rdx)
 	jmp	L(mm_return)
 
 L(mm_remaining_17_32_bytes_forward):
-	movdqu	(%r8), %xmm0
-	movdqu	-16(%r8, %rdx), %xmm1
-	movdqu	%xmm0, (%rdi)
-	movdqu	%xmm1, -16(%rdi, %rdx)
-	jmp	L(mm_return)
-
-L(mm_remaining_3_4_bytes_forward):
-	movzwl	-2(%r8,%rdx), %esi
-	movzwl	(%r8), %ebx
-	movw	%si, -2(%rdi,%rdx)
-	movw	%bx, (%rdi)
+	movdqu	(%r9), %xmm0
+	movdqu	-16(%r9, %rdx), %xmm1
+	movdqu	%xmm0, (%r8)
+	movdqu	%xmm1, -16(%r8, %rdx)
 	jmp	L(mm_return)
 
 L(mm_remaining_5_8_bytes_forward):
-	movl	(%r8), %esi
-	movl	-4(%r8,%rdx), %ebx
-	movl	%esi, (%rdi)
-	movl	%ebx, -4(%rdi,%rdx)
+	movl	(%r9), %esi
+	movl	-4(%r9,%rdx), %ebx
+	movl	%esi, (%r8)
+	movl	%ebx, -4(%r8,%rdx)
 	jmp	L(mm_return)
 
 L(mm_remaining_9_16_bytes_forward):
-	mov	(%r8), %rsi
-	mov	-8(%r8, %rdx), %rbx
-	mov	%rsi, (%rdi)
-	mov	%rbx, -8(%rdi, %rdx)
+	mov	(%r9), %rsi
+	mov	-8(%r9, %rdx), %rbx
+	mov	%rsi, (%r8)
+	mov	%rbx, -8(%r8, %rdx)
+	jmp	L(mm_return)
+
+L(mm_remaining_3_4_bytes_forward):
+	movzwl	-2(%r9,%rdx), %esi
+	movzwl	(%r9), %ebx
+	movw	%si, -2(%r8,%rdx)
+	movw	%bx, (%r8)
 	jmp	L(mm_return)
 
 L(mm_len_0_16_bytes_forward):
@@ -321,16 +315,21 @@
 	mov	%rsi, -8(%rdi, %rdx)
 	jmp	L(mm_return)
 
+L(mm_recalc_len):
+/* Compute in %rdx how many bytes are left to copy after
+	the main loop stops.  */
+	mov 	%rbx, %rdx
+	sub 	%rdi, %rdx
 /* The code for copying backwards.  */
 L(mm_len_0_or_more_backward):
 
-/* Now do checks for lengths. We do [0..16], [0..32], [0..64], [0..128]
+/* Now do checks for lengths. We do [0..16], [16..32], [32..64], [64..128]
 	separately.  */
 	cmp	$16, %rdx
 	jbe	L(mm_len_0_16_bytes_backward)
 
 	cmp	$32, %rdx
-	jg	L(mm_len_32_or_more_backward)
+	ja	L(mm_len_32_or_more_backward)
 
 /* Copy [0..32] and return.  */
 	movdqu	(%rsi), %xmm0
@@ -341,7 +340,7 @@
 
 L(mm_len_32_or_more_backward):
 	cmp	$64, %rdx
-	jg	L(mm_len_64_or_more_backward)
+	ja	L(mm_len_64_or_more_backward)
 
 /* Copy [0..64] and return.  */
 	movdqu	(%rsi), %xmm0
@@ -356,7 +355,7 @@
 
 L(mm_len_64_or_more_backward):
 	cmp	$128, %rdx
-	jg	L(mm_len_128_or_more_backward)
+	ja	L(mm_len_128_or_more_backward)
 
 /* Copy [0..128] and return.  */
 	movdqu	(%rsi), %xmm0
@@ -378,10 +377,6 @@
 	jmp	L(mm_return)
 
