The routines expect to be able to bias the count by a small number.
If the count is near -1ull, the count will overflow. Since we cannot
use the whole 64-bit address space, bound the count to LONG_MAX.
2012-06-06 Richard Henderson <rth@twiddle.net>
+ [BZ #13718]
+ * sysdeps/alpha/stxncmp.S: Bound count to LONG_MAX at startup.
+ * sysdeps/alpha/alphaev6/stxncmp.S: Likewise.
+
* sysdeps/alpha/fpu/e_sqrt.c: Include <math_private.h> before
redefining __ieee758_sqrt.
.align 4
__stxncpy:
/* Are source and destination co-aligned? */
+ lda t2, -1 # E :
xor a0, a1, t1 # E :
and a0, 7, t0 # E : find dest misalignment
- and t1, 7, t1 # E : (stall)
- addq a2, t0, a2 # E : bias count by dest misalignment (stall)
+ nop # E :
- subq a2, 1, a2 # E :
+ srl t2, 1, t2 # U :
+ and t1, 7, t1 # E :
+ cmovlt a2, t2, a2 # E : bound count to LONG_MAX (stall)
+ nop # E :
+
+ addq a2, t0, a2 # E : bias count by dest misalignment
+ subq a2, 1, a2 # E : (stall)
and a2, 7, t2 # E : (stall)
- srl a2, 3, a2 # U : a2 = loop counter = (count - 1)/8 (stall)
- addq zero, 1, t10 # E :
+ lda t10, 1 # E :
+ srl a2, 3, a2 # U : a2 = loop counter = (count - 1)/8
sll t10, t2, t10 # U : t10 = bitmask of last count byte
- bne t1, $unaligned # U :
+ nop # E :
+ bne t1, $unaligned # U : (stall)
/* We are co-aligned; take care of a partial first word. */
ldq_u t1, 0(a1) # L : load first src word
.align 3
__stxncpy:
/* Are source and destination co-aligned? */
- xor a0, a1, t1 # e0 :
- and a0, 7, t0 # .. e1 : find dest misalignment
- and t1, 7, t1 # e0 :
- addq a2, t0, a2 # .. e1 : bias count by dest misalignment
- subq a2, 1, a2 # e0 :
- and a2, 7, t2 # e1 :
- srl a2, 3, a2 # e0 : a2 = loop counter = (count - 1)/8
- addq zero, 1, t10 # .. e1 :
- sll t10, t2, t10 # e0 : t10 = bitmask of last count byte
- bne t1, $unaligned # .. e1 :
+ lda t2, -1
+ xor a0, a1, t1
+ srl t2, 1, t2
+ and a0, 7, t0 # find dest misalignment
+ cmovlt a2, t2, a2 # bound neg count to LONG_MAX
+ and t1, 7, t1
+ addq a2, t0, a2 # bias count by dest misalignment
+ subq a2, 1, a2
+ and a2, 7, t2
+ srl a2, 3, a2 # a2 = loop counter = (count - 1)/8
+ addq zero, 1, t10
+ sll t10, t2, t10 # t10 = bitmask of last count byte
+ bne t1, $unaligned
/* We are co-aligned; take care of a partial first word. */