|
 |
00db10 |
From c3fd92047b3cc1b66b9b241be0765fe1e72678a1 Mon Sep 17 00:00:00 2001
|
|
|
00db10 |
From: Stefan Liebler <stli@linux.vnet.ibm.com>
|
|
|
00db10 |
Date: Mon, 7 Nov 2016 15:29:26 +0100
|
|
|
00db10 |
Subject: [PATCH 03/17] S390: Optimize 8bit-generic iconv modules.
|
|
|
00db10 |
|
|
|
00db10 |
Upstream commit 4690dab084f854bf0013b5eaabcf90c2d5b692ff
|
|
|
00db10 |
|
|
|
00db10 |
This patch introduces a s390 specific 8bit-generic.c file which provides an
|
|
|
00db10 |
optimized version for z13 with translate-/vector-instructions, which will be
|
|
|
00db10 |
chosen at runtime via ifunc.
|
|
|
00db10 |
If the build-environment lacks vector support, then iconvdata/8bit-generic.c
|
|
|
00db10 |
is used wihtout any change. Otherwise iconvdata/8bit-generic.c is used to create
|
|
|
00db10 |
conversion loop routines without vector instructions as fallback, if vector
|
|
|
00db10 |
instructions aren't available at runtime.
|
|
|
00db10 |
|
|
|
00db10 |
The vector routines can only be used with charsets where the maximum UCS4 value
|
|
|
00db10 |
fits in 1 byte size. Then the hardware translate-instruction is used
|
|
|
00db10 |
to translate between up to 256 generic characters and "1 byte UCS4"
|
|
|
00db10 |
characters at once. The vector instructions are used to convert between
|
|
|
00db10 |
the "1 byte UCS4" and UCS4.
|
|
|
00db10 |
|
|
|
00db10 |
The gen-8bit.sh script in sysdeps/s390/multiarch generates the conversion
|
|
|
00db10 |
table to_ucs1. Therefore in sysdeps/s390/multiarch/Makefile is added an
|
|
|
00db10 |
override define generate-8bit-table, which is originally defined in
|
|
|
00db10 |
iconvdata/Makefile. This version calls the gen-8bit.sh in iconvdata folder
|
|
|
00db10 |
and the s390 one.
|
|
|
00db10 |
|
|
|
00db10 |
ChangeLog:
|
|
|
00db10 |
|
|
|
00db10 |
* sysdeps/s390/multiarch/8bit-generic.c: New File.
|
|
|
00db10 |
* sysdeps/s390/multiarch/gen-8bit.sh: New File.
|
|
|
00db10 |
* sysdeps/s390/multiarch/Makefile (generate-8bit-table):
|
|
|
00db10 |
New override define.
|
|
|
00db10 |
* sysdeps/s390/multiarch/iconv/skeleton.c: Likewise.
|
|
|
00db10 |
---
|
|
|
00db10 |
sysdeps/s390/multiarch/8bit-generic.c | 415 ++++++++++++++++++++++++++++++++
|
|
|
00db10 |
sysdeps/s390/multiarch/Makefile | 10 +
|
|
|
00db10 |
sysdeps/s390/multiarch/gen-8bit.sh | 6 +
|
|
|
00db10 |
sysdeps/s390/multiarch/iconv/skeleton.c | 21 ++
|
|
|
00db10 |
4 files changed, 452 insertions(+)
|
|
|
00db10 |
create mode 100644 sysdeps/s390/multiarch/8bit-generic.c
|
|
|
00db10 |
create mode 100644 sysdeps/s390/multiarch/gen-8bit.sh
|
|
|
00db10 |
create mode 100644 sysdeps/s390/multiarch/iconv/skeleton.c
|
|
|
00db10 |
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/8bit-generic.c b/sysdeps/s390/multiarch/8bit-generic.c
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..93565e1
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/8bit-generic.c
|
|
|
00db10 |
@@ -0,0 +1,415 @@
|
|
|
00db10 |
+/* Generic conversion to and from 8bit charsets - S390 version.
