diff options
| author | Pavel Kozlov <pavel.kozlov@synopsys.com> | 2022-10-07 13:43:45 +0400 |
|---|---|---|
| committer | Waldemar Brodkorb <wbx@openadk.org> | 2022-10-14 09:47:02 +0200 |
| commit | 663b8a0497c40a20668258bd69db13924c569c41 (patch) | |
| tree | c494a97dedbfa9ae8aa72b3c7f25b05490ec8130 /libc/string/arc/memcmp.S | |
| parent | de6be7bc60f190a0d746945a3a5a143bc93a1a65 (diff) | |
arc: add optimized string functions for ARCv3
Add ability to use optimized versions of string functions for ARCv3 32-bit
CPUs with UCLIBC_HAS_STRING_ARCH_OPT option. Add optimized
memcpy/memset/memcmp code for ARCv3 CPUs based on the code from newlib
and adapt for ARCv3 existed optimized strchr/strcmp/strcpy/strlen.
Link to the Synopsys newlib repo with code for ARCv3 on GitHub:
https://github.com/foss-for-synopsys-dwc-arc-processors/newlib
Signed-off-by: Pavel Kozlov <pavel.kozlov@synopsys.com>
Diffstat (limited to 'libc/string/arc/memcmp.S')
| -rw-r--r-- | libc/string/arc/memcmp.S | 94 |
1 files changed, 93 insertions, 1 deletions
diff --git a/libc/string/arc/memcmp.S b/libc/string/arc/memcmp.S index a60757e7a..20122a296 100644 --- a/libc/string/arc/memcmp.S +++ b/libc/string/arc/memcmp.S @@ -1,5 +1,5 @@ /* - * Copyright (C) 2013 Synopsys, Inc. (www.synopsys.com) + * Copyright (C) 2013, 2022 Synopsys, Inc. (www.synopsys.com) * Copyright (C) 2007 ARC International (UK) LTD * * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB in this tarball. @@ -17,6 +17,8 @@ #endif ENTRY(memcmp) + +#if defined(__ARC700__) || defined(__ARCHS__) or r12,r0,r1 asl_s r12,r12,30 sub r3,r2,1 @@ -149,6 +151,96 @@ ENTRY(memcmp) .Lnil: j_s.d [blink] mov r0,0 + +#elif (__ARC64_ARCH32__) + ;; Based on Synopsys code from newlib's arc64/memcmp.S + cmp r2, 32 + bls.d @.L_compare_1_bytes + mov r3, r0 ; "r0" will be used as return value + + lsr r12, r2, 4 ; counter for 16-byte chunks + xor r13, r13, r13 ; the mask showing inequal registers + +.L_compare_16_bytes: + ld.ab r4, [r3, +4] + ld.ab r5, [r1, +4] + ld.ab r6, [r3, +4] + ld.ab r7, [r1, +4] + ld.ab r8, [r3, +4] + ld.ab r9, [r1, +4] + ld.ab r10, [r3, +4] + ld.ab r11, [r1, +4] + xor.f 0, r4, r5 + xor.ne r13, r13, 0b0001 + xor.f 0, r6, r7 + xor.ne r13, r13, 0b0010 + xor.f 0, r8, r9 + xor.ne r13, r13, 0b0100 + xor.f 0, r10, r11 + xor.ne r13, r13, 0b1000 + brne r13, 0, @.L_unequal_find + dbnz r12, @.L_compare_16_bytes + + ;; Adjusting the pointers because of the extra loads in the end + sub r1, r1, 4 + sub r3, r3, 4 + bmsk_s r2, r2, 3 ; any remaining bytes to compare + +.L_compare_1_bytes: + cmp r2, 0 + jeq.d [blink] + xor_s r0, r0, r0 + +2: + ldb.ab r4, [r3, +1] + ldb.ab r5, [r1, +1] + sub.f r0, r4, r5 + jne [blink] + dbnz r2, @2b + j_s [blink] + + ;; At this point, we want to find the _first_ comparison that marked the + ;; inequality of "lhs" and "rhs" +.L_unequal_find: + ffs r13, r13 + asl r13, r13, 2 + bi [r13] +.L_unequal_r4r5: + mov r1, r4 + b.d @.L_diff_byte_in_regs + mov r2, r5 + nop +.L_unequal_r6r7: + mov r1, r6 + b.d @.L_diff_byte_in_regs + mov r2, r7 + nop +.L_unequal_r8r9: + mov r1, r8 + b.d @.L_diff_byte_in_regs + mov r2, r9 + nop +.L_unequal_r10r11: + mov r1, r10 + mov r2, r11 + + ;; fall-through + ;; If we're here, that means the two operands are not equal. +.L_diff_byte_in_regs: + xor r0, r1, r2 + ffs r0, r0 + and r0, r0, 0x18 + lsr r1, r1, r0 + lsr r2, r2, r0 + bmsk_s r1, r1, 7 + bmsk_s r2, r2, 7 + j_s.d [blink] + sub r0, r1, r2 + +#else +#error "Unsupported ARC CPU type" +#endif + END(memcmp) libc_hidden_def(memcmp) |