diff options
| author | Paul E. Murphy <murphyp@linux.vnet.ibm.com> | 2016-04-25 09:11:02 -0500 |
|---|---|---|
| committer | Paul E. Murphy <murphyp@linux.vnet.ibm.com> | 2016-04-25 09:11:02 -0500 |
| commit | 8f1b841e452dbb083112fd036033b7f4af506ba0 (patch) | |
| tree | 96374cca2c14ef7e066c316cf90b97eff4db6ea3 /sysdeps/powerpc/powerpc64/power8/strspn.S | |
| parent | fdcf1c9480342d9f5fc2d23f142d621bcb4d00a4 (diff) | |
| download | glibc-8f1b841e452dbb083112fd036033b7f4af506ba0.tar.gz glibc-8f1b841e452dbb083112fd036033b7f4af506ba0.tar.xz glibc-8f1b841e452dbb083112fd036033b7f4af506ba0.zip | |
powerpc: Add optimized strcspn for P8
A few minor adjustments to the P8 strspn gives us an almost equally optimized P8 strcspn.
Diffstat (limited to 'sysdeps/powerpc/powerpc64/power8/strspn.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/power8/strspn.S | 45 |
1 files changed, 34 insertions, 11 deletions
diff --git a/sysdeps/powerpc/powerpc64/power8/strspn.S b/sysdeps/powerpc/powerpc64/power8/strspn.S index 0dda437a2e..011081dd34 100644 --- a/sysdeps/powerpc/powerpc64/power8/strspn.S +++ b/sysdeps/powerpc/powerpc64/power8/strspn.S @@ -33,6 +33,21 @@ #include "sysdep.h" +#ifndef USE_AS_STRCSPN +# define USE_AS_STRCSPN 0 +# ifndef STRSPN +# define STRSPN strspn +# endif +# define INITIAL_MASK 0 +# define UPDATE_MASK(RA, RS, RB) or RA, RS, RB +#else +# ifndef STRSPN +# define STRSPN strcspn +# endif +# define INITIAL_MASK -1 +# define UPDATE_MASK(RA, RS, RB) andc RA, RS, RB +#endif + /* Simple macro to use VSX instructions in overlapping VR's. */ #define XXVR(insn, vrt, vra, vrb) \ insn 32+vrt, 32+vra, 32+vrb @@ -53,7 +68,7 @@ /* This can be updated to power8 once the minimum version of binutils supports power8 and the above instructions. */ .machine power7 -EALIGN(strspn, 4, 0) +EALIGN(STRSPN, 4, 0) CALL_MCOUNT 2 /* Generate useful constants for later on. */ @@ -66,10 +81,18 @@ EALIGN(strspn, 4, 0) /* Prepare to compute 256b mask. */ addi r4, r4, -1 - li r5, 0 - li r6, 0 - li r7, 0 - li r8, 0 + li r5, INITIAL_MASK + li r6, INITIAL_MASK + li r7, INITIAL_MASK + li r8, INITIAL_MASK + +#if USE_AS_STRCSPN + /* Ensure the null character never matches by clearing ISA bit 0 in + in r5 which is the bit which will check for it in the later usage + of vbpermq. */ + srdi r5, r5, 1 +#endif + li r11, 1 sldi r11, r11, 63 @@ -97,14 +120,14 @@ L(next_needle): /* Now, or the value into the correct GPR. */ bge cr1,L(needle_gt128) - or r5, r5, r10 /* 0 - 63. */ - or r6, r6, r12 /* 64 - 127. */ + UPDATE_MASK (r5, r5, r10) /* 0 - 63. */ + UPDATE_MASK (r6, r6, r12) /* 64 - 127. */ b L(next_needle) .align 4 L(needle_gt128): - or r7, r7, r10 /* 128 - 191. */ - or r8, r8, r12 /* 192 - 255. */ + UPDATE_MASK (r7, r7, r10) /* 128 - 191. */ + UPDATE_MASK (r8, r8, r12) /* 192 - 255. */ b L(next_needle) @@ -175,5 +198,5 @@ L(done): add r3, r3, r10 blr -END(strspn) -libc_hidden_builtin_def (strspn) +END(STRSPN) +libc_hidden_builtin_def (STRSPN) |