devel

devel@uclibc-ng.org

2 participants
1165 discussions

[PATCH] arc: Merge ARCv2 string routines in generic ARC .S files
by Alexey Brodkin 24 Aug '17

24 Aug '17

In cde74b83f9b2 "ARC: remove special CFLAGS/LDFLAGS handling" we got rid of CONFIG_ARC_CPU_HS which was used to select ARCv2-specific implementation of optimized string routines. So now ARCv2-tuned memset/memcpy/strcmp are not used, instead those for ARC700 used for both ARC700 and ARCHS. Without uClibc config option we may only tell which CPU type we're targeting by built-in defines of GCC. I.e. no more conditional file inclusion in Makefiles. That leaves us only one option - merge both implementations in 1 file and use ifdefs. Signed-off-by: Alexey Brodkin <abrodkin(a)synopsys.com> --- FWIW I run that change through LMbench, LTP and uClibc-ng-testsuite and didn't see any significant differences or regressions. extra/Configs/Config.in | 1 - libc/string/arc/arcv2/memcpy.S | 236 ---------------------------------------- libc/string/arc/arcv2/memset.S | 115 -------------------- libc/string/arc/arcv2/strcmp.S | 83 -------------- libc/string/arc/memcpy.S | 238 ++++++++++++++++++++++++++++++++++++++++- libc/string/arc/memset.S | 115 +++++++++++++++++++- libc/string/arc/strcmp.S | 81 +++++++++++++- 7 files changed, 428 insertions(+), 441 deletions(-) delete mode 100644 libc/string/arc/arcv2/memcpy.S delete mode 100644 libc/string/arc/arcv2/memset.S delete mode 100644 libc/string/arc/arcv2/strcmp.S diff --git a/extra/Configs/Config.in b/extra/Configs/Config.in index 59ef31c47dec..ce832b55bcc7 100644 --- a/extra/Configs/Config.in +++ b/extra/Configs/Config.in @@ -250,7 +250,6 @@ config TARGET_SUBARCH default "i486" if CONFIG_486 default "i586" if CONFIG_586 default "i686" if CONFIG_686 - default "arcv2" if CONFIG_ARC_CPU_HS default "" source "extra/Configs/Config.in.arch" diff --git a/libc/string/arc/arcv2/memcpy.S b/libc/string/arc/arcv2/memcpy.S deleted file mode 100644 index ba29e8790721..000000000000 --- a/libc/string/arc/arcv2/memcpy.S +++ /dev/null @@ -1,236 +0,0 @@ -/* - * Copyright (C) 2014-15 Synopsys, Inc. (www.synopsys.com) - * - * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB in this tarball. - */ - -#include <features.h> -#include <sysdep.h> - -#ifdef __LITTLE_ENDIAN__ -# define SHIFT_1(RX,RY,IMM) asl RX, RY, IMM ; << -# define SHIFT_2(RX,RY,IMM) lsr RX, RY, IMM ; >> -# define MERGE_1(RX,RY,IMM) asl RX, RY, IMM -# define MERGE_2(RX,RY,IMM) -# define EXTRACT_1(RX,RY,IMM) and RX, RY, 0xFFFF -# define EXTRACT_2(RX,RY,IMM) lsr RX, RY, IMM -#else -# define SHIFT_1(RX,RY,IMM) lsr RX, RY, IMM ; >> -# define SHIFT_2(RX,RY,IMM) asl RX, RY, IMM ; << -# define MERGE_1(RX,RY,IMM) asl RX, RY, IMM ; << -# define MERGE_2(RX,RY,IMM) asl RX, RY, IMM ; << -# define EXTRACT_1(RX,RY,IMM) lsr RX, RY, IMM -# define EXTRACT_2(RX,RY,IMM) lsr RX, RY, 0x08 -#endif - -#if defined(__LL64__) || defined(__ARC_LL64__) -# define PREFETCH_READ(RX) prefetch [RX, 56] -# define PREFETCH_WRITE(RX) prefetchw [RX, 64] -# define LOADX(DST,RX) ldd.ab DST, [RX, 8] -# define STOREX(SRC,RX) std.ab SRC, [RX, 8] -# define ZOLSHFT 5 -# define ZOLAND 0x1F -#else -# define PREFETCH_READ(RX) prefetch [RX, 28] -# define PREFETCH_WRITE(RX) prefetchw [RX, 32] -# define LOADX(DST,RX) ld.ab DST, [RX, 4] -# define STOREX(SRC,RX) st.ab SRC, [RX, 4] -# define ZOLSHFT 4 -# define ZOLAND 0xF -#endif - -ENTRY(memcpy) - prefetch [r1] ; Prefetch the read location - prefetchw [r0] ; Prefetch the write location - mov.f 0, r2 -;;; if size is zero - jz.d [blink] - mov r3, r0 ; don't clobber ret val - -;;; if size <= 8 - cmp r2, 8 - bls.d @.Lsmallchunk - mov.f lp_count, r2 - - and.f r4, r0, 0x03 - rsub lp_count, r4, 4 - lpnz @.Laligndestination - ;; LOOP BEGIN - ldb.ab r5, [r1,1] - sub r2, r2, 1 - stb.ab r5, [r3,1] -.Laligndestination: - -;;; Check the alignment of the source - and.f r4, r1, 0x03 - bnz.d @.Lsourceunaligned - -;;; CASE 0: Both source and destination are 32bit aligned -;;; Convert len to Dwords, unfold x4 - lsr.f lp_count, r2, ZOLSHFT - lpnz @.Lcopy32_64bytes - ;; LOOP START - LOADX (r6, r1) - PREFETCH_READ (r1) - PREFETCH_WRITE (r3) - LOADX (r8, r1) - LOADX (r10, r1) - LOADX (r4, r1) - STOREX (r6, r3) - STOREX (r8, r3) - STOREX (r10, r3) - STOREX (r4, r3) -.Lcopy32_64bytes: - - and.f lp_count, r2, ZOLAND ;Last remaining 31 bytes -.Lsmallchunk: - lpnz @.Lcopyremainingbytes - ;; LOOP START - ldb.ab r5, [r1,1] - stb.ab r5, [r3,1] -.Lcopyremainingbytes: - - j [blink] -;;; END CASE 0 - -.Lsourceunaligned: - cmp r4, 2 - beq.d @.LunalignedOffby2 - sub r2, r2, 1 - - bhi.d @.LunalignedOffby3 - ldb.ab r5, [r1, 1] - -;;; CASE 1: The source is unaligned, off by 1 - ;; Hence I need to read 1 byte for a 16bit alignment - ;; and 2bytes to reach 32bit alignment - ldh.ab r6, [r1, 2] - sub r2, r2, 2 - ;; Convert to words, unfold x2 - lsr.f lp_count, r2, 3 - MERGE_1 (r6, r6, 8) - MERGE_2 (r5, r5, 24) - or r5, r5, r6 - - ;; Both src and dst are aligned - lpnz @.Lcopy8bytes_1 - ;; LOOP START - ld.ab r6, [r1, 4] - prefetch [r1, 28] ;Prefetch the next read location - ld.ab r8, [r1,4] - prefetchw [r3, 32] ;Prefetch the next write location - - SHIFT_1 (r7, r6, 24) - or r7, r7, r5 - SHIFT_2 (r5, r6, 8) - - SHIFT_1 (r9, r8, 24) - or r9, r9, r5 - SHIFT_2 (r5, r8, 8) - - st.ab r7, [r3, 4] - st.ab r9, [r3, 4] -.Lcopy8bytes_1: - - ;; Write back the remaining 16bits - EXTRACT_1 (r6, r5, 16) - sth.ab r6, [r3, 2] - ;; Write back the remaining 8bits - EXTRACT_2 (r5, r5, 16) - stb.ab r5, [r3, 1] - - and.f lp_count, r2, 0x07 ;Last 8bytes - lpnz @.Lcopybytewise_1 - ;; LOOP START - ldb.ab r6, [r1,1] - stb.ab r6, [r3,1] -.Lcopybytewise_1: - j [blink] - -.LunalignedOffby2: -;;; CASE 2: The source is unaligned, off by 2 - ldh.ab r5, [r1, 2] - sub r2, r2, 1 - - ;; Both src and dst are aligned - ;; Convert to words, unfold x2 - lsr.f lp_count, r2, 3 -#ifdef __BIG_ENDIAN__ - asl.nz r5, r5, 16 -#endif - lpnz @.Lcopy8bytes_2 - ;; LOOP START - ld.ab r6, [r1, 4] - prefetch [r1, 28] ;Prefetch the next read location - ld.ab r8, [r1,4] - prefetchw [r3, 32] ;Prefetch the next write location - - SHIFT_1 (r7, r6, 16) - or r7, r7, r5 - SHIFT_2 (r5, r6, 16) - - SHIFT_1 (r9, r8, 16) - or r9, r9, r5 - SHIFT_2 (r5, r8, 16) - - st.ab r7, [r3, 4] - st.ab r9, [r3, 4] -.Lcopy8bytes_2: - -#ifdef __BIG_ENDIAN__ - lsr.nz r5, r5, 16 -#endif - sth.ab r5, [r3, 2] - - and.f lp_count, r2, 0x07 ;Last 8bytes - lpnz @.Lcopybytewise_2 - ;; LOOP START - ldb.ab r6, [r1,1] - stb.ab r6, [r3,1] -.Lcopybytewise_2: - j [blink] - -.LunalignedOffby3: -;;; CASE 3: The source is unaligned, off by 3 -;;; Hence, I need to read 1byte for achieve the 32bit alignment - - ;; Both src and dst are aligned - ;; Convert to words, unfold x2 - lsr.f lp_count, r2, 3 -#ifdef __BIG_ENDIAN__ - asl.ne r5, r5, 24 -#endif - lpnz @.Lcopy8bytes_3 - ;; LOOP START - ld.ab r6, [r1, 4] - prefetch [r1, 28] ;Prefetch the next read location - ld.ab r8, [r1,4] - prefetchw [r3, 32] ;Prefetch the next write location - - SHIFT_1 (r7, r6, 8) - or r7, r7, r5 - SHIFT_2 (r5, r6, 24) - - SHIFT_1 (r9, r8, 8) - or r9, r9, r5 - SHIFT_2 (r5, r8, 24) - - st.ab r7, [r3, 4] - st.ab r9, [r3, 4] -.Lcopy8bytes_3: - -#ifdef __BIG_ENDIAN__ - lsr.nz r5, r5, 24 -#endif - stb.ab r5, [r3, 1] - - and.f lp_count, r2, 0x07 ;Last 8bytes - lpnz @.Lcopybytewise_3 - ;; LOOP START - ldb.ab