|
|
ce426f |
From 835c3bf23a119a7fcb8c70734d4fdf49461d8195 Mon Sep 17 00:00:00 2001
|
|
|
ce426f |
From: Stefan Liebler <stli@linux.vnet.ibm.com>
|
|
|
ce426f |
Date: Mon, 7 Nov 2016 16:14:07 +0100
|
|
|
ce426f |
Subject: [PATCH 09/17] S390: Optimize utf8-utf32 module.
|
|
|
ce426f |
|
|
|
ce426f |
Upstream commit 421c5278d83e72740150259960a431706ac343f9
|
|
|
ce426f |
|
|
|
ce426f |
This patch reworks the s390 specific module to convert between utf8 and utf32.
|
|
|
ce426f |
Now ifunc is used to choose either the c or etf3eh (with convert utf
|
|
|
ce426f |
instruction) variants at runtime.
|
|
|
ce426f |
Furthermore a new vector variant for z13 is introduced which will be build
|
|
|
ce426f |
and chosen if vector support is available at build / runtime.
|
|
|
ce426f |
The vector variants optimize input of 1byte utf8 characters. The convert utf
|
|
|
ce426f |
instruction is used if a multibyte utf8 character is found.
|
|
|
ce426f |
|
|
|
ce426f |
This patch also fixes some whitespace errors. The c variants are rejecting
|
|
|
ce426f |
UTF-16 surrogates and values above 0x10ffff now.
|
|
|
ce426f |
Furthermore, the etf3eh variants are handling the "UTF-xx//IGNORE" case now.
|
|
|
ce426f |
Before they ignored the ignore-case and always stopped at an error.
|
|
|
ce426f |
|
|
|
ce426f |
ChangeLog:
|
|
|
ce426f |
|
|
|
ce426f |
* sysdeps/s390/s390-64/utf8-utf32-z9.c: Use ifunc to select c, etf3eh
|
|
|
ce426f |
or new vector loop-variant.
|
|
|
ce426f |
---
|
|
|
ce426f |
sysdeps/s390/s390-64/utf8-utf32-z9.c | 664 +++++++++++++++++++++++++----------
|
|
|
ce426f |
1 file changed, 480 insertions(+), 184 deletions(-)
|
|
|
ce426f |
|
|
|
ce426f |
diff --git a/sysdeps/s390/s390-64/utf8-utf32-z9.c b/sysdeps/s390/s390-64/utf8-utf32-z9.c
|
|
|
ce426f |
index 721279e..1ce5ac5 100644
|
|
|
ce426f |
--- a/sysdeps/s390/s390-64/utf8-utf32-z9.c
|
|
|
ce426f |
+++ b/sysdeps/s390/s390-64/utf8-utf32-z9.c
|
|
|
ce426f |
@@ -30,35 +30,25 @@
|
|
|
ce426f |
#include <dl-procinfo.h>
|
|
|
ce426f |
#include <gconv.h>
|
|
|
ce426f |
|
|
|
ce426f |
-/* UTF-32 big endian byte order mark. */
|
|
|
ce426f |
-#define BOM 0x0000feffu
|
|
|
ce426f |
+#if defined HAVE_S390_VX_GCC_SUPPORT
|
|
|
ce426f |
+# define ASM_CLOBBER_VR(NR) , NR
|
|
|
ce426f |
+#else
|
|
|
ce426f |
+# define ASM_CLOBBER_VR(NR)
|
|
|
ce426f |
+#endif
|
|
|
ce426f |
|
|
|
ce426f |
+/* Defines for skeleton.c. */
|
|
|
ce426f |
#define DEFINE_INIT 0
|
|
|
ce426f |
#define DEFINE_FINI 0
|
|
|
ce426f |
-/* These definitions apply to the UTF-8 to UTF-32 direction. The
|
|
|
ce426f |
- software implementation for UTF-8 still supports multibyte
|
|
|
ce426f |
- characters up to 6 bytes whereas the hardware variant does not. */
|
|
|
ce426f |
#define MIN_NEEDED_FROM 1
|
|
|
ce426f |
#define MAX_NEEDED_FROM 6
|
|
|
ce426f |
#define MIN_NEEDED_TO 4
|
|
|
ce426f |
-#define FROM_LOOP from_utf8_loop
|
|
|
ce426f |
-#define TO_LOOP to_utf8_loop
|
|
|
ce426f |
+#define FROM_LOOP __from_utf8_loop
|
|
|
ce426f |
+#define TO_LOOP __to_utf8_loop
|
|
|
ce426f |
#define FROM_DIRECTION (dir == from_utf8)
|
|
|
ce426f |
#define ONE_DIRECTION 0
|
|
|
ce426f |
-#define PREPARE_LOOP \
|
|
|
ce426f |
- enum direction dir = ((struct utf8_data *) step->__data)->dir; \
|
|
|
ce426f |
- int emit_bom = ((struct utf8_data *) step->__data)->emit_bom; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- if (emit_bom && !data->__internal_use \
|
|
|
ce426f |
- && data->__invocation_counter == 0) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* Emit the Byte Order Mark. */ \
|
|
|
ce426f |
- if (__glibc_unlikely (outbuf + 4 > outend)) \
|
|
|
ce426f |
- return __GCONV_FULL_OUTPUT; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- put32u (outbuf, BOM); \
|
|
|
ce426f |
- outbuf += 4; \
|
|
|
ce426f |
- }
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* UTF-32 big endian byte order mark. */
|
|
|
ce426f |
+#define BOM 0x0000feffu
|
|
|
ce426f |
|
|
|
ce426f |
/* Direction of the transformation. */
|
|
|
ce426f |
enum direction
|
|
|
ce426f |
@@ -155,16 +145,16 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
register unsigned long long outlen __asm__("11") = outend - outptr; \
|
|
|
ce426f |
uint64_t cc = 0; \
|
|
|
ce426f |
\
|
|
|
ce426f |
- __asm__ volatile (".machine push \n\t" \
|
|
|
ce426f |
- ".machine \"z9-109\" \n\t" \
|
|
|
ce426f |
- "0: " INSTRUCTION " \n\t" \
|
|
|
ce426f |
- ".machine pop \n\t" \
|
|
|
ce426f |
- " jo 0b \n\t" \
|
|
|
ce426f |
- " ipm %2 \n" \
|
|
|
ce426f |
- : "+a" (pOutput), "+a" (pInput), "+d" (cc), \
|
|
|
ce426f |
- "+d" (outlen), "+d" (inlen) \
|
|
|
ce426f |
- : \
|
|
|
ce426f |
- : "cc", "memory"); \
|
|
|
ce426f |
