diff options
Diffstat (limited to '')
-rw-r--r-- | arch/sh/lib/Makefile | 16 | ||||
-rw-r--r-- | arch/sh/lib/ashiftrt.S | 149 | ||||
-rw-r--r-- | arch/sh/lib/ashldi3.c | 29 | ||||
-rw-r--r-- | arch/sh/lib/ashlsi3.S | 193 | ||||
-rw-r--r-- | arch/sh/lib/ashrdi3.c | 31 | ||||
-rw-r--r-- | arch/sh/lib/ashrsi3.S | 185 | ||||
-rw-r--r-- | arch/sh/lib/libgcc.h | 26 | ||||
-rw-r--r-- | arch/sh/lib/lshrdi3.c | 29 | ||||
-rw-r--r-- | arch/sh/lib/lshrsi3.S | 193 | ||||
-rw-r--r-- | arch/sh/lib/mcount.S | 90 | ||||
-rw-r--r-- | arch/sh/lib/movmem.S | 238 | ||||
-rw-r--r-- | arch/sh/lib/udiv_qrnnd.S | 81 | ||||
-rw-r--r-- | arch/sh/lib/udivsi3.S | 87 | ||||
-rw-r--r-- | arch/sh/lib/udivsi3_i4i-Os.S | 149 | ||||
-rw-r--r-- | arch/sh/lib/udivsi3_i4i.S | 666 | ||||
-rw-r--r-- | arch/sh/lib64/Makefile | 8 | ||||
-rw-r--r-- | arch/sh/lib64/c-checksum.c | 4 | ||||
-rw-r--r-- | arch/sh/lib64/memcpy.S | 201 | ||||
-rw-r--r-- | arch/sh/lib64/memcpy.c | 81 | ||||
-rw-r--r-- | arch/sh/lib64/memset.S | 91 | ||||
-rw-r--r-- | arch/sh/lib64/sdivsi3.S | 131 | ||||
-rw-r--r-- | arch/sh/lib64/strcpy.S | 97 | ||||
-rw-r--r-- | arch/sh/lib64/strlen.S | 33 | ||||
-rw-r--r-- | arch/sh/lib64/udivdi3.S | 120 | ||||
-rw-r--r-- | arch/sh/lib64/udivsi3.S | 59 |
25 files changed, 2900 insertions, 87 deletions
diff --git a/arch/sh/lib/Makefile b/arch/sh/lib/Makefile index 8596cc78e18d..aaea580b65bb 100644 --- a/arch/sh/lib/Makefile +++ b/arch/sh/lib/Makefile @@ -5,12 +5,26 @@ lib-y = delay.o memset.o memmove.o memchr.o \ checksum.o strlen.o div64.o div64-generic.o +# Extracted from libgcc +lib-y += movmem.o ashldi3.o ashrdi3.o lshrdi3.o \ + ashlsi3.o ashrsi3.o ashiftrt.o lshrsi3.o \ + udiv_qrnnd.o + +udivsi3-y := udivsi3_i4i-Os.o + +ifneq ($(CONFIG_CC_OPTIMIZE_FOR_SIZE),y) +udivsi3-$(CONFIG_CPU_SH3) := udivsi3_i4i.o +udivsi3-$(CONFIG_CPU_SH4) := udivsi3_i4i.o +endif +udivsi3-y += udivsi3.o + obj-y += io.o memcpy-y := memcpy.o memcpy-$(CONFIG_CPU_SH4) := memcpy-sh4.o lib-$(CONFIG_MMU) += copy_page.o clear_page.o -lib-y += $(memcpy-y) +lib-$(CONFIG_FUNCTION_TRACER) += mcount.o +lib-y += $(memcpy-y) $(udivsi3-y) EXTRA_CFLAGS += -Werror diff --git a/arch/sh/lib/ashiftrt.S b/arch/sh/lib/ashiftrt.S new file mode 100644 index 000000000000..45ce86558f46 --- /dev/null +++ b/arch/sh/lib/ashiftrt.S @@ -0,0 +1,149 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + + .global __ashiftrt_r4_0 + .global __ashiftrt_r4_1 + .global __ashiftrt_r4_2 + .global __ashiftrt_r4_3 + .global __ashiftrt_r4_4 + .global __ashiftrt_r4_5 + .global __ashiftrt_r4_6 + .global __ashiftrt_r4_7 + .global __ashiftrt_r4_8 + .global __ashiftrt_r4_9 + .global __ashiftrt_r4_10 + .global __ashiftrt_r4_11 + .global __ashiftrt_r4_12 + .global __ashiftrt_r4_13 + .global __ashiftrt_r4_14 + .global __ashiftrt_r4_15 + .global __ashiftrt_r4_16 + .global __ashiftrt_r4_17 + .global __ashiftrt_r4_18 + .global __ashiftrt_r4_19 + .global __ashiftrt_r4_20 + .global __ashiftrt_r4_21 + .global __ashiftrt_r4_22 + .global __ashiftrt_r4_23 + .global __ashiftrt_r4_24 + .global __ashiftrt_r4_25 + .global __ashiftrt_r4_26 + .global __ashiftrt_r4_27 + .global __ashiftrt_r4_28 + .global __ashiftrt_r4_29 + .global __ashiftrt_r4_30 + .global __ashiftrt_r4_31 + .global __ashiftrt_r4_32 + + .align 1 +__ashiftrt_r4_32: +__ashiftrt_r4_31: + rotcl r4 + rts + subc r4,r4 +__ashiftrt_r4_30: + shar r4 +__ashiftrt_r4_29: + shar r4 +__ashiftrt_r4_28: + shar r4 +__ashiftrt_r4_27: + shar r4 +__ashiftrt_r4_26: + shar r4 +__ashiftrt_r4_25: + shar r4 +__ashiftrt_r4_24: + shlr16 r4 + shlr8 r4 + rts + exts.b r4,r4 +__ashiftrt_r4_23: + shar r4 +__ashiftrt_r4_22: + shar r4 +__ashiftrt_r4_21: + shar r4 +__ashiftrt_r4_20: + shar r4 +__ashiftrt_r4_19: + shar r4 +__ashiftrt_r4_18: + shar r4 +__ashiftrt_r4_17: + shar r4 +__ashiftrt_r4_16: + shlr16 r4 + rts + exts.w r4,r4 +__ashiftrt_r4_15: + shar r4 +__ashiftrt_r4_14: + shar r4 +__ashiftrt_r4_13: + shar r4 +__ashiftrt_r4_12: + shar r4 +__ashiftrt_r4_11: + shar r4 +__ashiftrt_r4_10: + shar r4 +__ashiftrt_r4_9: + shar r4 +__ashiftrt_r4_8: + shar r4 +__ashiftrt_r4_7: + shar r4 +__ashiftrt_r4_6: + shar r4 +__ashiftrt_r4_5: + shar r4 +__ashiftrt_r4_4: + shar r4 +__ashiftrt_r4_3: + shar r4 +__ashiftrt_r4_2: + shar r4 +__ashiftrt_r4_1: + rts + shar r4 +__ashiftrt_r4_0: + rts + nop diff --git a/arch/sh/lib/ashldi3.c b/arch/sh/lib/ashldi3.c new file mode 100644 index 000000000000..beb80f316095 --- /dev/null +++ b/arch/sh/lib/ashldi3.c @@ -0,0 +1,29 @@ +#include <linux/module.h> + +#include "libgcc.h" + +long long __ashldi3(long long u, word_type b) +{ + DWunion uu, w; + word_type bm; + + if (b == 0) + return u; + + uu.ll = u; + bm = 32 - b; + + if (bm <= 0) { + w.s.low = 0; + w.s.high = (unsigned int) uu.s.low << -bm; + } else { + const unsigned int carries = (unsigned int) uu.s.low >> bm; + + w.s.low = (unsigned int) uu.s.low << b; + w.s.high = ((unsigned int) uu.s.high << b) | carries; + } + + return w.ll; +} + +EXPORT_SYMBOL(__ashldi3); diff --git a/arch/sh/lib/ashlsi3.S b/arch/sh/lib/ashlsi3.S new file mode 100644 index 000000000000..bd47e9b403a5 --- /dev/null +++ b/arch/sh/lib/ashlsi3.S @@ -0,0 +1,193 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + +! +! __ashlsi3 +! +! Entry: +! +! r4: Value to shift +! r5: Shifts +! +! Exit: +! +! r0: Result +! +! Destroys: +! +! (none) +! + .global __ashlsi3 + + .align 2 +__ashlsi3: + mov #31,r0 + and r0,r5 + mova ashlsi3_table,r0 + mov.b @(r0,r5),r5 +#ifdef __sh1__ + add r5,r0 + jmp @r0 +#else + braf r5 +#endif + mov r4,r0 + + .align 2 +ashlsi3_table: + .byte ashlsi3_0-ashlsi3_table + .byte ashlsi3_1-ashlsi3_table + .byte ashlsi3_2-ashlsi3_table + .byte ashlsi3_3-ashlsi3_table + .byte ashlsi3_4-ashlsi3_table + .byte ashlsi3_5-ashlsi3_table + .byte ashlsi3_6-ashlsi3_table + .byte ashlsi3_7-ashlsi3_table + .byte ashlsi3_8-ashlsi3_table + .byte ashlsi3_9-ashlsi3_table + .byte ashlsi3_10-ashlsi3_table + .byte ashlsi3_11-ashlsi3_table + .byte ashlsi3_12-ashlsi3_table + .byte ashlsi3_13-ashlsi3_table + .byte ashlsi3_14-ashlsi3_table + .byte ashlsi3_15-ashlsi3_table + .byte ashlsi3_16-ashlsi3_table + .byte ashlsi3_17-ashlsi3_table + .byte ashlsi3_18-ashlsi3_table + .byte ashlsi3_19-ashlsi3_table + .byte ashlsi3_20-ashlsi3_table + .byte ashlsi3_21-ashlsi3_table + .byte ashlsi3_22-ashlsi3_table + .byte ashlsi3_23-ashlsi3_table + .byte ashlsi3_24-ashlsi3_table + .byte ashlsi3_25-ashlsi3_table + .byte ashlsi3_26-ashlsi3_table + .byte ashlsi3_27-ashlsi3_table + .byte ashlsi3_28-ashlsi3_table + .byte ashlsi3_29-ashlsi3_table + .byte ashlsi3_30-ashlsi3_table + .byte ashlsi3_31-ashlsi3_table + +ashlsi3_6: + shll2 r0 +ashlsi3_4: + shll2 r0 +ashlsi3_2: + rts + shll2 r0 + +ashlsi3_7: + shll2 r0 +ashlsi3_5: + shll2 r0 +ashlsi3_3: + shll2 r0 +ashlsi3_1: + rts + shll r0 + +ashlsi3_14: + shll2 r0 +ashlsi3_12: + shll2 r0 +ashlsi3_10: + shll2 r0 +ashlsi3_8: + rts + shll8 r0 + +ashlsi3_15: + shll2 r0 +ashlsi3_13: + shll2 r0 +ashlsi3_11: + shll2 r0 +ashlsi3_9: + shll8 r0 + rts + shll r0 + +ashlsi3_22: + shll2 r0 +ashlsi3_20: + shll2 r0 +ashlsi3_18: + shll2 