[Gmp-commit] /var/hg/gmp: 3 new changesets

[mercurial at gmplib.org]

details:   /var/hg/gmp/rev/9f380d652c92
changeset: 17318:9f380d652c92
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sat Mar 04 21:11:18 2017 +0100
description:
Fix comment typo.

details:   /var/hg/gmp/rev/da9d95e598b2
changeset: 17319:da9d95e598b2
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sat Mar 04 21:12:32 2017 +0100
description:
Fix comment typo.

details:   /var/hg/gmp/rev/fd5014d03e44
changeset: 17320:fd5014d03e44
user:      Torbjorn Granlund <tg at gmplib.org>
date:      Sat Mar 04 21:15:44 2017 +0100
description:
Rewrite ARM64 mpn_sqr_diag_addlsh1.

diffstat:

 mpn/arm64/sqr_diag_addlsh1.asm                |  87 +++++++++++++++++---------
 mpn/generic/toom22_mul.c                      |   2 +-
 mpn/sparc64/ultrasparct3/sqr_diag_addlsh1.asm |   2 +-
 3 files changed, 57 insertions(+), 34 deletions(-)

diffs (135 lines):

diff -r 043088c95083 -r fd5014d03e44 mpn/arm64/sqr_diag_addlsh1.asm
--- a/mpn/arm64/sqr_diag_addlsh1.asm	Tue Feb 28 14:35:40 2017 +0100
+++ b/mpn/arm64/sqr_diag_addlsh1.asm	Sat Mar 04 21:15:44 2017 +0100
@@ -1,8 +1,8 @@
-dnl  ARM64 mpn_sqr_dial_addlsh1.
+dnl  ARM64 mpn_sqr_diag_addlsh1.
 
 dnl  Contributed to the GNU project by Torbjörn Granlund.
 
-dnl  Copyright 2016 Free Software Foundation, Inc.
+dnl  Copyright 2016, 2017 Free Software Foundation, Inc.
 
 dnl  This file is part of the GNU MP Library.
 dnl
@@ -33,9 +33,11 @@
 include(`../config.m4')
 
 C	     cycles/limb
-C Cortex-A53	 6.85
+C Cortex-A53	 5.65
 C Cortex-A57	 3.5
-C X-Gene	 4.1
+C X-Gene	 3.38
+
+changecom(blah)
 
 define(`rp', `x0')
 define(`tp', `x1')
@@ -44,36 +46,57 @@
 
 ASM_START()
 PROLOGUE(mpn_sqr_diag_addlsh1)
-	ldp	x4, x5, [up], #16
-	mul	x10, x4, x4
-	umulh	x13, x4, x4
-	str	x10, [rp], #8
-
-	ldp	x6, x7, [tp], #16
-	mul	x4, x5, x5
-	adds	x10, x13, xzr
-	sub	n, n, #1
-	b	L(dm)
+	ldr	x15, [up],#8
+	lsr	x18, n, #1
+	tbz	n, #0, L(bx0)
 
-L(top):	ldr	x5, [up], #8
-	adds	x10, x6, x10
-	adcs	x11, x7, x4
-	ldp	x6, x7, [tp], #16
-	mul	x4, x5, x5
-	stp	x10, x11, [rp], #16
-	add	x10, x13, x12
-L(dm):	adcs	x6, x6, x6
-	sub	n, n, #1
-	adcs	x7, x7, x7
-	umulh	x13, x5, x5
-	adc	x12, xzr, xzr
-	cbnz	n, L(top)
+L(bx1):	adds	x7, xzr, xzr
+	mul	x12, x15, x15
+	ldr	x16, [up],#8
+	ldp	x4, x5, [tp],#16
+	umulh	x11, x15, x15
+	b	L(mid)
 
-	adds	x10, x6, x10
-	adcs	x11, x7, x4
-	stp	x10, x11, [rp], #16
-	adc	x13, x12, x13
-	str	x13, [rp]
+L(bx0):	adds	x5, xzr, xzr
+	mul	x12, x15, x15
+	ldr	x17, [up],#16
+	ldp	x6, x7, [tp],#32
+	umulh	x11, x15, x15
+	sub	x18, x18, #1
+	cbz	x18, L(end)
+
+	ALIGN(16)
+L(top):	extr	x9, x6, x5, #63
+	mul	x10, x17, x17
+	ldr	x16, [up,#-8]
+	adcs	x13, x9, x11
+	ldp	x4, x5, [tp,#-16]
+	umulh	x11, x17, x17
+	extr	x8, x7, x6, #63
+	stp	x12, x13, [rp],#16
+	adcs	x12, x8, x10
+L(mid):	extr	x9, x4, x7, #63
+	mul	x10, x16, x16
+	ldr	x17, [up],#16
+	adcs	x13, x9, x11
+	ldp	x6, x7, [tp],#32
+	umulh	x11, x16, x16
+	extr	x8, x5, x4, #63
+	stp	x12, x13, [rp],#16
+	adcs	x12, x8, x10
+	sub	x18, x18, #1
+	cbnz	x18, L(top)
+
+L(end):	extr	x9, x6, x5, #63
+	mul	x10, x17, x17
+	adcs	x13, x9, x11
+	umulh	x11, x17, x17
+	extr	x8, x7, x6, #63
+	stp	x12, x13, [rp]
+	adcs	x12, x8, x10
+	extr	x9, xzr, x7, #63
+	adcs	x13, x9, x11
+	stp	x12, x13, [rp,#16]
 
 	ret
 EPILOGUE()
diff -r 043088c95083 -r fd5014d03e44 mpn/generic/toom22_mul.c
--- a/mpn/generic/toom22_mul.c	Tue Feb 28 14:35:40 2017 +0100
+++ b/mpn/generic/toom22_mul.c	Sat Mar 04 21:15:44 2017 +0100
@@ -203,7 +203,7 @@
 
   MPN_INCR_U (pp + 2 * n, s + t, cy2);
   if (LIKELY (cy <= 2))
-    /* if s+t==n, cy is zero, but we should not acces pp[3*n] at all. */
+    /* if s+t==n, cy is zero, but we should not access pp[3*n] at all. */
     MPN_INCR_U (pp + 3 * n, s + t - n, cy);
   else
     MPN_DECR_U (pp + 3 * n, s + t - n, 1);
diff -r 043088c95083 -r fd5014d03e44 mpn/sparc64/ultrasparct3/sqr_diag_addlsh1.asm
--- a/mpn/sparc64/ultrasparct3/sqr_diag_addlsh1.asm	Tue Feb 28 14:35:40 2017 +0100
+++ b/mpn/sparc64/ultrasparct3/sqr_diag_addlsh1.asm	Sat Mar 04 21:15:44 2017 +0100
@@ -1,4 +1,4 @@
-dnl  SPARC v9 mpn_sqr_dial_addlsh1 for T3/T4/T5.
+dnl  SPARC v9 mpn_sqr_diag_addlsh1 for T3/T4/T5.
 
 dnl  Contributed to the GNU project by Torbjörn Granlund.