1 /* SPDX-License-Identifier: GPL-2.0-only */ << 2 /* 1 /* 3 * Copyright (c) 2011, The Linux Foundation. A !! 2 * This file is subject to the terms and conditions of the GNU General Public >> 3 * License. See the file "COPYING" in the main directory of this archive >> 4 * for more details. >> 5 * >> 6 * Copyright (C) 1998, 1999, 2000 by Ralf Baechle >> 7 * Copyright (C) 1999, 2000 Silicon Graphics, Inc. >> 8 * Copyright (C) 2007 Maciej W. Rozycki 4 */ 9 */ >> 10 #include <asm/asm.h> >> 11 #include <asm/asm-offsets.h> >> 12 #include <asm/regdef.h> >> 13 >> 14 #if LONGSIZE == 4 >> 15 #define LONG_S_L swl >> 16 #define LONG_S_R swr >> 17 #else >> 18 #define LONG_S_L sdl >> 19 #define LONG_S_R sdr >> 20 #endif 5 21 6 !! 22 #define EX(insn,reg,addr,handler) \ 7 /* HEXAGON assembly optimized memset */ !! 23 9: insn reg, addr; \ 8 /* Replaces the standard library function mems !! 24 .section __ex_table,"a"; \ 9 !! 25 PTR 9b, handler; \ 10 !! 26 .previous 11 .macro HEXAGON_OPT_FUNC_BEGIN name !! 27 12 .text !! 28 .macro f_fill64 dst, offset, val, fixup 13 .p2align 4 !! 29 EX(LONG_S, \val, (\offset + 0 * LONGSIZE)(\dst), \fixup) 14 .globl \name !! 30 EX(LONG_S, \val, (\offset + 1 * LONGSIZE)(\dst), \fixup) 15 .type \name, @function !! 31 EX(LONG_S, \val, (\offset + 2 * LONGSIZE)(\dst), \fixup) 16 \name: !! 32 EX(LONG_S, \val, (\offset + 3 * LONGSIZE)(\dst), \fixup) >> 33 EX(LONG_S, \val, (\offset + 4 * LONGSIZE)(\dst), \fixup) >> 34 EX(LONG_S, \val, (\offset + 5 * LONGSIZE)(\dst), \fixup) >> 35 EX(LONG_S, \val, (\offset + 6 * LONGSIZE)(\dst), \fixup) >> 36 EX(LONG_S, \val, (\offset + 7 * LONGSIZE)(\dst), \fixup) >> 37 #if LONGSIZE == 4 >> 38 EX(LONG_S, \val, (\offset + 8 * LONGSIZE)(\dst), \fixup) >> 39 EX(LONG_S, \val, (\offset + 9 * LONGSIZE)(\dst), \fixup) >> 40 EX(LONG_S, \val, (\offset + 10 * LONGSIZE)(\dst), \fixup) >> 41 EX(LONG_S, \val, (\offset + 11 * LONGSIZE)(\dst), \fixup) >> 42 EX(LONG_S, \val, (\offset + 12 * LONGSIZE)(\dst), \fixup) >> 43 EX(LONG_S, \val, (\offset + 13 * LONGSIZE)(\dst), \fixup) >> 44 EX(LONG_S, \val, (\offset + 14 * LONGSIZE)(\dst), \fixup) >> 45 EX(LONG_S, \val, (\offset + 15 * LONGSIZE)(\dst), \fixup) >> 46 #endif 17 .endm 47 .endm 18 48 19 .macro HEXAGON_OPT_FUNC_FINISH name !! 49 /* 20 .size \name, . - \name !! 50 * memset(void *s, int c, size_t n) 21 .endm !! 51 * >> 52 * a0: start of area to clear >> 53 * a1: char to fill with >> 54 * a2: size of area to clear >> 55 */ >> 56 .set noreorder >> 57 .align 5 >> 58 LEAF(memset) >> 59 beqz a1, 1f >> 60 move v0, a0 /* result */ >> 61 >> 62 andi a1, 0xff /* spread fillword */ >> 63 LONG_SLL t1, a1, 8 >> 64 or a1, t1 >> 65 LONG_SLL t1, a1, 16 >> 66 #if LONGSIZE == 8 >> 67 or a1, t1 >> 68 LONG_SLL t1, a1, 32 >> 69 #endif >> 70 or a1, t1 >> 71 1: 22 72 23 /* FUNCTION: memset (v2 version) */ !! 