r6, [r1,1] - stb.ab r6, [r3,1] -.Lcopybytewise_3: - j [blink] - -END(memcpy) -libc_hidden_def(memcpy) diff --git a/libc/string/arc/arcv2/memset.S b/libc/string/arc/arcv2/memset.S deleted file mode 100644 index 343cfaf81c6d..000000000000 --- a/libc/string/arc/arcv2/memset.S +++ /dev/null @@ -1,115 +0,0 @@ - -/* - * Copyright (C) 2014-15 Synopsys, Inc. (www.synopsys.com) - * - * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB in this tarball. - */ - -#include <features.h> -#include <sysdep.h> - -#ifdef DONT_USE_PREALLOC -#define PREWRITE(A,B) prefetchw [(A),(B)] -#else -#define PREWRITE(A,B) prealloc [(A),(B)] -#endif - -ENTRY(memset) - prefetchw [r0] ; Prefetch the write location - mov.f 0, r2 -;;; if size is zero - jz.d [blink] - mov r3, r0 ; don't clobber ret val - -;;; if length < 8 - brls.d.nt r2, 8, .Lsmallchunk - mov.f lp_count,r2 - - and.f r4, r0, 0x03 - rsub lp_count, r4, 4 - lpnz @.Laligndestination - ;; LOOP BEGIN - stb.ab r1, [r3,1] - sub r2, r2, 1 -.Laligndestination: - -;;; Destination is aligned - and r1, r1, 0xFF - asl r4, r1, 8 - or r4, r4, r1 - asl r5, r4, 16 - or r5, r5, r4 - mov r4, r5 - - sub3 lp_count, r2, 8 - cmp r2, 64 - bmsk.hi r2, r2, 5 - mov.ls lp_count, 0 - add3.hi r2, r2, 8 - -;;; Convert len to Dwords, unfold x8 - lsr.f lp_count, lp_count, 6 - lpnz @.Lset64bytes - ;; LOOP START - PREWRITE(r3, 64) ;Prefetch the next write location -#if defined(__LL64__) || defined(__ARC_LL64__) - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] -#else - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] -#endif -.Lset64bytes: - - lsr.f lp_count, r2, 5 ;Last remaining max 124 bytes - lpnz .Lset32bytes - ;; LOOP START - prefetchw [r3, 32] ;Prefetch the next write location -#if defined(__LL64__) || defined(__ARC_LL64__) - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] - std.ab r4, [r3, 8] -#else - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] - st.ab r4, [r3, 4] -#endif -.Lset32bytes: - - and.f lp_count, r2, 0x1F ;Last remaining 31 bytes -.Lsmallchunk: - lpnz .Lcopy3bytes - ;; LOOP START - stb.ab r1, [r3, 1] -.Lcopy3bytes: - - j [blink] - -END(memset) -libc_hidden_def(memset) diff --git a/libc/string/arc/arcv2/strcmp.S b/libc/string/arc/arcv2/strcmp.S deleted file mode 100644 index 2e0e64a0c394..000000000000 --- a/libc/string/arc/arcv2/strcmp.S +++ /dev/null @@ -1,83 +0,0 @@ -/* - * Copyright (C) 2014-15 Synopsys, Inc. (www.synopsys.com) - * - * Licensed under the LGPL v2.1 or later, see the file COPYING.LIB in this tarball. - */ - -#include <features.h> -#include <sysdep.h> - -ENTRY(strcmp) - or r2, r0, r1 - bmsk_s r2, r2, 1 - brne r2, 0, @.Lcharloop - -;;; s1 and s2 are word aligned - ld.ab r2, [r0, 4] - - mov_s r12, 0x01010101 - ror r11, r12 - .align 4 -.LwordLoop: - ld.ab r3, [r1, 4] - ;; Detect NULL char in str1 - sub r4, r2, r12 - ld.ab r5, [r0, 4] - bic r4, r4, r2 - and r4, r4, r11 - brne.d.nt r4, 0, .LfoundNULL - ;; Check if the read locations are the same - cmp r2, r3 - beq.d .LwordLoop - mov.eq r2, r5 - - ;; A match is found, spot it out -#ifdef __LITTLE_ENDIAN__ - swape r3, r3 - mov_s r0, 1 - swape r2, r2 -#else - mov_s r0, 1 -#endif - cmp_s r2, r3 - j_s.d [blink] - bset.lo r0, r0, 31 - - .align 4 -.LfoundNULL: -#ifdef __BIG_ENDIAN__ - swape r4, r4 - swape r2, r2 - swape r3, r3 -#endif - ;; Find null byte - ffs r0, r4 - bmsk r2, r2, r0 - bmsk r3, r3, r0 - swape r2, r2 - swape r3, r3 - ;; make the return value - sub.f r0, r2, r3 - mov.hi r0, 1 - j_s.d [blink] - bset.lo r0, r0, 31 - - .align 4 -.Lcharloop: - ldb.ab r2, [r0, 1] - ldb.ab r3, [r1, 1] - nop - breq r2, 0, .Lcmpend - breq r2, r3, .Lcharloop - - .align 4 -.Lcmpend: - j_s.d [blink] - sub