+ __asm__ __volatile__ (".machine push \n\t" \
|
|
|
ce426f |
+ ".machine \"z9-109\" \n\t" \
|
|
|
ce426f |
+ "0: " INSTRUCTION " \n\t" \
|
|
|
ce426f |
+ ".machine pop \n\t" \
|
|
|
ce426f |
+ " jo 0b \n\t" \
|
|
|
ce426f |
+ " ipm %2 \n" \
|
|
|
ce426f |
+ : "+a" (pOutput), "+a" (pInput), "+d" (cc), \
|
|
|
ce426f |
+ "+d" (outlen), "+d" (inlen) \
|
|
|
ce426f |
+ : \
|
|
|
ce426f |
+ : "cc", "memory"); \
|
|
|
ce426f |
\
|
|
|
ce426f |
inptr = pInput; \
|
|
|
ce426f |
outptr = pOutput; \
|
|
|
ce426f |
@@ -173,49 +163,150 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
if (cc == 1) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
result = __GCONV_FULL_OUTPUT; \
|
|
|
ce426f |
- break; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
else if (cc == 2) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
result = __GCONV_ILLEGAL_INPUT; \
|
|
|
ce426f |
- break; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
}
|
|
|
ce426f |
|
|
|
ce426f |
+#define PREPARE_LOOP \
|
|
|
ce426f |
+ enum direction dir = ((struct utf8_data *) step->__data)->dir; \
|
|
|
ce426f |
+ int emit_bom = ((struct utf8_data *) step->__data)->emit_bom; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ if (emit_bom && !data->__internal_use \
|
|
|
ce426f |
+ && data->__invocation_counter == 0) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* Emit the Byte Order Mark. */ \
|
|
|
ce426f |
+ if (__glibc_unlikely (outbuf + 4 > outend)) \
|
|
|
ce426f |
+ return __GCONV_FULL_OUTPUT; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ put32u (outbuf, BOM); \
|
|
|
ce426f |
+ outbuf += 4; \
|
|
|
ce426f |
+ }
|
|
|
ce426f |
+
|
|
|
ce426f |
/* Conversion function from UTF-8 to UTF-32 internal/BE. */
|
|
|
ce426f |
|
|
|
ce426f |
-#define MIN_NEEDED_INPUT MIN_NEEDED_FROM
|
|
|
ce426f |
-#define MAX_NEEDED_INPUT MAX_NEEDED_FROM
|
|
|
ce426f |
-#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO
|
|
|
ce426f |
-#define LOOPFCT FROM_LOOP
|
|
|
ce426f |
-/* The software routine is copied from gconv_simple.c. */
|
|
|
ce426f |
-#define BODY \
|
|
|
ce426f |
+#define STORE_REST_COMMON \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* We store the remaining bytes while converting them into the UCS4 \
|
|
|
ce426f |
+ format. We can assume that the first byte in the buffer is \
|
|
|
ce426f |
+ correct and that it requires a larger number of bytes than there \
|
|
|
ce426f |
+ are in the input buffer. */ \
|
|
|
ce426f |
+ wint_t ch = **inptrp; \
|
|
|
ce426f |
+ size_t cnt, r; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ state->__count = inend - *inptrp; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ assert (ch != 0xc0 && ch != 0xc1); \
|
|
|
ce426f |
+ if (ch >= 0xc2 && ch < 0xe0) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* We expect two bytes. The first byte cannot be 0xc0 or \
|
|
|
ce426f |
+ 0xc1, otherwise the wide character could have been \
|
|
|
ce426f |
+ represented using a single byte. */ \
|
|
|
ce426f |
+ cnt = 2; \
|
|
|
ce426f |
+ ch &= 0x1f; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* We expect three bytes. */ \
|
|
|
ce426f |
+ cnt = 3; \
|
|
|
ce426f |
+ ch &= 0x0f; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* We expect four bytes. */ \
|
|
|
ce426f |
+ cnt = 4; \
|
|
|
ce426f |
+ ch &= 0x07; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ else if (__glibc_likely ((ch & 0xfc) == 0xf8)) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* We expect five bytes. */ \
|
|
|
ce426f |
+ cnt = 5; \
|
|
|
ce426f |
+ ch &= 0x03; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ else \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* We expect six bytes. */ \
|
|
|
ce426f |
+ cnt = 6; \
|
|
|
ce426f |
+ ch &= 0x01; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ /* The first byte is already consumed. */ \
|
|
|
ce426f |
+ r = cnt - 1; \
|
|
|
ce426f |
+ while (++(*inptrp) < inend) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ ch <<= 6; \
|
|
|
ce426f |
+ ch |= **inptrp & 0x3f; \
|
|
|
ce426f |
+ --r; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ /* Shift for the so far missing bytes. */ \
|
|
|
ce426f |
+ ch <<= r * 6; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ /* Store the number of bytes expected for the entire sequence. */ \
|
|
|
ce426f |
+ state->__count |= cnt << 8; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ /* Store the value. */ \
|
|
|
ce426f |
+ state->__value.__wch = ch; \
|
|
|
ce426f |
+ }
|
|
|
ce426f |
+
|
|
|
ce426f |
+#define UNPACK_BYTES_COMMON \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ static const unsigned char inmask[5] = { 0xc0, 0xe0, 0xf0, 0xf8, 0xfc }; \
|