r0 +ashlsi3_16: + rts + shll16 r0 + +ashlsi3_23: + shll2 r0 +ashlsi3_21: + shll2 r0 +ashlsi3_19: + shll2 r0 +ashlsi3_17: + shll16 r0 + rts + shll r0 + +ashlsi3_30: + shll2 r0 +ashlsi3_28: + shll2 r0 +ashlsi3_26: + shll2 r0 +ashlsi3_24: + shll16 r0 + rts + shll8 r0 + +ashlsi3_31: + shll2 r0 +ashlsi3_29: + shll2 r0 +ashlsi3_27: + shll2 r0 +ashlsi3_25: + shll16 r0 + shll8 r0 + rts + shll r0 + +ashlsi3_0: + rts + nop diff --git a/arch/sh/lib/ashrdi3.c b/arch/sh/lib/ashrdi3.c new file mode 100644 index 000000000000..c884a912b660 --- /dev/null +++ b/arch/sh/lib/ashrdi3.c @@ -0,0 +1,31 @@ +#include <linux/module.h> + +#include "libgcc.h" + +long long __ashrdi3(long long u, word_type b) +{ + DWunion uu, w; + word_type bm; + + if (b == 0) + return u; + + uu.ll = u; + bm = 32 - b; + + if (bm <= 0) { + /* w.s.high = 1..1 or 0..0 */ + w.s.high = + uu.s.high >> 31; + w.s.low = uu.s.high >> -bm; + } else { + const unsigned int carries = (unsigned int) uu.s.high << bm; + + w.s.high = uu.s.high >> b; + w.s.low = ((unsigned int) uu.s.low >> b) | carries; + } + + return w.ll; +} + +EXPORT_SYMBOL(__ashrdi3); diff --git a/arch/sh/lib/ashrsi3.S b/arch/sh/lib/ashrsi3.S new file mode 100644 index 000000000000..6f3cf46b77c2 --- /dev/null +++ b/arch/sh/lib/ashrsi3.S @@ -0,0 +1,185 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + +! +! __ashrsi3 +! +! Entry: +! +! r4: Value to shift +! r5: Shifts +! +! Exit: +! +! r0: Result +! +! Destroys: +! +! (none) +! + + .global __ashrsi3 + + .align 2 +__ashrsi3: + mov #31,r0 + and r0,r5 + mova ashrsi3_table,r0 + mov.b @(r0,r5),r5 +#ifdef __sh1__ + add r5,r0 + jmp @r0 +#else + braf r5 +#endif + mov r4,r0 + + .align 2 +ashrsi3_table: + .byte ashrsi3_0-ashrsi3_table + .byte ashrsi3_1-ashrsi3_table + .byte ashrsi3_2-ashrsi3_table + .byte ashrsi3_3-ashrsi3_table + .byte ashrsi3_4-ashrsi3_table + .byte ashrsi3_5-ashrsi3_table + .byte ashrsi3_6-ashrsi3_table + .byte ashrsi3_7-ashrsi3_table + .byte ashrsi3_8-ashrsi3_table + .byte ashrsi3_9-ashrsi3_table + .byte ashrsi3_10-ashrsi3_table + .byte ashrsi3_11-ashrsi3_table + .byte ashrsi3_12-ashrsi3_table + .byte ashrsi3_13-ashrsi3_table + .byte ashrsi3_14-ashrsi3_table + .byte ashrsi3_15-ashrsi3_table + .byte ashrsi3_16-ashrsi3_table + .byte ashrsi3_17-ashrsi3_table + .byte ashrsi3_18-ashrsi3_table + .byte ashrsi3_19-ashrsi3_table + .byte ashrsi3_20-ashrsi3_table + .byte ashrsi3_21-ashrsi3_table + .byte ashrsi3_22-ashrsi3_table + .byte ashrsi3_23-ashrsi3_table + .byte ashrsi3_24-ashrsi3_table + .byte ashrsi3_25-ashrsi3_table + .byte ashrsi3_26-ashrsi3_table + .byte ashrsi3_27-ashrsi3_table + .byte ashrsi3_28-ashrsi3_table + .byte ashrsi3_29-ashrsi3_table + .byte ashrsi3_30-ashrsi3_table + .byte ashrsi3_31-ashrsi3_table + +ashrsi3_31: + rotcl r0 + rts + subc r0,r0 + +ashrsi3_30: + shar r0 +ashrsi3_29: + shar r0 +ashrsi3_28: + shar r0 +ashrsi3_27: + shar r0 +ashrsi3_26: + shar r0 +ashrsi3_25: + shar r0 +ashrsi3_24: + shlr16 r0 + shlr8 r0 + rts + exts.b r0,r0 + +ashrsi3_23: + shar r0 +ashrsi3_22: + shar r0 +ashrsi3_21: + shar r0 +ashrsi3_20: + shar r0 +ashrsi3_19: + shar r0 +ashrsi3_18: + shar r0 +ashrsi3_17: + shar r0 +ashrsi3_16: + shlr16 r0 + rts + exts.w r0,r0 + +ashrsi3_15: + shar r0 +ashrsi3_14: + shar r0 +ashrsi3_13: + shar r0 +ashrsi3_12: + shar r0 +ashrsi3_11: + shar r0 +ashrsi3_10: + shar r0 +ashrsi3_9: + shar r0 +ashrsi3_8: + shar r0 +ashrsi3_7: + shar r0 +ashrsi3_6: + shar r0 +ashrsi3_5: + shar r0 +ashrsi3_4: + shar r0 +ashrsi3_3: + shar r0 +ashrsi3_2: + shar r0 +ashrsi3_1: + rts + shar r0 + +ashrsi3_0: + rts + nop diff --git a/arch/sh/lib/libgcc.h b/arch/sh/lib/libgcc.h new file mode 100644 index 000000000000..3f19d1c5d942 --- /dev/null +++ b/arch/sh/lib/libgcc.h @@ -0,0 +1,26 @@ +#ifndef __ASM_LIBGCC_H +#define __ASM_LIBGCC_H + +#include <asm/byteorder.h> + +typedef int word_type __attribute__ ((mode (__word__))); + +#ifdef __BIG_ENDIAN +struct DWstruct { + int high, low; +}; +#elif defined(__LITTLE_ENDIAN) +struct DWstruct { + int low, high; +}; +#else +#error I feel sick. +#endif + +typedef union +{ + struct DWstruct s; + long long ll; +} DWunion; + +#endif /* __ASM_LIBGCC_H */ diff --git a/arch/sh/lib/lshrdi3.c b/arch/sh/lib/lshrdi3.c new file mode 100644 index 000000000000..dcf8d6810b7c --- /dev/null +++ b/arch/sh/lib/lshrdi3.c @@ -0,0 +1,29 @@ +#include <linux/module.h> + +#include "libgcc.h" + +long long __lshrdi3(long long u, word_type b) +{ + DWunion uu, w; + word_type bm; + + if (b == 0) + return u; + + uu.ll = u; + bm = 32 - b; + + if (bm <= 0) { + w.s.high = 0; + w.s.low = (unsigned int) uu.s.high >> -bm; + } else { + const unsigned int carries = (unsigned int) uu.s.high << bm; + + w.s.high = (unsigned int) uu.s.high >> b; + w.s.low = ((unsigned int) uu.s.low >> b) | carries; + } + + return w.ll; +} + +EXPORT_SYMBOL(__lshrdi3); diff --git a/arch/sh/lib/lshrsi3.S b/arch/sh/lib/lshrsi3.S new file mode 100644 index 000000000000..1e7aaa557130 --- /dev/null +++ b/arch/sh/lib/lshrsi3.S @@ -0,0 +1,193 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + +! +! __lshrsi3 +! +! Entry: +! +! r4: Value to shift +! r5: Shifts +! +! Exit: +! +! r0: Result +! +! Destroys: +! +! (none) +! + .global __lshrsi3 + + .align 2 +__lshrsi3: + mov #31,r0 + and r0,r5 + mova lshrsi3_table,r0 + mov.b @(r0,r5),r5 +#ifdef __sh1__ + add r5,r0 + jmp @r0 +#else + braf r5 +#endif + mov r4,r0 + + .align 2 +lshrsi3_table: + .byte lshrsi3_0-lshrsi3_table + .byte lshrsi3_1-lshrsi3_table + .byte lshrsi3_2-lshrsi3_table + .byte lshrsi3_3-lshrsi3_table + .byte lshrsi3_4-lshrsi3_table + .byte lshrsi3_5-lshrsi3_table + .byte lshrsi3_6-lshrsi3_table + .byte lshrsi3_7-lshrsi3_table + .byte lshrsi3_8-lshrsi3_table + .byte lshrsi3_9-lshrsi3_table + .byte lshrsi3_10-lshrsi3_table + .byte lshrsi3_11-lshrsi3_table + .byte lshrsi3_12-lshrsi3_table + .byte lshrsi3_13-lshrsi3_table + .byte lshrsi3_14-lshrsi3_table + .byte lshrsi3_15-lshrsi3_table + .byte lshrsi3_16-lshrsi3_table + .byte lshrsi3_17-lshrsi3_table + .byte lshrsi3_18-lshrsi3_table + .byte lshrsi3_19-lshrsi3_table + .byte lshrsi3_20-lshrsi3_table + .byte lshrsi3_21-lshrsi3_table + .byte lshrsi3_22-lshrsi3_table + .byte lshrsi3_23-lshrsi3_table + .byte lshrsi3_24-lshrsi3_table + .byte lshrsi3_25-lshrsi3_table + .byte lshrsi3_26-lshrsi3_table + .byte lshrsi3_27-lshrsi3_table + .byte lshrsi3_28-lshrsi3_table + .byte lshrsi3_29-lshrsi3_table + .byte lshrsi3_30-lshrsi3_table + .byte lshrsi3_31-lshrsi3_table + +lshrsi3_6: + shlr2 r0 +lshrsi3_4: + shlr2 r0 +lshrsi3_2: + rts + shlr2 r0 + +lshrsi3_7: + shlr2 r0 +lshrsi3_5: + shlr2 r0 +lshrsi3_3: + shlr2 r0 +lshrsi3_1: + rts + shlr r0 + +lshrsi3_14: + shlr2 r0 +lshrsi3_12: + shlr2 r0 +lshrsi3_10: + shlr2 r0 +lshrsi3_8: + rts + shlr8 r0 + +lshrsi3_15: + shlr2 r0 +lshrsi3_13: + shlr2 r0 +lshrsi3_11: + shlr2 r0 +lshrsi3_9: + shlr8 r0 + rts + shlr r0 + +lshrsi3_22: + shlr2 r0 +lshrsi3_20: + shlr2 r0 +lshrsi3_18: + shlr2 r0 +lshrsi3_16: + rts + shlr16 r0 + +lshrsi3_23: + shlr2 r0 +lshrsi3_21: + shlr2 r0 +lshrsi3_19: + shlr2 r0 +lshrsi3_17: + shlr16 r0 + rts + shlr r0 + +lshrsi3_30: + shlr2 r0 +lshrsi3_28: + shlr2 r0 +lshrsi3_26: + shlr2 r0 +lshrsi3_24: + shlr16 r0 + rts + shlr8 r0 + +lshrsi3_31: + shlr2 r0 +lshrsi3_29: + shlr2 r0 +lshrsi3_27: + shlr2 r0 +lshrsi3_25: + shlr16 r0 + shlr8 r0 + rts + shlr r0 + +lshrsi3_0: + rts + nop diff --git a/arch/sh/lib/mcount.S b/arch/sh/lib/mcount.S new file mode 100644 index 000000000000..110fbfe1831f --- /dev/null +++ b/arch/sh/lib/mcount.S @@ -0,0 +1,90 @@ +/* + * arch/sh/lib/mcount.S + * + * Copyright (C) 2008 Paul Mundt + * Copyright (C) 2008 Matt Fleming + * + * This file is subject to the terms and conditions of the GNU General Public + * License. See the file "COPYING" in the main directory of this archive + * for more details. + */ +#include <asm/ftrace.h> + +#define MCOUNT_ENTER() \ + mov.l r4, @-r15; \ + mov.l r5, @-r15; \ + mov.l r6, @-r15; \ + mov.l r7, @-r15; \ + sts.l pr, @-r15; \ + \ + mov.l @(20,r15),r4; \ + sts pr, r5 + +#define MCOUNT_LEAVE() \ + lds.l @r15+, pr; \ + mov.l @r15+, r7; \ + mov.l @r15+, r6; \ + mov.l @r15+, r5; \ + rts; \ + mov.l @r15+, r4 + + .align 2 + .globl _mcount + .type _mcount,@function + .globl mcount + .type mcount,@function +_mcount: +mcount: + MCOUNT_ENTER() + +#ifdef CONFIG_DYNAMIC_FTRACE + .globl mcount_call +mcount_call: + mov.l .Lftrace_stub, r6 +#else + mov.l .Lftrace_trace_function, r6 + mov.l ftrace_stub, r7 + cmp/eq r6, r7 + bt skip_trace + mov.l @r6, r6 +#endif + + jsr @r6 + nop + +skip_trace: + MCOUNT_LEAVE() + + .align 2 +.Lftrace_trace_function: + .long ftrace_trace_function + +#ifdef CONFIG_DYNAMIC_FTRACE + .globl ftrace_caller +ftrace_caller: + MCOUNT_ENTER() + + .globl ftrace_call +ftrace_call: + mov.l .Lftrace_stub, r6 + jsr @r6 + nop + + MCOUNT_LEAVE() +#endif /* CONFIG_DYNAMIC_FTRACE */ + +/* + * NOTE: From here on the locations of the .Lftrace_stub label and + * ftrace_stub itself are fixed. Adding additional data here will skew + * the displacement for the memory table and break the block replacement. + * Place new labels either after the ftrace_stub body, or before + * ftrace_caller. You have been warned. + */ + .align 2 +.Lftrace_stub: + .long ftrace_stub + + .globl ftrace_stub +ftrace_stub: + rts + nop diff --git a/arch/sh/lib/movmem.S b/arch/sh/lib/movmem.S new file mode 100644 index 000000000000..62075f6bc67c --- /dev/null +++ b/arch/sh/lib/movmem.S @@ -0,0 +1,238 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + + .text + .balign 4 + .global __movmem + .global __movstr + .set __movstr, __movmem + /* This would be a lot simpler if r6 contained the byte count + minus 64, and we wouldn't be called here for a byte count of 64. */ +__movmem: + sts.l pr,@-r15 + shll2 r6 + bsr __movmemSI52+2 + mov.l @(48,r5),r0 + .balign 4 +movmem_loop: /* Reached with rts */ + mov.l @(60,r5),r0 + add #-64,r6 + mov.l r0,@(60,r4) + tst r6,r6 + mov.l @(56,r5),r0 + bt movmem_done + mov.l r0,@(56,r4) + cmp/pl r6 + mov.l @(52,r5),r0 + add #64,r5 + mov.l r0,@(52,r4) + add #64,r4 + bt __movmemSI52 +! done all the large groups, do the remainder +! jump to movmem+ + mova __movmemSI4+4,r0 + add r6,r0 + jmp @r0 +movmem_done: ! share slot insn, works out aligned. + lds.l @r15+,pr + mov.l r0,@(56,r4) + mov.l @(52,r5),r0 + rts + mov.l r0,@(52,r4) + .balign 4 + + .global __movmemSI64 + .global __movstrSI64 + .set __movstrSI64, __movmemSI64 +__movmemSI64: + mov.l @(60,r5),r0 + mov.l r0,@(60,r4) + .global __movmemSI60 + .global __movstrSI60 + .set __movstrSI60, __movmemSI60 +__movmemSI60: + mov.l @(56,r5),r0 + mov.l r0,@(56,r4) + .global __movmemSI56 + .global __movstrSI56 + .set __movstrSI56, __movmemSI56 +__movmemSI56: + mov.l @(52,r5),r0 + mov.l r0,@(52,r4) + .global __movmemSI52 + .global __movstrSI52 + .set __movstrSI52, __movmemSI52 +__movmemSI52: + mov.l @(48,r5),r0 + mov.l r0,@(48,r4) + .global __movmemSI48 + .global __movstrSI48 + .set __movstrSI48, __movmemSI48 +__movmemSI48: + mov.l @(44,r5),r0 + mov.l r0,@(44,r4) + .global __movmemSI44 + .global __movstrSI44 + .set __movstrSI44, __movmemSI44 +__movmemSI44: + mov.l @(40,r5),r0 + mov.l r0,@(40,r4) + .global __movmemSI40 + .global __movstrSI40 + .set __movstrSI40, __movmemSI40 +__movmemSI40: + mov.l @(36,r5),r0 + mov.l r0,@(36,r4) + .global __movmemSI36 + .global __movstrSI36 + .set __movstrSI36, __movmemSI36 +__movmemSI36: + mov.l @(32,r5),r0 + mov.l r0,@(32,r4) + .global __movmemSI32 + .global __movstrSI32 + .set __movstrSI32, __movmemSI32 +__movmemSI32: + mov.l @(28,r5),r0 + mov.l r0,@(28,r4) + .global __movmemSI28 + .global __movstrSI28 + .set __movstrSI28, __movmemSI28 +__movmemSI28: + mov.l @(24,r5),r0 + mov.l r0,@(24,r4) + .global __movmemSI24 + .global __movstrSI24 + .set __movstrSI24, __movmemSI24 +__movmemSI24: + mov.l @(20,r5),r0 + mov.l r0,@(20,r4) + .global __movmemSI20 + .global __movstrSI20 + .set __movstrSI20, __movmemSI20 +__movmemSI20: + mov.l @(16,r5),r0 + mov.l r0,@(16,r4) + .global __movmemSI16 + .global __movstrSI16 + .set __movstrSI16, __movmemSI16 +__movmemSI16: + mov.l @(12,r5),r0 + mov.l r0,@(12,r4) + .global __movmemSI12 + .global __movstrSI12 + .set __movstrSI12, __movmemSI12 +__movmemSI12: + mov.l @(8,r5),r0 + mov.l r0,@(8,r4) + .global __movmemSI8 + .global __movstrSI8 + .set __movstrSI8, __movmemSI8 +__movmemSI8: + mov.l @(4,r5),r0 + mov.l r0,@(4,r4) + .global __movmemSI4 + .global __movstrSI4 + .set __movstrSI4, __movmemSI4 +__movmemSI4: + mov.l @(0,r5),r0 + rts + mov.l r0,@(0,r4) + + .global __movmem_i4_even + .global __movstr_i4_even + .set __movstr_i4_even, __movmem_i4_even + + .global __movmem_i4_odd + .global __movstr_i4_odd + .set __movstr_i4_odd, __movmem_i4_odd + + .global __movmemSI12_i4 + .global __movstrSI12_i4 + .set __movstrSI12_i4, __movmemSI12_i4 + + .p2align 5 +L_movmem_2mod4_end: + mov.l r0,@(16,r4) + rts + mov.l r1,@(20,r4) + + .p2align 2 + +__movmem_i4_even: + mov.l @r5+,r0 + bra L_movmem_start_even + mov.l @r5+,r1 + +__movmem_i4_odd: + mov.l @r5+,r1 + add #-4,r4 + mov.l @r5+,r2 + mov.l @r5+,r3 + mov.l r1,@(4,r4) + mov.l r2,@(8,r4) + +L_movmem_loop: + mov.l r3,@(12,r4) + dt r6 + mov.l @r5+,r0 + bt/s L_movmem_2mod4_end + mov.l @r5+,r1 + add #16,r4 +L_movmem_start_even: + mov.l @r5+,r2 + mov.l @r5+,r3 + mov.l r0,@r4 + dt r6 + mov.l r1,@(4,r4) + bf/s L_movmem_loop + mov.l r2,@(8,r4) + rts + mov.l r3,@(12,r4) + + .p2align 4 +__movmemSI12_i4: + mov.l @r5,r0 + mov.l @(4,r5),r1 + mov.l @(8,r5),r2 + mov.l r0,@r4 + mov.l r1,@(4,r4) + rts + mov.l r2,@(8,r4) diff --git a/arch/sh/lib/udiv_qrnnd.S b/arch/sh/lib/udiv_qrnnd.S new file mode 100644 index 000000000000..32b9a36de943 --- /dev/null +++ b/arch/sh/lib/udiv_qrnnd.S @@ -0,0 +1,81 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + + /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */ + /* n1 < d, but n1 might be larger than d1. */ + .global __udiv_qrnnd_16 + .balign 8 +__udiv_qrnnd_16: + div0u + cmp/hi r6,r0 + bt .Lots + .rept 16 + div1 r6,r0 + .endr + extu.w r0,r1 + bt 0f + add r6,r0 +0: rotcl r1 + mulu.w r1,r5 + xtrct r4,r0 + swap.w r0,r0 + sts macl,r2 + cmp/hs r2,r0 + sub r2,r0 + bt 0f + addc r5,r0 + add #-1,r1 + bt 0f +1: add #-1,r1 + rts + add r5,r0 + .balign 8 +.Lots: + sub r5,r0 + swap.w r4,r1 + xtrct r0,r1 + clrt + mov r1,r0 + addc r5,r0 + mov #-1,r1 + bf/s 1b + shlr16 r1 +0: rts + nop diff --git a/arch/sh/lib/udivsi3.S b/arch/sh/lib/udivsi3.S new file mode 100644 index 000000000000..72157ab5c314 --- /dev/null +++ b/arch/sh/lib/udivsi3.S @@ -0,0 +1,87 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + + .balign 4 + .global __udivsi3 + .type __udivsi3, @function +div8: + div1 r5,r4 +div7: + div1 r5,r4; div1 r5,r4; div1 r5,r4 + div1 r5,r4; div1 r5,r4; div1 r5,r4; rts; div1 r5,r4 + +divx4: + div1 r5,r4; rotcl r0 + div1 r5,r4; rotcl r0 + div1 r5,r4; rotcl r0 + rts; div1 r5,r4 + +__udivsi3: + sts.l pr,@-r15 + extu.w r5,r0 + cmp/eq r5,r0 + bf/s large_divisor + div0u + swap.w r4,r0 + shlr16 r4 + bsr div8 + shll16 r5 + bsr div7 + div1 r5,r4 + xtrct r4,r0 + xtrct r0,r4 + bsr div8 + swap.w r4,r4 + bsr div7 + div1 r5,r4 + lds.l @r15+,pr + xtrct r4,r0 + swap.w r0,r0 + rotcl r0 + rts + shlr16 r5 + +large_divisor: + mov #0,r0 + xtrct r4,r0 + xtrct r0,r4 + bsr divx4 + rotcl r0 + bsr divx4 + rotcl r0 + bsr divx4 + rotcl r0 + bsr divx4 + rotcl r0 + lds.l @r15+,pr + rts + rotcl r0 diff --git a/arch/sh/lib/udivsi3_i4i-Os.S b/arch/sh/lib/udivsi3_i4i-Os.S new file mode 100644 index 000000000000..4835553e1ea9 --- /dev/null +++ b/arch/sh/lib/udivsi3_i4i-Os.S @@ -0,0 +1,149 @@ +/* Copyright (C) 2006 Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +/* Moderately Space-optimized libgcc routines for the Renesas SH / + STMicroelectronics ST40 CPUs. + Contributed by J"orn Rennecke joern.rennecke@st.com. */ + +/* Size: 186 bytes jointly for udivsi3_i4i and sdivsi3_i4i + sh4-200 run times: + udiv small divisor: 55 cycles + udiv large divisor: 52 cycles + sdiv small divisor, positive result: 59 cycles + sdiv large divisor, positive result: 56 cycles + sdiv small divisor, negative result: 65 cycles (*) + sdiv large divisor, negative result: 62 cycles (*) + (*): r2 is restored in the rts delay slot and has a lingering latency + of two more cycles. */ + .balign 4 + .global __udivsi3_i4i + .global __udivsi3_i4 + .set __udivsi3_i4, __udivsi3_i4i + .type __udivsi3_i4i, @function + .type __sdivsi3_i4i, @function +__udivsi3_i4i: + sts pr,r1 + mov.l r4,@-r15 + extu.w r5,r0 + cmp/eq r5,r0 + swap.w r4,r0 + shlr16 r4 + bf/s large_divisor + div0u + mov.l r5,@-r15 + shll16 r5 +sdiv_small_divisor: + div1 r5,r4 + bsr div6 + div1 r5,r4 + div1 r5,r4 + bsr div6 + div1 r5,r4 + xtrct r4,r0 + xtrct r0,r4 + bsr div7 + swap.w r4,r4 + div1 r5,r4 + bsr div7 + div1 r5,r4 + xtrct r4,r0 + mov.l @r15+,r5 + swap.w r0,r0 + mov.l @r15+,r4 + jmp @r1 + rotcl r0 +div7: + div1 r5,r4 +div6: + div1 r5,r4; div1 r5,r4; div1 r5,r4 + div1 r5,r4; div1 r5,r4; rts; div1 r5,r4 + +divx3: + rotcl r0 + div1 r5,r4 + rotcl r0 + div1 r5,r4 + rotcl r0 + rts + div1 r5,r4 + +large_divisor: + mov.l r5,@-r15 +sdiv_large_divisor: + xor r4,r0 + .rept 4 + rotcl r0 + bsr divx3 + div1 r5,r4 + .endr + mov.l @r15+,r5 + mov.l @r15+,r4 + jmp @r1 + rotcl r0 + + .global __sdivsi3_i4i + .global __sdivsi3_i4 + .global __sdivsi3 + .set __sdivsi3_i4, __sdivsi3_i4i + .set __sdivsi3, __sdivsi3_i4i +__sdivsi3_i4i: + mov.l r4,@-r15 + cmp/pz r5 + mov.l r5,@-r15 + bt/s pos_divisor + cmp/pz r4 + neg r5,r5 + extu.w r5,r0 + bt/s neg_result + cmp/eq r5,r0 + neg r4,r4 +pos_result: + swap.w r4,r0 + bra sdiv_check_divisor + sts pr,r1 +pos_divisor: + extu.w r5,r0 + bt/s pos_result + cmp/eq r5,r0 + neg r4,r4 +neg_result: + mova negate_result,r0 + ; + mov r0,r1 + swap.w r4,r0 + lds r2,macl + sts pr,r2 +sdiv_check_divisor: + shlr16 r4 + bf/s sdiv_large_divisor + div0u + bra sdiv_small_divisor + shll16 r5 + .balign 4 +negate_result: + neg r0,r0 + jmp @r2 + sts macl,r2 diff --git a/arch/sh/lib/udivsi3_i4i.S b/arch/sh/lib/udivsi3_i4i.S new file mode 100644 index 000000000000..f1a79d9c5015 --- /dev/null +++ b/arch/sh/lib/udivsi3_i4i.S @@ -0,0 +1,666 @@ +/* Copyright (C) 1994, 1995, 1997, 1998, 1999, 2000, 2001, 2002, 2003, + 2004, 2005, 2006 + Free Software Foundation, Inc. + +This file is free software; you can redistribute it and/or modify it +under the terms of the GNU General Public License as published by the +Free Software Foundation; either version 2, or (at your option) any +later version. + +In addition to the permissions in the GNU General Public License, the +Free Software Foundation gives you unlimited permission to link the +compiled version of this file into combinations with other programs, +and to distribute those combinations without any restriction coming +from the use of this file. (The General Public License restrictions +do apply in other respects; for example, they cover modification of +the file, and distribution when not linked into a combine +executable.) + +This file is distributed in the hope that it will be useful, but +WITHOUT ANY WARRANTY; without even the implied warranty of +MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +General Public License for more details. + +You should have received a copy of the GNU General Public License +along with this program; see the file COPYING. If not, write to +the Free Software Foundation, 51 Franklin Street, Fifth Floor, +Boston, MA 02110-1301, USA. */ + +!! libgcc routines for the Renesas / SuperH SH CPUs. +!! Contributed by Steve Chamberlain. +!! sac@cygnus.com + +!! ashiftrt_r4_x, ___ashrsi3, ___ashlsi3, ___lshrsi3 routines +!! recoded in assembly by Toshiyasu Morita +!! tm@netcom.com + +/* SH2 optimizations for ___ashrsi3, ___ashlsi3, ___lshrsi3 and + ELF local label prefixes by J"orn Rennecke + amylaar@cygnus.com */ + +/* This code used shld, thus is not suitable for SH1 / SH2. */ + +/* Signed / unsigned division without use of FPU, optimized for SH4. + Uses a lookup table for divisors in the range -128 .. +128, and + div1 with case distinction for larger divisors in three more ranges. + The code is lumped together with the table to allow the use of mova. */ +#ifdef CONFIG_CPU_LITTLE_ENDIAN +#define L_LSB 0 +#define L_LSWMSB 1 +#define L_MSWLSB 2 +#else +#define L_LSB 3 +#define L_LSWMSB 2 +#define L_MSWLSB 1 +#endif + + .balign 4 + .global __udivsi3_i4i + .global __udivsi3_i4 + .set __udivsi3_i4, __udivsi3_i4i + .type __udivsi3_i4i, @function +__udivsi3_i4i: + mov.w c128_w, r1 + div0u + mov r4,r0 + shlr8 r0 + cmp/hi r1,r5 + extu.w r5,r1 + bf udiv_le128 + cmp/eq r5,r1 + bf udiv_ge64k + shlr r0 + mov r5,r1 + shll16 r5 + mov.l r4,@-r15 + div1 r5,r0 + mov.l r1,@-r15 + div1 r5,r0 + div1 r5,r0 + bra udiv_25 + div1 r5,r0 + +div_le128: + mova div_table_ix,r0 + bra div_le128_2 + mov.b @(r0,r5),r1 +udiv_le128: + mov.l r4,@-r15 + mova div_table_ix,r0 + mov.b @(r0,r5),r1 + mov.l r5,@-r15 +div_le128_2: + mova div_table_inv,r0 + mov.l @(r0,r1),r1 + mov r5,r0 + tst #0xfe,r0 + mova div_table_clz,r0 + dmulu.l r1,r4 + mov.b @(r0,r5),r1 + bt/s div_by_1 + mov r4,r0 + mov.l @r15+,r5 + sts mach,r0 + /* clrt */ + addc r4,r0 + mov.l @r15+,r4 + rotcr r0 + rts + shld r1,r0 + +div_by_1_neg: + neg r4,r0 +div_by_1: + mov.l @r15+,r5 + rts + mov.l @r15+,r4 + +div_ge64k: + bt/s div_r8 + div0u + shll8 r5 + bra div_ge64k_2 + div1 r5,r0 +udiv_ge64k: + cmp/hi r0,r5 + mov r5,r1 + bt udiv_r8 + shll8 r5 + mov.l r4,@-r15 + div1 r5,r0 + mov.l r1,@-r15 +div_ge64k_2: + div1 r5,r0 + mov.l zero_l,r1 + .rept 4 + div1 r5,r0 + .endr + mov.l r1,@-r15 + div1 r5,r0 + mov.w m256_w,r1 + div1 r5,r0 + mov.b r0,@(L_LSWMSB,r15) + xor r4,r0 + and r1,r0 + bra div_ge64k_end + xor r4,r0 + +div_r8: + shll16 r4 + bra div_r8_2 + shll8 r4 +udiv_r8: + mov.l r4,@-r15 + shll16 r4 + clrt + shll8 r4 + mov.l r5,@-r15 +div_r8_2: + rotcl r4 + mov r0,r1 + div1 r5,r1 + mov r4,r0 + rotcl r0 + mov r5,r4 + div1 r5,r1 + .rept 5 + rotcl r0; div1 r5,r1 + .endr + rotcl r0 + mov.l @r15+,r5 + div1 r4,r1 + mov.l @r15+,r4 + rts + rotcl r0 + + .global __sdivsi3_i4i + .global __sdivsi3_i4 + .global __sdivsi3 + .set __sdivsi3_i4, __sdivsi3_i4i + .set __sdivsi3, __sdivsi3_i4i + .type __sdivsi3_i4i, @function + /* This is link-compatible with a __sdivsi3 call, + but we effectively clobber only r1. */ +__sdivsi3_i4i: + mov.l r4,@-r15 + cmp/pz r5 + mov.w c128_w, r1 + bt/s pos_divisor + cmp/pz r4 + mov.l r5,@-r15 + neg r5,r5 + bt/s neg_result + cmp/hi r1,r5 + neg r4,r4 +pos_result: + extu.w r5,r0 + bf div_le128 + cmp/eq r5,r0 + mov r4,r0 + shlr8 r0 + bf/s div_ge64k + cmp/hi r0,r5 + div0u + shll16 r5 + div1 r5,r0 + div1 r5,r0 + div1 r5,r0 +udiv_25: + mov.l zero_l,r1 + div1 r5,r0 + div1 r5,r0 + mov.l r1,@-r15 + .rept 3 + div1 r5,r0 + .endr + mov.b r0,@(L_MSWLSB,r15) + xtrct r4,r0 + swap.w r0,r0 + .rept 8 + div1 r5,r0 + .endr + mov.b r0,@(L_LSWMSB,r15) +div_ge64k_end: + .rept 8 + div1 r5,r0 + .endr + mov.l @r15+,r4 ! zero-extension and swap using LS unit. + extu.b r0,r0 + mov.l @r15+,r5 + or r4,r0 + mov.l @r15+,r4 + rts + rotcl r0 + +div_le128_neg: + tst #0xfe,r0 + mova div_table_ix,r0 + mov.b @(r0,r5),r1 + mova div_table_inv,r0 + bt/s div_by_1_neg + mov.l @(r0,r1),r1 + mova div_table_clz,r0 + dmulu.l r1,r4 + mov.b @(r0,r5),r1 + mov.l @r15+,r5 + sts mach,r0 + /* clrt */ + addc r4,r0 + mov.l @r15+,r4 + rotcr r0 + shld r1,r0 + rts + neg r0,r0 + +pos_divisor: + mov.l r5,@-r15 + bt/s pos_result + cmp/hi r1,r5 + neg r4,r4 +neg_result: + extu.w r5,r0 + bf div_le128_neg + cmp/eq r5,r0 + mov r4,r0 + shlr8 r0 + bf/s div_ge64k_neg + cmp/hi r0,r5 + div0u + mov.l zero_l,r1 + shll16 r5 + div1 r5,r0 + mov.l r1,@-r15 + .rept 7 + div1 r5,r0 + .endr + mov.b r0,@(L_MSWLSB,r15) + xtrct r4,r0 + swap.w r0,r0 + .rept 8 + div1 r5,r0 + .endr + mov.b r0,@(L_LSWMSB,r15) +div_ge64k_neg_end: + .rept 8 + div1 r5,r0 + .endr + mov.l @r15+,r4 ! zero-extension and swap using LS unit. + extu.b r0,r1 + mov.l @r15+,r5 + or r4,r1 +div_r8_neg_end: + mov.l @r15+,r4 + rotcl r1 + rts + neg r1,r0 + +div_ge64k_neg: + bt/s div_r8_neg + div0u + shll8 r5 + mov.l zero_l,r1 + .rept 6 + div1 r5,r0 + .endr + mov.l r1,@-r15 + div1 r5,r0 + mov.w m256_w,r1 + div1 r5,r0 + mov.b r0,@(L_LSWMSB,r15) + xor r4,r0 + and r1,r0 + bra div_ge64k_neg_end + xor r4,r0 + +c128_w: + .word 128 + +div_r8_neg: + clrt + shll16 r4 + mov r4,r1 + shll8 r1 + mov r5,r4 + .rept 7 + rotcl r1; div1 r5,r0 + .endr + mov.l @r15+,r5 + rotcl r1 + bra div_r8_neg_end + div1 r4,r0 + +m256_w: + .word 0xff00 +/* This table has been generated by divtab-sh4.c. */ + .balign 4 +div_table_clz: + .byte 0 + .byte 1 + .byte 0 + .byte -1 + .byte -1 + .byte -2 + .byte -2 + .byte -2 + .byte -2 + .byte -3 + .byte -3 + .byte -3 + .byte -3 + .byte -3 + .byte -3 + .byte -3 + .byte -3 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -4 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -5 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 + .byte -6 +/* Lookup table translating positive divisor to index into table of + normalized inverse. N.B. the '0' entry is also the last entry of the + previous table, and causes an unaligned access for division by zero. */ +div_table_ix: + .byte -6 + .byte -128 + .byte -128 + .byte 0 + .byte -128 + .byte -64 + .byte 0 + .byte 64 + .byte -128 + .byte -96 + .byte -64 + .byte -32 + .byte 0 + .byte 32 + .byte 64 + .byte 96 + .byte -128 + .byte -112 + .byte -96 + .byte -80 + .byte -64 + .byte -48 + .byte -32 + .byte -16 + .byte 0 + .byte 16 + .byte 32 + .byte 48 + .byte 64 + .byte 80 + .byte 96 + .byte 112 + .byte -128 + .byte -120 + .byte -112 + .byte -104 + .byte -96 + .byte -88 + .byte -80 + .byte -72 + .byte -64 + .byte -56 + .byte -48 + .byte -40 + .byte -32 + .byte -24 + .byte -16 + .byte -8 + .byte 0 + .byte 8 + .byte 16 + .byte 24 + .byte 32 + .byte 40 + .byte 48 + .byte 56 + .byte 64 + .byte 72 + .byte 80 + .byte 88 + .byte 96 + .byte 104 + .byte 112 + .byte 120 + .byte -128 + .byte -124 + .byte -120 + .byte -116 + .byte -112 + .byte -108 + .byte -104 + .byte -100 + .byte -96 + .byte -92 + .byte -88 + .byte -84 + .byte -80 + .byte -76 + .byte -72 + .byte -68 + .byte -64 + .byte -60 + .byte -56 + .byte -52 + .byte -48 + .byte -44 + .byte -40 + .byte -36 + .byte -32 + .byte -28 + .byte -24 + .byte -20 + .byte -16 + .byte -12 + .byte -8 + .byte -4 + .byte 0 + .byte 4 + .byte 8 + .byte 12 + .byte 16 + .byte 20 + .byte 24 + .byte 28 + .byte 32 + .byte 36 + .byte 40 + .byte 44 + .byte 48 + .byte 52 + .byte 56 + .byte 60 + .byte 64 + .byte 68 + .byte 72 + .byte 76 + .byte 80 + .byte 84 + .byte 88 + .byte 92 + .byte 96 + .byte 100 + .byte 104 + .byte 108 + .byte 112 + .byte 116 + .byte 120 + .byte 124 + .byte -128 +/* 1/64 .. 1/127, normalized. There is an implicit leading 1 in bit 32. */ + .balign 4 +zero_l: + .long 0x0 + .long 0xF81F81F9 + .long 0xF07C1F08 + .long 0xE9131AC0 + .long 0xE1E1E1E2 + .long 0xDAE6076C + .long 0xD41D41D5 + .long 0xCD856891 + .long 0xC71C71C8 + .long 0xC0E07039 + .long 0xBACF914D + .long 0xB4E81B4F + .long 0xAF286BCB + .long 0xA98EF607 + .long 0xA41A41A5 + .long 0x9EC8E952 + .long 0x9999999A + .long 0x948B0FCE + .long 0x8F9C18FA + .long 0x8ACB90F7 + .long 0x86186187 + .long 0x81818182 + .long 0x7D05F418 + .long 0x78A4C818 + .long 0x745D1746 + .long 0x702E05C1 + .long 0x6C16C16D + .long 0x68168169 + .long 0x642C8591 + .long 0x60581606 + .long 0x5C9882BA + .long 0x58ED2309 +div_table_inv: + .long 0x55555556 + .long 0x51D07EAF + .long 0x4E5E0A73 + .long 0x4AFD6A06 + .long 0x47AE147B + .long 0x446F8657 + .long 0x41414142 + .long 0x3E22CBCF + .long 0x3B13B13C + .long 0x38138139 + .long 0x3521CFB3 + .long 0x323E34A3 + .long 0x2F684BDB + .long 0x2C9FB4D9 + .long 0x29E4129F + .long 0x27350B89 + .long 0x24924925 + .long 0x21FB7813 + .long 0x1F7047DD + .long 0x1CF06ADB + .long 0x1A7B9612 + .long 0x18118119 + .long 0x15B1E5F8 + .long 0x135C8114 + .long 0x11111112 + .long 0xECF56BF + .long 0xC9714FC + .long 0xA6810A7 + .long 0x8421085 + .long 0x624DD30 + .long 0x4104105 + .long 0x2040811 + /* maximum error: 0.987342 scaled: 0.921875*/ diff --git a/arch/sh/lib64/Makefile b/arch/sh/lib64/Makefile index 9950966923a0..4bacb9e83478 100644 --- a/arch/sh/lib64/Makefile +++ b/arch/sh/lib64/Makefile @@ -2,7 +2,7 @@ # Makefile for the SH-5 specific library files.. # # Copyright (C) 2000, 2001 Paolo Alberelli -# Copyright (C) 2003 Paul Mundt +# Copyright (C) 2003 - 2008 Paul Mundt # # This file is subject to the terms and conditions of the GNU General Public # License. See the file "COPYING" in the main directory of this archive @@ -10,6 +10,8 @@ # # Panic should really be compiled as PIC -lib-y := udelay.o c-checksum.o dbg.o panic.o memcpy.o copy_user_memcpy.o \ - copy_page.o clear_page.o +lib-y := udelay.o c-checksum.o dbg.o panic.o memcpy.o memset.o \ + copy_user_memcpy.o copy_page.o clear_page.o strcpy.o strlen.o +# Extracted from libgcc +lib-y += udivsi3.o udivdi3.o sdivsi3.o diff --git a/arch/sh/lib64/c-checksum.c b/arch/sh/lib64/c-checksum.c index 5c284e0cff9c..73c0877e3a29 100644 --- a/arch/sh/lib64/c-checksum.c +++ b/arch/sh/lib64/c-checksum.c @@ -35,7 +35,7 @@ static inline unsigned short foldto16(unsigned long x) static inline unsigned short myfoldto16(unsigned long long x) { - /* Fold down to 32-bits so we don't loose in the typedef-less + /* Fold down to 32-bits so we don't lose in the typedef-less network stack. */ /* 64 to 33 */ x = (x & 0xffffffff) + (x >> 32); @@ -199,7 +199,7 @@ __wsum csum_tcpudp_nofold(__be32 saddr, __be32 daddr, result = (__force u64) saddr + (__force u64) daddr + (__force u64) sum + ((len + proto) << 8); - /* Fold down to 32-bits so we don't loose in the typedef-less + /* Fold down to 32-bits so we don't lose in the typedef-less network stack. */ /* 64 to 33 */ result = (result & 0xffffffff) + (result >> 32); diff --git a/arch/sh/lib64/memcpy.S b/arch/sh/lib64/memcpy.S new file mode 100644 index 000000000000..dd300c372ce1 --- /dev/null +++ b/arch/sh/lib64/memcpy.S @@ -0,0 +1,201 @@ +/* Cloned and hacked for uClibc by Paul Mundt, December 2003 */ +/* Modified by SuperH, Inc. September 2003 */ +! +! Fast SH memcpy +! +! by Toshiyasu Morita (tm@netcom.com) +! hacked by J"orn Rernnecke (joern.rennecke@superh.com) ("o for o-umlaut) +! SH5 code Copyright 2002 SuperH Ltd. +! +! Entry: ARG0: destination pointer +! ARG1: source pointer +! ARG2: byte count +! +! Exit: RESULT: destination pointer +! any other registers in the range r0-r7: trashed +! +! Notes: Usually one wants to do small reads and write a longword, but +! unfortunately it is difficult in some cases to concatanate bytes +! into a longword on the SH, so this does a longword read and small +! writes. +! +! This implementation makes two assumptions about how it is called: +! +! 1.: If the byte count is nonzero, the address of the last byte to be +! copied is unsigned greater than the address of the first byte to +! be copied. This could be easily swapped for a signed comparison, +! but the algorithm used needs some comparison. +! +! 2.: When there are two or three bytes in the last word of an 11-or-more +! bytes memory chunk to b copied, the rest of the word can be read +! without side effects. +! This could be easily changed by increasing the minumum size of +! a fast memcpy and the amount subtracted from r7 before L_2l_loop be 2, +! however, this would cost a few extra cyles on average. +! For SHmedia, the assumption is that any quadword can be read in its +! enirety if at least one byte is included in the copy. +! + + .section .text..SHmedia32,"ax" + .globl memcpy + .type memcpy, @function + .align 5 + +memcpy: + +#define LDUAQ(P,O,D0,D1) ldlo.q P,O,D0; ldhi.q P,O+7,D1 +#define STUAQ(P,O,D0,D1) stlo.q P,O,D0; sthi.q P,O+7,D1 +#define LDUAL(P,O,D0,D1) ldlo.l P,O,D0; ldhi.l P,O+3,D1 +#define STUAL(P,O,D0,D1) stlo.l P,O,D0; sthi.l P,O+3,D1 + + ld.b r3,0,r63 + pta/l Large,tr0 + movi 25,r0 + bgeu/u r4,r0,tr0 + nsb r4,r0 + shlli r0,5,r0 + movi (L1-L0+63*32 + 1) & 0xffff,r1 + sub r1, r0, r0 +L0: ptrel r0,tr0 + add r2,r4,r5 + ptabs r18,tr1 + add r3,r4,r6 + blink tr0,r63 + +/* Rearranged to make cut2 safe */ + .balign 8 +L4_7: /* 4..7 byte memcpy cntd. */ + stlo.l r2, 0, r0 + or r6, r7, r6 + sthi.l r5, -1, r6 + stlo.l r5, -4, r6 + blink tr1,r63 + + .balign 8 +L1: /* 0 byte memcpy */ + nop + blink tr1,r63 + nop + nop + nop + nop + +L2_3: /* 2 or 3 byte memcpy cntd. */ + st.b r5,-1,r6 + blink tr1,r63 + + /* 1 byte memcpy */ + ld.b r3,0,r0 + st.b r2,0,r0 + blink tr1,r63 + +L8_15: /* 8..15 byte memcpy cntd. */ + stlo.q r2, 0, r0 + or r6, r7, r6 + sthi.q r5, -1, r6 + stlo.q r5, -8, r6 + blink tr1,r63 + + /* 2 or 3 byte memcpy */ + ld.b r3,0,r0 + ld.b r2,0,r63 + ld.b r3,1,r1 + st.b r2,0,r0 + pta/l L2_3,tr0 + ld.b r6,-1,r6 + st.b r2,1,r1 + blink tr0, r63 + + /* 4 .. 7 byte memcpy */ + LDUAL (r3, 0, r0, r1) + pta L4_7, tr0 + ldlo.l r6, -4, r7 + or r0, r1, r0 + sthi.l r2, 3, r0 + ldhi.l r6, -1, r6 + blink tr0, r63 + + /* 8 .. 15 byte memcpy */ + LDUAQ (r3, 0, r0, r1) + pta L8_15, tr0 + ldlo.q r6, -8, r7 + or r0, r1, r0 + sthi.q r2, 7, r0 + ldhi.q r6, -1, r6 + blink tr0, r63 + + /* 16 .. 24 byte memcpy */ + LDUAQ (r3, 0, r0, r1) + LDUAQ (r3, 8, r8, r9) + or r0, r1, r0 + sthi.q r2, 7, r0 + or r8, r9, r8 + sthi.q r2, 15, r8 + ldlo.q r6, -8, r7 + ldhi.q r6, -1, r6 + stlo.q r2, 8, r8 + stlo.q r2, 0, r0 + or r6, r7, r6 + sthi.q r5, -1, r6 + stlo.q r5, -8, r6 + blink tr1,r63 + +Large: + ld.b r2, 0, r63 + pta/l Loop_ua, tr1 + ori r3, -8, r7 + sub r2, r7, r22 + sub r3, r2, r6 + add r2, r4, r5 + ldlo.q r3, 0, r0 + addi r5, -16, r5 + movi 64+8, r27 // could subtract r7 from that. + stlo.q r2, 0, r0 + sthi.q r2, 7, r0 + ldx.q r22, r6, r0 + bgtu/l r27, r4, tr1 + + addi r5, -48, r27 + pta/l Loop_line, tr0 + addi r6, 64, r36 + addi r6, -24, r19 + addi r6, -16, r20 + addi r6, -8, r21 + +Loop_line: + ldx.q r22, r36, r63 + alloco r22, 32 + addi r22, 32, r22 + ldx.q r22, r19, r23 + sthi.q r22, -25, r0 + ldx.q r22, r20, r24 + ldx.q r22, r21, r25 + stlo.q r22, -32, r0 + ldx.q r22, r6, r0 + sthi.q r22, -17, r23 + sthi.q r22, -9, r24 + sthi.q r22, -1, r25 + stlo.q r22, -24, r23 + stlo.q r22, -16, r24 + stlo.q r22, -8, r25 + bgeu r27, r22, tr0 + +Loop_ua: + addi r22, 8, r22 + sthi.q r22, -1, r0 + stlo.q r22, -8, r0 + ldx.q r22, r6, r0 + bgtu/l r5, r22, tr1 + + add r3, r4, r7 + ldlo.q r7, -8, r1 + sthi.q r22, 7, r0 + ldhi.q r7, -1, r7 + ptabs r18,tr1 + stlo.q r22, 0, r0 + or r1, r7, r1 + sthi.q r5, 15, r1 + stlo.q r5, 8, r1 + blink tr1, r63 + + .size memcpy,.