r0, r2, r3 -END(strcmp) -libc_hidden_def(strcmp) - -#ifndef __UCLIBC_HAS_LOCALE__ -strong_alias(strcmp,strcoll) -libc_hidden_def(strcoll) -#endif diff --git a/libc/string/arc/memcpy.S b/libc/string/arc/memcpy.S index 1c11951e40d2..69d7220b8a7e 100644 --- a/libc/string/arc/memcpy.S +++ b/libc/string/arc/memcpy.S @@ -1,5 +1,5 @@ /* - * Copyright (C) 2013 Synopsys, Inc. (www.synopsys.com) + * Copyright (C) 2013, 2014-2015, 2017 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. @@ -7,12 +7,18 @@ #include <sysdep.h> +#if !defined(__ARC700__) && !defined(__ARCHS__) +#error "Neither ARC700 nor ARCHS is defined!" +#endif + +ENTRY(memcpy) + +#ifdef __ARC700__ /* This memcpy implementation does not support objects of 1GB or larger - the check for alignment does not work then. */ /* We assume that most sources and destinations are aligned, and that also lengths are mostly a multiple of four, although to a lesser extent. */ -ENTRY(memcpy) or r3,r0,r1 asl_s r3,r3,30 mov_s r5,r0 @@ -67,5 +73,233 @@ ENTRY(memcpy) .Lendbloop: j_s.d [blink] stb r12,[r5,0] +#endif /* __ARC700__ */ + +#ifdef __ARCHS__ +#ifdef __LITTLE_ENDIAN__ +# define SHIFT_1(RX,RY,IMM) asl RX, RY, IMM ; << +# define SHIFT_2(RX,RY,IMM) lsr RX, RY, IMM ; >> +# define MERGE_1(RX,RY,IMM) asl RX, RY, IMM +# define MERGE_2(RX,RY,IMM) +# define EXTRACT_1(RX,RY,IMM) and RX, RY, 0xFFFF +# define EXTRACT_2(RX,RY,IMM) lsr RX, RY, IMM +#else +# define SHIFT_1(RX,RY,IMM) lsr RX, RY, IMM ; >> +# define SHIFT_2(RX,RY,IMM) asl RX, RY, IMM ; << +# define MERGE_1(RX,RY,IMM) asl RX, RY, IMM ; << +# define MERGE_2(RX,RY,IMM) asl RX, RY, IMM ; << +# define EXTRACT_1(RX,RY,IMM) lsr RX, RY, IMM +# define EXTRACT_2(RX,RY,IMM) lsr RX, RY, 0x08 +#endif + +#if defined(__LL64__) || defined(__ARC_LL64__) +# define PREFETCH_READ(RX) prefetch [RX, 56] +# define PREFETCH_WRITE(RX) prefetchw [RX, 64] +# define LOADX(DST,RX) ldd.ab DST, [RX, 8] +# define STOREX(SRC,RX) std.ab SRC, [RX, 8] +# define ZOLSHFT 5 +# define ZOLAND 0x1F +#else +# define PREFETCH_READ(RX) prefetch [RX, 28] +# define PREFETCH_WRITE(RX) prefetchw [RX, 32] +# define LOADX(DST,RX) ld.ab DST, [RX, 4] +# define STOREX(SRC,RX) st.ab SRC, [RX, 4] +# define ZOLSHFT 4 +# define ZOLAND 0xF +#endif + + prefetch [r1] ; Prefetch the read location + prefetchw [r0] ; Prefetch the write location + mov.f 0, r2 +;;; if size is zero + jz.d [blink] + mov r3, r0 ; don't clobber ret val + +;;; if size <= 8 + cmp r2, 8 + bls.d @.Lsmallchunk + mov.f lp_count, r2 + + and.f r4, r0, 0x03 + rsub lp_count, r4, 4 + lpnz @.Laligndestination + ;; LOOP BEGIN + ldb.ab r5, [r1,1] + sub r2, r2, 1 + stb.ab r5, [r3,1] +.Laligndestination: + +;;; Check the alignment of the source + and.f r4, r1, 0x03 + bnz.d @.Lsourceunaligned + +;;; CASE 0: Both source and destination are 32bit aligned +;;; Convert len to Dwords, unfold x4 + lsr.f lp_count, r2, ZOLSHFT + lpnz @.Lcopy32_64bytes + ;; LOOP START + LOADX (r6, r1) + PREFETCH_READ (r1) + PREFETCH_WRITE (r3) + LOADX (r8, r1) + LOADX (r10, r1) + LOADX (r4, r1) + STOREX (r6, r3) + STOREX (r8, r3) + STOREX (r10, r3) + STOREX (r4, r3) +.Lcopy32_64bytes: + + and.f lp_count, r2, ZOLAND ;Last remaining 31 bytes +.Lsmallchunk: + lpnz @.Lcopyremainingbytes + ;; LOOP START + ldb.ab r5, [r1,1] + stb.ab r5, [r3,1] +.Lcopyremainingbytes: + + j [blink] +;;; END CASE 0 + +.Lsourceunaligned: + cmp r4, 2 + beq.d @.LunalignedOffby2 + sub r2, r2, 1 + + bhi.d @.LunalignedOffby3 + ldb.ab r5, [r1, 1] + +;;; CASE 1: The source is unaligned, off by 1 + ;; Hence I need to read 1 byte for a 16bit alignment + ;; and 2bytes to reach 32bit alignment + ldh.ab r6, [r1, 2] + sub r2, r2, 2 + ;; Convert to