|
|
ce426f |
+ wint_t wch = state->__value.__wch; \
|
|
|
ce426f |
+ size_t ntotal = state->__count >> 8; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ inlen = state->__count & 255; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ bytebuf[0] = inmask[ntotal - 2]; \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ do \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ if (--ntotal < inlen) \
|
|
|
ce426f |
+ bytebuf[ntotal] = 0x80 | (wch & 0x3f); \
|
|
|
ce426f |
+ wch >>= 6; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ while (ntotal > 1); \
|
|
|
ce426f |
+ \
|
|
|
ce426f |
+ bytebuf[0] |= wch; \
|
|
|
ce426f |
+ }
|
|
|
ce426f |
+
|
|
|
ce426f |
+#define CLEAR_STATE_COMMON \
|
|
|
ce426f |
+ state->__count = 0
|
|
|
ce426f |
+
|
|
|
ce426f |
+#define BODY_FROM_HW(ASM) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
- if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- HARDWARE_CONVERT ("cu14 %0, %1, 1"); \
|
|
|
ce426f |
+ ASM; \
|
|
|
ce426f |
+ if (__glibc_likely (inptr == inend) \
|
|
|
ce426f |
+ || result == __GCONV_FULL_OUTPUT) \
|
|
|
ce426f |
+ break; \
|
|
|
ce426f |
\
|
|
|
ce426f |
- if (inptr != inend) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- int i; \
|
|
|
ce426f |
- for (i = 1; inptr + i < inend; ++i) \
|
|
|
ce426f |
- if ((inptr[i] & 0xc0) != 0x80) \
|
|
|
ce426f |
- break; \
|
|
|
ce426f |
+ int i; \
|
|
|
ce426f |
+ for (i = 1; inptr + i < inend && i < 5; ++i) \
|
|
|
ce426f |
+ if ((inptr[i] & 0xc0) != 0x80) \
|
|
|
ce426f |
+ break; \
|
|
|
ce426f |
\
|
|
|
ce426f |
- if (__glibc_likely (inptr + i == inend)) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- result = __GCONV_INCOMPLETE_INPUT; \
|
|
|
ce426f |
- break; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- STANDARD_FROM_LOOP_ERR_HANDLER (i); \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- continue; \
|
|
|
ce426f |
+ if (__glibc_likely (inptr + i == inend \
|
|
|
ce426f |
+ && result == __GCONV_EMPTY_INPUT)) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ result = __GCONV_INCOMPLETE_INPUT; \
|
|
|
ce426f |
+ break; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
- \
|
|
|
ce426f |
+ STANDARD_FROM_LOOP_ERR_HANDLER (i); \
|
|
|
ce426f |
+ }
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* This hardware routine uses the Convert UTF8 to UTF32 (cu14) instruction. */
|
|
|
ce426f |
+#define BODY_FROM_ETF3EH BODY_FROM_HW (HARDWARE_CONVERT ("cu14 %0, %1, 1"))
|
|
|
ce426f |
+
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* The software routine is copied from gconv_simple.c. */
|
|
|
ce426f |
+#define BODY_FROM_C \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
/* Next input byte. */ \
|
|
|
ce426f |
uint32_t ch = *inptr; \
|
|
|
ce426f |
\
|
|
|
ce426f |
- if (__glibc_likely (ch < 0x80)) \
|
|
|
ce426f |
+ if (__glibc_likely (ch < 0x80)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* One byte sequence. */ \
|
|
|
ce426f |
++inptr; \
|
|
|
ce426f |
@@ -233,30 +324,18 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
cnt = 2; \
|
|
|
ce426f |
ch &= 0x1f; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \
|
|
|
ce426f |
+ else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* We expect three bytes. */ \
|
|
|
ce426f |
cnt = 3; \
|
|
|
ce426f |
ch &= 0x0f; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \
|
|
|
ce426f |
+ else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* We expect four bytes. */ \
|
|
|
ce426f |
cnt = 4; \
|
|
|
ce426f |
ch &= 0x07; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xfc) == 0xf8)) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect five bytes. */ \
|
|
|
ce426f |
- cnt = 5; \
|
|
|
ce426f |
- ch &= 0x03; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xfe) == 0xfc)) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect six bytes. */ \
|
|
|
ce426f |
- cnt = 6; \
|
|
|
ce426f |
- ch &= 0x01; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
else \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* Search the end of this ill-formed UTF-8 character. This \
|
|
|
ce426f |
@@ -272,7 +351,7 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
STANDARD_FROM_LOOP_ERR_HANDLER (i); \
|
|
|
ce426f |
} \
|
|
|
ce426f |
\
|
|
|
ce426f |
- if (__glibc_unlikely (inptr + cnt > inend)) \
|
|
|
ce426f |
+ if (__glibc_unlikely (inptr + cnt > inend)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* We don't have enough input. But before we report \
|
|
|
ce426f |
that check that all the bytes are correct. */ \
|
|
|
ce426f |
@@ -280,7 +359,7 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
if ((inptr[i] & 0xc0) != 0x80) \
|
|
|
ce426f |
break; \
|
|
|