73 FEXPORT(__bzero) 24 #if __HEXAGON_ARCH__ < 3 !! 74 sltiu t0, a2, LONGSIZE /* very small region? */ 25 HEXAGON_OPT_FUNC_BEGIN memset !! 75 bnez t0, .Lsmall_memset 26 { !! 76 andi t0, a0, LONGMASK /* aligned? */ 27 r6 = #8 !! 77 28 r7 = extractu(r0, #3 , #0) !! 78 #ifndef CONFIG_CPU_DADDI_WORKAROUNDS 29 p0 = cmp.eq(r2, #0) !! 79 beqz t0, 1f 30 p1 = cmp.gtu(r2, #7) !! 80 PTR_SUBU t0, LONGSIZE /* alignment in bytes */ 31 } !! 81 #else 32 { !! 82 .set noat 33 r4 = vsplatb(r1) !! 83 li AT, LONGSIZE 34 r8 = r0 /* leave r0 !! 84 beqz t0, 1f 35 r9 = sub(r6, r7) /* bytes unt !! 85 PTR_SUBU t0, AT /* alignment in bytes */ 36 if p0 jumpr r31 /* count == !! 86 .set at 37 } << 38 { << 39 r3 = #0 << 40 r7 = #0 << 41 p0 = tstbit(r9, #0) << 42 if p1 jump 2f /* skip byte loo << 43 } << 44 << 45 /* less than 8 bytes to set, so just set a byt << 46 << 47 loop0(1f, r2) /* byte loop */ << 48 .falign << 49 1: /* byte loop */ << 50 { << 51 memb(r8++#1) = r4 << 52 }:endloop0 << 53 jumpr r31 << 54 .falign << 55 2: /* skip byte loop */ << 56 { << 57 r6 = #1 << 58 p0 = tstbit(r9, #1) << 59 p1 = cmp.eq(r2, #1) << 60 if !p0 jump 3f /* skip initial << 61 } << 62 { << 63 memb(r8++#1) = r4 << 64 r3:2 = sub(r3:2, r7:6) << 65 if p1 jumpr r31 << 66 } << 67 .falign << 68 3: /* skip initial byte store */ << 69 { << 70 r6 = #2 << 71 p0 = tstbit(r9, #2) << 72 p1 = cmp.eq(r2, #2) << 73 if !p0 jump 4f /* skip initial << 74 } << 75 { << 76 memh(r8++#2) = r4 << 77 r3:2 = sub(r3:2, r7:6) << 78 if p1 jumpr r31 << 79 } << 80 .falign << 81 4: /* skip initial half store */ << 82 { << 83 r6 = #4 << 84 p0 = cmp.gtu(r2, #7) << 85 p1 = cmp.eq(r2, #4) << 86 if !p0 jump 5f /* skip initial << 87 } << 88 { << 89 memw(r8++#4) = r4 << 90 r3:2 = sub(r3:2, r7:6) << 91 p0 = cmp.gtu(r2, #11) << 92 if p1 jumpr r31 << 93 } << 94 .falign << 95 5: /* skip initial word store */ << 96 { << 97 r10 = lsr(r2, #3) << 98 p1 = cmp.eq(r3, #1) << 99 if !p0 jump 7f /* skip double << 100 } << 101 { << 102 r5 = r4 << 103 r6 = #8 << 104 loop0(6f, r10) /* double loop << 105 } << 106 << 107 /* set bytes a double word at a time */ << 108 << 109 .falign << 110 6: /* double loop */ << 111 { << 112 memd(r8++#8) = r5:4 << 113 r3:2 = sub(r3:2, r7:6) << 114 p1 = cmp.eq(r2, #8) << 115 }:endloop0 << 116 .falign << 117 7: /* skip double loop */ << 118 { << 119 p0 = tstbit(r2, #2) << 120 if p1 jumpr r31 << 121 } << 122 { << 123 r6 = #4 << 124 p0 = tstbit(r2, #1) << 125 p1 = cmp.eq(r2, #4) << 126 if !p0 jump 8f /* skip final w << 127 } << 128 { << 129 memw(r8++#4) = r4 << 130 r3:2 = sub(r3:2, r7:6) << 131 if p1 jumpr r31 << 132 } << 133 .falign << 134 8: /* skip final word store */ << 135 { << 136 p1 = cmp.eq(r2, #2) << 137 if !p0 jump 9f /* skip final h << 138 } << 139 { << 140 memh(r8++#2) = r4 << 141 if p1 jumpr r31 << 142 } << 143 .falign << 144 9: /* skip final half store */ << 145 { << 146 memb(r8++#1) = r4 << 147 jumpr r31 << 148 } << 149 HEXAGON_OPT_FUNC_FINISH memset << 150 #endif 87 #endif 151 88 >> 89 R10KCBARRIER(0(ra)) >> 90 #ifdef __MIPSEB__ >> 91 EX(LONG_S_L, a1, (a0), .Lfirst_fixup) /* make word/dword aligned */ >> 92 #endif >> 93 #ifdef __MIPSEL__ >> 94 EX(LONG_S_R, a1, (a0), .Lfirst_fixup) /* make word/dword aligned */ >> 95 #endif >> 96 PTR_SUBU a0, t0 /* long align ptr */ >> 97 PTR_ADDU a2, t0 /* correct size */ 152 98 153 /* FUNCTION: memset (v3 and higher version) !! 99 1: ori t1, a2, 0x3f /* # of full blocks */ 154 #if __HEXAGON_ARCH__ >= 3 !! 100 xori t1, 0x3f 155 HEXAGON_OPT_FUNC_BEGIN memset !! 101 beqz t1, .Lmemset_partial /* no block to fill */ 156 { !! 102 andi t0, a2, 0x40-LONGSIZE 157 r7=vsplatb(r1) !! 103 158 r6 = r0 !! 104 PTR_ADDU t1, a0 /* end address */ 159 if (r2==#0) jump:nt .L1 !! 105 .set reorder 160 } !! 106 1: PTR_ADDIU a0, 64 161 { !! 107 R10KCBARRIER(0(ra)) 162 r5:4=combine(r7,r7) !! 108 f_fill64 a0, -64, a1, .Lfwd_fixup 163 p0 = cmp.gtu(r2,#8) !! 109 bne t1, a0, 1b 164 if (p0.new) jump:nt .L3 !! 110 .set noreorder 165 } !! 111 166 { !! 112 .Lmemset_partial: 167 r3 = r0 !! 113 R10KCBARRIER(0(ra)) 168 loop0(.L47,r2) !! 114 PTR_LA t1, 2f /* where to start */ 169 } !! 115 #if LONGSIZE == 4 170 .falign !! 116 PTR_SUBU t1, t0 171 .L47: !! 117 #else 172 { !! 118 .set noat 173 memb(r3++#1) = r1 !! 119 LONG_SRL AT, t0, 1 174 }:endloop0 /* start=.L47 */ !! 120 PTR_SUBU t1, AT 175 jumpr r31 !! 121 .set at 176 .L3: << 177 { << 178 p0 = tstbit(r0,#0) << 179 if (!p0.new) jump:nt .L8 << 180 p1 = cmp.eq(r2, #1) << 181 } << 182 { << 183 r6 = add(r0, #1) << 184 r2 = add(r2,#-1) << 185 memb(r0) = r1 << 186 if (p1) jump .L1 << 187 } << 188 .L8: << 189 { << 190 p0 = tstbit(r6,#1) << 191 if (!p0.new) jump:nt .L10 << 192 } << 193 { << 194 r2 = add(r2,#-2) << 195 memh(r6++#2) = r7 << 196 p0 = cmp.eq(r2, #2) << 197 if (p0.new) jump:nt .L1 << 198 } << 199 .L10: << 200 { << 201 p0 = tstbit(r6,#2) << 202 if (!p0.new) jump:nt .L12 << 203 } << 204 { << 205 r2 = add(r2,#-4) << 206 memw(r6++#4) = r7 << 207 p0 = cmp.eq(r2, #4) << 208 if (p0.new) jump:nt .L1 << 209 } << 210 .L12: << 211 { << 212 p0 = cmp.gtu(r2,#127) << 213 if (!p0.new) jump:nt .L14 << 214 } << 215 r3 = and(r6,#31) << 216 if (r3==#0) jump:nt .L17 << 217 { << 218 memd(r6++#8) = r5:4 << 219 r2 = add(r2,#-8) << 220 } << 221 r3 = and(r6,#31) << 222 if (r3==#0) jump:nt .L17 << 223 { << 224 memd(r6++#8) = r5:4 << 225 r2 = add(r2,#-8) << 226 } << 227 r3 = and(r6,#31) << 228 if (r3==#0) jump:nt .L17 << 229 { << 230 memd(r6++#8) = r5:4 << 231 r2 = add(r2,#-8) << 232 } << 233 .L17: << 234 { << 235 r3 = lsr(r2,#5) << 236 if (r1!