powerpc: Handle worstcase behavior in strstr() for POWER7

Instead of checking needle length, constant 'n' number of comparisons
is checked to fall back to default implementation.  This patch is tested
on powerpc64 and powerpc64le.

2015-08-25  Rajalakshmi Srinivasaraghavan  <raji@linux.vnet.ibm.com>

	* sysdeps/powerpc/powerpc64/power7/strstr.S: Handle worst case.

diff --git a/ChangeLog b/ChangeLog
index a442ee1e12d96e0ec3b5c9e9304c24ea878531f3..13fc3fb98a01b8d1a46e01f5c5f69ff0b73204a5 100644 (file)
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,7 @@
+2015-08-25  Rajalakshmi Srinivasaraghavan  <raji@linux.vnet.ibm.com>
+
+       * sysdeps/powerpc/powerpc64/power7/strstr.S: Handle worst case.
+
 2015-08-25  H.J. Lu  <hongjiu.lu@intel.com>
 
        * sysdeps/x86_64/strlen.S: Replace %xmm[8-12] with %xmm[0-4].

diff --git a/sysdeps/powerpc/powerpc64/power7/strstr.S b/sysdeps/powerpc/powerpc64/power7/strstr.S
index bfb0c4992a82bec4cc274468dc85eb81c5cf58e3..fb3c810ac66ba370ff25890b909a9c663e5407b3 100644 (file)
--- a/sysdeps/powerpc/powerpc64/power7/strstr.S
+++ b/sysdeps/powerpc/powerpc64/power7/strstr.S
@@ -23,6 +23,8 @@
 /* The performance gain is obtained using aligned memory access, load
  * doubleword and usage of cmpb instruction for quicker comparison.  */
 
+#define ITERATIONS     64
+
 #ifndef STRLEN
 /* For builds with no IFUNC support, local calls should be made to internal
    GLIBC symbol (created by libc_hidden_builtin_def).  */
@@ -62,6 +64,8 @@ EALIGN (strstr, 4, 0)
        cfi_offset(r30, -16)
        std     r29, -24(r1)            /* Save callers register r29.  */
        cfi_offset(r29, -24)
+       std     r28, -32(r1)            /* Save callers register r28.  */
+       cfi_offset(r28, -32)
        std     r0, 16(r1)              /* Store the link register.  */
        cfi_offset(lr, 16)
        stdu    r1, -FRAMESIZE(r1)      /* Create the stack frame.  */
@@ -69,7 +73,6 @@ EALIGN (strstr, 4, 0)
 
        dcbt    0, r3
        dcbt    0, r4
-
        cmpdi   cr7, r3, 0
        beq     cr7, L(retnull)
        cmpdi   cr7, r4, 0
@@ -84,10 +87,6 @@ EALIGN (strstr, 4, 0)
        cmpdi   cr7, r3, 0      /* If search str is null.  */
        beq     cr7, L(ret_r3)
 
-       /* Call __strstr_ppc if needle len > 2048 */
-       cmpdi   cr7, r3, 2048
-       bgt     cr7, L(default)
-
        mr      r31, r3
        mr      r4, r3
        mr      r3, r29
@@ -105,7 +104,8 @@ EALIGN (strstr, 4, 0)
        /* If first char of search str is not present.  */
        cmpdi   cr7, r3, 0
        ble     cr7, L(end)
-
+       /* Reg r28 is used to count the number of iterations. */
+       li      r28, 0
        rldicl  r8, r3, 0, 52   /* Page cross check.  */
        cmpldi  cr7, r8, 4096-16
        bgt     cr7, L(bytebybyte)
@@ -324,6 +324,10 @@ L(return4):
        .align  4
 L(begin):
        mr      r3, r8
+       /* When our iterations exceed ITERATIONS,fall back to default. */
+       addi    r28, r28, 1
+       cmpdi   cr7, r28, ITERATIONS
+       beq     cr7, L(default)
        lbz     r4, 0(r30)
        bl      STRCHR
        nop
@@ -423,6 +427,10 @@ L(nextbyte):
        cmpdi   cr7, r9, -1
        beq     cr7, L(end)
        addi    r3, r4, 1
+       /* When our iterations exceed ITERATIONS,fall back to default. */
+       addi    r28, r28, 1
+       cmpdi   cr7, r28, ITERATIONS
+       beq     cr7, L(default)
        lbz     r4, 0(r30)
        bl      STRCHR
        nop
@@ -490,7 +498,6 @@ L(retnull):
 
        .align  4
 L(default):
-       mr      r3, r29
        mr      r4, r30
        bl      __strstr_ppc
        nop
@@ -500,6 +507,7 @@ L(end):
        addi    r1, r1, FRAMESIZE       /* Restore stack pointer.  */
        cfi_adjust_cfa_offset(-FRAMESIZE)
        ld      r0, 16(r1)      /* Restore the saved link register.  */
+       ld      r28, -32(r1)    /* Restore callers save register r28.  */
        ld      r29, -24(r1)    /* Restore callers save register r29.  */
        ld      r30, -16(r1)    /* Restore callers save register r30.  */
        ld      r31, -8(r1)     /* Restore callers save register r31.  */