words, unfold x2 + lsr.f lp_count, r2, 3 + MERGE_1 (r6, r6, 8) + MERGE_2 (r5, r5, 24) + or r5, r5, r6 + + ;; Both src and dst are aligned + lpnz @.Lcopy8bytes_1 + ;; LOOP START + ld.ab r6, [r1, 4] + prefetch [r1, 28] ;Prefetch the next read location + ld.ab r8, [r1,4] + prefetchw [r3, 32] ;Prefetch the next write location + + SHIFT_1 (r7, r6, 24) + or r7, r7, r5 + SHIFT_2 (r5, r6, 8) + + SHIFT_1 (r9, r8, 24) + or r9, r9, r5 + SHIFT_2 (r5, r8, 8) + + st.ab r7, [r3, 4] + st.ab r9, [r3, 4] +.Lcopy8bytes_1: + + ;; Write back the remaining 16bits + EXTRACT_1 (r6, r5, 16) + sth.ab r6, [r3, 2] + ;; Write back the remaining 8bits + EXTRACT_2 (r5, r5, 16) + stb.ab r5, [r3, 1] + + and.f lp_count, r2, 0x07 ;Last 8bytes + lpnz @.Lcopybytewise_1 + ;; LOOP START + ldb.ab r6, [r1,1] + stb.ab r6, [r3,1] +.Lcopybytewise_1: + j [blink] + +.LunalignedOffby2: +;;; CASE 2: The source is unaligned, off by 2 + ldh.ab r5, [r1, 2] + sub r2, r2, 1 + + ;; Both src and dst are aligned + ;; Convert to words, unfold x2 + lsr.f lp_count, r2, 3 +#ifdef __BIG_ENDIAN__ + asl.nz r5, r5, 16 +#endif + lpnz @.Lcopy8bytes_2 + ;; LOOP START + ld.ab r6, [r1, 4] + prefetch [r1, 28] ;Prefetch the next read location + ld.ab r8, [r1,4] + prefetchw [r3, 32] ;Prefetch the next write location + + SHIFT_1 (r7, r6, 16) + or r7, r7, r5 + SHIFT_2 (r5, r6, 16) + + SHIFT_1 (r9, r8, 16) + or r9, r9, r5 + SHIFT_2 (r5, r8, 16) + + st.ab r7, [r3, 4] + st.ab r9, [r3, 4] +.Lcopy8bytes_2: + +#ifdef __BIG_ENDIAN__ + lsr.nz r5, r5, 16 +#endif + sth.ab r5, [r3, 2] + + and.f lp_count, r2, 0x07 ;Last 8bytes + lpnz @.Lcopybytewise_2 + ;; LOOP START + ldb.ab r6, [r1,1] + stb.ab r6, [r3,1] +.Lcopybytewise_2: + j [blink] + +.LunalignedOffby3: +;;; CASE 3: The source is unaligned, off by 3 +;;; Hence, I need to read 1byte for achieve the 32bit alignment + + ;; Both src and dst are aligned + ;; Convert to words, unfold x2 + lsr.f lp_count, r2, 3 +#ifdef __BIG_ENDIAN__ + asl.ne r5, r5, 24 +#endif + lpnz @.Lcopy8bytes_3 + ;; LOOP START + ld.ab r6, [r1, 4] + prefetch [r1, 28] ;Prefetch the next read location + ld.ab r8, [r1,4] + prefetchw [r3, 32] ;Prefetch the next write location + + SHIFT_1 (r7, r6, 8) + or r7, r7, r5 + SHIFT_2 (r5, r6, 24) + + SHIFT_1 (r9, r8, 8) + or r9, r9, r5 + SHIFT_2 (r5, r8, 24) + + st.ab r7, [r3, 4] + st.ab r9, [r3, 4] +.Lcopy8bytes_3: + +#ifdef __BIG_ENDIAN__ + lsr.nz r5, r5, 24 +#endif + stb.ab r5, [r3, 1] + + and.f lp_count, r2, 0x07 ;Last 8bytes + lpnz @.Lcopybytewise_3 + ;; LOOP START + ldb.ab r6, [r1,1] + stb.ab r6, [r3,1] +.Lcopybytewise_3: + j [blink] +#endif /* __ARCHS__ */ + END(memcpy) libc_hidden_def(memcpy) diff --git a/libc/string/arc/memset.S b/libc/string/arc/memset.S index f4048455a433..0b74ddc7fca8 100644 --- a/libc/string/arc/memset.S +++ b/libc/string/arc/memset.S @@ -1,5 +1,5 @@ /* - * Copyright (C) 2013 Synopsys, Inc. (www.synopsys.com) + * Copyright (C) 2013, 2014-2015, 2017 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. @@ -7,10 +7,15 @@ #include <sysdep.h> -#define SMALL 7 /* Must be at least 6 to deal with alignment/loop issues. */ +#if !defined(__ARC700__) && !defined(__ARCHS__) +#error "Neither ARC700 nor ARCHS is defined!" +#endif ENTRY(memset) +#ifdef __ARC700__ +#define SMALL 7 /* Must be at least 6 to deal with alignment/loop issues. */ + mov_s r4,r0 or r12,r0,r2 bmsk.f r12,r12,1 @@ -47,5 +52,111 @@ ENTRY(memset) stb.ab r1,[r4,1] .Ltiny_end: j_s [blink] +#endif /* __ARC700__ */ + +#ifdef __ARCHS__ +#ifdef DONT_USE_PREALLOC +#define PREWRITE(A,B) prefetchw [(A),(B)] +#else +#define PREWRITE(A,B) prealloc [(A),(B)] +#endif + + prefetchw [r0] ; Prefetch the write location + mov.f 0, r2 +;;; if size is