diff options
Diffstat (limited to 'sysdeps/alpha/strncmp.S')
| -rw-r--r-- | sysdeps/alpha/strncmp.S | 33 |
1 files changed, 28 insertions, 5 deletions
diff --git a/sysdeps/alpha/strncmp.S b/sysdeps/alpha/strncmp.S index e2b4ebf857..ff199eb743 100644 --- a/sysdeps/alpha/strncmp.S +++ b/sysdeps/alpha/strncmp.S @@ -61,8 +61,10 @@ $aligned: ornot t0, t3, t0 # .. e1 : cmpbge zero, t1, t7 # e0 : bits set iff null found beq a2, $eoc # .. e1 : check end of count - subq a2, 1, a2 # e0 : + unop # e0 : bne t7, $eos # .. e1 : + unop # e0 : + beq t10, $ant_loop # .. e1 : /* Aligned compare main loop. On entry to this basic block: @@ -74,13 +76,30 @@ $a_loop: bne t2, $wordcmp # .. e1 (zdb) ldq_u t1, 8(a1) # e0 : ldq_u t0, 8(a0) # .. e1 : + subq a2, 1, a2 # e0 : + addq a1, 8, a1 # .. e1 : + addq a0, 8, a0 # e0 : + beq a2, $eoc # .. e1 : + cmpbge zero, t1, t7 # e0 : + beq t7, $a_loop # .. e1 : + unop # e0 : + br $eos # .. e1 : + + /* Alternate aligned compare loop, for when there's no trailing + bytes on the count. We have to avoid reading too much data. */ +$ant_loop: + xor t0, t1, t2 # e0 : + bne t2, $wordcmp # .. e1 (zdb) + subq a2, 1, a2 # e0 : + beq a2, $zerolength # .. e1 : + ldq_u t1, 8(a1) # e0 : + ldq_u t0, 8(a0) # .. e1 : addq a1, 8, a1 # e0 : addq a0, 8, a0 # .. e1 : cmpbge zero, t1, t7 # e0 : - beq a2, $eoc # .. e1 : - subq a2, 1, a2 # e0 : - beq t7, $a_loop # .. e1 : - br $eos # e1 : + beq t7, $ant_loop # .. e1 : + unop # e0 : + br $eos # .. e1 : /* The two strings are not co-aligned. Align s1 and cope. */ $unaligned: @@ -184,6 +203,8 @@ $u_final: $eoc: mskql t0, t10, t0 mskql t1, t10, t1 + unop + cmpbge zero, t1, t7 /* We've found a zero somewhere in a word we just read. On entry to this basic block: @@ -203,6 +224,7 @@ $eos: /* Here we have two differing co-aligned words in t0 & t1. Bytewise compare them and return (t0 > t1 ? 1 : -1). */ + .align 3 $wordcmp: cmpbge t0, t1, t2 # e0 : comparison yields bit mask of ge cmpbge t1, t0, t3 # .. e1 : @@ -216,6 +238,7 @@ $wordcmp: $done: ret # e1 : + .align 3 $zerolength: clr v0 ret |