diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64/power4/memcpy.S')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/power4/memcpy.S | 18 |
1 files changed, 9 insertions, 9 deletions
diff --git a/sysdeps/powerpc/powerpc64/power4/memcpy.S b/sysdeps/powerpc/powerpc64/power4/memcpy.S index 9d0b478029..734434af07 100644 --- a/sysdeps/powerpc/powerpc64/power4/memcpy.S +++ b/sysdeps/powerpc/powerpc64/power4/memcpy.S @@ -28,11 +28,11 @@ with the appropriate combination of byte and halfword load/stores. There is minimal effort to optimize the alignment of short moves. The 64-bit implementations of POWER3 and POWER4 do a reasonable job - of handling unligned load/stores that do not cross 32-byte boundries. + of handling unaligned load/stores that do not cross 32-byte boundaries. Longer moves (>= 32-bytes) justify the effort to get at least the destination doubleword (8-byte) aligned. Further optimization is - posible when both source and destination are doubleword aligned. + possible when both source and destination are doubleword aligned. Each case has a optimized unrolled loop. */ .machine power4 @@ -44,9 +44,9 @@ EALIGN (BP_SYM (memcpy), 5, 0) std 3,-16(1) std 31,-8(1) cfi_offset(31,-8) - andi. 11,3,7 /* check alignement of dst. */ + andi. 11,3,7 /* check alignment of dst. */ clrldi 0,0,61 /* Number of bytes until the 1st doubleword of dst. */ - clrldi 10,4,61 /* check alignement of src. */ + clrldi 10,4,61 /* check alignment of src. */ cmpldi cr6,5,8 ble- cr1,.L2 /* If move < 32 bytes use short move code. */ cmpld cr6,10,11 @@ -57,7 +57,7 @@ EALIGN (BP_SYM (memcpy), 5, 0) beq .L0 subf 31,0,5 - /* Move 0-7 bytes as needed to get the destination doubleword alligned. */ + /* Move 0-7 bytes as needed to get the destination doubleword aligned. */ 1: bf 31,2f lbz 6,0(12) addi 12,12,1 @@ -74,10 +74,10 @@ EALIGN (BP_SYM (memcpy), 5, 0) stw 6,0(3) addi 3,3,4 0: - clrldi 10,12,61 /* check alignement of src again. */ + clrldi 10,12,61 /* check alignment of src again. */ srdi 9,31,3 /* Number of full double words remaining. */ - /* Copy doublewords from source to destination, assumpting the + /* Copy doublewords from source to destination, assuming the destination is aligned on a doubleword boundary. At this point we know there are at least 25 bytes left (32-7) to copy. @@ -154,7 +154,7 @@ EALIGN (BP_SYM (memcpy), 5, 0) add 12,12,0 /* At this point we have a tail of 0-7 bytes and we know that the - destiniation is double word aligned. */ + destination is double word aligned. */ 4: bf 29,2f lwz 6,0(12) addi 12,12,4 @@ -284,7 +284,7 @@ EALIGN (BP_SYM (memcpy), 5, 0) bne cr6,4f /* Would have liked to use use ld/std here but the 630 processors are slow for load/store doubles that are not at least word aligned. - Unaligned Load/Store word execute with only a 1 cycle penaltity. */ + Unaligned Load/Store word execute with only a 1 cycle penalty. */ lwz 6,0(4) lwz 7,4(4) stw 6,0(3) |