ce426f |
\
|
|
|
ce426f |
- if (__glibc_likely (inptr + i == inend)) \
|
|
|
ce426f |
+ if (__glibc_likely (inptr + i == inend)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
result = __GCONV_INCOMPLETE_INPUT; \
|
|
|
ce426f |
break; \
|
|
|
ce426f |
@@ -305,7 +384,10 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
/* If i < cnt, some trail byte was not >= 0x80, < 0xc0. \
|
|
|
ce426f |
If cnt > 2 and ch < 2^(5*cnt-4), the wide character ch could \
|
|
|
ce426f |
have been represented with fewer than cnt bytes. */ \
|
|
|
ce426f |
- if (i < cnt || (cnt > 2 && (ch >> (5 * cnt - 4)) == 0)) \
|
|
|
ce426f |
+ if (i < cnt || (cnt > 2 && (ch >> (5 * cnt - 4)) == 0) \
|
|
|
ce426f |
+ /* Do not accept UTF-16 surrogates. */ \
|
|
|
ce426f |
+ || (ch >= 0xd800 && ch <= 0xdfff) \
|
|
|
ce426f |
+ || (ch > 0x10ffff)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* This is an illegal encoding. */ \
|
|
|
ce426f |
goto errout; \
|
|
|
ce426f |
@@ -318,137 +400,212 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
*((uint32_t *) outptr) = ch; \
|
|
|
ce426f |
outptr += sizeof (uint32_t); \
|
|
|
ce426f |
}
|
|
|
ce426f |
-#define LOOP_NEED_FLAGS
|
|
|
ce426f |
|
|
|
ce426f |
-#define STORE_REST \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We store the remaining bytes while converting them into the UCS4 \
|
|
|
ce426f |
- format. We can assume that the first byte in the buffer is \
|
|
|
ce426f |
- correct and that it requires a larger number of bytes than there \
|
|
|
ce426f |
- are in the input buffer. */ \
|
|
|
ce426f |
- wint_t ch = **inptrp; \
|
|
|
ce426f |
- size_t cnt, r; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- state->__count = inend - *inptrp; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- if (ch >= 0xc2 && ch < 0xe0) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect two bytes. The first byte cannot be 0xc0 or \
|
|
|
ce426f |
- 0xc1, otherwise the wide character could have been \
|
|
|
ce426f |
- represented using a single byte. */ \
|
|
|
ce426f |
- cnt = 2; \
|
|
|
ce426f |
- ch &= 0x1f; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xf0) == 0xe0)) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect three bytes. */ \
|
|
|
ce426f |
- cnt = 3; \
|
|
|
ce426f |
- ch &= 0x0f; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xf8) == 0xf0)) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect four bytes. */ \
|
|
|
ce426f |
- cnt = 4; \
|
|
|
ce426f |
- ch &= 0x07; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- else if (__glibc_likely ((ch & 0xfc) == 0xf8)) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect five bytes. */ \
|
|
|
ce426f |
- cnt = 5; \
|
|
|
ce426f |
- ch &= 0x03; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- else \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- /* We expect six bytes. */ \
|
|
|
ce426f |
- cnt = 6; \
|
|
|
ce426f |
- ch &= 0x01; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- /* The first byte is already consumed. */ \
|
|
|
ce426f |
- r = cnt - 1; \
|
|
|
ce426f |
- while (++(*inptrp) < inend) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- ch <<= 6; \
|
|
|
ce426f |
- ch |= **inptrp & 0x3f; \
|
|
|
ce426f |
- --r; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- /* Shift for the so far missing bytes. */ \
|
|
|
ce426f |
- ch <<= r * 6; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- /* Store the number of bytes expected for the entire sequence. */ \
|
|
|
ce426f |
- state->__count |= cnt << 8; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- /* Store the value. */ \
|
|
|
ce426f |
- state->__value.__wch = ch; \
|
|
|
ce426f |
+#define HW_FROM_VX \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ register const unsigned char* pInput asm ("8") = inptr; \
|
|
|
ce426f |
+ register size_t inlen asm ("9") = inend - inptr; \
|
|
|
ce426f |
+ register unsigned char* pOutput asm ("10") = outptr; \
|
|
|
ce426f |
+ register size_t outlen asm("11") = outend - outptr; \
|
|
|
ce426f |
+ unsigned long tmp, tmp2, tmp3; \
|
|
|
ce426f |
+ asm volatile (".machine push\n\t" \
|
|
|
ce426f |
+ ".machine \"z13\"\n\t" \
|
|
|
ce426f |
+ ".machinemode \"zarch_nohighgprs\"\n\t" \
|
|
|
ce426f |
+ " vrepib %%v30,0x7f\n\t" /* For compare > 0x7f. */ \
|
|
|
ce426f |
+ " vrepib %%v31,0x20\n\t" \
|
|
|
ce426f |
+ /* Loop which handles UTF-8 chars <=0x7f. */ \
|
|
|
ce426f |
+ "0: clgijl %[R_INLEN],16,20f\n\t" \
|
|
|
ce426f |
+ " clgijl %[R_OUTLEN],64,20f\n\t" \
|
|
|
ce426f |
+ "1: vl %%v16,0(%[R_IN])\n\t" \
|
|
|
ce426f |
+ " vstrcbs %%v17,%%v16,%%v30,%%v31\n\t" \
|
|
|
ce426f |
+ " jno 10f\n\t" /* Jump away if not all bytes are 1byte \
|
|
|
