diff options
| author | Alan Modra <amodra@gmail.com> | 2013-08-17 18:47:59 +0930 |
|---|---|---|
| committer | Alan Modra <amodra@gmail.com> | 2013-10-04 10:41:35 +0930 |
| commit | 3be87c77d24c4456ccca4034363b6d1814cd0c84 (patch) | |
| tree | c81589a21d74b94d5ae20235bc6d81715eca858c /sysdeps/powerpc/powerpc32 | |
| parent | 759cfef3ac4c07dba1ece0bbc1207e099348816d (diff) | |
| download | glibc-3be87c77d24c4456ccca4034363b6d1814cd0c84.tar.gz glibc-3be87c77d24c4456ccca4034363b6d1814cd0c84.tar.xz glibc-3be87c77d24c4456ccca4034363b6d1814cd0c84.zip | |
PowerPC LE memset
http://sourceware.org/ml/libc-alpha/2013-08/msg00104.html
One of the things I noticed when looking at power7 timing is that rlwimi
is cracked and the two resulting insns have a register dependency.
That makes it a little slower than the equivalent rldimi.
* sysdeps/powerpc/powerpc64/memset.S: Replace rlwimi with
insrdi. Formatting.
* sysdeps/powerpc/powerpc64/power4/memset.S: Likewise.
* sysdeps/powerpc/powerpc64/power6/memset.S: Likewise.
* sysdeps/powerpc/powerpc64/power7/memset.S: Likewise.
* sysdeps/powerpc/powerpc32/power4/memset.S: Likewise.
* sysdeps/powerpc/powerpc32/power6/memset.S: Likewise.
* sysdeps/powerpc/powerpc32/power7/memset.S: Likewise.
Diffstat (limited to 'sysdeps/powerpc/powerpc32')
| -rw-r--r-- | sysdeps/powerpc/powerpc32/power4/memset.S | 4 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc32/power6/memset.S | 4 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc32/power7/memset.S | 4 |
3 files changed, 6 insertions, 6 deletions
diff --git a/sysdeps/powerpc/powerpc32/power4/memset.S b/sysdeps/powerpc/powerpc32/power4/memset.S index c2d288b38b..4fd9d8cb4a 100644 --- a/sysdeps/powerpc/powerpc32/power4/memset.S +++ b/sysdeps/powerpc/powerpc32/power4/memset.S @@ -50,7 +50,7 @@ L(_memset): /* Align to word boundary. */ cmplwi cr5, rLEN, 31 - rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ + insrdi rCHR, rCHR, 8, 48 /* Replicate byte to halfword. */ beq+ L(aligned) mtcrf 0x01, rMEMP0 subfic rALIGN, rALIGN, 4 @@ -65,7 +65,7 @@ L(g0): /* Handle the case of size < 31. */ L(aligned): mtcrf 0x01, rLEN - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ ble cr5, L(medium) /* Align to 32-byte boundary. */ andi. rALIGN, rMEMP, 0x1C diff --git a/sysdeps/powerpc/powerpc32/power6/memset.S b/sysdeps/powerpc/powerpc32/power6/memset.S index 8c23c8d136..a4b002a960 100644 --- a/sysdeps/powerpc/powerpc32/power6/memset.S +++ b/sysdeps/powerpc/powerpc32/power6/memset.S @@ -48,7 +48,7 @@ L(_memset): ble- cr1, L(small) /* Align to word boundary. */ cmplwi cr5, rLEN, 31 - rlwimi rCHR, rCHR, 8, 16, 23 /* Replicate byte to halfword. */ + insrdi rCHR, rCHR, 8, 48 /* Replicate byte to halfword. */ beq+ L(aligned) mtcrf 0x01, rMEMP0 subfic rALIGN, rALIGN, 4 @@ -64,7 +64,7 @@ L(g0): /* Handle the case of size < 31. */ L(aligned): mtcrf 0x01, rLEN - rlwimi rCHR, rCHR, 16, 0, 15 /* Replicate halfword to word. */ + insrdi rCHR, rCHR, 16, 32 /* Replicate halfword to word. */ ble cr5, L(medium) /* Align to 32-byte boundary. */ andi. rALIGN, rMEMP, 0x1C diff --git a/sysdeps/powerpc/powerpc32/power7/memset.S b/sysdeps/powerpc/powerpc32/power7/memset.S index 360ea717f4..aadda2558f 100644 --- a/sysdeps/powerpc/powerpc32/power7/memset.S +++ b/sysdeps/powerpc/powerpc32/power7/memset.S @@ -35,8 +35,8 @@ L(_memset): cfi_offset(31,-8) /* Replicate byte to word. */ - rlwimi 4,4,8,16,23 - rlwimi 4,4,16,0,15 + insrdi 4,4,8,48 + insrdi 4,4,16,32 ble cr6,L(small) /* If length <= 8, use short copy code. */ |