zero + jz.d [blink] + mov r3, r0 ; don't clobber ret val + +;;; if length < 8 + brls.d.nt r2, 8, .Lsmallchunk + mov.f lp_count,r2 + + and.f r4, r0, 0x03 + rsub lp_count, r4, 4 + lpnz @.Laligndestination + ;; LOOP BEGIN + stb.ab r1, [r3,1] + sub r2, r2, 1 +.Laligndestination: + +;;; Destination is aligned + and r1, r1, 0xFF + asl r4, r1, 8 + or r4, r4, r1 + asl r5, r4, 16 + or r5, r5, r4 + mov r4, r5 + + sub3 lp_count, r2, 8 + cmp r2, 64 + bmsk.hi r2, r2, 5 + mov.ls lp_count, 0 + add3.hi r2, r2, 8 + +;;; Convert len to Dwords, unfold x8 + lsr.f lp_count, lp_count, 6 + lpnz @.Lset64bytes + ;; LOOP START + PREWRITE(r3, 64) ;Prefetch the next write location +#if defined(__LL64__) || defined(__ARC_LL64__) + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] +#else + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] +#endif +.Lset64bytes: + + lsr.f lp_count, r2, 5 ;Last remaining max 124 bytes + lpnz .Lset32bytes + ;; LOOP START + prefetchw [r3, 32] ;Prefetch the next write location +#if defined(__LL64__) || defined(__ARC_LL64__) + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] + std.ab r4, [r3, 8] +#else + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] + st.ab r4, [r3, 4] +#endif +.Lset32bytes: + + and.f lp_count, r2, 0x1F ;Last remaining 31 bytes +.Lsmallchunk: + lpnz .Lcopy3bytes + ;; LOOP START + stb.ab r1, [r3, 1] +.Lcopy3bytes: + + j [blink] +#endif /* __ARCHS__ */ + END(memset) libc_hidden_def(memset) diff --git a/libc/string/arc/strcmp.S b/libc/string/arc/strcmp.S index 5a0e56045c44..ad38d9e00c8a 100644 --- a/libc/string/arc/strcmp.S +++ b/libc/string/arc/strcmp.S @@ -1,5 +1,5 @@ /* - * Copyright (C) 2013 Synopsys, Inc. (www.synopsys.com) + * Copyright (C) 2013, 2014-2015, 2017 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. @@ -8,6 +8,13 @@ #include <features.h> #include <sysdep.h> +#if !defined(__ARC700__) && !defined(__ARCHS__) +#error "Neither ARC700 nor ARCHS is defined!" +#endif + +ENTRY(strcmp) + +#ifdef __ARC700__ /* This is optimized primarily for the ARC700. It would be possible to speed up the loops by one cycle / word respective one cycle / byte by forcing double source 1 alignment, unrolling @@ -15,7 +22,6 @@ source 1; however, that would increase the overhead for loop setup / finish, and strcmp might often terminate early. */ -ENTRY(strcmp) or r2,r0,r1 bmsk_s r2,r2,1 brne r2,0,.Lcharloop @@ -93,6 +99,77 @@ ENTRY(strcmp) .Lcmpend: j_s.d [blink] sub r0,r2,r3 +#endif /* __ARC700__ */ + +#ifdef __ARCHS__ + or r2, r0, r1 + bmsk_s r2, r2, 1 + brne r2, 0, @.Lcharloop + +;;; s1 and s2 are word aligned + ld.ab r2, [r0, 4] + + mov_s r12, 0x01010101 + ror r11, r12 + .align 4 +.LwordLoop: + ld.ab r3, [r1, 4] + ;; Detect NULL char in str1 + sub r4, r2, r12 + ld.ab r5, [r0, 4] + bic r4, r4, r2 + and r4, r4, r11 + brne.d.nt r4, 0, .LfoundNULL + ;; Check if the read locations are the same + cmp r2, r3 + beq.d .LwordLoop + mov.eq r2, r5 + + ;; A match is found, spot it out +#ifdef __LITTLE_ENDIAN__ + swape r3, r3 + mov_s r0, 1 + swape r2, r2 +#else + mov_s r0, 1 +#endif + cmp_s r2, r3 + j_s.d [blink] + bset.lo r0, r0, 31 + + .align 4 +.LfoundNULL: +#ifdef __BIG_ENDIAN__ + swape r4, r4 + swape r2, r2 + swape r3, r3 +#endif + ;; Find null byte + ffs r0, r4 + bmsk r2, r2, r0 + bmsk r3, r3, r0 + swape r2, r2 + swape r3, r3 + ;; make the return value + sub.f r0, r2, r3 + mov.hi r0, 1 + j_s.d [blink] + bset.lo r0, r0, 31 + + .align 4 +.Lcharloop: + ldb.ab r2, [r0, 1] + ldb.ab r3, [r1, 1] + nop + breq r2, 0, .Lcmpend + breq r2, r3, .Lcharloop + + .align 4 +.Lcmpend: + j_s.d [blink] + sub r0, r2, r3 +#endif /* __ARCHS__ */ + END(strcmp) libc_hidden_def(strcmp) -- 2.7.5

