From: Mark Wooding <mdw@distorted.org.uk>
Date: Wed, 18 May 2016 09:29:03 +0000 (+0100)
Subject: symm/rijndael-x86-aesni.S: Use xmm5 instead of xmm7.
X-Git-Tag: 2.2.3~1^2~22^2~4
X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/commitdiff_plain/8d6ca5549ea58b040ba25dacafeb45a309de57dc

symm/rijndael-x86-aesni.S: Use xmm5 instead of xmm7.

The only reason is that (stupidly) the Windows 64-bit ABI designates
(the bottom 128 bits of) xmm7 as being callee-saved.
---

diff --git a/symm/rijndael-x86-aesni.S b/symm/rijndael-x86-aesni.S
index eba7b058..13c1ad56 100644
--- a/symm/rijndael-x86-aesni.S
+++ b/symm/rijndael-x86-aesni.S
@@ -267,7 +267,7 @@ FUNC(rijndael_setup_x86_aesni)
 
 	// Find the byte-reordering table.
 	ldgot	ecx
-	movdqa	xmm7, [INTADDR(endswap_tab, ecx)]
+	movdqa	xmm5, [INTADDR(endswap_tab, ecx)]
 
 	// Calculate the number of subkey words again.  (It's a good job
 	// we've got a fast multiplier.)
@@ -295,9 +295,9 @@ FUNC(rijndael_setup_x86_aesni)
 	.align	16
 endswap_block:
 	// End-swap ECX words starting at ESI.  The end-swapping table is
-	// already loaded into XMM7; and it's OK to work in 16-byte chunks.
+	// already loaded into XMM5; and it's OK to work in 16-byte chunks.
 	movdqu	xmm1, [esi]
-	pshufb	xmm1, xmm7
+	pshufb	xmm1, xmm5
 	movdqu	[esi], xmm1
 	add	esi, 16
 	sub	ecx, 4
@@ -318,13 +318,13 @@ FUNC(rijndael_eblk_x86_aesni)
 
 	// Find the magic endianness-swapping table.
 	ldgot	ecx
-	movdqa	xmm7, [INTADDR(endswap_tab, ecx)]
+	movdqa	xmm5, [INTADDR(endswap_tab, ecx)]
 
 	// Load the input block and end-swap it.  Also, start loading the
 	// keys.
 	mov	eax, [esp + 8]
 	movdqu	xmm0, [eax]
-	pshufb	xmm0, xmm7
+	pshufb	xmm0, xmm5
 	mov	eax, [esp + 4]
 	lea	edx, [eax + w]
 	mov	eax, [eax + nr]
@@ -409,7 +409,7 @@ er10:	movdqu	xmm1, [edx]
 	aesenclast xmm0, xmm1
 
 	// Unpermute the ciphertext block and store it.
-	pshufb	xmm0, xmm7
+	pshufb	xmm0, xmm5
 	mov	eax, [esp + 12]
 	movdqu	[eax], xmm0
 
@@ -427,13 +427,13 @@ FUNC(rijndael_dblk_x86_aesni)
 
 	// Find the magic endianness-swapping table.
 	ldgot	ecx
-	movdqa	xmm7, [INTADDR(endswap_tab, ecx)]
+	movdqa	xmm5, [INTADDR(endswap_tab, ecx)]
 
 	// Load the input block and end-swap it.  Also, start loading the
 	// keys.
 	mov	eax, [esp + 8]
 	movdqu	xmm0, [eax]
-	pshufb	xmm0, xmm7
+	pshufb	xmm0, xmm5
 	mov	eax, [esp + 4]
 	lea	edx, [eax + wi]
 	mov	eax, [eax + nr]
@@ -518,7 +518,7 @@ dr10:	movdqu	xmm1, [edx]
 	aesdeclast xmm0, xmm1
 
 	// Unpermute the ciphertext block and store it.
-	pshufb	xmm0, xmm7
+	pshufb	xmm0, xmm5
 	mov	eax, [esp + 12]
 	movdqu	[eax], xmm0