Linux/arch/loongarch/lib/memset.S

Version: ~ [ linux-6.12-rc7 ] ~ [ linux-6.11.7 ] ~ [ linux-6.10.14 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.60 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.116 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.171 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.229 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.285 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.323 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.12 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

Diff markup

1 /* SPDX-License-Identifier: GPL-2.0 */ 1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* !! 2 /* linux/arch/sparc/lib/memset.S: Sparc optimized memset, bzero and clear_user code 3 * Copyright (C) 2020-2022 Loongson Technology !! 3 * Copyright (C) 1991,1996 Free Software Foundation 4 */ !! 4 * Copyright (C) 1996,1997 Jakub Jelinek (jj@sunsite.mff.cuni.cz) 5 !! 5 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu) 6 #include <linux/export.h> << 7 #include <asm/alternative-asm.h> << 8 #include <asm/asm.h> << 9 #include <asm/asmmacro.h> << 10 #include <asm/cpu.h> << 11 #include <asm/regdef.h> << 12 #include <asm/unwind_hints.h> << 13 << 14 .macro fill_to_64 r0 << 15 bstrins.d \r0, \r0, 15, 8 << 16 bstrins.d \r0, \r0, 31, 16 << 17 bstrins.d \r0, \r0, 63, 32 << 18 .endm << 19 << 20 .section .noinstr.text, "ax" << 21 << 22 SYM_FUNC_START(memset) << 23 /* << 24 * Some CPUs support hardware unaligne << 25 */ << 26 ALTERNATIVE "b __memset_generic", << 27 "b __memset_fast", CPU << 28 SYM_FUNC_END(memset) << 29 SYM_FUNC_ALIAS(__memset, memset) << 30 << 31 EXPORT_SYMBOL(memset) << 32 EXPORT_SYMBOL(__memset) << 33 << 34 _ASM_NOKPROBE(memset) << 35 _ASM_NOKPROBE(__memset) << 36 << 37 /* << 38 * void *__memset_generic(void *s, int c, size << 39 * 6 * 40 * a0: s !! 7 * Calls to memset returns initial %o0. Calls to bzero returns 0, if ok, and 41 * a1: c !! 8 * number of bytes not yet set if exception occurs and we were called as 42 * a2: n !! 9 * clear_user. 43 */ 10 */ 44 SYM_FUNC_START(__memset_generic) << 45 move a3, a0 << 46 beqz a2, 2f << 47 << 48 1: st.b a1, a0, 0 << 49 addi.d a0, a0, 1 << 50 addi.d a2, a2, -1 << 51 bgt a2, zero, 1b << 52 << 53 2: move a0, a3 << 54 jr ra << 55 SYM_FUNC_END(__memset_generic) << 56 _ASM_NOKPROBE(__memset_generic) << 57 << 58 /* << 59 * void *__memset_fast(void *s, int c, size_t << 60 * << 61 * a0: s << 62 * a1: c << 63 * a2: n << 64 */ << 65 SYM_FUNC_START(__memset_fast) << 66 /* fill a1 to 64 bits */ << 67 fill_to_64 a1 << 68 << 69 sltui t0, a2, 9 << 70 bnez t0, .Lsmall << 71 << 72 add.d a2, a0, a2 << 73 st.d a1, a0, 0 << 74 << 75 /* align up address */ << 76 addi.d a3, a0, 8 << 77 bstrins.d a3, zero, 2, 0 << 78 << 79 addi.d a4, a2, -64 << 80 bgeu a3, a4, .Llt64 << 81 << 82 /* set 64 bytes at a time */ << 83 .Lloop64: << 84 st.d a1, a3, 0 << 85 st.d a1, a3, 8 << 86 st.d a1, a3, 16 << 87 st.d a1, a3, 24 << 88 st.d a1, a3, 32 << 89 st.d a1, a3, 40 << 90 st.d a1, a3, 48 << 91 st.d a1, a3, 56 << 92 addi.d a3, a3, 64 << 93 bltu a3, a4, .Lloop64 << 94 << 95 /* set