This is the mail archive of the
libc-alpha@sources.redhat.com
mailing list for the glibc project.
powerpc32/fpu cleanup
- From: Alan Modra <amodra at bigpond dot net dot au>
- To: libc-alpha at sources dot redhat dot com
- Date: Tue, 17 May 2005 23:26:04 +0930
- Subject: powerpc32/fpu cleanup
While fiddling with glibc support for the new ppc32 non-exec GOT and
PLT, I found a number of things that needed cleaning up. This patch
deletes some unused constants, fixes four functions that emitted code to
.rodata.cst4 or .rodata.cst8, and does some formatting.
2005-05-17 Alan Modra <amodra@bigpond.net.au>
* sysdeps/powerpc/powerpc32/fpu/s_ceil.S (TWO52.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_ceilf.S (TWO23.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_floor.S (TWO52.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_floorf.S (TWO23.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_lround.S: Comment typo.
(NEGZERO.0, POINTFIVE.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_rint.S (TWO52.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_rintf.S (TWO23.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_round.S (__round): Code in .text.
(TWO52.0, POINTFIVE.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_roundf.S (__roundf): Code in .text.
(TWO23.0, POINTFIVE.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_trunc.S (__trunc): Code in .text.
(TWO52.0): Delete.
* sysdeps/powerpc/powerpc32/fpu/s_truncf.S (__truncf): Code in .text.
(TWO23.0): Delete.
* sysdeps/powerpc/powerpc32/memset.S (memset): Formatting.
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_ceil.S libc1/sysdeps/powerpc/powerpc32/fpu/s_ceil.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_ceil.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_ceil.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,14 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 3
- .type TWO52.0,@object
- .size TWO52.0,8
-TWO52.0:
- .long 0x43300000
- .long 0
-
.section .rodata.cst8,"aM",@progbits,8
.align 3
.LC0: /* 2**52 */
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S libc1/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_ceilf.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,13 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 2
- .type TWO23.0,@object
- .size TWO23.0,4
-TWO23.0:
- .long 0x4b000000
-
.section .rodata.cst4,"aM",@progbits,4
.align 2
.LC0: /* 2**23 */
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_floor.S libc1/sysdeps/powerpc/powerpc32/fpu/s_floor.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_floor.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_floor.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,14 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 3
- .type TWO52.0,@object
- .size TWO52.0,8
-TWO52.0:
- .long 0x43300000
- .long 0
-
.section .rodata.cst8,"aM",@progbits,8
.align 3
.LC0: /* 2**52 */
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_floorf.S libc1/sysdeps/powerpc/powerpc32/fpu/s_floorf.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_floorf.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_floorf.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,13 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 2
- .type TWO23.0,@object
- .size TWO23.0,4
-TWO23.0:
- .long 0x4b000000
-
.section .rodata.cst4,"aM",@progbits,4
.align 2
.LC0: /* 2**23 */
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_lround.S libc1/sysdeps/powerpc/powerpc32/fpu/s_lround.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_lround.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_lround.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,19 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 3
- .type NEGZERO.0,@object
- .size NEGZERO.0,8
-NEGZERO.0:
- .long 0x00000000
- .long 0
- .type POINTFIVE.0,@object
- .size POINTFIVE.0,8
-POINTFIVE.0:
- .long 0x3fe00000
- .long 0
-
.section .rodata.cst8,"aM",@progbits,8
.align 3
.LC0: /* 0.0 */
@@ -78,8 +65,8 @@ ENTRY (__lround)
.L9:
fctiwz fp2,fp1 /* Convert To Integer DW lround toward 0. */
stfd fp2,-8(r1)
- nop /* Insure the following load is in a different dispatch group */
- nop /* to avoid pipe stall on POWER4&5. */
+ nop /* Ensure the following load is in a different dispatch */
+ nop /* group to avoid pipe stall on POWER4&5. */
nop
lwz r3,-4(r1)
blr
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_rint.S libc1/sysdeps/powerpc/powerpc32/fpu/s_rint.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_rint.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_rint.S 2005-05-16 22:00:59.000000000 +0930