ce426f |
+ UTF8 chars. */ \
|
|
|
ce426f |
+ /* Enlarge to UCS4. */ \
|
|
|
ce426f |
+ " vuplhb %%v18,%%v16\n\t" \
|
|
|
ce426f |
+ " vupllb %%v19,%%v16\n\t" \
|
|
|
ce426f |
+ " la %[R_IN],16(%[R_IN])\n\t" \
|
|
|
ce426f |
+ " vuplhh %%v20,%%v18\n\t" \
|
|
|
ce426f |
+ " aghi %[R_INLEN],-16\n\t" \
|
|
|
ce426f |
+ " vupllh %%v21,%%v18\n\t" \
|
|
|
ce426f |
+ " aghi %[R_OUTLEN],-64\n\t" \
|
|
|
ce426f |
+ " vuplhh %%v22,%%v19\n\t" \
|
|
|
ce426f |
+ " vupllh %%v23,%%v19\n\t" \
|
|
|
ce426f |
+ /* Store 64 bytes to buf_out. */ \
|
|
|
ce426f |
+ " vstm %%v20,%%v23,0(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " la %[R_OUT],64(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " clgijl %[R_INLEN],16,20f\n\t" \
|
|
|
ce426f |
+ " clgijl %[R_OUTLEN],64,20f\n\t" \
|
|
|
ce426f |
+ " j 1b\n\t" \
|
|
|
ce426f |
+ "10: \n\t" \
|
|
|
ce426f |
+ /* At least one byte is > 0x7f. \
|
|
|
ce426f |
+ Store the preceding 1-byte chars. */ \
|
|
|
ce426f |
+ " vlgvb %[R_TMP],%%v17,7\n\t" \
|
|
|
ce426f |
+ " sllk %[R_TMP2],%[R_TMP],2\n\t" /* Compute highest \
|
|
|
ce426f |
+ index to store. */ \
|
|
|
ce426f |
+ " llgfr %[R_TMP3],%[R_TMP2]\n\t" \
|
|
|
ce426f |
+ " ahi %[R_TMP2],-1\n\t" \
|
|
|
ce426f |
+ " jl 20f\n\t" \
|
|
|
ce426f |
+ " vuplhb %%v18,%%v16\n\t" \
|
|
|
ce426f |
+ " vuplhh %%v20,%%v18\n\t" \
|
|
|
ce426f |
+ " vstl %%v20,%[R_TMP2],0(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " ahi %[R_TMP2],-16\n\t" \
|
|
|
ce426f |
+ " jl 11f\n\t" \
|
|
|
ce426f |
+ " vupllh %%v21,%%v18\n\t" \
|
|
|
ce426f |
+ " vstl %%v21,%[R_TMP2],16(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " ahi %[R_TMP2],-16\n\t" \
|
|
|
ce426f |
+ " jl 11f\n\t" \
|
|
|
ce426f |
+ " vupllb %%v19,%%v16\n\t" \
|
|
|
ce426f |
+ " vuplhh %%v22,%%v19\n\t" \
|
|
|
ce426f |
+ " vstl %%v22,%[R_TMP2],32(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " ahi %[R_TMP2],-16\n\t" \
|
|
|
ce426f |
+ " jl 11f\n\t" \
|
|
|
ce426f |
+ " vupllh %%v23,%%v19\n\t" \
|
|
|
ce426f |
+ " vstl %%v23,%[R_TMP2],48(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ "11: \n\t" \
|
|
|
ce426f |
+ /* Update pointers. */ \
|
|
|
ce426f |
+ " la %[R_IN],0(%[R_TMP],%[R_IN])\n\t" \
|
|
|
ce426f |
+ " slgr %[R_INLEN],%[R_TMP]\n\t" \
|
|
|
ce426f |
+ " la %[R_OUT],0(%[R_TMP3],%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " slgr %[R_OUTLEN],%[R_TMP3]\n\t" \
|
|
|
ce426f |
+ /* Handle multibyte utf8-char with convert instruction. */ \
|
|
|
ce426f |
+ "20: cu14 %[R_OUT],%[R_IN],1\n\t" \
|
|
|
ce426f |
+ " jo 0b\n\t" /* Try vector implemenation again. */ \
|
|
|
ce426f |
+ " lochil %[R_RES],%[RES_OUT_FULL]\n\t" /* cc == 1. */ \
|
|
|
ce426f |
+ " lochih %[R_RES],%[RES_IN_ILL]\n\t" /* cc == 2. */ \
|
|
|
ce426f |
+ ".machine pop" \
|
|
|
ce426f |
+ : /* outputs */ [R_IN] "+a" (pInput) \
|
|
|
ce426f |
+ , [R_INLEN] "+d" (inlen), [R_OUT] "+a" (pOutput) \
|
|
|
ce426f |
+ , [R_OUTLEN] "+d" (outlen), [R_TMP] "=a" (tmp) \
|
|
|
ce426f |
+ , [R_TMP2] "=d" (tmp2), [R_TMP3] "=a" (tmp3) \
|
|
|
ce426f |
+ , [R_RES] "+d" (result) \
|
|
|
ce426f |
+ : /* inputs */ \
|
|
|
ce426f |
+ [RES_OUT_FULL] "i" (__GCONV_FULL_OUTPUT) \
|
|
|
ce426f |
+ , [RES_IN_ILL] "i" (__GCONV_ILLEGAL_INPUT) \
|
|
|
ce426f |
+ : /* clobber list */ "memory", "cc" \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v20") ASM_CLOBBER_VR ("v21") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v22") ASM_CLOBBER_VR ("v30") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v31") \
|
|
|
ce426f |
+ ); \
|
|
|
ce426f |
+ inptr = pInput; \
|
|
|
ce426f |
+ outptr = pOutput; \
|
|
|
ce426f |
}
|
|
|
ce426f |
+#define BODY_FROM_VX BODY_FROM_HW (HW_FROM_VX)
|
|
|
ce426f |
|
|
|
ce426f |
-#define UNPACK_BYTES \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- static const unsigned char inmask[5] = { 0xc0, 0xe0, 0xf0, 0xf8, 0xfc }; \
|
|
|
ce426f |
- wint_t wch = state->__value.__wch; \
|
|
|
ce426f |
- size_t ntotal = state->__count >> 8; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- inlen = state->__count & 255; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- bytebuf[0] = inmask[ntotal - 2]; \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- do \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- if (--ntotal < inlen) \
|
|
|
ce426f |
- bytebuf[ntotal] = 0x80 | (wch & 0x3f); \
|
|
|
ce426f |
- wch >>= 6; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- while (ntotal > 1); \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- bytebuf[0] |= wch; \
|
|
|
ce426f |
- }
|
|
|
ce426f |
+/* These definitions apply to the UTF-8 to UTF-32 direction. The
|
|
|
ce426f |
+ software implementation for UTF-8 still supports multibyte
|
|
|