the remaining bytes */ << 96 .Llt64: << 97 addi.d a4, a2, -32 << 98 bgeu a3, a4, .Llt32 << 99 st.d a1, a3, 0 << 100 st.d a1, a3, 8 << 101 st.d a1, a3, 16 << 102 st.d a1, a3, 24 << 103 addi.d a3, a3, 32 << 104 << 105 .Llt32: << 106 addi.d a4, a2, -16 << 107 bgeu a3, a4, .Llt16 << 108 st.d a1, a3, 0 << 109 st.d a1, a3, 8 << 110 addi.d a3, a3, 16 << 111 << 112 .Llt16: << 113 addi.d a4, a2, -8 << 114 bgeu a3, a4, .Llt8 << 115 st.d a1, a3, 0 << 116 << 117 .Llt8: << 118 st.d a1, a2, -8 << 119 << 120 /* return */ << 121 jr ra << 122 << 123 .align 4 << 124 .Lsmall: << 125 pcaddi t0, 4 << 126 slli.d a2, a2, 4 << 127 add.d t0, t0, a2 << 128 jr t0 << 129 << 130 .align 4 << 131 0: jr ra << 132 11 133 .align 4 !! 12 #include <linux/export.h> 134 1: st.b a1, a0, 0 !! 13 #include <asm/ptrace.h> 135 jr ra << 136 << 137 .align 4 << 138 2: st.h a1, a0, 0 << 139 jr ra << 140 << 141 .align 4 << 142 3: st.h a1, a0, 0 << 143 st.b a1, a0, 2 << 144 jr ra << 145 << 146 .align 4 << 147 4: st.w a1, a0, 0 << 148 jr ra << 149 << 150 .align 4 << 151 5: st.w a1, a0, 0 << 152 st.b a1, a0, 4 << 153 jr ra << 154 << 155 .align 4 << 156 6: st.w a1, a0, 0 << 157 st.h a1, a0, 4 << 158 jr ra << 159 << 160 .align 4 << 161 7: st.w a1, a0, 0 << 162 st.w a1, a0, 3 << 163 jr ra << 164 14 165 .align 4 !! 15 /* Work around cpp -rob */ 166 8: st.d a1, a0, 0 !! 16 #define ALLOC #alloc 167 jr ra !! 17 #define EXECINSTR #execinstr 168 SYM_FUNC_END(__memset_fast) !! 18 #define EX(x,y,a,b) \ 169 _ASM_NOKPROBE(__memset_fast) !! 19 98: x,y; \ >> 20 .section .fixup,ALLOC,EXECINSTR; \ >> 21 .align 4; \ >> 22 99: retl; \ >> 23 a, b, %o0; \ >> 24 .section __ex_table,ALLOC; \ >> 25 .align 4; \ >> 26 .word 98b, 99b; \ >> 27 .text; \ >> 28 .align 4 >> 29 >> 30 #define STORE(source, base, offset, n) \ >> 31 98: std source, [base + offset + n]; \ >> 32 .section .fixup,ALLOC,EXECINSTR; \ >> 33 .align 4; \ >> 34 99: ba 30f; \ >> 35 sub %o3, n - offset, %o3; \ >> 36 .section __ex_table,ALLOC; \ >> 37 .align 4; \ >> 38 .word 98b, 99b; \ >> 39 .text; \ >> 40 .align 4; >> 41 >> 42 #define STORE_LAST(source, base, offset, n) \ >> 43 EX(std source, [base - offset - n], \ >> 44 add %o1, offset + n); >> 45 >> 46 /* Please don't change these macros, unless you change the logic >> 47 * in the .fixup section below as well. >> 48 * Store 64 bytes at (BASE + OFFSET) using value SOURCE. */ >> 49 #define ZERO_BIG_BLOCK(base, offset, source) \ >> 50 STORE(source, base, offset, 0x00); \ >> 51 STORE(source, base, offset, 0x08); \ >> 52 STORE(source, base, offset, 0x10); \ >> 53 STORE(source, base, offset, 0x18); \ >> 54 STORE(source, base, offset, 0x20); \ >> 55 STORE(source, base, offset, 0x28); \ >> 56 STORE(source, base, offset, 0x30); \ >> 57 STORE(source, base, offset, 0x38); >> 58 >> 59 #define ZERO_LAST_BLOCKS(base, offset, source) \ >> 60 STORE_LAST(source, base, offset, 0x38); \ >> 61 STORE_LAST(source, base, offset, 0x30); \ >> 62 STORE_LAST(source, base, offset, 0x28); \ >> 63 STORE_LAST(source, base, offset, 0x20); \ >> 64 STORE_LAST(source, base, offset, 0x18); \ >> 65 STORE_LAST(source, base, offset, 0x10); \ >> 66 