aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorWill Schmidt <will_schmidt@vnet.ibm.com>2011-08-18 11:01:44 -0500
committerRyan S. Arnold <rsa@us.ibm.com>2011-08-18 11:01:44 -0500
commita7e0baec8c61a6bdf3b8fcb4ccb725477254f1d3 (patch)
tree313fa7049fd3e214a72a554a1cee7ae9b5b01b32
parent0c6ff7937931f1171a6aba3970c4aea7978a1b86 (diff)
downloadglibc-a7e0baec8c61a6bdf3b8fcb4ccb725477254f1d3.zip
glibc-a7e0baec8c61a6bdf3b8fcb4ccb725477254f1d3.tar.gz
glibc-a7e0baec8c61a6bdf3b8fcb4ccb725477254f1d3.tar.bz2
Provide a throughput boost to the 64-bit power7 strncmp code of approx
15%. The 32-bit throughput is not notably affected by this change, so the change to the 32-bit code is done to keep the two files in sync with each other.
-rw-r--r--ChangeLog7
-rw-r--r--sysdeps/powerpc/powerpc32/power7/strncmp.S4
-rw-r--r--sysdeps/powerpc/powerpc64/power7/strncmp.S4
3 files changed, 13 insertions, 2 deletions
diff --git a/ChangeLog b/ChangeLog
index f1a4004..0638748 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,10 @@
+2011-08-04 Will Schmidt <will_schmidt@vnet.ibm.com>
+
+ * sysdeps/powerpc/powerpc32/power7/strncmp.S: Adjust the alignment
+ and add nop instructions for throughput optimization.
+ * sysdeps/powerpc/powerpc64/power7/strncmp.S: Adjust the alignment
+ and nop instructions for throughput optimization.
+
2011-08-01 Adhemerval Zanella <azanella@linux.vnet.ibm.com>
* sysdeps/powerpc/fpu/e_hypot.c: New file: hypot optimized for POWER.
diff --git a/sysdeps/powerpc/powerpc32/power7/strncmp.S b/sysdeps/powerpc/powerpc32/power7/strncmp.S
index 7ee9e03..db466f0 100644
--- a/sysdeps/powerpc/powerpc32/power7/strncmp.S
+++ b/sysdeps/powerpc/powerpc32/power7/strncmp.S
@@ -27,7 +27,7 @@
const char *s2 [r4],
size_t size [r5]) */
-EALIGN (BP_SYM(strncmp),4,0)
+EALIGN (BP_SYM(strncmp),5,0)
#define rTMP r0
#define rRTN r3
@@ -47,9 +47,11 @@ EALIGN (BP_SYM(strncmp),4,0)
#define rBITDIF r11 /* bits that differ in s1 & s2 words */
dcbt 0,rSTR1
+ nop
or rTMP,rSTR2,rSTR1
lis r7F7F,0x7f7f
dcbt 0,rSTR2
+ nop
clrlwi. rTMP,rTMP,30
cmplwi cr1,rN,0
lis rFEFE,-0x101
diff --git a/sysdeps/powerpc/powerpc64/power7/strncmp.S b/sysdeps/powerpc/powerpc64/power7/strncmp.S
index 5ee5e2e..eace179 100644
--- a/sysdeps/powerpc/powerpc64/power7/strncmp.S
+++ b/sysdeps/powerpc/powerpc64/power7/strncmp.S
@@ -27,7 +27,7 @@
const char *s2 [r4],
size_t size [r5]) */
-EALIGN (BP_SYM(strncmp),4,0)
+EALIGN (BP_SYM(strncmp),5,0)
CALL_MCOUNT 3
#define rTMP r0
@@ -48,9 +48,11 @@ EALIGN (BP_SYM(strncmp),4,0)
#define rBITDIF r11 /* bits that differ in s1 & s2 words */
dcbt 0,rSTR1
+ nop
or rTMP,rSTR2,rSTR1
lis r7F7F,0x7f7f
dcbt 0,rSTR2
+ nop
clrldi. rTMP,rTMP,61
cmpldi cr1,rN,0
lis rFEFE,-0x101