[PATCH 3/3] aarch64: Import memrchr.S
Sebastian Huber <[email protected]>
| Newsgroups | gmane.comp.lib.newlib |
|---|---|
| Message-ID | <[email protected]> |
Import memrchr.S for AArch64 from: https://github.com/ARM-software/optimized-routines commit 88b050e7d88d8dff4ffda97d9cf13a1c1b448eb7 Author: Joe Ramsay <[email protected]> Date: Tue Jul 25 10:01:52 2023 +0100 pl/math: Use width-agnostic intrinsic forms No effect on performance, but a small improvement in readability. --- newlib/Makefile.in | 56 +++++++++-- newlib/libc/machine/aarch64/Makefile.inc | 2 + newlib/libc/machine/aarch64/memrchr-stub.c | 11 ++ newlib/libc/machine/aarch64/memrchr.S | 112 +++++++++++++++++++++ 4 files changed, 173 insertions(+), 8 deletions(-) create mode 100644 newlib/libc/machine/aarch64/memrchr-stub.c create mode 100644 newlib/libc/machine/aarch64/memrchr.S diff --git a/newlib/Makefile.in b/newlib/Makefile.in index 4d2b5e7825..302dbeea29 100644 --- a/newlib/Makefile.in +++ b/newlib/Makefile.in @@ -594,6 +594,8 @@ check_PROGRAMS = @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memcpy.S \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memmove-stub.c \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memmove.S \ +@HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memrchr-stub.c \ +@HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memrchr.S \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memset-stub.c \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/memset.S \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/rawmemchr.S \ @@ -911,18 +913,18 @@ check_PROGRAMS = @NEWLIB_HW_FP_FALSE@am__append_136 = libm/math/math.tex @HAVE_LONG_DOUBLE_TRUE@am__append_137 = $(libm_common_lsrc) @HAVE_FPMATH_H_TRUE@@HAVE_LONG_DOUBLE_TRUE@am__append_138 = $(libm_ld_lsrc) -@HAVE_FPMATH_H_TRUE@am__append_139 = -@HAVE_FPMATH_H_TRUE@am__append_140 = +@HAVE_FPMATH_H_TRUE@am__append_139 = +@HAVE_FPMATH_H_TRUE@am__append_140 = @HAVE_LIBM_MACHINE_AARCH64_TRUE@am__append_141 = $(libm_machine_aarch64_src) @HAVE_LIBM_MACHINE_AARCH64_TRUE@@HAVE_LONG_DOUBLE_TRUE@am__append_142 = $(libm_ld128_lsrc) -@HAVE_LIBM_MACHINE_AARCH64_TRUE@am__append_143 = -@HAVE_LIBM_MACHINE_AARCH64_TRUE@am__append_144 = +@HAVE_LIBM_MACHINE_AARCH64_TRUE@am__append_143 = +@HAVE_LIBM_MACHINE_AARCH64_TRUE@am__append_144 = @HAVE_LIBM_MACHINE_AMDGCN_TRUE@am__append_145 = $(libm_machine_amdgcn_src) @HAVE_LIBM_MACHINE_ARM_TRUE@am__append_146 = $(libm_machine_arm_src) @HAVE_LIBM_MACHINE_I386_TRUE@am__append_147 = $(libm_machine_i386_src) @HAVE_LIBM_MACHINE_I386_TRUE@@HAVE_LONG_DOUBLE_TRUE@am__append_148 = $(libm_ld80_lsrc) -@HAVE_LIBM_MACHINE_I386_TRUE@am__append_149 = -@HAVE_LIBM_MACHINE_I386_TRUE@am__append_150 = +@HAVE_LIBM_MACHINE_I386_TRUE@am__append_149 = +@HAVE_LIBM_MACHINE_I386_TRUE@am__append_150 = @HAVE_LIBM_MACHINE_MIPS_TRUE@am__append_151 = $(libm_machine_mips_src) @HAS_NDS32_FPU_SP_TRUE@@HAVE_LIBM_MACHINE_NDS32_TRUE@am__append_152 = libm/machine/nds32/wf_sqrt.S @HAS_NDS32_FPU_DP_TRUE@@HAVE_LIBM_MACHINE_NDS32_TRUE@am__append_153 = libm/machine/nds32/w_sqrt.S @@ -934,8 +936,8 @@ check_PROGRAMS = @HAVE_LIBM_MACHINE_RISCV_TRUE@am__append_159 = $(libm_machine_riscv_src) @HAVE_LIBM_MACHINE_X86_64_TRUE@am__append_160 = $(libm_machine_x86_64_src) @HAVE_LIBM_MACHINE_X86_64_TRUE@@HAVE_LONG_DOUBLE_TRUE@am__append_161 = $(libm_ld80_lsrc) -@HAVE_LIBM_MACHINE_X86_64_TRUE@am__append_162 = -@HAVE_LIBM_MACHINE_X86_64_TRUE@am__append_163 = +@HAVE_LIBM_MACHINE_X86_64_TRUE@am__append_162 = +@HAVE_LIBM_MACHINE_X86_64_TRUE@am__append_163 = subdir = . ACLOCAL_M4 = $(top_srcdir)/aclocal.m4 am__aclocal_m4_deps = $(top_srcdir)/../config/depstand.m4 \ @@ -1831,6 +1833,8 @@ am__objects_51 = libc/ssp/libc_a-chk_fail.$(OBJEXT) \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memcpy.$(OBJEXT) \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memmove-stub.$(OBJEXT) \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memmove.$(OBJEXT) \ +@HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memrchr-stub.$(OBJEXT) \ +@HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memrchr.$(OBJEXT) \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memset-stub.$(OBJEXT) \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-memset.$(OBJEXT) \ @HAVE_LIBC_MACHINE_AARCH64_TRUE@ libc/machine/aarch64/libc_a-rawmemchr.$(OBJEXT) \ @@ -7948,6 +7952,12 @@ libc/machine/aarch64/libc_a-memmove-stub.$(OBJEXT): \ libc/machine/aarch64/libc_a-memmove.$(OBJEXT): \ libc/machine/aarch64/$(am__dirstamp) \ libc/machine/aarch64/$(DEPDIR)/$(am__dirstamp) +libc/machine/aarch64/libc_a-memrchr-stub.$(OBJEXT): \ + libc/machine/aarch64/$(am__dirstamp) \ + libc/machine/aarch64/$(DEPDIR)/$(am__dirstamp) +libc/machine/aarch64/libc_a-memrchr.$(OBJEXT): \ + libc/machine/aarch64/$(am__dirstamp) \ + libc/machine/aarch64/$(DEPDIR)/$(am__dirstamp) libc/machine/aarch64/libc_a-memset-stub.$(OBJEXT): \ libc/machine/aarch64/$(am__dirstamp) \ libc/machine/aarch64/$(DEPDIR)/$(am__dirstamp) @@ -12588,6 +12598,8 @@ distclean-compile: @AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memcpy.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memmove-stub.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memmove.Po@am__quote@ +@AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Po@am__quote@ +@AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memset-stub.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-memset.Po@am__quote@ @AMDEP_TRUE@@am__include@ @am__quote@libc/machine/aarch64/$(DEPDIR)/libc_a-rawmemchr-stub.Po@am__quote@ @@ -16547,6 +16559,20 @@ libc/machine/aarch64/libc_a-memmove.obj: libc/machine/aarch64/memmove.S @AMDEP_TRUE@@am__fastdepCCAS_FALSE@ DEPDIR=$(DEPDIR) $(CCASDEPMODE) $(depcomp) @AMDEPBACKSLASH@ @am__fastdepCCAS_FALSE@ $(AM_V_CPPAS@am__nodep@)$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CCASFLAGS) $(CCASFLAGS) -c -o libc/machine/aarch64/libc_a-memmove.obj `if test -f 'libc/machine/aarch64/memmove.S'; then $(CYGPATH_W) 'libc/machine/aarch64/memmove.S'; else $(CYGPATH_W) '$(srcdir)/libc/machine/aarch64/memmove.S'; fi` +libc/machine/aarch64/libc_a-memrchr.o: libc/machine/aarch64/memrchr.S +@am__fastdepCCAS_TRUE@ $(AM_V_CPPAS)$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CCASFLAGS) $(CCASFLAGS) -MT libc/machine/aarch64/libc_a-memrchr.o -MD -MP -MF libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Tpo -c -o libc/machine/aarch64/libc_a-memrchr.o `test -f 'libc/machine/aarch64/memrchr.S' || echo '$(srcdir)/'`libc/machine/aarch64/memrchr.S +@am__fastdepCCAS_TRUE@ $(AM_V_at)$(am__mv) libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Tpo libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Po +@AMDEP_TRUE@@am__fastdepCCAS_FALSE@ $(AM_V_CPPAS)source='libc/machine/aarch64/memrchr.S' object='libc/machine/aarch64/libc_a-memrchr.o' libtool=no @AMDEPBACKSLASH@ +@AMDEP_TRUE@@am__fastdepCCAS_FALSE@ DEPDIR=$(DEPDIR) $(CCASDEPMODE) $(depcomp) @AMDEPBACKSLASH@ +@am__fastdepCCAS_FALSE@ $(AM_V_CPPAS@am__nodep@)$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CCASFLAGS) $(CCASFLAGS) -c -o libc/machine/aarch64/libc_a-memrchr.o `test -f 'libc/machine/aarch64/memrchr.S' || echo '$(srcdir)/'`libc/machine/aarch64/memrchr.S + +libc/machine/aarch64/libc_a-memrchr.obj: libc/machine/aarch64/memrchr.S +@am__fastdepCCAS_TRUE@ $(AM_V_CPPAS)$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CCASFLAGS) $(CCASFLAGS) -MT libc/machine/aarch64/libc_a-memrchr.obj -MD -MP -MF libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Tpo -c -o libc/machine/aarch64/libc_a-memrchr.obj `if test -f 'libc/machine/aarch64/memrchr.S'; then $(CYGPATH_W) 'libc/machine/aarch64/memrchr.S'; else $(CYGPATH_W) '$(srcdir)/libc/machine/aarch64/memrchr.S'; fi` +@am__fastdepCCAS_TRUE@ $(AM_V_at)$(am__mv) libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Tpo libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr.Po +@AMDEP_TRUE@@am__fastdepCCAS_FALSE@ $(AM_V_CPPAS)source='libc/machine/aarch64/memrchr.S' object='libc/machine/aarch64/libc_a-memrchr.obj' libtool=no @AMDEPBACKSLASH@ +@AMDEP_TRUE@@am__fastdepCCAS_FALSE@ DEPDIR=$(DEPDIR) $(CCASDEPMODE) $(depcomp) @AMDEPBACKSLASH@ +@am__fastdepCCAS_FALSE@ $(AM_V_CPPAS@am__nodep@)$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CCASFLAGS) $(CCASFLAGS) -c -o libc/machine/aarch64/libc_a-memrchr.obj `if test -f 'libc/machine/aarch64/memrchr.S'; then $(CYGPATH_W) 'libc/machine/aarch64/memrchr.S'; else $(CYGPATH_W) '$(srcdir)/libc/machine/aarch64/memrchr.S'; fi` + libc/machine/aarch64/libc_a-memset.o: libc/machine/aarch64/memset.S @am__fastdepCCAS_TRUE@ $(AM_V_CPPAS)$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CCASFLAGS) $(CCASFLAGS) -MT libc/machine/aarch64/libc_a-memset.o -MD -MP -MF libc/machine/aarch64/$(DEPDIR)/libc_a-memset.Tpo -c -o libc/machine/aarch64/libc_a-memset.o `test -f 'libc/machine/aarch64/memset.S' || echo '$(srcdir)/'`libc/machine/aarch64/memset.S @am__fastdepCCAS_TRUE@ $(AM_V_at)$(am__mv) libc/machine/aarch64/$(DEPDIR)/libc_a-memset.Tpo libc/machine/aarch64/$(DEPDIR)/libc_a-memset.Po @@ -32649,6 +32675,20 @@ libc/machine/aarch64/libc_a-memmove-stub.obj: libc/machine/aarch64/memmove-stub. @AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ @am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CFLAGS) $(CFLAGS) -c -o libc/machine/aarch64/libc_a-memmove-stub.obj `if test -f 'libc/machine/aarch64/memmove-stub.c'; then $(CYGPATH_W) 'libc/machine/aarch64/memmove-stub.c'; else $(CYGPATH_W) '$(srcdir)/libc/machine/aarch64/memmove-stub.c'; fi` +libc/machine/aarch64/libc_a-memrchr-stub.o: libc/machine/aarch64/memrchr-stub.c +@am__fastdepCC_TRUE@ $(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CFLAGS) $(CFLAGS) -MT libc/machine/aarch64/libc_a-memrchr-stub.o -MD -MP -MF libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Tpo -c -o libc/machine/aarch64/libc_a-memrchr-stub.o `test -f 'libc/machine/aarch64/memrchr-stub.c' || echo '$(srcdir)/'`libc/machine/aarch64/memrchr-stub.c +@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Tpo libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Po +@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='libc/machine/aarch64/memrchr-stub.c' object='libc/machine/aarch64/libc_a-memrchr-stub.o' libtool=no @AMDEPBACKSLASH@ +@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ +@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CFLAGS) $(CFLAGS) -c -o libc/machine/aarch64/libc_a-memrchr-stub.o `test -f 'libc/machine/aarch64/memrchr-stub.c' || echo '$(srcdir)/'`libc/machine/aarch64/memrchr-stub.c + +libc/machine/aarch64/libc_a-memrchr-stub.obj: libc/machine/aarch64/memrchr-stub.c +@am__fastdepCC_TRUE@ $(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CFLAGS) $(CFLAGS) -MT libc/machine/aarch64/libc_a-memrchr-stub.obj -MD -MP -MF libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Tpo -c -o libc/machine/aarch64/libc_a-memrchr-stub.obj `if test -f 'libc/machine/aarch64/memrchr-stub.c'; then $(CYGPATH_W) 'libc/machine/aarch64/memrchr-stub.c'; else $(CYGPATH_W) '$(srcdir)/libc/machine/aarch64/memrchr-stub.c'; fi` +@am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Tpo libc/machine/aarch64/$(DEPDIR)/libc_a-memrchr-stub.Po +@AMDEP_TRUE@@am__fastdepCC_FALSE@ $(AM_V_CC)source='libc/machine/aarch64/memrchr-stub.c' object='libc/machine/aarch64/libc_a-memrchr-stub.obj' libtool=no @AMDEPBACKSLASH@ +@AMDEP_TRUE@@am__fastdepCC_FALSE@ DEPDIR=$(DEPDIR) $(CCDEPMODE) $(depcomp) @AMDEPBACKSLASH@ +@am__fastdepCC_FALSE@ $(AM_V_CC@am__nodep@)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CFLAGS) $(CFLAGS) -c -o libc/machine/aarch64/libc_a-memrchr-stub.obj `if test -f 'libc/machine/aarch64/memrchr-stub.c'; then $(CYGPATH_W) 'libc/machine/aarch64/memrchr-stub.c'; else $(CYGPATH_W) '$(srcdir)/libc/machine/aarch64/memrchr-stub.c'; fi` + libc/machine/aarch64/libc_a-memset-stub.o: libc/machine/aarch64/memset-stub.c @am__fastdepCC_TRUE@ $(AM_V_CC)$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(libc_a_CPPFLAGS) $(CPPFLAGS) $(libc_a_CFLAGS) $(CFLAGS) -MT libc/machine/aarch64/libc_a-memset-stub.o -MD -MP -MF libc/machine/aarch64/$(DEPDIR)/libc_a-memset-stub.Tpo -c -o libc/machine/aarch64/libc_a-memset-stub.o `test -f 'libc/machine/aarch64/memset-stub.c' || echo '$(srcdir)/'`libc/machine/aarch64/memset-stub.c @am__fastdepCC_TRUE@ $(AM_V_at)$(am__mv) libc/machine/aarch64/$(DEPDIR)/libc_a-memset-stub.Tpo libc/machine/aarch64/$(DEPDIR)/libc_a-memset-stub.Po diff --git a/newlib/libc/machine/aarch64/Makefile.inc b/newlib/libc/machine/aarch64/Makefile.inc index 063a2a84ae..c749b0d575 100644 --- a/newlib/libc/machine/aarch64/Makefile.inc +++ b/newlib/libc/machine/aarch64/Makefile.inc @@ -7,6 +7,8 @@ libc_a_SOURCES += \ %D%/memcpy.S \ %D%/memmove-stub.c \ %D%/memmove.S \ + %D%/memrchr-stub.c \ + %D%/memrchr.S \ %D%/memset-stub.c \ %D%/memset.S \ %D%/rawmemchr.S \ diff --git a/newlib/libc/machine/aarch64/memrchr-stub.c b/newlib/libc/machine/aarch64/memrchr-stub.c new file mode 100644 index 0000000000..d80ca88ab0 --- /dev/null +++ b/newlib/libc/machine/aarch64/memrchr-stub.c @@ -0,0 +1,11 @@ +/* + * Copyright (C) 2023 embedded brains GmbH & Co. KG + * + * SPDX-License-Identifier: BSD-2-Clause + */ + +#if defined (__OPTIMIZE_SIZE__) || defined (PREFER_SIZE_OVER_SPEED) +#include "../../string/memrchr.c" +#else +/* See memrchr.S */ +#endif diff --git a/newlib/libc/machine/aarch64/memrchr.S b/newlib/libc/machine/aarch64/memrchr.S new file mode 100644 index 0000000000..817e96dfdd --- /dev/null +++ b/newlib/libc/machine/aarch64/memrchr.S @@ -0,0 +1,112 @@ +/* + * memrchr - find last character in a memory zone. + * + * Copyright (c) 2020-2022, Arm Limited. + * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception + */ + +/* Assumptions: + * + * ARMv8-a, AArch64, Advanced SIMD. + * MTE compatible. + */ + +#include "asmdefs.h" + +#define srcin x0 +#define chrin w1 +#define cntin x2 +#define result x0 + +#define src x3 +#define cntrem x4 +#define synd x5 +#define shift x6 +#define tmp x7 +#define end x8 +#define endm1 x9 + +#define vrepchr v0 +#define qdata q1 +#define vdata v1 +#define vhas_chr v2 +#define vend v3 +#define dend d3 + +/* + Core algorithm: + For each 16-byte chunk we calculate a 64-bit nibble mask value with four bits + per byte. We take 4 bits of every comparison byte with shift right and narrow + by 4 instruction. Since the bits in the nibble mask reflect the order in + which things occur in the original string, counting leading zeros identifies + exactly which byte matched. */ + +ENTRY (memrchr) + PTR_ARG (0) + add end, srcin, cntin + sub endm1, end, 1 + bic src, endm1, 15 + cbz cntin, L(nomatch) + ld1 {vdata.16b}, [src] + dup vrepchr.16b, chrin + cmeq vhas_chr.16b, vdata.16b, vrepchr.16b + neg shift, end, lsl 2 + shrn vend.8b, vhas_chr.8h, 4 /* 128->64 */ + fmov synd, dend + lsl synd, synd, shift + cbz synd, L(start_loop) + + clz synd, synd + sub result, endm1, synd, lsr 2 + cmp cntin, synd, lsr 2 + csel result, result, xzr, hi + ret + + nop +L(start_loop): + subs cntrem, src, srcin + b.ls L(nomatch) + + /* Make sure that it won't overread by a 16-byte chunk */ + sub cntrem, cntrem, 1 + tbz cntrem, 4, L(loop32_2) + add src, src, 16 + + .p2align 5 +L(loop32): + ldr qdata, [src, -32]! + cmeq vhas_chr.16b, vdata.16b, vrepchr.16b + umaxp vend.16b, vhas_chr.16b, vhas_chr.16b /* 128->64 */ + fmov synd, dend + cbnz synd, L(end) + +L(loop32_2): + ldr qdata, [src, -16] + subs cntrem, cntrem, 32 + cmeq vhas_chr.16b, vdata.16b, vrepchr.16b + b.lo L(end_2) + umaxp vend.16b, vhas_chr.16b, vhas_chr.16b /* 128->64 */ + fmov synd, dend + cbz synd, L(loop32) +L(end_2): + sub src, src, 16 +L(end): + shrn vend.8b, vhas_chr.8h, 4 /* 128->64 */ + fmov synd, dend + + add tmp, src, 15 +#ifdef __AARCH64EB__ + rbit synd, synd +#endif + clz synd, synd + sub tmp, tmp, synd, lsr 2 + cmp tmp, srcin + csel result, tmp, xzr, hs + ret + +L(nomatch): + mov result, 0 + ret + +END (memrchr) + -- 2.35.3