From patchwork Thu Apr 21 14:50:13 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Stefan Liebler X-Patchwork-Id: 11840 Received: (qmail 57588 invoked by alias); 21 Apr 2016 15:20:18 -0000 Mailing-List: contact libc-alpha-help@sourceware.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Subscribe: List-Archive: List-Post: List-Help: , Sender: libc-alpha-owner@sourceware.org Delivered-To: mailing list libc-alpha@sourceware.org Received: (qmail 57578 invoked by uid 89); 21 Apr 2016 15:20:18 -0000 Authentication-Results: sourceware.org; auth=none X-Virus-Found: No X-Spam-SWARE-Status: No, score=-3.2 required=5.0 tests=AWL, BAYES_00, RCVD_IN_DNSWL_LOW, RP_MATCHES_RCVD, SPF_HELO_PASS, SPF_PASS autolearn=ham version=3.3.2 spammy=28023 X-HELO: plane.gmane.org To: libc-alpha@sourceware.org From: Stefan Liebler Subject: Re: [PATCH 08/14] S390: Optimize utf8-utf16 module. Date: Thu, 21 Apr 2016 16:50:13 +0200 Lines: 1534 Message-ID: References: <1456219278-5258-1-git-send-email-stli@linux.vnet.ibm.com> <1456219278-5258-9-git-send-email-stli@linux.vnet.ibm.com> Mime-Version: 1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.7.0 In-Reply-To: <1456219278-5258-9-git-send-email-stli@linux.vnet.ibm.com> Here is an updated patch, where the labels in inline assemblies are out-dented as suggested by Florian. On 02/23/2016 10:21 AM, Stefan Liebler wrote: > This patch reworks the s390 specific module to convert between utf8 and utf16. > Now ifunc is used to choose either the c or etf3eh (with convert utf instruction) > variants at runtime. Furthermore a new vector variant for z13 is introduced > which will be build and chosen if vector support is available at build / runtime. > > In case of converting utf 8 to utf16, the vector variant optimizes input of > 1byte utf8 characters. The convert utf instruction is used if a multibyte utf8 > character is found. > > For the other direction utf16 to utf8, the cu21 instruction can't be re-enabled, > because it does not report an error, if the input-stream consists of a single > low surrogate utf16 char (e.g. 0xdc00). This applies to the newest z13, too. > Thus there is only the c or the new vector variant, which can handle 1..4 byte > utf8 characters. > > The c variant from utf16 to utf8 has beed fixed. If a high surrogate was at the > end of the input-buffer, then errno was set to EINVAL and the input-pointer > pointed just after the high surrogate. Now it points to the beginning of the > high surrogate. > > This patch also fixes some whitespace errors. The c variant from utf8 to utf16 > is now checking that tail-bytes starts with 0b10... and the value is not in > range of an utf16 surrogate. > > Furthermore, the etf3eh variants are handling the "UTF-xx//IGNORE" case now. > Before they ignored the ignore-case and always stopped at an error. > > ChangeLog: > > * sysdeps/s390/s390-64/utf8-utf16-z9.c: Use ifunc to select c, > etf3eh or new vector loop-variant. > --- > sysdeps/s390/s390-64/utf8-utf16-z9.c | 547 ++++++++++++++++++++++++++++------- > 1 file changed, 441 insertions(+), 106 deletions(-) > > diff --git a/sysdeps/s390/s390-64/utf8-utf16-z9.c b/sysdeps/s390/s390-64/utf8-utf16-z9.c > index 4148ed7..76625d0 100644 > --- a/sysdeps/s390/s390-64/utf8-utf16-z9.c > +++ b/sysdeps/s390/s390-64/utf8-utf16-z9.c > @@ -30,33 +30,27 @@ > #include > #include > > -/* UTF-16 big endian byte order mark. */ > -#define BOM_UTF16 0xfeff > +#if defined HAVE_S390_VX_GCC_SUPPORT > +# define ASM_CLOBBER_VR(NR) , NR > +#else > +# define ASM_CLOBBER_VR(NR) > +#endif > > +/* Defines for skeleton.c. */ > #define DEFINE_INIT 0 > #define DEFINE_FINI 0 > #define MIN_NEEDED_FROM 1 > #define MAX_NEEDED_FROM 4 > #define MIN_NEEDED_TO 2 > #define MAX_NEEDED_TO 4 > -#define FROM_LOOP from_utf8_loop > -#define TO_LOOP to_utf8_loop > +#define FROM_LOOP __from_utf8_loop > +#define TO_LOOP __to_utf8_loop > #define FROM_DIRECTION (dir == from_utf8) > #define ONE_DIRECTION 0 > -#define PREPARE_LOOP \ > - enum direction dir = ((struct utf8_data *) step->__data)->dir; \ > - int emit_bom = ((struct utf8_data *) step->__data)->emit_bom; \ > - \ > - if (emit_bom && !data->__internal_use \ > - && data->__invocation_counter == 0) \ > - { \ > - /* Emit the UTF-16 Byte Order Mark. */ \ > - if (__glibc_unlikely (outbuf + 2 > outend)) \ > - return __GCONV_FULL_OUTPUT; \ > - \ > - put16u (outbuf, BOM_UTF16); \ > - outbuf += 2; \ > - } > + > + > +/* UTF-16 big endian byte order mark. */ > +#define BOM_UTF16 0xfeff > > /* Direction of the transformation. */ > enum direction > @@ -151,16 +145,16 @@ gconv_end (struct __gconv_step *data) > register unsigned long long outlen __asm__("11") = outend - outptr; \ > uint64_t cc = 0; \ > \ > - __asm__ volatile (".machine push \n\t" \ > - ".machine \"z9-109\" \n\t" \ > - "0: " INSTRUCTION " \n\t" \ > - ".machine pop \n\t" \ > - " jo 0b \n\t" \ > - " ipm %2 \n" \ > - : "+a" (pOutput), "+a" (pInput), "+d" (cc), \ > - "+d" (outlen), "+d" (inlen) \ > - : \ > - : "cc", "memory"); \ > + __asm__ __volatile__ (".machine push \n\t" \ > + ".machine \"z9-109\" \n\t" \ > + "0: " INSTRUCTION " \n\t" \ > + ".machine pop \n\t" \ > + " jo 0b \n\t" \ > + " ipm %2 \n" \ > + : "+a" (pOutput), "+a" (pInput), "+d" (cc), \ > + "+d" (outlen), "+d" (inlen) \ > + : \ > + : "cc", "memory"); \ > \ > inptr = pInput; \ > outptr = pOutput; \ > @@ -169,50 +163,135 @@ gconv_end (struct __gconv_step *data) > if (cc == 1) \ > { \ > result = __GCONV_FULL_OUTPUT; \ > - break; \ > } \ > else if (cc == 2) \ > { \ > result = __GCONV_ILLEGAL_INPUT; \ > - break; \ > } \ > } > > +#define PREPARE_LOOP \ > + enum direction dir = ((struct utf8_data *) step->__data)->dir; \ > + int emit_bom = ((struct utf8_data *) step->__data)->emit_bom; \ > + \ > + if (emit_bom && !data->__internal_use \ > + && data->__invocation_counter == 0) \ > + { \ > + /* Emit the UTF-16 Byte Order Mark. */ \ > + if (__glibc_unlikely (outbuf + 2 > outend)) \ > + return __GCONV_FULL_OUTPUT; \ > + \ > + put16u (outbuf, BOM_UTF16); \ > + outbuf += 2; \ > + } > + > /* Conversion function from UTF-8 to UTF-16. */ > +#define BODY_FROM_HW(ASM) \ > + { \ > + ASM; \ > + if (__glibc_likely (inptr == inend) \ > + || result == __GCONV_FULL_OUTPUT) \ > + break; \ > + \ > + int