@@ -22,14 +22,6 @@
#include <sysdep.h>
- .section .rodata
- .align 3
- .type TWO52.0,@object
- .size TWO52.0,8
-TWO52.0:
- .long 0x43300000
- .long 0
-
.section .rodata.cst8,"aM",@progbits,8
.align 3
.LC0: /* 2**52 */
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_rintf.S libc1/sysdeps/powerpc/powerpc32/fpu/s_rintf.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_rintf.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_rintf.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,14 +19,6 @@
#include <sysdep.h>
-
- .section .rodata
- .align 2
- .type TWO23.0,@object
- .size TWO23.0,4
-TWO23.0:
- .long 0x4b000000
-
.section .rodata.cst4,"aM",@progbits,4
.align 2
.LC0: /* 2**23 */
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_round.S libc1/sysdeps/powerpc/powerpc32/fpu/s_round.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_round.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_round.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,19 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 3
- .type TWO52.0,@object
- .size TWO52.0,8
-TWO52.0:
- .long 0x43300000
- .long 0
- .type POINTFIVE.0,@object
- .size POINTFIVE.0,8
-POINTFIVE.0:
- .long 0x3fe00000
- .long 0
-
.section .rodata.cst8,"aM",@progbits,8
.align 3
.LC0: /* 2**52 */
@@ -51,6 +38,7 @@ POINTFIVE.0:
"Round toward Zero" mode and round by adding +-0.5 before rounding
to the integer value. */
+ .section ".text"
ENTRY (__round)
mffs fp11 /* Save current FPU rounding mode. */
#ifdef SHARED
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_roundf.S libc1/sysdeps/powerpc/powerpc32/fpu/s_roundf.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_roundf.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_roundf.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,17 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 2
- .type TWO23.0,@object
- .size TWO23.0,4
-TWO23.0:
- .long 0x4b000000
- .type POINTFIVE.0,@object
- .size POINTFIVE.0,4
-POINTFIVE.0:
- .long 0x3f000000
-
.section .rodata.cst4,"aM",@progbits,4
.align 2
.LC0: /* 2**23 */
@@ -47,6 +36,7 @@ POINTFIVE.0:
"Round toward Zero" mode and round by adding +-0.5 before rounding
to the integer value. */
+ .section ".text"
ENTRY (__roundf )
mffs fp11 /* Save current FPU rounding mode. */
#ifdef SHARED
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_trunc.S libc1/sysdeps/powerpc/powerpc32/fpu/s_trunc.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_trunc.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_trunc.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,14 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 3
- .type TWO52.0,@object
- .size TWO52.0,8
-TWO52.0:
- .long 0x43300000
- .long 0
-
.section .rodata.cst8,"aM",@progbits,8
.align 3
.LC0: /* 2**52 */
@@ -40,6 +32,7 @@ TWO52.0:
We set "round toward Zero" mode and trunc by adding +-2**52 then
subtracting +-2**52. */
+ .section ".text"
ENTRY (__trunc)
mffs fp11 /* Save current FPU rounding mode. */
#ifdef SHARED
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_truncf.S libc1/sysdeps/powerpc/powerpc32/fpu/s_truncf.S
--- libc-virgin/sysdeps/powerpc/powerpc32/fpu/s_truncf.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/fpu/s_truncf.S 2005-05-16 22:00:59.000000000 +0930
@@ -19,13 +19,6 @@
#include <sysdep.h>
- .section .rodata
- .align 2
- .type TWO23.0,@object
- .size TWO23.0,2
-TWO23.0:
- .long 0x4b000000
-
.section .rodata.cst4,"aM",@progbits,4
.align 2
.LC0: /* 2**23 */
@@ -38,6 +31,7 @@ TWO23.0:
We set "round toward Zero" mode and trunc by adding +-2**23 then
subtracting +-2**23. */
+ .section ".text"
ENTRY (__truncf)
mffs fp11 /* Save current FPU rounding mode. */
#ifdef SHARED
diff -urp -xCVS libc-virgin/sysdeps/powerpc/powerpc32/memset.S libc1/sysdeps/powerpc/powerpc32/memset.S
--- libc-virgin/sysdeps/powerpc/powerpc32/memset.S 2005-05-16 21:21:10.000000000 +0930
+++ libc1/sysdeps/powerpc/powerpc32/memset.S 2005-05-16 22:00:59.000000000 +0930
@@ -140,7 +140,7 @@ L(nondcbz):
/* We can't use dcbz here as we don't know the cache line size. We can
use "data cache block touch for store", which is safe. */
-L(c3): dcbtst rNEG64, rMEMP
+L(c3): dcbtst rNEG64, rMEMP
stw rCHR, -4(rMEMP)
stw rCHR, -8(rMEMP)
stw rCHR, -12(rMEMP)
@@ -166,7 +166,7 @@ L(cloopdone):
add rMEMP, rMEMP, rALIGN
b L(medium_tail2) /* 72nd instruction from .align */
- .align 5
+ .align 5
nop
/* Clear cache lines of memory in 128-byte chunks.
This code is optimized for processors with 32-byte cache lines.
