X-Git-Url: https://git.distorted.org.uk/~mdw/catacomb/blobdiff_plain/0923a413958b0e778a3f059c76355ab58e5be414..b9b279b4105524d5d4e5dcd389141645d904aa0c:/symm/salsa20-x86ish-sse2.S

diff --git a/symm/salsa20-x86ish-sse2.S b/symm/salsa20-x86ish-sse2.S
index ca677f17..76ac0ed9 100644
--- a/symm/salsa20-x86ish-sse2.S
+++ b/symm/salsa20-x86ish-sse2.S
@@ -33,21 +33,29 @@
 ///--------------------------------------------------------------------------
 /// Main code.
 
-	.arch pentium4
 	.text
 
+FUNC(salsa20_core_x86ish_avx)
+	.arch	.avx
+	vzeroupper
+  endprologue
+	// drop through...
+ENDFUNC
+
+	.arch	pentium4
+
 FUNC(salsa20_core_x86ish_sse2)
 
 	// Initial setup.
 
 #if CPUFAM_X86
 	// Arguments come in on the stack, and will need to be collected.  We
-	// we can get away with just the scratch registers for integer work,
-	// but we'll run out of XMM registers and will need some properly
-	// aligned space which we'll steal from the stack.  I don't trust the
-	// stack pointer's alignment, so I'll have to mask the stack pointer,
-	// which in turn means I'll need to keep track of the old value.
-	// Hence I'm making a full i386-style stack frame here.
+	// can get away with just the scratch registers for integer work, but
+	// we'll run out of XMM registers and will need some properly aligned
+	// space which we'll steal from the stack.  I don't trust the stack
+	// pointer's alignment, so I'll have to mask the stack pointer, which
+	// in turn means I'll need to keep track of the old value.  Hence I'm
+	// making a full i386-style stack frame here.
 	//
 	// The Windows and SysV ABIs are sufficiently similar that we don't
 	// need to worry about the differences here.