STORE_LAST(source, base, offset, 0x08); \ >> 67 STORE_LAST(source, base, offset, 0x00); >> 68 >> 69 .text >> 70 .align 4 >> 71 >> 72 .globl __bzero_begin >> 73 __bzero_begin: >> 74 >> 75 .globl __bzero >> 76 .type __bzero,#function >> 77 .globl memset >> 78 EXPORT_SYMBOL(__bzero) >> 79 EXPORT_SYMBOL(memset) >> 80 memset: >> 81 mov %o0, %g1 >> 82 mov 1, %g4 >> 83 and %o1, 0xff, %g3 >> 84 sll %g3, 8, %g2 >> 85 or %g3, %g2, %g3 >> 86 sll %g3, 16, %g2 >> 87 or %g3, %g2, %g3 >> 88 b 1f >> 89 mov %o2, %o1 >> 90 3: >> 91 cmp %o2, 3 >> 92 be 2f >> 93 EX(stb %g3, [%o0], sub %o1, 0) >> 94 >> 95 cmp %o2, 2 >> 96 be 2f >> 97 EX(stb %g3, [%o0 + 0x01], sub %o1, 1) >> 98 >> 99 EX(stb %g3, [%o0 + 0x02], sub %o1, 2) >> 100 2: >> 101 sub %o2, 4, %o2 >> 102 add %o1, %o2, %o1 >> 103 b 4f >> 104 sub %o0, %o2, %o0 >> 105 >> 106 __bzero: >> 107 clr %g4 >> 108 mov %g0, %g3 >> 109 1: >> 110 cmp %o1, 7 >> 111 bleu 7f >> 112 andcc %o0, 3, %o2 >> 113 >> 114 bne 3b >> 115 4: >> 116 andcc %o0, 4, %g0 >> 117 >> 118 be 2f >> 119 mov %g3, %g2 >> 120 >> 121 EX(st %g3, [%o0], sub %o1, 0) >> 122 sub %o1, 4, %o1 >> 123 add %o0, 4, %o0 >> 124 2: >> 125 andcc %o1, 0xffffff80, %o3 ! Now everything is 8 aligned and o1 is len to run >> 126 be 9f >> 127 andcc %o1, 0x78, %o2 >> 128 10: >> 129 ZERO_BIG_BLOCK(%o0, 0x00, %g2) >> 130 subcc %o3, 128, %o3 >> 131 ZERO_BIG_BLOCK(%o0, 0x40, %g2) >> 132 bne 10b >> 133 add %o0, 128, %o0 >> 134 >> 135 orcc %o2, %g0, %g0 >> 136 9: >> 137 be 13f >> 138 andcc %o1, 7, %o1 >> 139 >> 140 srl %o2, 1, %o3 >> 141 set 13f, %o4 >> 142 sub %o4, %o3, %o4 >> 143 jmp %o4 >> 144 add %o0, %o2, %o0 >> 145 >> 146 ZERO_LAST_BLOCKS(%o0, 0x48, %g2) >> 147 ZERO_LAST_BLOCKS(%o0, 0x08, %g2) >> 148 13: >> 149 be 8f >> 150 andcc %o1, 4, %g0 >> 151 >> 152 be 1f >> 153 andcc %o1, 2, %g0 >> 154 >> 155 EX(st %g3, [%o0], and %o1, 7) >> 156 add %o0, 4, %o0 >> 157 1: >> 158 be 1f >> 159 andcc %o1, 1, %g0 >> 160 >> 161 EX(sth %g3, [%o0], and %o1, 3) >> 162 add %o0, 2, %o0 >> 163 1: >> 164 bne,a 8f >> 165 EX(stb %g3, [%o0], and %o1, 1) >> 166 8: >> 167 b 0f >> 168 nop >> 169 7: >> 170 be 13b >> 171 orcc %o1, 0, %g0 >> 172 >> 173 be 0f >> 174 8: >> 175 add %o0, 1, %o0 >> 176 subcc %o1, 1, %o1 >> 177 bne 8b >> 178 EX(stb %g3, [%o0 - 1], add %o1, 1) >> 179 0: >> 180 andcc %g4, 1, %g0 >> 181 be 5f >> 182 nop >> 183 retl >> 184 mov %g1, %o0 >> 185 5: >> 186 retl >> 187 clr %o0 >> 188 >> 189 .section .fixup,#alloc,#execinstr >> 190 .align 4 >> 191 30: >> 192 and %o1, 0x7f, %o1 >> 193 retl >> 194 add %o3, %o1, %o0 170 195 171 STACK_FRAME_NON_STANDARD __memset_fast !! 196 .globl __bzero_end >> 197 __bzero_end:

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

TOMOYO Linux Cross Reference
Linux/arch/loongarch/lib/memset.S

Diff markup

Differences between /arch/loongarch/lib/memset.S (Architecture i386) and /arch/sparc/lib/memset.S (Architecture sparc)

TOMOYO Linux Cross Reference Linux/arch/loongarch/lib/memset.S

Diff markup

Differences between /arch/loongarch/lib/memset.S (Architecture i386) and /arch/sparc/lib/memset.S (Architecture sparc)

TOMOYO Linux Cross Reference
Linux/arch/loongarch/lib/memset.S