4 4

Compile fix when debugging is enabled
by Bogdan Harjoc 19 Aug '17

19 Aug '17

Pthread pid caching was removed recently but an assert is still present which checks pthread->pid, and this breaks the build when debugging is enabled: http://repo.or.cz/uclibc-ng.git/commit/bddde5860ffb8a78587854cc8e3e914bd692… http://repo.or.cz/uclibc-ng.git/blob/HEAD:/libpthread/nptl/pthread_getattr_… Line 69 should be removed from pthread_getattr_np.c . Cheers, Bogdan

2 1

buildroot uClibc-ng build failure
by Waldemar Brodkorb 10 Aug '17

10 Aug '17

Hi André, we are seeing a build error on your machine when locales in uClibc-ng are enabled: http://autobuild.buildroot.net/results/d5b/d5ba81eea9223569ba5b363551c4a2f7… Can you show us locale -a from your machine? For a working tolower(), the uClibc-ng code needs to use setlocale on the host. But it seems only these are tried to use: en_US en_US en_GB best regards Waldemar

6 12

[PATCH] towlower and locales
by Eugene Yudin 02 Aug '17

02 Aug '17

Hi, The function towlower doesn't work with locales diffrent from C. Issue was introduced in this commit: https://cgit.openadk.org/cgi/cgit/uclibc-ng.git/commit/?id= 8cde3a9bf2856dcb9a759dec7ecb04a68e712254 Call to setlocale is needed for correct generation of the table uplow_diff. I received the compile time error "range assumption error" after uncommenting the call. Similar problem described here: http://lists.uclibc.org/pipermail/uclibc/2015-March/048852.html The attached patch fix the problem by using int32_t values. Test program: $ cat test.c #include <locale.h> #include <stdio.h> #include <wchar.h> #include <wctype.h> int main(int argc, char *argv[]) { int i = 0; wchar_t str[] = L"ТЕСТОВАЯ СТРОКА"; wchar_t c; setlocale(LC_ALL, "ru_RU.utf-8"); wprintf(L"Input:\t\"%ls\"\n", str); wprintf(L"Output:\t\""); while (str[i]) { c = str[i]; putwchar(towlower(c)); i++; } wprintf(L"\"\n"); return 0; } Output (without patch): $ ./test Input: "ТЕСТОВАЯ СТРОКА" Output: "ТЕСТОВАЯ СТРОКА" Output (with patch): $ ./test Input: "ТЕСТОВАЯ СТРОКА" Output: "тестовая строка" -- Best regards, Eugene

3 4

[PATCH] valloc declaration in stdlib.h
by Eugene Yudin 29 Jul '17

29 Jul '17

Hi, Current uClibc-ng have issue with several different valloc declarations. malloc.h: #ifdef __UCLIBC_SUSV2_LEGACY__ /* Allocate SIZE bytes on a page boundary. */ extern __malloc_ptr_t valloc __MALLOC_P ((size_t __size)) __attribute_malloc__; #endif stdlib.h: #if defined __USE_BSD || defined __USE_XOPEN_EXTENDED /* Allocate SIZE bytes on a page boundary. The storage cannot be freed. */ extern void *valloc (size_t __size) __THROW __attribute_malloc__ __wur; #endif The second declaration doesn't use the define __UCLIBC_SUSV2_LEGACY__. That leads to compile time problems. The attached patch fixes this diffrence. -- Best regards, Eugene