ce426f |
+ characters up to 6 bytes whereas the hardware variant does not. */
|
|
|
ce426f |
+#define MIN_NEEDED_INPUT MIN_NEEDED_FROM
|
|
|
ce426f |
+#define MAX_NEEDED_INPUT MAX_NEEDED_FROM
|
|
|
ce426f |
+#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO
|
|
|
ce426f |
+#define LOOPFCT __from_utf8_loop_c
|
|
|
ce426f |
|
|
|
ce426f |
-#define CLEAR_STATE \
|
|
|
ce426f |
- state->__count = 0
|
|
|
ce426f |
+#define LOOP_NEED_FLAGS
|
|
|
ce426f |
|
|
|
ce426f |
+#define STORE_REST STORE_REST_COMMON
|
|
|
ce426f |
+#define UNPACK_BYTES UNPACK_BYTES_COMMON
|
|
|
ce426f |
+#define CLEAR_STATE CLEAR_STATE_COMMON
|
|
|
ce426f |
+#define BODY BODY_FROM_C
|
|
|
ce426f |
#include <iconv/loop.c>
|
|
|
ce426f |
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* Generate loop-function with hardware utf-convert instruction. */
|
|
|
ce426f |
+#define MIN_NEEDED_INPUT MIN_NEEDED_FROM
|
|
|
ce426f |
+#define MAX_NEEDED_INPUT MAX_NEEDED_FROM
|
|
|
ce426f |
+#define MIN_NEEDED_OUTPUT MIN_NEEDED_TO
|
|
|
ce426f |
+#define LOOPFCT __from_utf8_loop_etf3eh
|
|
|
ce426f |
+
|
|
|
ce426f |
+#define LOOP_NEED_FLAGS
|
|
|
ce426f |
+
|
|
|
ce426f |
+#define STORE_REST STORE_REST_COMMON
|
|
|
ce426f |
+#define UNPACK_BYTES UNPACK_BYTES_COMMON
|
|
|
ce426f |
+#define CLEAR_STATE CLEAR_STATE_COMMON
|
|
|
ce426f |
+#define BODY BODY_FROM_ETF3EH
|
|
|
ce426f |
+#include <iconv/loop.c>
|
|
|
ce426f |
+
|
|
|
ce426f |
+#if defined HAVE_S390_VX_ASM_SUPPORT
|
|
|
ce426f |
+/* Generate loop-function with hardware vector instructions. */
|
|
|
ce426f |
+# define MIN_NEEDED_INPUT MIN_NEEDED_FROM
|
|
|
ce426f |
+# define MAX_NEEDED_INPUT MAX_NEEDED_FROM
|
|
|
ce426f |
+# define MIN_NEEDED_OUTPUT MIN_NEEDED_TO
|
|
|
ce426f |
+# define LOOPFCT __from_utf8_loop_vx
|
|
|
ce426f |
+
|
|
|
ce426f |
+# define LOOP_NEED_FLAGS
|
|
|
ce426f |
+
|
|
|
ce426f |
+# define STORE_REST STORE_REST_COMMON
|
|
|
ce426f |
+# define UNPACK_BYTES UNPACK_BYTES_COMMON
|
|
|
ce426f |
+# define CLEAR_STATE CLEAR_STATE_COMMON
|
|
|
ce426f |
+# define BODY BODY_FROM_VX
|
|
|
ce426f |
+# include <iconv/loop.c>
|
|
|
ce426f |
+#endif
|
|
|
ce426f |
+
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* Generate ifunc'ed loop function. */
|
|
|
ce426f |
+__typeof(__from_utf8_loop_c)
|
|
|
ce426f |
+__attribute__ ((ifunc ("__from_utf8_loop_resolver")))
|
|
|
ce426f |
+__from_utf8_loop;
|
|
|
ce426f |
+
|
|
|
ce426f |
+static void *
|
|
|
ce426f |
+__from_utf8_loop_resolver (unsigned long int dl_hwcap)
|
|
|
ce426f |
+{
|
|
|
ce426f |
+#if defined HAVE_S390_VX_ASM_SUPPORT
|
|
|
ce426f |
+ if (dl_hwcap & HWCAP_S390_VX)
|
|
|
ce426f |
+ return __from_utf8_loop_vx;
|
|
|
ce426f |
+ else
|
|
|
ce426f |
+#endif
|
|
|
ce426f |
+ if (dl_hwcap & HWCAP_S390_ETF3EH)
|
|
|
ce426f |
+ return __from_utf8_loop_etf3eh;
|
|
|
ce426f |
+ else
|
|
|
ce426f |
+ return __from_utf8_loop_c;
|
|
|
ce426f |
+}
|
|
|
ce426f |
+
|
|
|
ce426f |
+strong_alias (__from_utf8_loop_c_single, __from_utf8_loop_single)
|
|
|
ce426f |
+
|
|
|
ce426f |
+
|
|
|
ce426f |
/* Conversion from UTF-32 internal/BE to UTF-8. */
|
|
|
ce426f |
+#define BODY_TO_HW(ASM) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ ASM; \
|
|
|
ce426f |
+ if (__glibc_likely (inptr == inend) \
|
|
|
ce426f |
+ || result == __GCONV_FULL_OUTPUT) \
|
|
|
ce426f |
+ break; \
|
|
|
ce426f |
+ if (inptr + 4 > inend) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ result = __GCONV_INCOMPLETE_INPUT; \
|
|
|
ce426f |
+ break; \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
+ STANDARD_TO_LOOP_ERR_HANDLER (4); \
|
|
|
ce426f |
+ }
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* The hardware routine uses the S/390 cu41 instruction. */
|
|
|
ce426f |
+#define BODY_TO_ETF3EH BODY_TO_HW (HARDWARE_CONVERT ("cu41 %0, %1"))
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* The hardware routine uses the S/390 vector and cu41 instructions. */
|
|
|
ce426f |
+#define BODY_TO_VX BODY_TO_HW (HW_TO_VX)
|
|
|
ce426f |
|
|
|
ce426f |
-#define MIN_NEEDED_INPUT MIN_NEEDED_TO
|
|
|
ce426f |
-#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM
|
|
|
ce426f |
-#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM
|
|
|
ce426f |
-#define LOOPFCT TO_LOOP
|
|
|
ce426f |
/* The software routine mimics the S/390 cu41 instruction. */
|
|
|
ce426f |
-#define BODY \
|
|
|
ce426f |
+#define BODY_TO_C \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
- if (GLRO (dl_hwcap) & HWCAP_S390_ETF3EH) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- HARDWARE_CONVERT ("cu41 %0, %1"); \
|
|
|
ce426f |
- \
|
|
|
ce426f |
- if (inptr != inend) \
|
|
|
ce426f |
- { \
|
|
|
ce426f |
- result = __GCONV_INCOMPLETE_INPUT; \