-memcpy diff --git a/arch/sh/lib64/memcpy.c b/arch/sh/lib64/memcpy.c deleted file mode 100644 index fba436a92bfa..000000000000 --- a/arch/sh/lib64/memcpy.c +++ /dev/null @@ -1,81 +0,0 @@ -/* - * Copyright (C) 2002 Mark Debbage (Mark.Debbage@superh.com) - * - * May be copied or modified under the terms of the GNU General Public - * License. See linux/COPYING for more information. - * - */ - -#include <linux/types.h> -#include <asm/string.h> - -// This is a simplistic optimization of memcpy to increase the -// granularity of access beyond one byte using aligned -// loads and stores. This is not an optimal implementation -// for SH-5 (especially with regard to prefetching and the cache), -// and a better version should be provided later ... - -void *memcpy(void *dest, const void *src, size_t count) -{ - char *d = (char *) dest, *s = (char *) src; - - if (count >= 32) { - int i = 8 - (((unsigned long) d) & 0x7); - - if (i != 8) - while (i-- && count--) { - *d++ = *s++; - } - - if (((((unsigned long) d) & 0x7) == 0) && - ((((unsigned long) s) & 0x7) == 0)) { - while (count >= 32) { - unsigned long long t1, t2, t3, t4; - t1 = *(unsigned long long *) (s); - t2 = *(unsigned long long *) (s + 8); - t3 = *(unsigned long long *) (s + 16); - t4 = *(unsigned long long *) (s + 24); - *(unsigned long long *) (d) = t1; - *(unsigned long long *) (d + 8) = t2; - *(unsigned long long *) (d + 16) = t3; - *(unsigned long long *) (d + 24) = t4; - d += 32; - s += 32; - count -= 32; - } - while (count >= 8) { - *(unsigned long long *) d = - *(unsigned long long *) s; - d += 8; - s += 8; - count -= 8; - } - } - - if (((((unsigned long) d) & 0x3) == 0) && - ((((unsigned long) s) & 0x3) == 0)) { - while (count >= 4) { - *(unsigned long *) d = *(unsigned long *) s; - d += 4; - s += 4; - count -= 4; - } - } - - if (((((unsigned long) d) & 0x1) == 0) && - ((((unsigned long) s) & 0x1) == 0)) { - while (count >= 2) { - *(unsigned short *) d = *(unsigned short *) s; - d += 2; - s += 2; - count -= 2; - } - } - } - - while (count--) { - *d++ = *s++; - } - - return d; -} diff --git a/arch/sh/lib64/memset.S b/arch/sh/lib64/memset.S new file mode 100644 index 000000000000..2d37b0488552 --- /dev/null +++ b/arch/sh/lib64/memset.S @@ -0,0 +1,91 @@ +/* Cloned and hacked for uClibc by Paul Mundt, December 2003 */ +/* Modified by SuperH, Inc. September 2003 */ +! +! Fast SH memset +! +! by Toshiyasu Morita (tm@netcom.com) +! +! SH5 code by J"orn Rennecke (joern.rennecke@superh.com) +! Copyright 2002 SuperH Ltd. +! + +#if __BYTE_ORDER == __LITTLE_ENDIAN +#define SHHI shlld +#define SHLO shlrd +#else +#define SHHI shlrd +#define SHLO shlld +#endif + + .section .text..SHmedia32,"ax" + .globl memset + .type memset, @function + + .align 5 + +memset: + pta/l multiquad, tr0 + andi r2, 7, r22 + ptabs r18, tr2 + mshflo.b r3,r3,r3 + add r4, r22, r23 + mperm.w r3, r63, r3 // Fill pattern now in every byte of r3 + + movi 8, r9 + bgtu/u r23, r9, tr0 // multiquad + + beqi/u r4, 0, tr2 // Return with size 0 - ensures no mem accesses + ldlo.q r2, 0, r7 + shlli r4, 2, r4 + movi -1, r8 + SHHI r8, r4, r8 + SHHI r8, r4, r8 + mcmv r7, r8, r3 + stlo.q r2, 0, r3 + blink tr2, r63 + +multiquad: + pta/l lastquad, tr0 + stlo.q r2, 0, r3 + shlri r23, 3, r24 + add r2, r4, r5 + beqi/u r24, 1, tr0 // lastquad + pta/l loop, tr1 + sub r2, r22, r25 + andi r5, -8, r20 // calculate end address and + addi r20, -7*8, r8 // loop end address; This might overflow, so we need + // to use a different test before we start the loop + bge/u r24, r9, tr1 // loop + st.q r25, 8, r3 + st.q r20, -8, r3 + shlri r24, 1, r24 + beqi/u r24, 1, tr0 // lastquad + st.q r25, 16, r3 + st.q r20, -16, r3 + beqi/u r24, 2, tr0 // lastquad + st.q r25, 24, r3 + st.q r20, -24, r3 +lastquad: + sthi.q r5, -1, r3 + blink tr2,r63 + +loop: +!!! alloco r25, 32 // QQQ comment out for short-term fix to SHUK #3895. + // QQQ commenting out is locically correct, but sub-optimal + // QQQ Sean McGoogan - 4th April 2003. + st.q r25, 8, r3 + st.q r25, 16, r3 + st.q r25, 24, r3 + st.q r25, 32, r3 + addi r25, 32, r25 + bgeu/l r8, r25, tr1 // loop + + st.q r20, -40, r3 + st.q r20, -32, r3 + st.q r20, -24, r3 + st.q r20, -16, r3 + st.q r20, -8, r3 + sthi.q r5, -1, r3 + blink tr2,r63 + + .size memset,.-memset diff --git a/arch/sh/lib64/sdivsi3.S b/arch/sh/lib64/sdivsi3.S new file mode 100644 index 000000000000..6a800c6a4904 --- /dev/null +++ b/arch/sh/lib64/sdivsi3.S @@ -0,0 +1,131 @@ + .global __sdivsi3 + .section .text..SHmedia32,"ax" + .align 2 + + /* inputs: r4,r5 */ + /* clobbered: r1,r18,r19,r20,r21,r25,tr0 */ + /* result in r0 */ +__sdivsi3: + ptb __div_table,tr0 + + nsb r5, r1 + shlld r5, r1, r25 /* normalize; [-2 ..1, 1..2) in s2.62 */ + shari r25, 58, r21 /* extract 5(6) bit index (s2.4 with hole -1..1) */ + /* bubble */ + gettr tr0,r20 + ldx.ub r20, r21, r19 /* u0.8 */ + shari r25, 32, r25 /* normalize to s2.30 */ + shlli r21, 1, r21 + muls.l r25, r19, r19 /* s2.38 */ + ldx.w r20, r21, r21 /* s2.14 */ + ptabs r18, tr0 + shari r19, 24, r19 /* truncate to s2.14 */ + sub r21, r19, r19 /* some 11 bit inverse in s1.14 */ + muls.l r19, r19, r21 /* u0.28 */ + sub r63, r1, r1 + addi r1, 92, r1 + muls.l r25, r21, r18 /* s2.58 */ + shlli r19, 45, r19 /* multiply by two and convert to s2.58 */ + /* bubble */ + sub r19, r18, r18 + shari r18, 28, r18 /* some 22 bit inverse in s1.30 */ + muls.l r18, r25, r0 /* s2.60 */ + muls.l r18, r4, r25 /* s32.30 */ + /* bubble */ + shari r0, 16, r19 /* s-16.44 */ + muls.l r19, r18, r19 /* s-16.74 */ + shari r25, 63, r0 + shari r4, 14, r18 /* s19.-14 */ + shari r19, 30, r19 /* s-16.44 */ + muls.l r19, r18, r19 /* s15.30 */ + xor r21, r0, r21 /* You could also use the constant 1 << 27. */ + add r21, r25, r21 + sub r21, r19, r21 + shard r21, r1, r21 + sub r21, r0, r0 + blink tr0, r63 + +/* This table has been generated by divtab.c . +Defects for bias -330: + Max defect: 6.081536e-07 at -1.000000e+00 + Min defect: 2.849516e-08 at 1.030651e+00 + Max 2nd step defect: 9.606539e-12 at -1.000000e+00 + Min 2nd step defect: 0.000000e+00 at 0.000000e+00 + Defect at 1: 1.238659e-07 + Defect at -2: 1.061708e-07 */ + + .balign 2 + .type __div_table,@object + .size __div_table,128 +/* negative division constants */ + .word -16638 + .word -17135 + .word -17737 + .word -18433 + .word -19103 + .word -19751 + .word -20583 + .word -21383 + .word -22343 + .word -23353 + .word -24407 + .word -25582 + .word -26863 + .word -28382 + .word -29965 + .word -31800 +/* negative division factors */ + .byte 66 + .byte 70 + .byte 75 + .byte 81 + .byte 87 + .byte 93 + .byte 101 + .byte 109 + .byte 119 + .byte 130 + .byte 142 + .byte 156 + .byte 172 + .byte 192 + .byte 214 + .byte 241 + .skip 16 + .global __div_table +__div_table: + .skip 16 +/* positive division factors */ + .byte 241 + .byte 214 + .byte 192 + .byte 172 + .byte 156 + .byte 142 + .byte 130 + .byte 119 + .byte 109 + .byte 101 + .byte 93 + .byte 87 + .byte 81 + .byte 75 + .byte 70 + .byte 66 +/* positive division constants */ + .word 31801 + .word 29966 + .word 28383 + .word 26864 + .word 25583 + .word 24408 + .word 23354 + .word 22344 + .word 21384 + .word 20584 + .word 19752 + .word 19104 + .word 18434 + .word 17738 + .word 17136 + .word 16639 diff --git a/arch/sh/lib64/strcpy.S b/arch/sh/lib64/strcpy.S new file mode 100644 index 000000000000..ea7c9c533eea --- /dev/null +++ b/arch/sh/lib64/strcpy.S @@ -0,0 +1,97 @@ +/* Cloned and hacked for uClibc by Paul Mundt, December 2003 */ +/* Modified by SuperH, Inc. September 2003 */ +! Entry: arg0: destination +! arg1: source +! Exit: result: destination +! +! SH5 code Copyright 2002 SuperH Ltd. + +#if __BYTE_ORDER == __LITTLE_ENDIAN +#define SHHI shlld +#define SHLO shlrd +#else +#define SHHI shlrd +#define SHLO shlld +#endif + + .section .text..SHmedia32,"ax" + .globl strcpy + .type strcpy, @function + .align 5 + +strcpy: + + pta/l shortstring,tr1 + ldlo.q r3,0,r4 + ptabs r18,tr4 + shlli r3,3,r7 + addi r2, 8, r0 + mcmpeq.b r4,r63,r6 + SHHI r6,r7,r6 + bnei/u r6,0,tr1 // shortstring + pta/l no_lddst, tr2 + ori r3,-8,r23 + sub r2, r23, r0 + sub r3, r2, r21 + addi r21, 8, r20 + ldx.q r0, r21, r5 + pta/l loop, tr0 + ori r2,-8,r22 + mcmpeq.b r5, r63, r6 + bgt/u r22, r23, tr2 // no_lddst + + // r22 < r23 : Need to do a load from the destination. + // r22 == r23 : Doesn't actually need to load from destination, + // but still can be handled here. + ldlo.q r2, 0, r9 + movi -1, r8 + SHLO r8, r7, r8 + mcmv r4, r8, r9 + stlo.q r2, 0, r9 + beqi/l r6, 0, tr0 // loop + + add r5, r63, r4 + addi r0, 8, r0 + blink tr1, r63 // shortstring +no_lddst: + // r22 > r23: note that for r22 == r23 the sthi.q would clobber + // bytes before the destination region. + stlo.q r2, 0, r4 + SHHI r4, r7, r4 + sthi.q r0, -1, r4 + beqi/l r6, 0, tr0 // loop + + add r5, r63, r4 + addi r0, 8, r0 +shortstring: +#if __BYTE_ORDER != __LITTLE_ENDIAN + pta/l shortstring2,tr1 + byterev r4,r4 +#endif +shortstring2: + st.b r0,-8,r4 + andi r4,0xff,r5 + shlri r4,8,r4 + addi r0,1,r0 + bnei/l r5,0,tr1 + blink tr4,r63 // return + + .balign 8 +loop: + stlo.q r0, 0, r5 + ldx.q r0, r20, r4 + addi r0, 16, r0 + sthi.q r0, -9, r5 + mcmpeq.b r4, r63, r6 + bnei/u r6, 0, tr1 // shortstring + ldx.q r0, r21, r5 + stlo.q r0, -8, r4 + sthi.q r0, -1, r4 + mcmpeq.b r5, r63, r6 + beqi/l r6, 0, tr0 // loop + + add r5, r63, r4 + addi r0, 8, r0 + blink tr1, r63 // shortstring + + .size strcpy,.-strcpy diff --git a/arch/sh/lib64/strlen.S b/arch/sh/lib64/strlen.S new file mode 100644 index 000000000000..cbc0d912e5f3 --- /dev/null +++ b/arch/sh/lib64/strlen.S @@ -0,0 +1,33 @@ +/* + * Simplistic strlen() implementation for SHmedia. + * + * Copyright (C) 2003 Paul Mundt <lethal@linux-sh.org> + */ + + .section .text..SHmedia32,"ax" + .globl strlen + .type strlen,@function + + .balign 16 +strlen: + ptabs r18, tr4 + + /* + * Note: We could easily deal with the NULL case here with a simple + * sanity check, though it seems that the behavior we want is to fault + * in the event that r2 == NULL, so we don't bother. + */ +/* beqi r2, 0, tr4 */ ! Sanity check + + movi -1, r0 + pta/l loop, tr0 +loop: + ld.b r2, 0, r1 + addi r2, 1, r2 + addi r0, 1, r0 + bnei/l r1, 0, tr0 + + or r0, r63, r2 + blink tr4, r63 + + .size strlen,.-strlen diff --git a/arch/sh/lib64/udivdi3.S b/arch/sh/lib64/udivdi3.S new file mode 100644 index 000000000000..6895c0225b85 --- /dev/null +++ b/arch/sh/lib64/udivdi3.S @@ -0,0 +1,120 @@ + .section .text..SHmedia32,"ax" + .align 2 + .global __udivdi3 +__udivdi3: + shlri r3,1,r4 + nsb r4,r22 + shlld r3,r22,r6 + shlri r6,49,r5 + movi 0xffffffffffffbaf1,r21 /* .l shift count 17. */ + sub r21,r5,r1 + mmulfx.w r1,r1,r4 + mshflo.w r1,r63,r1 + sub r63,r22,r20 // r63 == 64 % 64 + mmulfx.w r5,r4,r4 + pta large_divisor,tr0 + addi r20,32,r9 + msub.w r1,r4,r1 + madd.w r1,r1,r1 + mmulfx.w r1,r1,r4 + shlri r6,32,r7 + bgt/u r9,r63,tr0 // large_divisor + mmulfx.w r5,r4,r4 + shlri r2,32+14,r19 + addi r22,-31,r0 + msub.w r1,r4,r1 + + mulu.l r1,r7,r4 + addi r1,-3,r5 + mulu.l r5,r19,r5 + sub r63,r4,r4 // Negate to make sure r1 ends up <= 1/r2 + shlri r4,2,r4 /* chop off leading %0000000000000000 001.00000000000 - or, as + the case may be, %0000000000000000 000.11111111111, still */ + muls.l r1,r4,r4 /* leaving at least one sign bit. */ + mulu.l r5,r3,r8 + mshalds.l r1,r21,r1 + shari r4,26,r4 + shlld r8,r0,r8 + add r1,r4,r1 // 31 bit unsigned reciprocal now in r1 (msb equiv. 0.5) + sub r2,r8,r2 + /* Can do second step of 64 : 32 div now, using r1 and the rest in r2. */ + + shlri r2,22,r21 + mulu.l r21,r1,r21 + shlld r5,r0,r8 + addi r20,30-22,r0 + shlrd r21,r0,r21 + mulu.l r21,r3,r5 + add r8,r21,r8 + mcmpgt.l r21,r63,r21 // See Note 1 + addi r20,30,r0 + mshfhi.l r63,r21,r21 + sub r2,r5,r2 + andc r2,r21,r2 + + /* small divisor: need a third divide step */ + mulu.l r2,r1,r7 + ptabs r18,tr0 + addi r2,1,r2 + shlrd r7,r0,r7 + mulu.l r7,r3,r5 + add r8,r7,r8 + sub r2,r3,r2 + cmpgt r2,r5,r5 + add r8,r5,r2 + /* could test r3 here to check for divide by zero. */ + blink tr0,r63 + +large_divisor: + mmulfx.w r5,r4,r4 + shlrd r2,r9,r25 + shlri r25,32,r8 + msub.w r1,r4,r1 + + mulu.l r1,r7,r4 + addi r1,-3,r5 + mulu.l r5,r8,r5 + sub r63,r4,r4 // Negate to make sure r1 ends up <= 1/r2 + shlri r4,2,r4 /* chop off leading %0000000000000000 001.00000000000 - or, as + the case may be, %0000000000000000 000.11111111111, still */ + muls.l r1,r4,r4 /* leaving at least one sign bit. */ + shlri r5,14-1,r8 + mulu.l r8,r7,r5 + mshalds.l r1,r21,r1 + shari r4,26,r4 + add r1,r4,r1 // 31 bit unsigned reciprocal now in r1 (msb equiv. 0.5) + sub r25,r5,r25 + /* Can do second step of 64 : 32 div now, using r1 and the rest in r25. */ + + shlri r25,22,r21 + mulu.l r21,r1,r21 + pta no_lo_adj,tr0 + addi r22,32,r0 + shlri r21,40,r21 + mulu.l r21,r7,r5 + add r8,r21,r8 + shlld r2,r0,r2 + sub r25,r5,r25 + bgtu/u r7,r25,tr0 // no_lo_adj + addi r8,1,r8 + sub r25,r7,r25 +no_lo_adj: + mextr4 r2,r25,r2 + + /* large_divisor: only needs a few adjustments. */ + mulu.l r8,r6,r5 + ptabs r18,tr0 + /* bubble */ + cmpgtu r5,r2,r5 + sub r8,r5,r2 + blink tr0,r63 + +/* Note 1: To shift the result of the second divide stage so that the result + always fits into 32 bits, yet we still reduce the rest sufficiently + would require a lot of instructions to do the shifts just right. Using + the full 64 bit shift result to multiply with the divisor would require + four extra instructions for the upper 32 bits (shift / mulu / shift / sub). + Fortunately, if the upper 32 bits of the shift result are nonzero, we + know that the rest after taking this partial result into account will + fit into 32 bits. So we just clear the upper 32 bits of the rest if the + upper 32 bits of the partial result are nonzero. */ diff --git a/arch/sh/lib64/udivsi3.S b/arch/sh/lib64/udivsi3.S new file mode 100644 index 000000000000..e68120e4b847 --- /dev/null +++ b/arch/sh/lib64/udivsi3.S @@ -0,0 +1,59 @@ + .global __udivsi3 + .section .text..SHmedia32,"ax" + .align 2 + +/* + inputs: r4,r5 + clobbered: r18,r19,r20,r21,r22,r25,tr0 + result in r0. + */ +__udivsi3: + addz.l r5,r63,r22 + nsb r22,r0 + shlld r22,r0,r25 + shlri r25,48,r25 + movi 0xffffffffffffbb0c,r20 /* shift count eqiv 76 */ + sub r20,r25,r21 + mmulfx.w r21,r21,r19 + mshflo.w r21,r63,r21 + ptabs r18,tr0 + mmulfx.w r25,r19,r19 + sub r20,r0,r0 + /* bubble */ + msub.w r21,r19,r19 + + /* + * It would be nice for scheduling to do this add to r21 before + * the msub.w, but we need a different value for r19 to keep + * errors under control. + */ + addi r19,-2,r21 + mulu.l r4,r21,r18 + mmulfx.w r19,r19,r19 + shlli r21,15,r21 + shlrd r18,r0,r18 + mulu.l r18,r22,r20 + mmacnfx.wl r25,r19,r21 + /* bubble */ + sub r4,r20,r25 + + mulu.l r25,r21,r19 + addi r0,14,r0 + /* bubble */ + shlrd r19,r0,r19 + mulu.l r19,r22,r20 + add r18,r19,r18 + /* bubble */ + sub.l r25,r20,r25 + + mulu.l r25,r21,r19 + addz.l r25,r63,r25 + sub r25,r22,r25 + shlrd r19,r0,r19 + mulu.l r19,r22,r20 + addi r25,1,r25 + add r18,r19,r18 + + cmpgt r25,r20,r25 + add.l r18,r25,r0 + blink tr0,r63 |