2 3

syscall cancel simplification patches
by Waldemar Brodkorb 29 Jul '17

29 Jul '17

Hi, I would like to get some feedback for following three patches. The test suite runs over all architectures showed no regressions. The goal is to simplify the addition of new architectures and remove the need to add assembly code to sysdep-cancel.h. Most of the work was done in 2011 in commit 9f68f0cbf8c8eea6a7f9e195e4617bbaa808d7c6, but it was not finished for NPTL and RT components of uClibc. best regards Waldemar

3 3

segfault in dlclose() when lots of shared objects are dlopened - the story continues
by Bernd Kuhls 20 Jul '17

20 Jul '17

Hi, investigating a segfault (gdb) bt full #0 0x00007fffec28dd90 in __deregister_frame_info () from /lib/ libgcc_s.so.1 No symbol table info available. #1 0x00007fffebf8af66 in __do_global_dtors_aux () from /usr/lib/ libgcrypt.so.20 when starting Apache with mod_php activated on a buildroot-built x86_64- system with uclibc-1.0.19 I came across discussions from 2012 & 2014 on the uClibc mailinglist: 2012: http://lists.uclibc.org/pipermail/uclibc/2012-October/047059.html 2014: http://lists.busybox.net/pipermail/uclibc/2014-December/048737.html Building mod_php without libgcrypt.so, which means disabling BR2_PACKAGE_PHP_EXT_XSL, does not fix the problem, the segfault will occur in another shared lib. Quoting the message from 2014: > In my particular case, this meant dlopening and dlclosing some 47 > shared objects as reported by "LD_DEBUG=1 /usr/bin/gdk-pixbuf-query- > loaders ./libpixbufloader-svg.la 2>&1 | grep ^do_dlopen | grep ctors > | wc -l" My system opens even more shared libs when starting Apache & mod_php: # LD_DEBUG=1 /usr/bin/httpd -t 2>&1 | grep ^do_dlopen | grep ctors | wc -l 64 Segmentation fault Please note that using Apache without mod_php works fine, php-cgi itself also works fine. In 2014 Anthony suggested to revert https://git.busybox.net/uClibc/commit/ldso? h=0.9.33&id=9b42da7d0558884e2a3cc9a8674ccfc752369610 which still solves the Apache/mod_php segfault today. Uclibc-1.0.19 includes all commits from 2016-09-26 and older in https://cgit.openadk.org/cgi/cgit/uclibc-ng.git/log/ldso/libdl it therefore includes https://cgit.openadk.org/cgi/cgit/uclibc-ng.git/ commit/ldso/libdl?id=cc04ab27ba6341f46bbe094478c9af3e3706f411 which refers to fix the problem from the 2012 message, but it did not. Regards, Bernd

2 1

Missing git tags at github mirror
by Anton Kolesov 13 Jul '17

13 Jul '17

Hi, It seems that GitHub mirror at https://github.com/wbx-github/uclibc-ng/tree/master is missing git tags for releases starting with v1.0.23. Mirror at http://repo.or.cz/uclibc-ng.git OTOH has those tags. Anton

2 1

[PATCH] netinet/udp: add missing include
by Romain Naour 12 Jul '17

12 Jul '17

Since the refactoring introduced by [1] any check using netinet/udp.h fail with the following error: [...]/usr/include/netinet/udp.h:62:7: error: unknown type name 'uint16_t' See the error in netinet/udp.h check: [efl] http://autobuild.buildroot.net/results/8c6/8c629a1c5c4a1d78c8bf09a091a332a4… [trinity] http://autobuild.buildroot.net/results/c8f/c8f34aaaecf931e0044ef858ef587325… [1] 58a5ba12bffad5916d9897c2870fc483f1db8282 Signed-off-by: Romain Naour <romain.naour(a)gmail.com> --- include/netinet/udp.h | 1 + 1 file changed, 1 insertion(+) diff --git a/include/netinet/udp.h b/include/netinet/udp.h index ac6f234..f03e2dc 100644 --- a/include/netinet/udp.h +++ b/include/netinet/udp.h @@ -48,6 +48,7 @@ #define __NETINET_UDP_H 1 #include <features.h> +#include <stdint.h> #include <sys/types.h> -- 2.9.4

2 2

ld.so cleanup
by Waldemar Brodkorb 29 Jun '17

29 Jun '17

Hi, I would suggest following patch, which removes the special communication between ld.so and gdb. While debugging a ld.so issue with gdb 8.0 on alpha, I recognized that the debugger stalled when trying to execute this code. It thought this code was added to help debugging ld.so, but it seems nowadays it prevents to do so. May be bitrotted? Does anyone found it useful while debugging ld.so with gdb in the past? best regards Waldemar

3 3

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

devel