|
|
|
ce426f |
- break; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- continue; \
|
|
|
ce426f |
- } \
|
|
|
ce426f |
- \
|
|
|
ce426f |
uint32_t wc = *((const uint32_t *) inptr); \
|
|
|
ce426f |
\
|
|
|
ce426f |
- if (__glibc_likely (wc <= 0x7f)) \
|
|
|
ce426f |
+ if (__glibc_likely (wc <= 0x7f)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
- /* Single UTF-8 char. */ \
|
|
|
ce426f |
- *outptr = (uint8_t)wc; \
|
|
|
ce426f |
+ /* Single UTF-8 char. */ \
|
|
|
ce426f |
+ *outptr = (uint8_t)wc; \
|
|
|
ce426f |
outptr++; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
else if (wc <= 0x7ff) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
- /* Two UTF-8 chars. */ \
|
|
|
ce426f |
- if (__glibc_unlikely (outptr + 2 > outend)) \
|
|
|
ce426f |
+ /* Two UTF-8 chars. */ \
|
|
|
ce426f |
+ if (__glibc_unlikely (outptr + 2 > outend)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* Overflow in the output buffer. */ \
|
|
|
ce426f |
result = __GCONV_FULL_OUTPUT; \
|
|
|
ce426f |
break; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
\
|
|
|
ce426f |
- outptr[0] = 0xc0; \
|
|
|
ce426f |
+ outptr[0] = 0xc0; \
|
|
|
ce426f |
outptr[0] |= wc >> 6; \
|
|
|
ce426f |
\
|
|
|
ce426f |
outptr[1] = 0x80; \
|
|
|
ce426f |
@@ -459,12 +616,18 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
else if (wc <= 0xffff) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* Three UTF-8 chars. */ \
|
|
|
ce426f |
- if (__glibc_unlikely (outptr + 3 > outend)) \
|
|
|
ce426f |
+ if (__glibc_unlikely (outptr + 3 > outend)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* Overflow in the output buffer. */ \
|
|
|
ce426f |
result = __GCONV_FULL_OUTPUT; \
|
|
|
ce426f |
break; \
|
|
|
ce426f |
} \
|
|
|
ce426f |
+ if (wc >= 0xd800 && wc < 0xdc00) \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ /* Do not accept UTF-16 surrogates. */ \
|
|
|
ce426f |
+ result = __GCONV_ILLEGAL_INPUT; \
|
|
|
ce426f |
+ STANDARD_TO_LOOP_ERR_HANDLER (4); \
|
|
|
ce426f |
+ } \
|
|
|
ce426f |
outptr[0] = 0xe0; \
|
|
|
ce426f |
outptr[0] |= wc >> 12; \
|
|
|
ce426f |
\
|
|
|
ce426f |
@@ -479,7 +642,7 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
else if (wc <= 0x10ffff) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* Four UTF-8 chars. */ \
|
|
|
ce426f |
- if (__glibc_unlikely (outptr + 4 > outend)) \
|
|
|
ce426f |
+ if (__glibc_unlikely (outptr + 4 > outend)) \
|
|
|
ce426f |
{ \
|
|
|
ce426f |
/* Overflow in the output buffer. */ \
|
|
|
ce426f |
result = __GCONV_FULL_OUTPUT; \
|
|
|
ce426f |
@@ -505,7 +668,140 @@ gconv_end (struct __gconv_step *data)
|
|
|
ce426f |
} \
|
|
|
ce426f |
inptr += 4; \
|
|
|
ce426f |
}
|
|
|
ce426f |
+
|
|
|
ce426f |
+#define HW_TO_VX \
|
|
|
ce426f |
+ { \
|
|
|
ce426f |
+ register const unsigned char* pInput asm ("8") = inptr; \
|
|
|
ce426f |
+ register size_t inlen asm ("9") = inend - inptr; \
|
|
|
ce426f |
+ register unsigned char* pOutput asm ("10") = outptr; \
|
|
|
ce426f |
+ register size_t outlen asm("11") = outend - outptr; \
|
|
|
ce426f |
+ unsigned long tmp, tmp2; \
|
|
|
ce426f |
+ asm volatile (".machine push\n\t" \
|
|
|
ce426f |
+ ".machine \"z13\"\n\t" \
|
|
|
ce426f |
+ ".machinemode \"zarch_nohighgprs\"\n\t" \
|
|
|
ce426f |
+ " vleif %%v20,127,0\n\t" /* element 0: 127 */ \
|
|
|
ce426f |
+ " vzero %%v21\n\t" \
|
|
|
ce426f |
+ " vleih %%v21,8192,0\n\t" /* element 0: > */ \
|
|
|
ce426f |
+ " vleih %%v21,-8192,2\n\t" /* element 1: =<> */ \
|
|
|
ce426f |
+ /* Loop which handles UTF-32 chars <=0x7f. */ \
|
|
|
ce426f |
+ "0: clgijl %[R_INLEN],64,20f\n\t" \
|
|
|
ce426f |
+ " clgijl %[R_OUTLEN],16,20f\n\t" \
|
|
|
ce426f |
+ "1: vlm %%v16,%%v19,0(%[R_IN])\n\t" \
|
|
|
ce426f |
+ " lghi %[R_TMP],0\n\t" \
|
|
|
ce426f |
+ /* Shorten to byte values. */ \
|
|
|
ce426f |
+ " vpkf %%v23,%%v16,%%v17\n\t" \
|
|
|
ce426f |
+ " vpkf %%v24,%%v18,%%v19\n\t" \
|
|
|
ce426f |
+ " vpkh %%v23,%%v23,%%v24\n\t" \
|
|
|
ce426f |
+ /* Checking for values > 0x7f. */ \
|
|
|
ce426f |
+ " vstrcfs %%v22,%%v16,%%v20,%%v21\n\t" \
|
|
|
ce426f |
+ " jno 10f\n\t" \
|
|
|
ce426f |
+ " vstrcfs %%v22,%%v17,%%v20,%%v21\n\t" \
|
|
|
ce426f |
+ " jno 11f\n\t" \
|
|
|
ce426f |
+ " vstrcfs %%v22,%%v18,%%v20,%%v21\n\t" \
|
|
|
ce426f |
+ " jno 12f\n\t" \
|
|
|
ce426f |
+ " vstrcfs %%v22,%%v19,%%v20,%%v21\n\t" \
|
|
|
ce426f |
+ " jno 13f\n\t" \
|
|
|
ce426f |
+ /* Store 16bytes to outptr. */ \
|
|
|
ce426f |
+ " vst %%v23,0(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " aghi %[R_INLEN],-64\n\t" \
|
|
|
ce426f |
+ " aghi %[R_OUTLEN],-16\n\t" \
|
|
|
ce426f |
+ " la %[R_IN],64(%[R_IN])\n\t" \
|
|
|
ce426f |