|
|
|
00db10 |
+ Copyright (C) 2016 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#if defined HAVE_S390_VX_ASM_SUPPORT
|
|
|
00db10 |
+
|
|
|
00db10 |
+# if defined HAVE_S390_VX_GCC_SUPPORT
|
|
|
00db10 |
+# define ASM_CLOBBER_VR(NR) , NR
|
|
|
00db10 |
+# else
|
|
|
00db10 |
+# define ASM_CLOBBER_VR(NR)
|
|
|
00db10 |
+# endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Generate the conversion loop routines without vector instructions as
|
|
|
00db10 |
+ fallback, if vector instructions aren't available at runtime. */
|
|
|
00db10 |
+# define IGNORE_ICONV_SKELETON
|
|
|
00db10 |
+# define from_generic __from_generic_c
|
|
|
00db10 |
+# define to_generic __to_generic_c
|
|
|
00db10 |
+# include "iconvdata/8bit-generic.c"
|
|
|
00db10 |
+# undef IGNORE_ICONV_SKELETON
|
|
|
00db10 |
+# undef from_generic
|
|
|
00db10 |
+# undef to_generic
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Generate the converion routines with vector instructions. The vector
|
|
|
00db10 |
+ routines can only be used with charsets where the maximum UCS4 value
|
|
|
00db10 |
+ fits in 1 byte size. Then the hardware translate-instruction is used
|
|
|
00db10 |
+ to translate between multiple generic characters and "1 byte UCS4"
|
|
|
00db10 |
+ characters at once. The vector instructions are used to convert between
|
|
|
00db10 |
+ the "1 byte UCS4" and UCS4. */
|
|
|
00db10 |
+# include <unistd.h>
|
|
|
00db10 |
+# include <dl-procinfo.h>
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef FROM_LOOP
|
|
|
00db10 |
+# undef TO_LOOP
|
|
|
00db10 |
+# define FROM_LOOP __from_generic_vx
|
|
|
00db10 |
+# define TO_LOOP __to_generic_vx
|
|
|
00db10 |
+
|
|
|
00db10 |
+# define MIN_NEEDED_FROM 1
|
|
|
00db10 |
+# define MIN_NEEDED_TO 4
|
|
|
00db10 |
+# define ONE_DIRECTION 0
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* First define the conversion function from the 8bit charset to UCS4. */
|
|
|
00db10 |
+# define MIN_NEEDED_INPUT MIN_NEEDED_FROM
|
|
|
00db10 |
+# define MIN_NEEDED_OUTPUT MIN_NEEDED_TO
|
|
|
00db10 |
+# define LOOPFCT FROM_LOOP
|
|
|
00db10 |
+# define BODY_FROM_ORIG \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ uint32_t ch = to_ucs4[*inptr]; \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ if (HAS_HOLES && __builtin_expect (ch == L'\0', 0) && *inptr != '\0') \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* This is an illegal character. */ \
|
|
|
00db10 |
+ STANDARD_FROM_LOOP_ERR_HANDLER (1); \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ put32 (outptr, ch); \
|
|
|
00db10 |
+ outptr += 4; \
|
|
|
00db10 |
+ ++inptr; \
|
|
|
00db10 |
+ }
|
|
|
00db10 |
+
|
|
|
00db10 |
+# define BODY \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ if (__builtin_expect (inend - inptr < 16, 1) \
|
|
|
00db10 |
+ || outend - outptr < 64) \
|
|
|
00db10 |
+ /* Convert remaining bytes with c code. */ \
|
|
|
00db10 |
+ BODY_FROM_ORIG \
|
|
|
00db10 |
+ else \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* Convert 16 ... 256 bytes at once with tr-instruction. */ \
|
|
|
00db10 |
+ size_t index; \
|
|
|
00db10 |
+ char buf[256]; \
|
|
|
00db10 |
+ size_t loop_count = (inend - inptr) / 16; \
|
|
|
00db10 |
+ if (loop_count > (outend - outptr) / 64) \
|
|
|
00db10 |
