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

[mercurial at gmplib.org]

details:   /var/hg/gmp/rev/2e5831d3f70b
changeset: 15167:2e5831d3f70b
user:      Torbjorn Granlund <tege at gmplib.org>
date:      Sat Dec 08 12:03:48 2012 +0100
description:
Fix a comment typo.

details:   /var/hg/gmp/rev/5d118b8463f8
changeset: 15168:5d118b8463f8
user:      Torbjorn Granlund <tege at gmplib.org>
date:      Sat Dec 08 12:05:06 2012 +0100
description:
Add item about --disable-assembly.

details:   /var/hg/gmp/rev/d0579520edca
changeset: 15169:d0579520edca
user:      Torbjorn Granlund <tege at gmplib.org>
date:      Sat Dec 08 12:13:19 2012 +0100
description:
Add some ARM A15 cycle numbers.

diffstat:

 NEWS                    |  3 +++
 mpn/arm/aors_n.asm      |  2 +-
 mpn/arm/aorsmul_1.asm   |  2 +-
 mpn/arm/bdiv_dbm1c.asm  |  2 +-
 mpn/arm/v5/mod_1_1.asm  |  2 +-
 mpn/arm/v5/mod_1_2.asm  |  2 +-
 mpn/arm/v6/addmul_1.asm |  2 +-
 mpn/arm/v6/addmul_2.asm |  2 +-
 mpn/x86_64/lshift.asm   |  2 +-
 9 files changed, 11 insertions(+), 8 deletions(-)

diffs (109 lines):

diff -r 73e97279c978 -r d0579520edca NEWS
--- a/NEWS	Thu Dec 06 11:10:48 2012 +0100
+++ b/NEWS	Sat Dec 08 12:13:19 2012 +0100
@@ -39,6 +39,9 @@
   * The fat binary mechanism is now more robust in its CPU recognition.
 
   MISC
+  * Inclusion of assembly code is now controlled by the configure options
+    --enable-assembly and --disable-assembly.  The "none" CPU targets is gone.
+
   * In C++, the conversions mpq_class->mpz_class, mpf_class->mpz_class and
     mpf_class->mpq_class are now explicit.
 
diff -r 73e97279c978 -r d0579520edca mpn/arm/aors_n.asm
--- a/mpn/arm/aors_n.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/aors_n.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -26,7 +26,7 @@
 C XScale	 ?
 C Cortex-A8	 ?
 C Cortex-A9	 2.5	slightly fluctuating
-C Cortex-A15	 ?
+C Cortex-A15	 2.25
 
 define(`rp', `r0')
 define(`up', `r1')
diff -r 73e97279c978 -r d0579520edca mpn/arm/aorsmul_1.asm
--- a/mpn/arm/aorsmul_1.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/aorsmul_1.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -24,7 +24,7 @@
 C XScale	 ?
 C Cortex-A8	 ?
 C Cortex-A9	 5.25
-C Cortex-A15	 ?
+C Cortex-A15	 4
 
 define(`rp', `r0')
 define(`up', `r1')
diff -r 73e97279c978 -r d0579520edca mpn/arm/bdiv_dbm1c.asm
--- a/mpn/arm/bdiv_dbm1c.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/bdiv_dbm1c.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -24,7 +24,7 @@
 C XScale	 ?
 C Cortex-A8	 ?
 C Cortex-A9	 4.25
-C Cortex-A15	 ?
+C Cortex-A15	 2.5
 
 C TODO
 C  * Try using umlal or umaal.
diff -r 73e97279c978 -r d0579520edca mpn/arm/v5/mod_1_1.asm
--- a/mpn/arm/v5/mod_1_1.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/v5/mod_1_1.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -26,7 +26,7 @@
 C XScale	 ?
 C Cortex-A8	 ?
 C Cortex-A9	 7
-C Cortex-A15	 ?
+C Cortex-A15	 6
 
 define(`ap', `r0')
 define(`n',  `r1')
diff -r 73e97279c978 -r d0579520edca mpn/arm/v5/mod_1_2.asm
--- a/mpn/arm/v5/mod_1_2.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/v5/mod_1_2.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -26,7 +26,7 @@
 C XScale	 ?
 C Cortex-A8	 ?
 C Cortex-A9	 4.25
-C Cortex-A15	 ?
+C Cortex-A15	 3
 
 define(`ap', `r0')
 define(`n',  `r1')
diff -r 73e97279c978 -r d0579520edca mpn/arm/v6/addmul_1.asm
--- a/mpn/arm/v6/addmul_1.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/v6/addmul_1.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -24,7 +24,7 @@
 C XScale	 -
 C Cortex-A8	 ?
 C Cortex-A9	 3.25
-C Cortex-A15	 ?
+C Cortex-A15	 4
 
 C TODO
 C  * Micro-optimise feed-in code.
diff -r 73e97279c978 -r d0579520edca mpn/arm/v6/addmul_2.asm
--- a/mpn/arm/v6/addmul_2.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/arm/v6/addmul_2.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -26,7 +26,7 @@
 C XScale	 -
 C Cortex-A8	 ?
 C Cortex-A9	 2.38
-C Cortex-A15	 ?
+C Cortex-A15	 2.5
 
 C TODO
 C  * Consider using more registers for the r[] loads, allowing better load-use
diff -r 73e97279c978 -r d0579520edca mpn/x86_64/lshift.asm
--- a/mpn/x86_64/lshift.asm	Thu Dec 06 11:10:48 2012 +0100
+++ b/mpn/x86_64/lshift.asm	Sat Dec 08 12:13:19 2012 +0100
@@ -49,7 +49,7 @@
 	jne	L(gen)
 
 C For cnt=1 we want to work from lowest limb towards higher limbs.
-C Check for bad overlap (up=rp is OK!) up=1..rp+n-1 is bad.
+C Check for bad overlap (up=rp is OK!) up=rp+1..rp+n-1 is bad.
 C FIXME: this could surely be done more cleverly.
 
 	mov    rp, %rax