=#0) jump:nt .L18 << 237 } << 238 { << 239 r8 = r3 << 240 r3 = r6 << 241 loop0(.L46,r3) << 242 } << 243 .falign << 244 .L46: << 245 { << 246 dczeroa(r6) << 247 r6 = add(r6,#32) << 248 r2 = add(r2,#-32) << 249 }:endloop0 /* start=.L46 */ << 250 .L14: << 251 { << 252 p0 = cmp.gtu(r2,#7) << 253 if (!p0.new) jump:nt .L28 << 254 r8 = lsr(r2,#3) << 255 } << 256 loop0(.L44,r8) << 257 .falign << 258 .L44: << 259 { << 260 memd(r6++#8) = r5:4 << 261 r2 = add(r2,#-8) << 262 }:endloop0 /* start=.L44 */ << 263 .L28: << 264 { << 265 p0 = tstbit(r2,#2) << 266 if (!p0.new) jump:nt .L33 << 267 } << 268 { << 269 r2 = add(r2,#-4) << 270 memw(r6++#4) = r7 << 271 } << 272 .L33: << 273 { << 274 p0 = tstbit(r2,#1) << 275 if (!p0.new) jump:nt .L35 << 276 } << 277 { << 278 r2 = add(r2,#-2) << 279 memh(r6++#2) = r7 << 280 } << 281 .L35: << 282 p0 = cmp.eq(r2,#1) << 283 if (p0) memb(r6) = r1 << 284 .L1: << 285 jumpr r31 << 286 .L18: << 287 loop0(.L45,r3) << 288 .falign << 289 .L45: << 290 dczeroa(r6) << 291 { << 292 memd(r6++#8) = r5:4 << 293 r2 = add(r2,#-32) << 294 } << 295 memd(r6++#8) = r5:4 << 296 memd(r6++#8) = r5:4 << 297 { << 298 memd(r6++#8) = r5:4 << 299 }:endloop0 /* start=.L45 */ << 300 jump .L14 << 301 HEXAGON_OPT_FUNC_FINISH memset << 302 #endif 122 #endif >> 123 jr t1 >> 124 PTR_ADDU a0, t0 /* dest ptr */ >> 125 >> 126 .set push >> 127 .set noreorder >> 128 .set nomacro >> 129 f_fill64 a0, -64, a1, .Lpartial_fixup /* ... but first do longs ... */ >> 130 2: .set pop >> 131 andi a2, LONGMASK /* At most one long to go */ >> 132 >> 133 beqz a2, 1f >> 134 PTR_ADDU a0, a2 /* What's left */ >> 135 R10KCBARRIER(0(ra)) >> 136 #ifdef __MIPSEB__ >> 137 EX(LONG_S_R, a1, -1(a0), .Llast_fixup) >> 138 #endif >> 139 #ifdef __MIPSEL__ >> 140 EX(LONG_S_L, a1, -1(a0), .Llast_fixup) >> 141 #endif >> 142 1: jr ra >> 143 move a2, zero >> 144 >> 145 .Lsmall_memset: >> 146 beqz a2, 2f >> 147 PTR_ADDU t1, a0, a2 >> 148 >> 149 1: PTR_ADDIU a0, 1 /* fill bytewise */ >> 150 R10KCBARRIER(0(ra)) >> 151 bne t1, a0, 1b >> 152 sb a1, -1(a0) >> 153 >> 154 2: jr ra /* done */ >> 155 move a2, zero >> 156 END(memset) >> 157 >> 158 .Lfirst_fixup: >> 159 jr ra >> 160 nop >> 161 >> 162 .Lfwd_fixup: >> 163 PTR_L t0, TI_TASK($28) >> 164 LONG_L t0, THREAD_BUADDR(t0) >> 165 andi a2, 0x3f >> 166 LONG_ADDU a2, t1 >> 167 jr ra >> 168 LONG_SUBU a2, t0 >> 169 >> 170 .Lpartial_fixup: >> 171 PTR_L t0, TI_TASK($28) >> 172 LONG_L t0, THREAD_BUADDR(t0) >> 173 andi a2, LONGMASK >> 174 LONG_ADDU a2, t1 >> 175 jr ra >> 176 LONG_SUBU a2, t0 >> 177 >> 178 .Llast_fixup: >> 179 jr ra >> 180 andi v1, a2, LONGMASK
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.