[Gmp-commit] /var/hg/gmp: Speed up loop code, simplify non-loop code.

[mercurial at gmplib.org]

details:   /var/hg/gmp/rev/7c7a9cba56d3
changeset: 14757:7c7a9cba56d3
user:      Torbjorn Granlund <tege at gmplib.org>
date:      Thu Mar 15 15:53:22 2012 +0100
description:
Speed up loop code, simplify non-loop code.

diffstat:

 ChangeLog                  |   5 +++
 mpn/x86_64/core2/gcd_1.asm |  62 ++++++++++++++++++++-------------------------
 2 files changed, 32 insertions(+), 35 deletions(-)

diffs (111 lines):

diff -r d16912b36656 -r 7c7a9cba56d3 ChangeLog
--- a/ChangeLog	Tue Mar 13 20:24:40 2012 +0100
+++ b/ChangeLog	Thu Mar 15 15:53:22 2012 +0100
@@ -1,3 +1,8 @@
+2012-03-15  Torbjorn Granlund  <tege at gmplib.org>
+
+	* mpn/x86_64/core2/gcd_1.asm: Speed up loop code, simplify non-loop
+	code.
+
 2012-03-13  Torbjorn Granlund  <tege at gmplib.org>
 
 	* mpn/x86_64/core2/gcd_1.asm: Add hack to support fat builds.
diff -r d16912b36656 -r 7c7a9cba56d3 mpn/x86_64/core2/gcd_1.asm
--- a/mpn/x86_64/core2/gcd_1.asm	Tue Mar 13 20:24:40 2012 +0100
+++ b/mpn/x86_64/core2/gcd_1.asm	Thu Mar 15 15:53:22 2012 +0100
@@ -25,21 +25,20 @@
 
 
 C	     cycles/bit (approx)
-C AMD K8,K9	 8.5
-C AMD K10	 5
-C AMD bd1	 5
-C AMD bobcat	10
-C Intel P4	18 
-C Intel core2	 4.3
-C Intel NHM	 5
-C Intel SBR	 5
-C Intel atom	17
-C VIA nano	 5.3
-C Numbers measured with: speed -CD -s1-64 mpn_gcd_1
+C AMD K8,K9	 9.79
+C AMD K10	 5.34
+C AMD bd1	 ?
+C AMD bobcat	11.3
+C Intel P4	20.8
+C Intel core2	 5.09
+C Intel NHM	 6.27
+C Intel SBR	 5.30
+C Intel atom	19.6
+C VIA nano	 6.75
+C Numbers measured with: speed -c -s64 mpn_gcd_1
 
 C TODO
-C  * Optimise inner-loop for specific CPUs.  The code relies too much on OoO
-C    execution.
+C  * Optimise inner-loop for specific CPUs.
 
 C INPUT PARAMETERS
 define(`up',    `%rdi')
@@ -64,20 +63,14 @@
 PROLOGUE(mpn_gcd_1)
 	DOS64_ENTRY(3)
 	mov	(%rdi), %r8		C src low limb
-	mov	%r8, %r10
 	or	%rdx, %r8		C x | y
-
-	bsf	%r8, %rcx
-
-	shr	R8(%rcx), %r10
-	shr	R8(%rcx), %rdx
-	mov	R32(%rcx), R32(%r8)	C common twos
+	bsf	%r8, %r8		C common twos
 
 	bsf	%rdx, %rcx
 	shr	R8(%rcx), %rdx
 
-	push	%r8
-	push	%rdx
+	push	%r8			C preserve common twos over call
+	push	%rdx			C preserve v0 argument over call
 	sub	$8, %rsp		C maintain ABI required rsp alignment
 
 IFDOS(`	mov	%rdx, %r8	')
@@ -96,25 +89,24 @@
 	pop	%r8
 
 	bsf	%rax, %rcx
-
 	test	%rax, %rax
 	jnz	L(mid)
+	jmp	L(end)
 
-	jmp	L(done)
+C FIXME: 1st sub to rdx would shorten path
 
-	ALIGN(16)		C		K10	BD	C2	NHM	SBR
-L(top):	cmovc	%r10, %rax	C if x-y < 0	0,6	0,5	0,6	0,5	0,6
-	cmovc	%r9, %rdx	C use x,y-x	0	0	2	1	1
-	bsf	%r10, %rcx	C		0	0	0	0	0
-L(mid):	mov	%rdx, %r10	C		1	1	4	3	3
-	shr	R8(%rcx), %rax	C		4	3	2	3	3
-	mov	%rax, %r9	C		5	4	3	4	5
-	sub	%rax, %r10	C		5	4	5	4	5
-	sub	%rdx, %rax	C		5	4	4	4	5
+	ALIGN(16)		C               K10   BD    C2    NHM   SBR
+L(top):	cmovc	%r10, %rax	C if x-y < 0    0,3   0,3   0,6   0,5   0,5
+	cmovc	%r9, %rdx	C use x,y-x     0,3   0,3   2,8   1,7   1,7
+L(mid):	shr	R8(%rcx), %rax	C               1,7   1,6   2,8   2,8   2,8
+	mov	%rdx, %r10	C               1     1     4     3     3
+	sub	%rax, %r10	C               2     2     5     4     4
+	bsf	%r10, %rcx	C               3     3     6     5     5
+	mov	%rax, %r9	C               2     2     3     3     4
+	sub	%rdx, %rax	C               2     2     4     3     4
 	jnz	L(top)		C
 
-L(done):
-	mov	%rdx, %rax
+L(end):	mov	%rdx, %rax
 	mov	%r8, %rcx
 	shl	R8(%rcx), %rax
 	DOS64_EXIT()