i; \ > + for (i = 1; inptr + i < inend && i < 5; ++i) \ > + if ((inptr[i] & 0xc0) != 0x80) \ > + break; \ > + \ > + if (__glibc_likely (inptr + i == inend \ > + && result == __GCONV_EMPTY_INPUT)) \ > + { \ > + result = __GCONV_INCOMPLETE_INPUT; \ > + break; \ > + } \ > + STANDARD_FROM_LOOP_ERR_HANDLER (i); \ > + } > + > +#define BODY_FROM_ETF3EH BODY_FROM_HW (HARDWARE_CONVERT ("cu12 %0, %1, 1")) > + > +#define HW_FROM_VX \ > + { \ > + register const unsigned char* pInput asm ("8") = inptr; \ > + register size_t inlen asm ("9") = inend - inptr; \ > + register unsigned char* pOutput asm ("10") = outptr; \ > + register size_t outlen asm("11") = outend - outptr; \ > + unsigned long tmp, tmp2, tmp3; \ > + asm volatile (".machine push\n\t" \ > + ".machine \"z13\"\n\t" \ > + ".machinemode \"zarch_nohighgprs\"\n\t" \ > + "vrepib %%v30,0x7f\n\t" /* For compare > 0x7f. */ \ > + "vrepib %%v31,0x20\n\t" \ > + /* Loop which handles UTF-8 chars <=0x7f. */ \ > + "0: clgijl %[R_INLEN],16,20f\n\t" \ > + "clgijl %[R_OUTLEN],32,20f\n\t" \ > + "1: vl %%v16,0(%[R_IN])\n\t" \ > + "vstrcbs %%v17,%%v16,%%v30,%%v31\n\t" \ > + "jno 10f\n\t" /* Jump away if not all bytes are 1byte \ > + UTF8 chars. */ \ > + /* Enlarge to UTF-16. */ \ > + "vuplhb %%v18,%%v16\n\t" \ > + "la %[R_IN],16(%[R_IN])\n\t" \ > + "vupllb %%v19,%%v16\n\t" \ > + "aghi %[R_INLEN],-16\n\t" \ > + /* Store 32 bytes to buf_out. */ \ > + "vstm %%v18,%%v19,0(%[R_OUT])\n\t" \ > + "aghi %[R_OUTLEN],-32\n\t" \ > + "la %[R_OUT],32(%[R_OUT])\n\t" \ > + "clgijl %[R_INLEN],16,20f\n\t" \ > + "clgijl %[R_OUTLEN],32,20f\n\t" \ > + "j 1b\n\t" \ > + "10:\n\t" \ > + /* At least one byte is > 0x7f. \ > + Store the preceding 1-byte chars. */ \ > + "vlgvb %[R_TMP],%%v17,7\n\t" \ > + "sllk %[R_TMP2],%[R_TMP],1\n\t" /* Compute highest \ > + index to store. */ \ > + "llgfr %[R_TMP3],%[R_TMP2]\n\t" \ > + "ahi %[R_TMP2],-1\n\t" \ > + "jl 20f\n\t" \ > + "vuplhb %%v18,%%v16\n\t" \ > + "vstl %%v18,%[R_TMP2],0(%[R_OUT])\n\t" \ > + "ahi %[R_TMP2],-16\n\t" \ > + "jl 11f\n\t" \ > + "vupllb %%v19,%%v16\n\t" \ > + "vstl %%v19,%[R_TMP2],16(%[R_OUT])\n\t" \ > + "11:\n\t" /* Update pointers. */ \ > + "la %[R_IN],0(%[R_TMP],%[R_IN])\n\t" \ > + "slgr %[R_INLEN],%[R_TMP]\n\t" \ > + "la %[R_OUT],0(%[R_TMP3],%[R_OUT])\n\t" \ > + "slgr %[R_OUTLEN],%[R_TMP3]\n\t" \ > + /* Handle multibyte utf8-char with convert instruction. */ \ > + "20: cu12 %[R_OUT],%[R_IN],1\n\t" \ > + "jo 0b\n\t" /* Try vector implemenation again. */ \ > + "lochil %[R_RES],%[RES_OUT_FULL]\n\t" /* cc == 1. */ \ > + "lochih %[R_RES],%[RES_IN_ILL]\n\t" /* cc == 2. */ \ > + ".machine pop" \ > + : /* outputs */ [R_IN] "+a" (pInput) \ > + , [R_INLEN] "+d" (inlen), [R_OUT] "+a" (pOutput) \ > + , [R_OUTLEN] "+d" (outlen), [R_TMP] "=a" (tmp) \ > + , [R_TMP2] "=d" (tmp2), [R_TMP3] "=a" (tmp3) \ > + , [R_RES] "+d" (result) \ > + : /* inputs */ \ > + [RES_OUT_FULL] "i" (__GCONV_FULL_OUTPUT) \ > + , [RES_IN_ILL] "i" (__GCONV_ILLEGAL_INPUT) \ > + : /* clobber list */ "memory", "cc" \ > + ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \ > + ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \ > + ASM_CLOBBER_VR ("v30") ASM_CLOBBER_VR ("v31") \ > + ); \ > + inptr = pInput; \ > + outptr = pOutput; \ > + } > +#define BODY_FROM_VX BODY_FROM_HW (HW_FROM_VX) > + > > -#define MIN_NEEDED_INPUT MIN_NEEDED_FROM > -#define MAX_NEEDED_INPUT MAX_NEEDED_FROM > -#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO > -#define MAX_NEEDED_OUTPUT MAX_NEEDED_TO > -#define LOOPFCT FROM_LOOP > /* The software implementation is based on the code in gconv_simple.c. */ > -#define BODY \ > +#define BODY_FROM_C \ > { \ > - if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH) \ > - { \ > - HARDWARE_CONVERT ("cu12 %0, %1, 1"); \ > - \ > - if (inptr != inend) \ > - { \ > - int i; \ > - for (i = 1; inptr + i < inend; ++i) \ > - if ((inptr[i] & 0xc0) != 0x80) \ > - break; \ > - \ > - if (__glibc_likely (inptr + i == inend)) \ > - { \ > - result = __GCONV_INCOMPLETE_INPUT; \ > - break; \ > - } \ > - STANDARD_FROM_LOOP_ERR_HANDLER (i); \ > - } \ > - continue; \ > - } \ > - \ > /* Next input byte. */ \ > uint16_t ch = *inptr; \ > \ > - if (__glibc_likely (ch < 0x80)) \ > + if (__glibc_likely (ch < 0x80)) \ > { \ > /* One byte sequence. */ \ > ++inptr; \ > @@ -230,13 +309,13 @@ gconv_end (struct __gconv_step *data) > cnt = 2; \ > ch &= 0x1f; \ > } \ > - else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \ > + else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \ > { \ > /* We expect three bytes. */ \ > cnt = 3; \ > ch &= 0x0f; \ > } \ > - else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \ > + else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \ > { \ > /* We expect four bytes. */ \ > cnt = 4; \ > @@ -257,7 +336,7 @@ gconv_end (struct __gconv_step *data) > STANDARD_FROM_LOOP_ERR_HANDLER (i); \ > } \ > \ > - if (__glibc_unlikely (inptr + cnt > inend)) \ > + if (__glibc_unlikely (inptr + cnt > inend)) \ > { \ > /* We don't have enough input. But before we report \ > that check that all the bytes are correct. */ \ > @@ -265,7 +344,7 @@ gconv_end (struct __gconv_step *data) > if ((inptr[i] & 0xc0) != 0x80) \ > break; \ > \ > - if (__glibc_likely (inptr + i == inend)) \ > + if (__glibc_likely (inptr + i == inend)) \ > { \ > result = __GCONV_INCOMPLETE_INPUT; \ > break; \ > @@ -280,23 +359,31 @@ gconv_end (struct __gconv_step *data) > low) are needed. */ \ > uint16_t zabcd, high, low; \ > \ > - if (__glibc_unlikely (outptr + 4 > outend)) \ > + if (__glibc_unlikely (outptr + 4 > outend)) \ > { \ > /* Overflow in the output buffer. */ \ > result = __GCONV_FULL_OUTPUT; \ > break; \ > } \ > \ > + /* Check if tail-bytes >= 0x80, < 0xc0. */ \ > + for (i = 1; i < cnt; ++i) \ > + { \ > + if ((inptr[i] & 0xc0) != 0x80) \ > + /* This is an illegal encoding. */ \ > + goto errout; \ > + } \ > + \ > /* See Principles of Operations cu12. */ \ > zabcd = (((inptr[0] & 0x7) << 2) | \ > - ((inptr[1] & 0x30) >> 4)) - 1; \ > + ((inptr[1] & 0x30) >> 4)) - 1; \ > \ > /* z-bit must be zero after subtracting 1. */ \ > if (zabcd & 0x10) \ > STANDARD_FROM_LOOP_ERR_HANDLER (4) \ > \ > high = (uint16_t)(0xd8 << 8); /* high surrogate id */ \ > - high |= zabcd << 6; /* abcd bits */ \ > + high |= zabcd << 6; /* abcd bits */ \ > high |= (inptr[1] & 0xf) << 2; /* efgh bits */ \ > high |= (inptr[2] & 0x30) >> 4; /* ij bits */ \ > \ > @@ -326,8 +413,19 @@ gconv_end (struct __gconv_step *data) > ch <<= 6; \ > ch |= byte & 0x3f; \ > } \ > - inptr += cnt; \ > \ > + /* If i < cnt, some trail byte was not >= 0x80, < 0xc0. \ > + If cnt > 2 and ch < 2^(5*cnt-4), the wide character ch could \ > + have been represented with fewer than cnt bytes. */ \ > + if (i < cnt || (cnt > 2 && (ch >> (5 * cnt - 4)) == 0) \ > + /* Do not accept UTF-16 surrogates. */ \ > + || (ch >= 0xd800 && ch <= 0xdfff)) \ > + { \ > + /* This is an illegal encoding. */ \ > + goto errout; \ > + } \ > + \ > + inptr += cnt; \ > } \ > } \ > /* Now adjust the pointers and store the result. */ \ > @@ -335,43 +433,70 @@ gconv_end (struct __gconv_step *data) > outptr += sizeof (uint16_t); \ > } > > +/* Generate loop-function with software implementation. */ > +#define MIN_NEEDED_INPUT MIN_NEEDED_FROM > +#define MAX_NEEDED_INPUT MAX_NEEDED_FROM > +#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO > +#define MAX_NEEDED_OUTPUT MAX_NEEDED_TO > +#define LOOPFCT __from_utf8_loop_c > +#define LOOP_NEED_FLAGS > +#define BODY BODY_FROM_C > +#include > + > +/* Generate loop-function with hardware utf-convert instruction. */ > +#define MIN_NEEDED_INPUT MIN_NEEDED_FROM > +#define MAX_NEEDED_INPUT MAX_NEEDED_FROM > +#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO > +#define MAX_NEEDED_OUTPUT MAX_NEEDED_TO > +#define LOOPFCT __from_utf8_loop_etf3eh > #define LOOP_NEED_FLAGS > +#define BODY BODY_FROM_ETF3EH > #include > > +#if defined HAVE_S390_VX_ASM_SUPPORT > +/* Generate loop-function with hardware vector and utf-convert instructions. */ > +# define MIN_NEEDED_INPUT MIN_NEEDED_FROM > +# define MAX_NEEDED_INPUT MAX_NEEDED_FROM > +# define MIN_NEEDED_OUTPUT MIN_NEEDED_TO > +# define MAX_NEEDED_OUTPUT MAX_NEEDED_TO > +# define LOOPFCT __from_utf8_loop_vx > +# define LOOP_NEED_FLAGS > +# define BODY BODY_FROM_VX > +# include > +#endif > + > + > +/* Generate ifunc'ed loop function. */ > +__typeof(__from_utf8_loop_c) > +__attribute__ ((ifunc ("__from_utf8_loop_resolver"))) > +__from_utf8_loop; > + > +static void * > +__from_utf8_loop_resolver (unsigned long int dl_hwcap) > +{ > +#if defined HAVE_S390_VX_ASM_SUPPORT > + if (dl_hwcap & HWCAP_S390_VX) > + return __from_utf8_loop_vx; > + else > +#endif > + if (dl_hwcap & HWCAP_S390_ETF3EH) > + return __from_utf8_loop_etf3eh; > + else > + return __from_utf8_loop_c; > +} > + > +strong_alias (__from_utf8_loop_c_single, __from_utf8_loop_single) > + > /* Conversion from UTF-16 to UTF-8. */ > > -#define MIN_NEEDED_INPUT MIN_NEEDED_TO > -#define MAX_NEEDED_INPUT MAX_NEEDED_TO > -#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM > -#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM > -#define LOOPFCT TO_LOOP > /* The software routine is based on the functionality of the S/390 > hardware instruction (cu21) as described in the Principles of > Operation. */ > -#define BODY \ > +#define BODY_TO_C \ > { \ > - /* The hardware instruction currently fails to report an error for \ > - isolated low surrogates so we have to disable the instruction \ > - until this gets resolved. */ \ > - if (0) /* (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH) */ \ > - { \ > - HARDWARE_CONVERT ("cu21 %0, %1, 1"); \ > - if (inptr != inend) \ > - { \ > - /* Check if the third byte is \ > - a valid start of a UTF-16 surrogate. */ \ > - if (inend - inptr == 3 && (inptr[3] & 0xfc) != 0xdc) \ > - STANDARD_TO_LOOP_ERR_HANDLER (3); \ > - \ > - result = __GCONV_INCOMPLETE_INPUT; \ > - break; \ > - } \ > - continue; \ > - } \ > - \ > uint16_t c = get16 (inptr); \ > \ > - if (__glibc_likely (c <= 0x007f)) \ > + if (__glibc_likely (c <= 0x007f)) \ > { \ > /* Single byte UTF-8 char. */ \ > *outptr = c & 0xff; \ > @@ -379,20 +504,20 @@ gconv_end (struct __gconv_step *data) > } \ > else if (c >= 0x0080 && c <= 0x07ff) \ > { \ > - /* Two byte UTF-8 char. */ \ > + /* Two byte UTF-8 char. */ \ > \ > - if (__glibc_unlikely (outptr + 2 > outend)) \ > + if (__glibc_unlikely (outptr + 2 > outend)) \ > { \ > /* Overflow in the output buffer. */ \ > result = __GCONV_FULL_OUTPUT; \ > break; \ > } \ > \ > - outptr[0] = 0xc0; \ > - outptr[0] |= c >> 6; \ > + outptr[0] = 0xc0; \ > + outptr[0] |= c >> 6; \ > \ > - outptr[1] = 0x80; \ > - outptr[1] |= c & 0x3f; \ > + outptr[1] = 0x80; \ > + outptr[1] |= c & 0x3f; \ > \ > outptr += 2; \ > } \ > @@ -400,7 +525,7 @@ gconv_end (struct __gconv_step *data) > { \ > /* Three byte UTF-8 char. */ \ > \ > - if (__glibc_unlikely (outptr + 3 > outend)) \ > + if (__glibc_unlikely (outptr + 3 > outend)) \ > { \ > /* Overflow in the output buffer. */ \ > result = __GCONV_FULL_OUTPUT; \ > @@ -419,22 +544,22 @@ gconv_end (struct __gconv_step *data) > } \ > else if (c >= 0xd800 && c <= 0xdbff) \ > { \ > - /* Four byte UTF-8 char. */ \ > + /* Four byte UTF-8 char. */ \ > uint16_t low, uvwxy; \ > \ > - if (__glibc_unlikely (outptr + 4 > outend)) \ > + if (__glibc_unlikely (outptr + 4 > outend)) \ > { \ > /* Overflow in the output buffer. */ \ > result = __GCONV_FULL_OUTPUT; \ > break; \ > } \ > - inptr += 2; \ > - if (__glibc_unlikely (inptr + 2 > inend)) \ > + if (__glibc_unlikely (inptr + 4 > inend)) \ > { \ > result = __GCONV_INCOMPLETE_INPUT; \ > break; \ > } \ > \ > + inptr += 2; \ > low = get16 (inptr); \ > \ > if ((low & 0xfc00) != 0xdc00) \ > @@ -461,11 +586,221 @@ gconv_end (struct __gconv_step *data) > } \ > else \ > { \ > - STANDARD_TO_LOOP_ERR_HANDLER (2); \ > + STANDARD_TO_LOOP_ERR_HANDLER (2); \ > } \ > inptr += 2; \ > } > -#define LOOP_NEED_FLAGS > -#include > + > +#define BODY_TO_VX \ > + { \ > + size_t inlen = inend - inptr; \ > + size_t outlen = outend - outptr; \ > + unsigned long tmp, tmp2, tmp3; \ > + asm volatile (".machine push\n\t" \ > + ".machine \"z13\"\n\t" \ > + ".machinemode \"zarch_nohighgprs\"\n\t" \ > + /* Setup to check for values <= 0x7f. */ \ > + "larl %[R_TMP],9f\n\t" \ > + "vlm %%v30,%%v31,0(%[R_TMP])\n\t" \ > + /* Loop which handles UTF-16 chars <=0x7f. */ \ > + "0: clgijl %[R_INLEN],32,2f\n\t" \ > + "clgijl %[R_OUTLEN],16,2f\n\t" \ > + "1: vlm %%v16,%%v17,0(%[R_IN])\n\t" \ > + "lghi %[R_TMP2],0\n\t" \ > + /* Check for > 1byte UTF-8 chars. */ \ > + "vstrchs %%v19,%%v16,%%v30,%%v31\n\t" \ > + "jno 10f\n\t" /* Jump away if not all bytes are 1byte \ > + UTF8 chars. */ \ > + "vstrchs %%v19,%%v17,%%v30,%%v31\n\t" \ > + "jno 11f\n\t" /* Jump away if not all bytes are 1byte \ > + UTF8 chars. */ \ > + /* Shorten to UTF-8. */ \ > + "vpkh %%v18,%%v16,%%v17\n\t" \ > + "la %[R_IN],32(%[R_IN])\n\t" \ > + "aghi %[R_INLEN],-32\n\t" \ > + /* Store 16 bytes to buf_out. */ \ > + "vst %%v18,0(%[R_OUT])\n\t" \ > + "aghi %[R_OUTLEN],-16\n\t" \ > + "la %[R_OUT],16(%[R_OUT])\n\t" \ > + "clgijl %[R_INLEN],32,2f\n\t" \ > + "clgijl %[R_OUTLEN],16,2f\n\t" \ > + "j 1b\n\t" \ > + /* Setup to check for ch > 0x7f. (v30, v31) */ \ > + "9: .short 0x7f,0x7f,0x0,0x0,0x0,0x0,0x0,0x0\n\t" \ > + ".short 0x2000,0x2000,0x0,0x0,0x0,0x0,0x0,0x0\n\t" \ > + /* At least one byte is > 0x7f. \ > + Store the preceding 1-byte chars. */ \ > + "11: lghi %[R_TMP2],16\n\t" /* match was found in v17. */ \ > + "10:\n\t" \ > + "vlgvb %[R_TMP],%%v19,7\n\t" \ > + /* Shorten to UTF-8. */ \ > + "vpkh %%v18,%%v16,%%v17\n\t" \ > + "ar %[R_TMP],%[R_TMP2]\n\t" /* Number of in bytes. */ \ > + "srlg %[R_TMP3],%[R_TMP],1\n\t" /* Number of out bytes. */ \ > + "ahik %[R_TMP2],%[R_TMP3],-1\n\t" /* Highest index to store. */ \ > + "jl 13f\n\t" \ > + "vstl %%v18,%[R_TMP2],0(%[R_OUT])\n\t" \ > + /* Update pointers. */ \ > + "la %[R_IN],0(%[R_TMP],%[R_IN])\n\t" \ > + "slgr %[R_INLEN],%[R_TMP]\n\t" \ > + "la %[R_OUT],0(%[R_TMP3],%[R_OUT])\n\t" \ > + "slgr %[R_OUTLEN],%[R_TMP3]\n\t" \ > + "13:\n\t" \ > + /* Calculate remaining uint16_t values in loaded vrs. */ \ > + "lghi %[R_TMP2],16\n\t" \ > + "slgr %[R_TMP2],%[R_TMP3]\n\t" \ > + "llh %[R_TMP],0(%[R_IN])\n\t" \ > + "aghi %[R_INLEN],-2\n\t" \ > + "j 22f\n\t" \ > + /* Handle remaining bytes. */ \ > + "2:\n\t" \ > + /* Zero, one or more bytes available? */ \ > + "clgfi %[R_INLEN],1\n\t" \ > + "locghie %[R_RES],%[RES_IN_FULL]\n\t" /* Only one byte. */ \ > + "jle 99f\n\t" /* End if less than two bytes. */ \ > + /* Calculate remaining uint16_t values in inptr. */ \ > + "srlg %[R_TMP2],%[R_INLEN],1\n\t" \ > + /* Handle multibyte utf8-char. */ \ > + "20: llh %[R_TMP],0(%[R_IN])\n\t" \ > + "aghi %[R_INLEN],-2\n\t" \ > + /* Test if ch is 1-byte UTF-8 char. */ \ > + "21: clijh %[R_TMP],0x7f,22f\n\t" \ > + /* Handle 1-byte UTF-8 char. */ \ > + "31: slgfi %[R_OUTLEN],1\n\t" \ > + "jl 90f \n\t" \ > + "stc %[R_TMP],0(%[R_OUT])\n\t" \ > + "la %[R_IN],2(%[R_IN])\n\t" \ > + "la %[R_OUT],1(%[R_OUT])\n\t" \ > + "brctg %[R_TMP2],20b\n\t" \ > + "j 0b\n\t" /* Switch to vx-loop. */ \ > + /* Test if ch is 2-byte UTF-8 char. */ \ > + "22: clfi %[R_TMP],0x7ff\n\t" \ > + "jh 23f\n\t" \ > + /* Handle 2-byte UTF-8 char. */ \ > + "32: slgfi %[R_OUTLEN],2\n\t" \ > + "jl 90f \n\t" \ > + "llill %[R_TMP3],0xc080\n\t" \ > + "la %[R_IN],2(%[R_IN])\n\t" \ > + "risbgn %[R_TMP3],%[R_TMP],51,55,2\n\t" /* 1. byte. */ \ > + "risbgn %[R_TMP3],%[R_TMP],58,63,0\n\t" /* 2. byte. */ \ > + "sth %[R_TMP3],0(%[R_OUT])\n\t" \ > + "la %[R_OUT],2(%[R_OUT])\n\t" \ > + "brctg %[R_TMP2],20b\n\t" \ > + "j 0b\n\t" /* Switch to vx-loop. */ \ > + /* Test if ch is 3-byte UTF-8 char. */ \ > + "23: clfi %[R_TMP],0xd7ff\n\t" \ > + "jh 24f\n\t" \ > + /* Handle 3-byte UTF-8 char. */ \ > + "33: slgfi %[R_OUTLEN],3\n\t" \ > + "jl 90f \n\t" \ > + "llilf %[R_TMP3],0xe08080\n\t" \ > + "la %[R_IN],2(%[R_IN])\n\t" \ > + "risbgn %[R_TMP3],%[R_TMP],44,47,4\n\t" /* 1. byte. */ \ > + "risbgn %[R_TMP3],%[R_TMP],50,55,2\n\t" /* 2. byte. */ \ > + "risbgn %[R_TMP3],%[R_TMP],58,63,0\n\t" /* 3. byte. */ \ > + "stcm %[R_TMP3],7,0(%[R_OUT])\n\t" \ > + "la %[R_OUT],3(%[R_OUT])\n\t" \ > + "brctg %[R_TMP2],20b\n\t" \ > + "j 0b\n\t" /* Switch to vx-loop. */ \ > + /* Test if ch is 4-byte UTF-8 char. */ \ > + "24: clfi %[R_TMP],0xdfff\n\t" \ > + "jh 33b\n\t" /* Handle this 3-byte UTF-8 char. */ \ > + "clfi %[R_TMP],0xdbff\n\t" \ > + "locghih %[R_RES],%[RES_IN_ILL]\n\t" \ > + "jh 99f\n\t" /* Jump away if this is a low surrogate \ > + without a preceding high surrogate. */ \ > + /* Handle 4-byte UTF-8 char. */ \ > + "34: slgfi %[R_OUTLEN],4\n\t" \ > + "jl 90f \n\t" \ > + "slgfi %[R_INLEN],2\n\t" \ > + "locghil %[R_RES],%[RES_IN_FULL]\n\t" \ > + "jl 99f\n\t" /* Jump away if low surrogate is missing. */ \ > + "llilf %[R_TMP3],0xf0808080\n\t" \ > + "aghi %[R_TMP],0x40\n\t" \ > + "risbgn %[R_TMP3],%[R_TMP],37,39,16\n\t" /* 1. byte: uvw */ \ > + "risbgn %[R_TMP3],%[R_TMP],42,43,14\n\t" /* 2. byte: xy */ \ > + "risbgn %[R_TMP3],%[R_TMP],44,47,14\n\t" /* 2. byte: efgh */ \ > + "risbgn %[R_TMP3],%[R_TMP],50,51,12\n\t" /* 3. byte: ij */ \ > + "llh %[R_TMP],2(%[R_IN])\n\t" /* Load low surrogate. */ \ > + "risbgn %[R_TMP3],%[R_TMP],52,55,2\n\t" /* 3. byte: klmn */ \ > + "risbgn %[R_TMP3],%[R_TMP],58,63,0\n\t" /* 4. byte: opqrst */ \ > + "nilf %[R_TMP],0xfc00\n\t" \ > + "clfi %[R_TMP],0xdc00\n\t" /* Check if it starts with 0xdc00. */ \ > + "locghine %[R_RES],%[RES_IN_ILL]\n\t" \ > + "jne 99f\n\t" /* Jump away if low surrogate is invalid. */ \ > + "st %[R_TMP3],0(%[R_OUT])\n\t" \ > + "la %[R_IN],4(%[R_IN])\n\t" \ > + "la %[R_OUT],4(%[R_OUT])\n\t" \ > + "aghi %[R_TMP2],-2\n\t" \ > + "jh 20b\n\t" \ > + "j 0b\n\t" /* Switch to vx-loop. */ \ > + /* Exit with __GCONV_FULL_OUTPUT. */ \ > + "90: lghi %[R_RES],%[RES_OUT_FULL]\n\t" \ > + "99:\n\t" \ > + ".machine pop" \ > + : /* outputs */ [R_IN] "+a" (inptr) \ > + , [R_INLEN] "+d" (inlen), [R_OUT] "+a" (outptr) \ > + , [R_OUTLEN] "+d" (outlen), [R_TMP] "=a" (tmp) \ > + , [R_TMP2] "=d" (tmp2), [R_TMP3] "=a" (tmp3) \ > + , [R_RES] "+d" (result) \ > + : /* inputs */ \ > + [RES_OUT_FULL] "i" (__GCONV_FULL_OUTPUT) \ > + , [RES_IN_ILL] "i" (__GCONV_ILLEGAL_INPUT) \ > + , [RES_IN_FULL] "i" (__GCONV_INCOMPLETE_INPUT) \ > + : /* clobber list */ "memory", "cc" \ > + ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \ > + ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \ > + ASM_CLOBBER_VR ("v30") ASM_CLOBBER_VR ("v31") \ > + ); \ > + if (__glibc_likely (inptr == inend) \ > + || result != __GCONV_ILLEGAL_INPUT) \ > + break; \ > + \ > + STANDARD_TO_LOOP_ERR_HANDLER (2); \ > + } > + > +/* Generate loop-function with software implementation. */ > +#define MIN_NEEDED_INPUT MIN_NEEDED_TO > +#define MAX_NEEDED_INPUT MAX_NEEDED_TO > +#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM > +#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM > +#if defined HAVE_S390_VX_ASM_SUPPORT > +# define LOOPFCT __to_utf8_loop_c > +# define BODY BODY_TO_C > +# define LOOP_NEED_FLAGS > +# include > + > +/* Generate loop-function with software implementation. */ > +# define MIN_NEEDED_INPUT MIN_NEEDED_TO > +# define MAX_NEEDED_INPUT MAX_NEEDED_TO > +# define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM > +# define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM > +# define LOOPFCT __to_utf8_loop_vx > +# define BODY BODY_TO_VX > +# define LOOP_NEED_FLAGS > +# include > + > +/* Generate ifunc'ed loop function. */ > +__typeof(__to_utf8_loop_c) > +__attribute__ ((ifunc ("__to_utf8_loop_resolver"))) > +__to_utf8_loop; > + > +static void * > +__to_utf8_loop_resolver (unsigned long int dl_hwcap) > +{ > + if (dl_hwcap & HWCAP_S390_VX) > + return __to_utf8_loop_vx; > + else > + return __to_utf8_loop_c; > +} > + > +strong_alias (__to_utf8_loop_c_single, __to_utf8_loop_single) > + > +#else > +# define LOOPFCT TO_LOOP > +# define BODY BODY_TO_C > +# define LOOP_NEED_FLAGS > +# include > +#endif /* !HAVE_S390_VX_ASM_SUPPORT */ > > #include > From 85018cce92ca04453f7b7177d6dd4100349f2ec3 Mon Sep 17 00:00:00 2001 From: Stefan Liebler Date: Thu, 21 Apr 2016 12:42:49 +0200 Subject: [PATCH 08/14] S390: Optimize utf8-utf16 module. This patch reworks the s390 specific module to convert between utf8 and utf16. Now ifunc is used to choose either the c or etf3eh (with convert utf instruction) variants at runtime. Furthermore a new vector variant for z13 is introduced which will be build and chosen if vector support is available at build / runtime. In case of converting utf 8 to utf16, the vector variant optimizes input of 1byte utf8 characters. The convert utf instruction is used if a multibyte utf8 character is found. For the other direction utf16 to utf8, the cu21 instruction can't be re-enabled, because it does not report an error, if the input-stream consists of a single low surrogate utf16 char (e.g. 0xdc00). This applies to the newest z13, too. Thus there is only the c or the new vector variant, which can handle 1..4 byte utf8 characters. The c variant from utf16 to utf8 has beed fixed. If a high surrogate was at the end of the input-buffer, then errno was set to EINVAL and the input-pointer pointed just after the high surrogate. Now it points to the beginning of the high surrogate. This patch also fixes some whitespace errors. The c variant from utf8 to utf16 is now checking that tail-bytes starts with 0b10... and the value is not in range of an utf16 surrogate. Furthermore, the etf3eh variants are handling the "UTF-xx//IGNORE" case now. Before they ignored the ignore-case and always stopped at an error. ChangeLog: * sysdeps/s390/s390-64/utf8-utf16-z9.c: Use ifunc to select c, etf3eh or new vector loop-variant. --- sysdeps/s390/s390-64/utf8-utf16-z9.c | 547 ++++++++++++++++++++++++++++------- 1 file changed, 441 insertions(+), 106 deletions(-) diff --git a/sysdeps/s390/s390-64/utf8-utf16-z9.c b/sysdeps/s390/s390-64/utf8-utf16-z9.c index 4148ed7..7520ef2 100644 --- a/sysdeps/s390/s390-64/utf8-utf16-z9.c +++ b/sysdeps/s390/s390-64/utf8-utf16-z9.c @@ -30,33 +30,27 @@ #include #include -/* UTF-16 big endian byte order mark. */ -#define BOM_UTF16 0xfeff +#if defined HAVE_S390_VX_GCC_SUPPORT +# define ASM_CLOBBER_VR(NR) , NR +#else +# define ASM_CLOBBER_VR(NR) +#endif +/* Defines for skeleton.c. */ #define DEFINE_INIT 0 #define DEFINE_FINI 0 #define MIN_NEEDED_FROM 1 #define MAX_NEEDED_FROM 4 #define MIN_NEEDED_TO 2 #define MAX_NEEDED_TO 4 -#define FROM_LOOP from_utf8_loop -#define TO_LOOP to_utf8_loop +#define FROM_LOOP __from_utf8_loop +#define TO_LOOP __to_utf8_loop #define FROM_DIRECTION (dir == from_utf8) #define ONE_DIRECTION 0 -#define PREPARE_LOOP \ - enum direction dir = ((struct utf8_data *) step->__data)->dir; \ - int emit_bom = ((struct utf8_data *) step->__data)->emit_bom; \ - \ - if (emit_bom && !data->__internal_use \ - && data->__invocation_counter == 0) \ - { \ - /* Emit the UTF-16 Byte Order Mark. */ \ - if (__glibc_unlikely (outbuf + 2 > outend)) \ - return __GCONV_FULL_OUTPUT; \ - \ - put16u (outbuf, BOM_UTF16); \ - outbuf += 2; \ - } + + +/* UTF-16 big endian byte order mark. */ +#define BOM_UTF16 0xfeff /* Direction of the transformation. */ enum direction @@ -151,16 +145,16 @@ gconv_end (struct __gconv_step *data) register unsigned long long outlen __asm__("11") = outend - outptr; \ uint64_t cc = 0; \ \ - __asm__ volatile (".machine push \n\t" \ - ".machine \"z9-109\" \n\t" \ - "0: " INSTRUCTION " \n\t" \ - ".machine pop \n\t" \ - " jo 0b \n\t" \ - " ipm %2 \n" \ - : "+a" (pOutput), "+a" (pInput), "+d" (cc), \ - "+d" (outlen), "+d" (inlen) \ - : \ - : "cc", "memory"); \ + __asm__ __volatile__ (".machine push \n\t" \ + ".machine \"z9-109\" \n\t" \ + "0: " INSTRUCTION " \n\t" \ + ".machine pop \n\t" \ + " jo 0b \n\t" \ + " ipm %2 \n" \ + : "+a" (pOutput), "+a" (pInput), "+d" (cc), \ + "+d" (outlen), "+d" (inlen) \ + : \ + : "cc", "memory"); \ \ inptr = pInput; \ outptr = pOutput; \ @@ -169,50 +163,135 @@ gconv_end (struct __gconv_step *data) if (cc == 1) \ { \ result = __GCONV_FULL_OUTPUT; \ - break; \ } \ else if (cc == 2) \ { \ result = __GCONV_ILLEGAL_INPUT; \ - break; \ } \ } +#define PREPARE_LOOP \ + enum direction dir = ((struct utf8_data *) step->__data)->dir; \ + int emit_bom = ((struct utf8_data *) step->__data)->emit_bom; \ + \ + if (emit_bom && !data->__internal_use \ + && data->__invocation_counter == 0) \ + { \ + /* Emit the UTF-16 Byte Order Mark. */ \ + if (__glibc_unlikely (outbuf + 2 > outend)) \ + return __GCONV_FULL_OUTPUT; \ + \ + put16u (outbuf, BOM_UTF16); \ + outbuf += 2; \ + } + /* Conversion function from UTF-8 to UTF-16. */ +#define BODY_FROM_HW(ASM) \ + { \ + ASM; \ + if (__glibc_likely (inptr == inend) \ + || result == __GCONV_FULL_OUTPUT) \ + break; \ + \ + int i; \ + for (i = 1; inptr + i < inend && i < 5; ++i) \ + if ((inptr[i] & 0xc0) != 0x80) \ + break; \ + \ + if (__glibc_likely (inptr + i == inend \ + && result == __GCONV_EMPTY_INPUT)) \ + { \ + result = __GCONV_INCOMPLETE_INPUT; \ + break; \ + } \ + STANDARD_FROM_LOOP_ERR_HANDLER (i); \ + } + +#define BODY_FROM_ETF3EH BODY_FROM_HW (HARDWARE_CONVERT ("cu12 %0, %1, 1")) + +#define HW_FROM_VX \ + { \ + register const unsigned char* pInput asm ("8") = inptr; \ + register size_t inlen asm ("9") = inend - inptr; \ + register unsigned char* pOutput asm ("10") = outptr; \ + register size_t outlen asm("11") = outend - outptr; \ + unsigned long tmp, tmp2, tmp3; \ + asm volatile (".machine push\n\t" \ + ".machine \"z13\"\n\t" \ + ".machinemode \"zarch_nohighgprs\"\n\t" \ + " vrepib %%v30,0x7f\n\t" /* For compare > 0x7f. */ \ + " vrepib %%v31,0x20\n\t" \ + /* Loop which handles UTF-8 chars <=0x7f. */ \ + "0: clgijl %[R_INLEN],16,20f\n\t" \ + " clgijl %[R_OUTLEN],32,20f\n\t" \ + "1: vl %%v16,0(%[R_IN])\n\t" \ + " vstrcbs %%v17,%%v16,%%v30,%%v31\n\t" \ + " jno 10f\n\t" /* Jump away if not all bytes are 1byte \ + UTF8 chars. */ \ + /* Enlarge to UTF-16. */ \ + " vuplhb %%v18,%%v16\n\t" \ + " la %[R_IN],16(%[R_IN])\n\t" \ + " vupllb %%v19,%%v16\n\t" \ + " aghi %[R_INLEN],-16\n\t" \ + /* Store 32 bytes to buf_out. */ \ + " vstm %%v18,%%v19,0(%[R_OUT])\n\t" \ + " aghi %[R_OUTLEN],-32\n\t" \ + " la %[R_OUT],32(%[R_OUT])\n\t" \ + " clgijl %[R_INLEN],16,20f\n\t" \ + " clgijl %[R_OUTLEN],32,20f\n\t" \ + " j 1b\n\t" \ + "10:\n\t" \ + /* At least one byte is > 0x7f. \ + Store the preceding 1-byte chars. */ \ + " vlgvb %[R_TMP],%%v17,7\n\t" \ + " sllk %[R_TMP2],%[R_TMP],1\n\t" /* Compute highest \ + index to store. */ \ + " llgfr %[R_TMP3],%[R_TMP2]\n\t" \ + " ahi %[R_TMP2],-1\n\t" \ + " jl 20f\n\t" \ + " vuplhb %%v18,%%v16\n\t" \ + " vstl %%v18,%[R_TMP2],0(%[R_OUT])\n\t" \ + " ahi %[R_TMP2],-16\n\t" \ + " jl 11f\n\t" \ + " vupllb %%v19,%%v16\n\t" \ + " vstl %%v19,%[R_TMP2],16(%[R_OUT])\n\t" \ + "11: \n\t" /* Update pointers. */ \ + " la %[R_IN],0(%[R_TMP],%[R_IN])\n\t" \ + " slgr %[R_INLEN],%[R_TMP]\n\t" \ + " la %[R_OUT],0(%[R_TMP3],%[R_OUT])\n\t" \ + " slgr %[R_OUTLEN],%[R_TMP3]\n\t" \ + /* Handle multibyte utf8-char with convert instruction. */ \ + "20: cu12 %[R_OUT],%[R_IN],1\n\t" \ + " jo 0b\n\t" /* Try vector implemenation again. */ \ + " lochil %[R_RES],%[RES_OUT_FULL]\n\t" /* cc == 1. */ \ + " lochih %[R_RES],%[RES_IN_ILL]\n\t" /* cc == 2. */ \ + ".machine pop" \ + : /* outputs */ [R_IN] "+a" (pInput) \ + , [R_INLEN] "+d" (inlen), [R_OUT] "+a" (pOutput) \ + , [R_OUTLEN] "+d" (outlen), [R_TMP] "=a" (tmp) \ + , [R_TMP2] "=d" (tmp2), [R_TMP3] "=a" (tmp3) \ + , [R_RES] "+d" (result) \ + : /* inputs */ \ + [RES_OUT_FULL] "i" (__GCONV_FULL_OUTPUT) \ + , [RES_IN_ILL] "i" (__GCONV_ILLEGAL_INPUT) \ + : /* clobber list */ "memory", "cc" \ + ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \ + ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \ + ASM_CLOBBER_VR ("v30") ASM_CLOBBER_VR ("v31") \ + ); \ + inptr = pInput; \ + outptr = pOutput; \ + } +#define BODY_FROM_VX BODY_FROM_HW (HW_FROM_VX) + -#define MIN_NEEDED_INPUT MIN_NEEDED_FROM -#define MAX_NEEDED_INPUT MAX_NEEDED_FROM -#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO -#define MAX_NEEDED_OUTPUT MAX_NEEDED_TO -#define LOOPFCT FROM_LOOP /* The software implementation is based on the code in gconv_simple.c. */ -#define BODY \ +#define BODY_FROM_C \ { \ - if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH) \ - { \ - HARDWARE_CONVERT ("cu12 %0, %1, 1"); \ - \ - if (inptr != inend) \ - { \ - int i; \ - for (i = 1; inptr + i < inend; ++i) \ - if ((inptr[i] & 0xc0) != 0x80) \ - break; \ - \ - if (__glibc_likely (inptr + i == inend)) \ - { \ - result = __GCONV_INCOMPLETE_INPUT; \ - break; \ - } \ - STANDARD_FROM_LOOP_ERR_HANDLER (i); \ - } \ - continue; \ - } \ - \ /* Next input byte. */ \ uint16_t ch = *inptr; \ \ - if (__glibc_likely (ch < 0x80)) \ + if (__glibc_likely (ch < 0x80)) \ { \ /* One byte sequence. */ \ ++inptr; \ @@ -230,13 +309,13 @@ gconv_end (struct __gconv_step *data) cnt = 2; \ ch &= 0x1f; \ } \ - else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \ + else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \ { \ /* We expect three bytes. */ \ cnt = 3; \ ch &= 0x0f; \ } \ - else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \ + else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \ { \ /* We expect four bytes. */ \ cnt = 4; \ @@ -257,7 +336,7 @@ gconv_end (struct __gconv_step *data) STANDARD_FROM_LOOP_ERR_HANDLER (i); \ } \ \ - if (__glibc_unlikely (inptr + cnt > inend)) \ + if (__glibc_unlikely (inptr + cnt > inend)) \ { \ /* We don't have enough input. But before we report \ that check that all the bytes are correct. */ \ @@ -265,7 +344,7 @@ gconv_end (struct __gconv_step *data) if ((inptr[i] & 0xc0) != 0x80) \ break; \ \ - if (__glibc_likely (inptr + i == inend)) \ + if (__glibc_likely (inptr + i == inend)) \ { \ result = __GCONV_INCOMPLETE_INPUT; \ break; \ @@ -280,23 +359,31 @@ gconv_end (struct __gconv_step *data) low) are needed. */ \ uint16_t zabcd, high, low; \ \ - if (__glibc_unlikely (outptr + 4 > outend)) \ + if (__glibc_unlikely (outptr + 4 > outend)) \ { \ /* Overflow in the output buffer. */ \ result = __GCONV_FULL_OUTPUT; \ break; \ } \ \ + /* Check if tail-bytes >= 0x80, < 0xc0. */ \ + for (i = 1; i < cnt; ++i) \ + { \ + if ((inptr[i] & 0xc0) != 0x80) \ + /* This is an illegal encoding. */ \ + goto errout; \ + } \ + \ /* See Principles of Operations cu12. */ \ zabcd = (((inptr[0] & 0x7) << 2) | \ - ((inptr[1] & 0x30) >> 4)) - 1; \ + ((inptr[1] & 0x30) >> 4)) - 1; \ \ /* z-bit must be zero after subtracting 1. */ \ if (zabcd & 0x10) \ STANDARD_FROM_LOOP_ERR_HANDLER (4) \ \ high = (uint16_t)(0xd8 << 8); /* high surrogate id */ \ - high |= zabcd << 6; /* abcd bits */ \ + high |= zabcd << 6; /* abcd bits */ \ high |= (inptr[1] & 0xf) << 2; /* efgh bits */ \ high |= (inptr[2] & 0x30) >> 4; /* ij bits */ \ \ @@ -326,8 +413,19 @@ gconv_end (struct __gconv_step *data) ch <<= 6; \ ch |= byte & 0x3f; \ } \ - inptr += cnt; \ \ + /* If i < cnt, some trail byte was not >= 0x80, < 0xc0. \ + If cnt > 2 and ch < 2^(5*cnt-4), the wide character ch could \ + have been represented with fewer than cnt bytes. */ \ + if (i < cnt || (cnt > 2 && (ch >> (5 * cnt - 4)) == 0) \ + /* Do not accept UTF-16 surrogates. */ \ + || (ch >= 0xd800 && ch <= 0xdfff)) \ + { \ + /* This is an illegal encoding. */ \ + goto errout; \ + } \ + \ + inptr += cnt; \ } \ } \ /* Now adjust the pointers and store the result. */ \ @@ -335,43 +433,70 @@ gconv_end (struct __gconv_step *data) outptr += sizeof (uint16_t); \ } +/* Generate loop-function with software implementation. */ +#define MIN_NEEDED_INPUT MIN_NEEDED_FROM +#define MAX_NEEDED_INPUT MAX_NEEDED_FROM +#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO +#define MAX_NEEDED_OUTPUT MAX_NEEDED_TO +#define LOOPFCT __from_utf8_loop_c +#define LOOP_NEED_FLAGS +#define BODY BODY_FROM_C +#include + +/* Generate loop-function with hardware utf-convert instruction. */ +#define MIN_NEEDED_INPUT MIN_NEEDED_FROM +#define MAX_NEEDED_INPUT MAX_NEEDED_FROM +#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO +#define MAX_NEEDED_OUTPUT MAX_NEEDED_TO +#define LOOPFCT __from_utf8_loop_etf3eh #define LOOP_NEED_FLAGS +#define BODY BODY_FROM_ETF3EH #include +#if defined HAVE_S390_VX_ASM_SUPPORT +/* Generate loop-function with hardware vector and utf-convert instructions. */ +# define MIN_NEEDED_INPUT MIN_NEEDED_FROM +# define MAX_NEEDED_INPUT MAX_NEEDED_FROM +# define MIN_NEEDED_OUTPUT MIN_NEEDED_TO +# define MAX_NEEDED_OUTPUT MAX_NEEDED_TO +# define LOOPFCT __from_utf8_loop_vx +# define LOOP_NEED_FLAGS +# define BODY BODY_FROM_VX +# include +#endif + + +/* Generate ifunc'ed loop function. */ +__typeof(__from_utf8_loop_c) +__attribute__ ((ifunc ("__from_utf8_loop_resolver"))) +__from_utf8_loop; + +static void * +__from_utf8_loop_resolver (unsigned long int dl_hwcap) +{ +#if defined HAVE_S390_VX_ASM_SUPPORT + if (dl_hwcap & HWCAP_S390_VX) + return __from_utf8_loop_vx; + else +#endif + if (dl_hwcap & HWCAP_S390_ETF3EH) + return __from_utf8_loop_etf3eh; + else + return __from_utf8_loop_c; +} + +strong_alias (__from_utf8_loop_c_single, __from_utf8_loop_single) + /* Conversion from UTF-16 to UTF-8. */ -#define MIN_NEEDED_INPUT MIN_NEEDED_TO -#define MAX_NEEDED_INPUT MAX_NEEDED_TO -#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM -#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM -#define LOOPFCT TO_LOOP /* The software routine is based on the functionality of the S/390 hardware instruction (cu21) as described in the Principles of Operation. */ -#define BODY \ +#define BODY_TO_C \ { \ - /* The hardware instruction currently fails to report an error for \ - isolated low surrogates so we have to disable the instruction \ - until this gets resolved. */ \ - if (0) /* (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH) */ \ - { \ - HARDWARE_CONVERT ("cu21 %0, %1, 1"); \ - if (inptr != inend) \ - { \ - /* Check if the third byte is \ - a valid start of a UTF-16 surrogate. */ \ - if (inend - inptr == 3 && (inptr[3] & 0xfc) != 0xdc) \ - STANDARD_TO_LOOP_ERR_HANDLER (3); \ - \ - result = __GCONV_INCOMPLETE_INPUT; \ - break; \ - } \ - continue; \ - } \ - \ uint16_t c = get16 (inptr); \ \ - if (__glibc_likely (c <= 0x007f)) \ + if (__glibc_likely (c <= 0x007f)) \ { \ /* Single byte UTF-8 char. */ \ *outptr = c & 0xff; \ @@ -379,20 +504,20 @@ gconv_end (struct __gconv_step *data) } \ else if (c >= 0x0080 && c <= 0x07ff) \ { \ - /* Two byte UTF-8 char. */ \ + /* Two byte UTF-8 char. */ \ \ - if (__glibc_unlikely (outptr + 2 > outend)) \ + if (__glibc_unlikely (outptr + 2 > outend)) \ { \ /* Overflow in the output buffer. */ \ result = __GCONV_FULL_OUTPUT; \ break; \ } \ \ - outptr[0] = 0xc0; \ - outptr[0] |= c >> 6; \ + outptr[0] = 0xc0; \ + outptr[0] |= c >> 6; \ \ - outptr[1] = 0x80; \ - outptr[1] |= c & 0x3f; \ + outptr[1] = 0x80; \ + outptr[1] |= c & 0x3f; \ \ outptr += 2; \ } \ @@ -400,7 +525,7 @@ gconv_end (struct __gconv_step *data) { \ /* Three byte UTF-8 char. */ \ \ - if (__glibc_unlikely (outptr + 3 > outend)) \ + if (__glibc_unlikely (outptr + 3 > outend)) \ { \ /* Overflow in the output buffer. */ \ result = __GCONV_FULL_OUTPUT; \ @@ -419,22 +544,22 @@ gconv_end (struct __gconv_step *data) } \ else if (c >= 0xd800 && c <= 0xdbff) \ { \ - /* Four byte UTF-8 char. */ \ + /* Four byte UTF-8 char. */ \ uint16_t low, uvwxy; \ \ - if (__glibc_unlikely (outptr + 4 > outend)) \ + if (__glibc_unlikely (outptr + 4 > outend)) \ { \ /* Overflow in the output buffer. */ \ result = __GCONV_FULL_OUTPUT; \ break; \ } \ - inptr += 2; \ - if (__glibc_unlikely (inptr + 2 > inend)) \ + if (__glibc_unlikely (inptr + 4 > inend)) \ { \ result = __GCONV_INCOMPLETE_INPUT; \ break; \ } \ \ + inptr += 2; \ low = get16 (inptr); \ \ if ((low & 0xfc00) != 0xdc00) \ @@ -461,11 +586,221 @@ gconv_end (struct __gconv_step *data) } \ else \ { \ - STANDARD_TO_LOOP_ERR_HANDLER (2); \ + STANDARD_TO_LOOP_ERR_HANDLER (2); \ } \ inptr += 2; \ } -#define LOOP_NEED_FLAGS -#include + +#define BODY_TO_VX \ + { \ + size_t inlen = inend - inptr; \ + size_t outlen = outend - outptr; \ + unsigned long tmp, tmp2, tmp3; \ + asm volatile (".machine push\n\t" \ + ".machine \"z13\"\n\t" \ + ".machinemode \"zarch_nohighgprs\"\n\t" \ + /* Setup to check for values <= 0x7f. */ \ + " larl %[R_TMP],9f\n\t" \ + " vlm %%v30,%%v31,0(%[R_TMP])\n\t" \ + /* Loop which handles UTF-16 chars <=0x7f. */ \ + "0: clgijl %[R_INLEN],32,2f\n\t" \ + " clgijl %[R_OUTLEN],16,2f\n\t" \ + "1: vlm %%v16,%%v17,0(%[R_IN])\n\t" \ + " lghi %[R_TMP2],0\n\t" \ + /* Check for > 1byte UTF-8 chars. */ \ + " vstrchs %%v19,%%v16,%%v30,%%v31\n\t" \ + " jno 10f\n\t" /* Jump away if not all bytes are 1byte \ + UTF8 chars. */ \ + " vstrchs %%v19,%%v17,%%v30,%%v31\n\t" \ + " jno 11f\n\t" /* Jump away if not all bytes are 1byte \ + UTF8 chars. */ \ + /* Shorten to UTF-8. */ \ + " vpkh %%v18,%%v16,%%v17\n\t" \ + " la %[R_IN],32(%[R_IN])\n\t" \ + " aghi %[R_INLEN],-32\n\t" \ + /* Store 16 bytes to buf_out. */ \ + " vst %%v18,0(%[R_OUT])\n\t" \ + " aghi %[R_OUTLEN],-16\n\t" \ + " la %[R_OUT],16(%[R_OUT])\n\t" \ + " clgijl %[R_INLEN],32,2f\n\t" \ + " clgijl %[R_OUTLEN],16,2f\n\t" \ + " j 1b\n\t" \ + /* Setup to check for ch > 0x7f. (v30, v31) */ \ + "9: .short 0x7f,0x7f,0x0,0x0,0x0,0x0,0x0,0x0\n\t" \ + " .short 0x2000,0x2000,0x0,0x0,0x0,0x0,0x0,0x0\n\t" \ + /* At least one byte is > 0x7f. \ + Store the preceding 1-byte chars. */ \ + "11: lghi %[R_TMP2],16\n\t" /* match was found in v17. */ \ + "10:\n\t" \ + " vlgvb %[R_TMP],%%v19,7\n\t" \ + /* Shorten to UTF-8. */ \ + " vpkh %%v18,%%v16,%%v17\n\t" \ + " ar %[R_TMP],%[R_TMP2]\n\t" /* Number of in bytes. */ \ + " srlg %[R_TMP3],%[R_TMP],1\n\t" /* Number of out bytes. */ \ + " ahik %[R_TMP2],%[R_TMP3],-1\n\t" /* Highest index to store. */ \ + " jl 13f\n\t" \ + " vstl %%v18,%[R_TMP2],0(%[R_OUT])\n\t" \ + /* Update pointers. */ \ + " la %[R_IN],0(%[R_TMP],%[R_IN])\n\t" \ + " slgr %[R_INLEN],%[R_TMP]\n\t" \ + " la %[R_OUT],0(%[R_TMP3],%[R_OUT])\n\t" \ + " slgr %[R_OUTLEN],%[R_TMP3]\n\t" \ + "13: \n\t" \ + /* Calculate remaining uint16_t values in loaded vrs. */ \ + " lghi %[R_TMP2],16\n\t" \ + " slgr %[R_TMP2],%[R_TMP3]\n\t" \ + " llh %[R_TMP],0(%[R_IN])\n\t" \ + " aghi %[R_INLEN],-2\n\t" \ + " j 22f\n\t" \ + /* Handle remaining bytes. */ \ + "2: \n\t" \ + /* Zero, one or more bytes available? */ \ + " clgfi %[R_INLEN],1\n\t" \ + " locghie %[R_RES],%[RES_IN_FULL]\n\t" /* Only one byte. */ \ + " jle 99f\n\t" /* End if less than two bytes. */ \ + /* Calculate remaining uint16_t values in inptr. */ \ + " srlg %[R_TMP2],%[R_INLEN],1\n\t" \ + /* Handle multibyte utf8-char. */ \ + "20: llh %[R_TMP],0(%[R_IN])\n\t" \ + " aghi %[R_INLEN],-2\n\t" \ + /* Test if ch is 1-byte UTF-8 char. */ \ + "21: clijh %[R_TMP],0x7f,22f\n\t" \ + /* Handle 1-byte UTF-8 char. */ \ + "31: slgfi %[R_OUTLEN],1\n\t" \ + " jl 90f \n\t" \ + " stc %[R_TMP],0(%[R_OUT])\n\t" \ + " la %[R_IN],2(%[R_IN])\n\t" \ + " la %[R_OUT],1(%[R_OUT])\n\t" \ + " brctg %[R_TMP2],20b\n\t" \ + " j 0b\n\t" /* Switch to vx-loop. */ \ + /* Test if ch is 2-byte UTF-8 char. */ \ + "22: clfi %[R_TMP],0x7ff\n\t" \ + " jh 23f\n\t" \ + /* Handle 2-byte UTF-8 char. */ \ + "32: slgfi %[R_OUTLEN],2\n\t" \ + " jl 90f \n\t" \ + " llill %[R_TMP3],0xc080\n\t" \ + " la %[R_IN],2(%[R_IN])\n\t" \ + " risbgn %[R_TMP3],%[R_TMP],51,55,2\n\t" /* 1. byte. */ \ + " risbgn %[R_TMP3],%[R_TMP],58,63,0\n\t" /* 2. byte. */ \ + " sth %[R_TMP3],0(%[R_OUT])\n\t" \ + " la %[R_OUT],2(%[R_OUT])\n\t" \ + " brctg %[R_TMP2],20b\n\t" \ + " j 0b\n\t" /* Switch to vx-loop. */ \ + /* Test if ch is 3-byte UTF-8 char. */ \ + "23: clfi %[R_TMP],0xd7ff\n\t" \ + " jh 24f\n\t" \ + /* Handle 3-byte UTF-8 char. */ \ + "33: slgfi %[R_OUTLEN],3\n\t" \ + " jl 90f \n\t" \ + " llilf %[R_TMP3],0xe08080\n\t" \ + " la %[R_IN],2(%[R_IN])\n\t" \ + " risbgn %[R_TMP3],%[R_TMP],44,47,4\n\t" /* 1. byte. */ \ + " risbgn %[R_TMP3],%[R_TMP],50,55,2\n\t" /* 2. byte. */ \ + " risbgn %[R_TMP3],%[R_TMP],58,63,0\n\t" /* 3. byte. */ \ + " stcm %[R_TMP3],7,0(%[R_OUT])\n\t" \ + " la %[R_OUT],3(%[R_OUT])\n\t" \ + " brctg %[R_TMP2],20b\n\t" \ + " j 0b\n\t" /* Switch to vx-loop. */ \ + /* Test if ch is 4-byte UTF-8 char. */ \ + "24: clfi %[R_TMP],0xdfff\n\t" \ + " jh 33b\n\t" /* Handle this 3-byte UTF-8 char. */ \ + " clfi %[R_TMP],0xdbff\n\t" \ + " locghih %[R_RES],%[RES_IN_ILL]\n\t" \ + " jh 99f\n\t" /* Jump away if this is a low surrogate \ + without a preceding high surrogate. */ \ + /* Handle 4-byte UTF-8 char. */ \ + "34: slgfi %[R_OUTLEN],4\n\t" \ + " jl 90f \n\t" \ + " slgfi %[R_INLEN],2\n\t" \ + " locghil %[R_RES],%[RES_IN_FULL]\n\t" \ + " jl 99f\n\t" /* Jump away if low surrogate is missing. */ \ + " llilf %[R_TMP3],0xf0808080\n\t" \ + " aghi %[R_TMP],0x40\n\t" \ + " risbgn %[R_TMP3],%[R_TMP],37,39,16\n\t" /* 1. byte: uvw */ \ + " risbgn %[R_TMP3],%[R_TMP],42,43,14\n\t" /* 2. byte: xy */ \ + " risbgn %[R_TMP3],%[R_TMP],44,47,14\n\t" /* 2. byte: efgh */ \ + " risbgn %[R_TMP3],%[R_TMP],50,51,12\n\t" /* 3. byte: ij */ \ + " llh %[R_TMP],2(%[R_IN])\n\t" /* Load low surrogate. */ \ + " risbgn %[R_TMP3],%[R_TMP],52,55,2\n\t" /* 3. byte: klmn */ \ + " risbgn %[R_TMP3],%[R_TMP],58,63,0\n\t" /* 4. byte: opqrst */ \ + " nilf %[R_TMP],0xfc00\n\t" \ + " clfi %[R_TMP],0xdc00\n\t" /* Check if it starts with 0xdc00. */ \ + " locghine %[R_RES],%[RES_IN_ILL]\n\t" \ + " jne 99f\n\t" /* Jump away if low surrogate is invalid. */ \ + " st %[R_TMP3],0(%[R_OUT])\n\t" \ + " la %[R_IN],4(%[R_IN])\n\t" \ + " la %[R_OUT],4(%[R_OUT])\n\t" \ + " aghi %[R_TMP2],-2\n\t" \ + " jh 20b\n\t" \ + " j 0b\n\t" /* Switch to vx-loop. */ \ + /* Exit with __GCONV_FULL_OUTPUT. */ \ + "90: lghi %[R_RES],%[RES_OUT_FULL]\n\t" \ + "99: \n\t" \ + ".machine pop" \ + : /* outputs */ [R_IN] "+a" (inptr) \ + , [R_INLEN] "+d" (inlen), [R_OUT] "+a" (outptr) \ + , [R_OUTLEN] "+d" (outlen), [R_TMP] "=a" (tmp) \ + , [R_TMP2] "=d" (tmp2), [R_TMP3] "=a" (tmp3) \ + , [R_RES] "+d" (result) \ + : /* inputs */ \ + [RES_OUT_FULL] "i" (__GCONV_FULL_OUTPUT) \ + , [RES_IN_ILL] "i" (__GCONV_ILLEGAL_INPUT) \ + , [RES_IN_FULL] "i" (__GCONV_INCOMPLETE_INPUT) \ + : /* clobber list */ "memory", "cc" \ + ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \ + ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \ + ASM_CLOBBER_VR ("v30") ASM_CLOBBER_VR ("v31") \ + ); \ + if (__glibc_likely (inptr == inend) \ + || result != __GCONV_ILLEGAL_INPUT) \ + break; \ + \ + STANDARD_TO_LOOP_ERR_HANDLER (2); \ + } + +/* Generate loop-function with software implementation. */ +#define MIN_NEEDED_INPUT MIN_NEEDED_TO +#define MAX_NEEDED_INPUT MAX_NEEDED_TO +#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM +#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM +#if defined HAVE_S390_VX_ASM_SUPPORT +# define LOOPFCT __to_utf8_loop_c +# define BODY BODY_TO_C +# define LOOP_NEED_FLAGS +# include + +/* Generate loop-function with software implementation. */ +# define MIN_NEEDED_INPUT MIN_NEEDED_TO +# define MAX_NEEDED_INPUT MAX_NEEDED_TO +# define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM +# define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM +# define LOOPFCT __to_utf8_loop_vx +# define BODY BODY_TO_VX +# define LOOP_NEED_FLAGS +# include + +/* Generate ifunc'ed loop function. */ +__typeof(__to_utf8_loop_c) +__attribute__ ((ifunc ("__to_utf8_loop_resolver"))) +__to_utf8_loop; + +static void * +__to_utf8_loop_resolver (unsigned long int dl_hwcap) +{ + if (dl_hwcap & HWCAP_S390_VX) + return __to_utf8_loop_vx; + else + return __to_utf8_loop_c; +} + +strong_alias (__to_utf8_loop_c_single, __to_utf8_loop_single) + +#else +# define LOOPFCT TO_LOOP +# define BODY BODY_TO_C +# define LOOP_NEED_FLAGS +# include +#endif /* !HAVE_S390_VX_ASM_SUPPORT */ #include -- 2.5.5