diff options
| author | Torbjorn Granlund <tege@gmplib.org> | 2009-12-28 14:54:27 +0100 |
|---|---|---|
| committer | Torbjorn Granlund <tege@gmplib.org> | 2009-12-28 14:54:27 +0100 |
| commit | 3c252d75bfead24d19a2d929bba4d2a1b59abb31 (patch) | |
| tree | 4baf7c7a2a6e26ad3cd6f5b430b2728906f3c2cb /mpn/x86_64/core2 | |
| parent | 7bae50bedb078dcca5716cb268d4a4e44bb59694 (diff) | |
| download | gmp-3c252d75bfead24d19a2d929bba4d2a1b59abb31.tar.gz | |
Optimise shifting for Core i.
Diffstat (limited to 'mpn/x86_64/core2')
| -rw-r--r-- | mpn/x86_64/core2/lshift.asm | 6 | ||||
| -rw-r--r-- | mpn/x86_64/core2/rshift.asm | 6 |
2 files changed, 6 insertions, 6 deletions
diff --git a/mpn/x86_64/core2/lshift.asm b/mpn/x86_64/core2/lshift.asm index 876b7433a..e3e400874 100644 --- a/mpn/x86_64/core2/lshift.asm +++ b/mpn/x86_64/core2/lshift.asm @@ -1,6 +1,6 @@ dnl x86-64 mpn_lshift optimized for "Core 2". -dnl Copyright 2007 Free Software Foundation, Inc. +dnl Copyright 2007, 2009 Free Software Foundation, Inc. dnl dnl This file is part of the GNU MP Library. dnl @@ -25,7 +25,7 @@ C K8,K9: 4.25 C K10: 4.25 C P4: 14.7 C P6 core2: 1.27 -C P6 corei7: 1.75 +C P6 corei7: 1.5 C INPUT PARAMETERS @@ -112,8 +112,8 @@ L(01): shld %cl, %r10, %r9 mov %r9, -16(rp) L(00): shld %cl, %r11, %r10 mov -24(up), %r9 - lea -32(up), up mov %r10, -24(rp) + add $-32, up lea -32(rp), rp sub $4, n jnc L(top) diff --git a/mpn/x86_64/core2/rshift.asm b/mpn/x86_64/core2/rshift.asm index 4405723ee..485fd4b0d 100644 --- a/mpn/x86_64/core2/rshift.asm +++ b/mpn/x86_64/core2/rshift.asm @@ -1,6 +1,6 @@ dnl x86-64 mpn_rshift optimized for "Core 2". -dnl Copyright 2007 Free Software Foundation, Inc. +dnl Copyright 2007, 2009 Free Software Foundation, Inc. dnl dnl This file is part of the GNU MP Library. dnl @@ -25,7 +25,7 @@ C K8,K9: 4.25 C K10: 4.25 C P4: 14.7 C P6 core2: 1.27 -C P6 corei7: 1.75 +C P6 corei7: 1.5 C INPUT PARAMETERS @@ -110,8 +110,8 @@ L(01): shrd %cl, %r10, %r9 mov %r9, 16(rp) L(00): shrd %cl, %r11, %r10 mov 16(up), %r9 - lea 32(up), up mov %r10, 24(rp) + add $32, up lea 32(rp), rp sub $4, n jnc L(top) |