 L(mm_len_128_or_more_backward):
-
-	cmp	$SHARED_CACHE_SIZE_HALF, %rdx
-	jae	L(mm_large_page_backward)
-
 /* Aligning the address of destination. We need to save
 	16 bits from the source in order not to overwrite them.  */
 	movdqu	-16(%rsi, %rdx), %xmm0
@@ -405,22 +400,19 @@
 	movdqu	%xmm2, -48(%rdi, %rdx)
 	movdqu	%xmm3, -64(%rdi, %rdx)
 	movdqa	%xmm4, -16(%r9)
-	movdqa	%xmm5, -32(%r9)
-	movdqa	%xmm6, -48(%r9)
-	movdqa	%xmm7, -64(%r9)
+	movaps	%xmm5, -32(%r9)
+	movaps	%xmm6, -48(%r9)
+	movaps	%xmm7, -64(%r9)
 	lea	-64(%r9), %r9
 
 	lea	64(%rdi), %rbx
 	and	$-64, %rbx
 
-/* Compute in %rdx how many bytes are left to copy after
-	the main loop stops.  */
-	mov 	%rbx, %rdx
-	sub 	%rdi, %rdx
-
 	cmp	%r9, %rbx
-	jb	L(mm_main_loop_backward)
-	jmp	L(mm_len_0_or_more_backward)
+	jae	L(mm_recalc_len)
+
+	cmp	$SHARED_CACHE_SIZE_HALF, %rdx
+	jae	L(mm_large_page_loop_backward)
 
 	.p2align 4
 L(mm_main_loop_backward):
@@ -432,13 +424,13 @@
 	movdqu	-32(%r9, %r8), %xmm2
 	movdqu	-16(%r9, %r8), %xmm3
 	movdqa	%xmm0, -64(%r9)
-	movdqa	%xmm1, -48(%r9)
-	movdqa	%xmm2, -32(%r9)
-	movdqa	%xmm3, -16(%r9)
+	movaps	%xmm1, -48(%r9)
+	movaps	%xmm2, -32(%r9)
+	movaps	%xmm3, -16(%r9)
 	lea	-64(%r9), %r9
 	cmp	%r9, %rbx
 	jb	L(mm_main_loop_backward)
-	jmp	L(mm_len_0_or_more_backward)
+	jmp	L(mm_recalc_len)
 
 /* Copy [0..16] and return.  */
 L(mm_len_0_16_bytes_backward):
@@ -485,138 +477,23 @@
 
 /* Big length copy forward part.  */
 
-L(mm_large_page_forward):
-/* Aligning the address of destination. We need to save
-	16 bits from the source in order not to overwrite them.  */
-
-	mov	%rsi, %r8
-	mov	%rdi, %r9
-
-	movdqu	(%rsi), %xmm0
-	movdqu	16(%rsi), %xmm1
-	movdqu	32(%rsi), %xmm2
-	movdqu	48(%rsi), %xmm3
-
-	lea	64(%r9), %rdi
-	and	$-64, %rdi      /* rdi = aligned dst */
-
-	sub	%r9, %rsi        /* rsi = diff */
-
-	movdqu	(%rdi, %rsi), %xmm4
-	movdqu	16(%rdi, %rsi), %xmm5
-	movdqu	32(%rdi, %rsi), %xmm6
-	movdqu	48(%rdi, %rsi), %xmm7
-
-	movdqu	%xmm0, (%r9)
-	movdqu	%xmm1, 16(%r9)
-	movdqu	%xmm2, 32(%r9)
-	movdqu	%xmm3, 48(%r9)
-	movntdq	%xmm4, (%rdi)
-	movntdq	%xmm5, 16(%rdi)
-	movntdq	%xmm6, 32(%rdi)
-	movntdq	%xmm7, 48(%rdi)
-	add	$64, %rdi
-
-	lea	(%r9, %rdx), %rbx
-	and	$-128, %rbx
-
-	cmp	%rdi, %rbx
-	jbe	L(mm_copy_remaining_forward)
-
 	.p2align 4
 L(mm_large_page_loop_forward):
-	movdqu	(%rdi, %rsi), %xmm0
-	movdqu	16(%rdi, %rsi), %xmm1
-	movdqu	32(%rdi, %rsi), %xmm2
-	movdqu	48(%rdi, %rsi), %xmm3
-	movdqu	64(%rdi, %rsi), %xmm4
-	movdqu	80(%rdi, %rsi), %xmm5
-	movdqu	96(%rdi, %rsi), %xmm6
-	movdqu	112(%rdi, %rsi), %xmm7
-	movntdq	%xmm0, (%rdi)
-	movntdq	%xmm1, 16(%rdi)
-	movntdq	%xmm2, 32(%rdi)
-	movntdq	%xmm3, 48(%rdi)
-	movntdq	%xmm4, 64(%rdi)
-	movntdq	%xmm5, 80(%rdi)
-	movntdq	%xmm6, 96(%rdi)
-	movntdq	%xmm7, 112(%rdi)
-	lea 	128(%rdi), %rdi
-	cmp	%rdi, %rbx
+	movdqu	(%r8, %rsi), %xmm0
+	movdqu	16(%r8, %rsi), %xmm1
+	movdqu	32(%r8, %rsi), %xmm2
+	movdqu	48(%r8, %rsi), %xmm3
+	movntdq	%xmm0, (%r8)
+	movntdq	%xmm1, 16(%r8)
+	movntdq	%xmm2, 32(%r8)
+	movntdq	%xmm3, 48(%r8)
+	lea 	64(%r8), %r8
+	cmp	%r8, %rbx
 	ja	L(mm_large_page_loop_forward)
 	sfence
-
-	add 	%r9, %rdx
-	sub 	%rdi, %rdx
-/* We copied all up till %rdi position in the dst.
-	In %rdx now is how many bytes are left to copy.
-	Now we need to advance %r8. */
-	lea 	(%rdi, %rsi), %r8
-
-	cmp	$64, %rdx
-	jb	L(mm_remaining_0_64_bytes_forward)
-
-	movdqu	(%r8), %xmm0
-	movdqu	16(%r8), %xmm1
-	movdqu	32(%r8), %xmm2
-	movdqu	48(%r8), %xmm3
-	movdqu	-64(%r8, %rdx), %xmm4
-	movdqu	-48(%r8, %rdx), %xmm5
-	movdqu	-32(%r8, %rdx), %xmm6
-	movdqu	-16(%r8, %rdx), %xmm7
-	movdqu	%xmm0, (%rdi)
-	movdqu	%xmm1, 16(%rdi)
-	movdqu	%xmm2, 32(%rdi)
-	movdqu	%xmm3, 48(%rdi)
-	movdqu	%xmm4, -64(%rdi, %rdx)
-	movdqu	%xmm5, -48(%rdi, %rdx)
-	movdqu	%xmm6, -32(%rdi, %rdx)
-	movdqu	%xmm7, -16(%rdi, %rdx)
-	jmp	L(mm_return)
-
+	jmp	L(mm_copy_remaining_forward)
 