+ " la %[R_OUT],16(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " clgijl %[R_INLEN],64,20f\n\t" \
|
|
|
ce426f |
+ " clgijl %[R_OUTLEN],16,20f\n\t" \
|
|
|
ce426f |
+ " j 1b\n\t" \
|
|
|
ce426f |
+ /* Found a value > 0x7f. */ \
|
|
|
ce426f |
+ "13: ahi %[R_TMP],4\n\t" \
|
|
|
ce426f |
+ "12: ahi %[R_TMP],4\n\t" \
|
|
|
ce426f |
+ "11: ahi %[R_TMP],4\n\t" \
|
|
|
ce426f |
+ "10: vlgvb %[R_I],%%v22,7\n\t" \
|
|
|
ce426f |
+ " srlg %[R_I],%[R_I],2\n\t" \
|
|
|
ce426f |
+ " agr %[R_I],%[R_TMP]\n\t" \
|
|
|
ce426f |
+ " je 20f\n\t" \
|
|
|
ce426f |
+ /* Store characters before invalid one... */ \
|
|
|
ce426f |
+ " slgr %[R_OUTLEN],%[R_I]\n\t" \
|
|
|
ce426f |
+ "15: aghi %[R_I],-1\n\t" \
|
|
|
ce426f |
+ " vstl %%v23,%[R_I],0(%[R_OUT])\n\t" \
|
|
|
ce426f |
+ /* ... and update pointers. */ \
|
|
|
ce426f |
+ " aghi %[R_I],1\n\t" \
|
|
|
ce426f |
+ " la %[R_OUT],0(%[R_I],%[R_OUT])\n\t" \
|
|
|
ce426f |
+ " sllg %[R_I],%[R_I],2\n\t" \
|
|
|
ce426f |
+ " la %[R_IN],0(%[R_I],%[R_IN])\n\t" \
|
|
|
ce426f |
+ " slgr %[R_INLEN],%[R_I]\n\t" \
|
|
|
ce426f |
+ /* Handle multibyte utf8-char with convert instruction. */ \
|
|
|
ce426f |
+ "20: cu41 %[R_OUT],%[R_IN]\n\t" \
|
|
|
ce426f |
+ " jo 0b\n\t" /* Try vector implemenation again. */ \
|
|
|
ce426f |
+ " lochil %[R_RES],%[RES_OUT_FULL]\n\t" /* cc == 1. */ \
|
|
|
ce426f |
+ " lochih %[R_RES],%[RES_IN_ILL]\n\t" /* cc == 2. */ \
|
|
|
ce426f |
+ ".machine pop" \
|
|
|
ce426f |
+ : /* outputs */ [R_IN] "+a" (pInput) \
|
|
|
ce426f |
+ , [R_INLEN] "+d" (inlen), [R_OUT] "+a" (pOutput) \
|
|
|
ce426f |
+ , [R_OUTLEN] "+d" (outlen), [R_TMP] "=d" (tmp) \
|
|
|
ce426f |
+ , [R_I] "=a" (tmp2) \
|
|
|
ce426f |
+ , [R_RES] "+d" (result) \
|
|
|
ce426f |
+ : /* inputs */ \
|
|
|
ce426f |
+ [RES_OUT_FULL] "i" (__GCONV_FULL_OUTPUT) \
|
|
|
ce426f |
+ , [RES_IN_ILL] "i" (__GCONV_ILLEGAL_INPUT) \
|
|
|
ce426f |
+ : /* clobber list */ "memory", "cc" \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v20") ASM_CLOBBER_VR ("v21") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v22") ASM_CLOBBER_VR ("v23") \
|
|
|
ce426f |
+ ASM_CLOBBER_VR ("v24") \
|
|
|
ce426f |
+ ); \
|
|
|
ce426f |
+ inptr = pInput; \
|
|
|
ce426f |
+ outptr = pOutput; \
|
|
|
ce426f |
+ }
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* Generate loop-function with software routing. */
|
|
|
ce426f |
+#define MIN_NEEDED_INPUT MIN_NEEDED_TO
|
|
|
ce426f |
+#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM
|
|
|
ce426f |
+#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM
|
|
|
ce426f |
+#define LOOPFCT __to_utf8_loop_c
|
|
|
ce426f |
+#define BODY BODY_TO_C
|
|
|
ce426f |
+#define LOOP_NEED_FLAGS
|
|
|
ce426f |
+#include <iconv/loop.c>
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* Generate loop-function with hardware utf-convert instruction. */
|
|
|
ce426f |
+#define MIN_NEEDED_INPUT MIN_NEEDED_TO
|
|
|
ce426f |
+#define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM
|
|
|
ce426f |
+#define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM
|
|
|
ce426f |
+#define LOOPFCT __to_utf8_loop_etf3eh
|
|
|
ce426f |
#define LOOP_NEED_FLAGS
|
|
|
ce426f |
+#define BODY BODY_TO_ETF3EH
|
|
|
ce426f |
#include <iconv/loop.c>
|
|
|
ce426f |
|
|
|
ce426f |
+#if defined HAVE_S390_VX_ASM_SUPPORT
|
|
|
ce426f |
+/* Generate loop-function with hardware vector and utf-convert instructions. */
|
|
|
ce426f |
+# define MIN_NEEDED_INPUT MIN_NEEDED_TO
|
|
|
ce426f |
+# define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM
|
|
|
ce426f |
+# define MAX_NEEDED_OUTPUT MAX_NEEDED_FROM
|
|
|
ce426f |
+# define LOOPFCT __to_utf8_loop_vx
|
|
|
ce426f |
+# define BODY BODY_TO_VX
|
|
|
ce426f |
+# define LOOP_NEED_FLAGS
|
|
|
ce426f |
+# include <iconv/loop.c>
|
|
|
ce426f |
+#endif
|
|
|
ce426f |
+
|
|
|
ce426f |
+/* Generate ifunc'ed loop function. */
|
|
|
ce426f |
+__typeof(__to_utf8_loop_c)
|
|
|
ce426f |
+__attribute__ ((ifunc ("__to_utf8_loop_resolver")))
|
|
|
ce426f |
+__to_utf8_loop;
|
|
|
ce426f |
+
|
|
|
ce426f |
+static void *
|
|
|
ce426f |
+__to_utf8_loop_resolver (unsigned long int dl_hwcap)
|
|
|
ce426f |
+{
|
|
|
ce426f |
+#if defined HAVE_S390_VX_ASM_SUPPORT
|
|
|
ce426f |
+ if (dl_hwcap & HWCAP_S390_VX)
|
|
|
ce426f |
+ return __to_utf8_loop_vx;
|
|
|
ce426f |
+ else
|
|
|
ce426f |
+#endif
|
|
|
ce426f |
+ if (dl_hwcap & HWCAP_S390_ETF3EH)
|
|
|
ce426f |
+ return __to_utf8_loop_etf3eh;
|
|
|
ce426f |
+ else
|
|
|
ce426f |
+ return __to_utf8_loop_c;
|
|
|
ce426f |
+}
|
|
|
ce426f |
+
|
|
|
ce426f |
+strong_alias (__to_utf8_loop_c_single, __to_utf8_loop_single)
|
|
|
ce426f |
+
|
|
|
ce426f |
+
|
|
|
ce426f |
#include <iconv/skeleton.c>
|
|
|
ce426f |
--
|
|
|
ce426f |
1.8.3.1
|
|
|
ce426f |
|