[Gmp-commit] /var/hg/gmp: core2/aorsmul_1.asm: Added mpn_addmul_1c and mpn_su...

[mercurial at gmplib.org]

details:   /var/hg/gmp/rev/c61e22c8e042
changeset: 14669:c61e22c8e042
user:      Niels M?ller <nisse at lysator.liu.se>
date:      Fri Feb 24 10:31:01 2012 +0100
description:
core2/aorsmul_1.asm: Added mpn_addmul_1c and mpn_submul_1c entry points.

diffstat:

 ChangeLog                      |   5 +++++
 mpn/x86_64/core2/aorsmul_1.asm |  31 ++++++++++++++++++++++++++++++-
 2 files changed, 35 insertions(+), 1 deletions(-)

diffs (73 lines):

diff -r a96b5b4c7ef2 -r c61e22c8e042 ChangeLog
--- a/ChangeLog	Fri Feb 24 10:06:38 2012 +0100
+++ b/ChangeLog	Fri Feb 24 10:31:01 2012 +0100
@@ -1,3 +1,8 @@
+2012-02-24  Niels Möller  <nisse at lysator.liu.se>
+
+	* mpn/x86_64/core2/aorsmul_1.asm: Added mpn_addmul_1c and
+	mpn_submul_1c entry points.
+
 2012-02-24 Marco Bodrato <bodrato at mail.dm.unipi.it>
 
 	* mpz/invert.c: Use ABSIZ, MPZ_EQUAL_1_P.
diff -r a96b5b4c7ef2 -r c61e22c8e042 mpn/x86_64/core2/aorsmul_1.asm
--- a/mpn/x86_64/core2/aorsmul_1.asm	Fri Feb 24 10:06:38 2012 +0100
+++ b/mpn/x86_64/core2/aorsmul_1.asm	Fri Feb 24 10:31:01 2012 +0100
@@ -41,20 +41,48 @@
 ifdef(`OPERATION_addmul_1',`
       define(`ADDSUB',        `add')
       define(`func',  `mpn_addmul_1')
+      define(`func_1c',  `mpn_addmul_1c')
 ')
 ifdef(`OPERATION_submul_1',`
       define(`ADDSUB',        `sub')
       define(`func',  `mpn_submul_1')
+      define(`func_1c',  `mpn_submul_1c')
 ')
 
-MULFUNC_PROLOGUE(mpn_addmul_1 mpn_submul_1)
+MULFUNC_PROLOGUE(mpn_addmul_1 mpn_addmul_1c mpn_submul_1 mpn_submul_1c)
 
 ABI_SUPPORT(DOS64)
 ABI_SUPPORT(STD64)
 
+	C For DOS, on the stack we have four saved registers, return
+	C address, space for four register arguments, and finally the
+	C carry input.
+	
+IFDOS(` define(`carry_in', `72(%rsp)')') dnl
+IFSTD(` define(`carry_in', `%r8')') dnl
+	
 ASM_START()
 	TEXT
 	ALIGN(16)
+PROLOGUE(func_1c)
+	DOS64_ENTRY(4)
+	push	%rbx
+	push	%rbp
+	lea	(%rdx), %rbx
+	neg	%rbx
+
+	mov	(up), %rax
+	mov	(rp), %r10
+
+	lea	-16(rp,%rdx,8), rp
+	lea	(up,%rdx,8), up
+	mul	%rcx
+	add	carry_in, %rax
+	adc	$0, %rdx
+	jmp	L(start_nc)
+EPILOGUE()
+
+	ALIGN(16)
 PROLOGUE(func)
 	DOS64_ENTRY(4)
 	push	%rbx
@@ -69,6 +97,7 @@
 	lea	(up,%rdx,8), up
 	mul	%rcx
 
+L(start_nc):
 	bt	$0, R32(%rbx)
 	jc	L(odd)