 /* Big length copy backward part.  */
-L(mm_large_page_backward):
-/* Aligning the address of destination. We need to save
-	16 bits from the source in order not to overwrite them.  */
-
-	movdqu	-16(%rsi, %rdx), %xmm0
-	movdqu	-32(%rsi, %rdx), %xmm1
-	movdqu	-48(%rsi, %rdx), %xmm2
-	movdqu	-64(%rsi, %rdx), %xmm3
-
-	lea	(%rdi, %rdx), %r9
-	and	$-64, %r9
-
-	mov 	%rsi, %r8
-	sub 	%rdi, %r8
-
-	movdqu	-16(%r9, %r8), %xmm4
-	movdqu	-32(%r9, %r8), %xmm5
-	movdqu	-48(%r9, %r8), %xmm6
-	movdqu	-64(%r9, %r8), %xmm7
-
-	movdqu	%xmm0, -16(%rdi, %rdx)
-	movdqu	%xmm1, -32(%rdi, %rdx)
-	movdqu	%xmm2, -48(%rdi, %rdx)
-	movdqu	%xmm3, -64(%rdi, %rdx)
-	movntdq	%xmm4, -16(%r9)
-	movntdq	%xmm5, -32(%r9)
-	movntdq	%xmm6, -48(%r9)
-	movntdq	%xmm7, -64(%r9)
-	lea 	-64(%r9), %r9
-
-	lea 	128(%rdi), %rbx
-	and 	$-64, %rbx
-
-/* Compute in %rdx how many bytes are left to copy after
-	the main loop stops.  */
-	mov 	%rbx, %rdx
-	sub 	%rdi, %rdx
-
-	cmp	%r9, %rbx
-	jae	L(mm_len_0_or_more_backward)
-
 	.p2align 4
 L(mm_large_page_loop_backward):
 	movdqu	-64(%r9, %r8), %xmm0
@@ -630,6 +507,7 @@
 	lea 	-64(%r9), %r9
 	cmp	%r9, %rbx
 	jb	L(mm_large_page_loop_backward)
-	jmp	L(mm_len_0_or_more_backward)
+	sfence
+	jmp	L(mm_recalc_len)
 
 END (MEMMOVE)