@@ -200,7 +200,7 @@ L(zloop):
beqlr cr5
b L(medium_tail2)
- .align 5
+ .align 5
L(small):
/* Memset of 4 bytes or less. */
cmplwi cr5, rLEN, 1
@@ -218,7 +218,7 @@ L(small):
blr
/* Memset of 0-31 bytes. */
- .align 5
+ .align 5
L(medium):
cmplwi cr1, rLEN, 16
L(medium_tail2):
@@ -258,70 +258,70 @@ L(medium_28t):
L(checklinesize):
#ifdef SHARED
- mflr rTMP
+ mflr rTMP
/* If the remaining length is less the 32 bytes then don't bother getting
- the cache line size. */
+ the cache line size. */
beq L(medium)
/* Establishes GOT addressability so we can load __cache_line_size
from static. This value was set from the aux vector during startup. */
- bl _GLOBAL_OFFSET_TABLE_@local-4
- mflr rGOT
- lwz rGOT,__cache_line_size@got(rGOT)
- lwz rCLS,0(rGOT)
- mtlr rTMP
+ bl _GLOBAL_OFFSET_TABLE_@local-4
+ mflr rGOT
+ lwz rGOT,__cache_line_size@got(rGOT)
+ lwz rCLS,0(rGOT)
+ mtlr rTMP
#else
/* Load __cache_line_size from static. This value was set from the
aux vector during startup. */
- lis rCLS,__cache_line_size@ha
+ lis rCLS,__cache_line_size@ha
/* If the remaining length is less the 32 bytes then don't bother getting
- the cache line size. */
+ the cache line size. */
beq L(medium)
- lwz rCLS,__cache_line_size@l(rCLS)
+ lwz rCLS,__cache_line_size@l(rCLS)
#endif
-/*If the cache line size was not set then goto to L(nondcbz), which is
- safe for any cache line size. */
- cmplwi cr1,rCLS,0
+/* If the cache line size was not set then goto to L(nondcbz), which is
+ safe for any cache line size. */
+ cmplwi cr1,rCLS,0
beq cr1,L(nondcbz)
/* If the cache line size is 32 bytes then goto to L(zloopstart),
- which is coded specificly for 32-byte lines (and 601). */
- cmplwi cr1,rCLS,32
+ which is coded specificly for 32-byte lines (and 601). */
+ cmplwi cr1,rCLS,32
beq cr1,L(zloopstart)
/* Now we know the cache line size and it is not 32-bytes. However
- we may not yet be aligned to the cache line and may have a partial
- line to fill. Touch it 1st to fetch the cache line. */
- dcbtst 0,rMEMP
+ we may not yet be aligned to the cache line and may have a partial
+ line to fill. Touch it 1st to fetch the cache line. */
+ dcbtst 0,rMEMP
- addi rCLM,rCLS,-1
+ addi rCLM,rCLS,-1
L(getCacheAligned):
- cmplwi cr1,rLEN,32
- and. rTMP,rCLM,rMEMP
- blt cr1,L(handletail32)
- beq L(cacheAligned)
+ cmplwi cr1,rLEN,32
+ and. rTMP,rCLM,rMEMP
+ blt cr1,L(handletail32)
+ beq L(cacheAligned)
/* We are not aligned to start of a cache line yet. Store 32-byte
of data and test again. */
- addi rMEMP,rMEMP,32
- addi rLEN,rLEN,-32
- stw rCHR,-32(rMEMP)
- stw rCHR,-28(rMEMP)
- stw rCHR,-24(rMEMP)
- stw rCHR,-20(rMEMP)
- stw rCHR,-16(rMEMP)
- stw rCHR,-12(rMEMP)
- stw rCHR,-8(rMEMP)
- stw rCHR,-4(rMEMP)
- b L(getCacheAligned)
+ addi rMEMP,rMEMP,32
+ addi rLEN,rLEN,-32
+ stw rCHR,-32(rMEMP)
+ stw rCHR,-28(rMEMP)
+ stw rCHR,-24(rMEMP)
+ stw rCHR,-20(rMEMP)
+ stw rCHR,-16(rMEMP)
+ stw rCHR,-12(rMEMP)
+ stw rCHR,-8(rMEMP)
+ stw rCHR,-4(rMEMP)
+ b L(getCacheAligned)
/* Now we are aligned to the cache line and can use dcbz. */
L(cacheAligned):
- cmplw cr1,rLEN,rCLS
- blt cr1,L(handletail32)
- dcbz 0,rMEMP
- subf rLEN,rCLS,rLEN
- add rMEMP,rMEMP,rCLS
- b L(cacheAligned)
+ cmplw cr1,rLEN,rCLS
+ blt cr1,L(handletail32)
+ dcbz 0,rMEMP
+ subf rLEN,rCLS,rLEN
+ add rMEMP,rMEMP,rCLS
+ b L(cacheAligned)
/* We are here because; the cache line size was set, it was not
32-bytes, and the remainder (rLEN) is now less than the actual cache
@@ -329,7 +329,7 @@ L(cacheAligned):
store the remaining bytes. */
L(handletail32):
clrrwi. rALIGN, rLEN, 5
- b L(nondcbz)
+ b L(nondcbz)
END (BP_SYM (memset))
libc_hidden_builtin_def (memset)
--
Alan Modra
IBM OzLabs - Linux Technology Centre