+ loop_count = (outend - outptr) / 64; \
|
|
|
00db10 |
+ if (loop_count > 16) \
|
|
|
00db10 |
+ loop_count = 16; \
|
|
|
00db10 |
+ __asm__ volatile (".machine push\n\t" \
|
|
|
00db10 |
+ ".machine \"z13\"\n\t" \
|
|
|
00db10 |
+ ".machinemode \"zarch_nohighgprs\"\n\t" \
|
|
|
00db10 |
+ " sllk %[R_I],%[R_LI],4\n\t" \
|
|
|
00db10 |
+ " ahi %[R_I],-1\n\t" \
|
|
|
00db10 |
+ /* Execute mvc and tr with correct len. */ \
|
|
|
00db10 |
+ " exrl %[R_I],21f\n\t" \
|
|
|
00db10 |
+ " exrl %[R_I],22f\n\t" \
|
|
|
00db10 |
+ /* Post-processing. */ \
|
|
|
00db10 |
+ " lghi %[R_I],0\n\t" \
|
|
|
00db10 |
+ " vzero %%v0\n\t" \
|
|
|
00db10 |
+ "0: \n\t" \
|
|
|
00db10 |
+ /* Find invalid character - value is zero. */ \
|
|
|
00db10 |
+ " vl %%v16,0(%[R_I],%[R_BUF])\n\t" \
|
|
|
00db10 |
+ " vceqbs %%v23,%%v0,%%v16\n\t" \
|
|
|
00db10 |
+ " jle 10f\n\t" \
|
|
|
00db10 |
+ "1: \n\t" \
|
|
|
00db10 |
+ /* Enlarge to UCS4. */ \
|
|
|
00db10 |
+ " vuplhb %%v17,%%v16\n\t" \
|
|
|
00db10 |
+ " vupllb %%v18,%%v16\n\t" \
|
|
|
00db10 |
+ " vuplhh %%v19,%%v17\n\t" \
|
|
|
00db10 |
+ " vupllh %%v20,%%v17\n\t" \
|
|
|
00db10 |
+ " vuplhh %%v21,%%v18\n\t" \
|
|
|
00db10 |
+ " vupllh %%v22,%%v18\n\t" \
|
|
|
00db10 |
+ /* Store 64bytes to buf_out. */ \
|
|
|
00db10 |
+ " vstm %%v19,%%v22,0(%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " aghi %[R_I],16\n\t" \
|
|
|
00db10 |
+ " la %[R_OUT],64(%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " brct %[R_LI],0b\n\t" \
|
|
|
00db10 |
+ " la %[R_IN],0(%[R_I],%[R_IN])\n\t" \
|
|
|
00db10 |
+ " j 20f\n\t" \
|
|
|
00db10 |
+ "21: mvc 0(1,%[R_BUF]),0(%[R_IN])\n\t" \
|
|
|
00db10 |
+ "22: tr 0(1,%[R_BUF]),0(%[R_TBL])\n\t" \
|
|
|
00db10 |
+ /* Possibly invalid character found. */ \
|
|
|
00db10 |
+ "10: \n\t" \
|
|
|
00db10 |
+ /* Test if input was zero, too. */ \
|
|
|
00db10 |
+ " vl %%v24,0(%[R_I],%[R_IN])\n\t" \
|
|
|
00db10 |
+ " vceqb %%v24,%%v0,%%v24\n\t" \
|
|
|
00db10 |
+ /* Zeros in buf (v23) and inptr (v24) are marked \
|
|
|
00db10 |
+ with one bits. After xor, invalid characters \
|
|
|
00db10 |
+ are marked as one bits. Proceed, if no \
|
|
|
00db10 |
+ invalid characters are found. */ \
|
|
|
00db10 |
+ " vx %%v24,%%v23,%%v24\n\t" \
|
|
|
00db10 |
+ " vfenebs %%v24,%%v24,%%v0\n\t" \
|
|
|
00db10 |
+ " jo 1b\n\t" \
|
|
|
00db10 |
+ /* Found an invalid translation. \
|
|
|
00db10 |
+ Store the preceding chars. */ \
|
|
|
00db10 |
+ " la %[R_IN],0(%[R_I],%[R_IN])\n\t" \
|
|
|
00db10 |
+ " vlgvb %[R_I],%%v24,7\n\t" \
|
|
|
00db10 |
+ " la %[R_IN],0(%[R_I],%[R_IN])\n\t" \
|
|
|
00db10 |
+ " sll %[R_I],2\n\t" \
|
|
|
00db10 |
+ " ahi %[R_I],-1\n\t" \
|
|
|
00db10 |
+ " jl 20f\n\t" \
|
|
|
00db10 |
+ " lgr %[R_LI],%[R_I]\n\t" \
|
|
|
00db10 |
+ " vuplhb %%v17,%%v16\n\t" \
|
|
|
00db10 |
+ " vuplhh %%v19,%%v17\n\t" \
|
|
|
00db10 |
+ " vstl %%v19,%[R_I],0(%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " ahi %[R_I],-16\n\t" \
|
|
|
00db10 |
+ " jl 11f\n\t" \
|
|
|
00db10 |
+ " vupllh %%v20,%%v17\n\t" \
|
|
|
00db10 |
+ " vstl %%v20,%[R_I],16(%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " ahi %[R_I],-16\n\t" \
|
|
|
00db10 |
+ " jl 11f\n\t" \
|
|
|
00db10 |
+ " vupllb %%v18,%%v16\n\t" \
|
|
|
00db10 |
+ " vuplhh %%v21,%%v18\n\t" \
|
|
|
00db10 |
+ " vstl %%v21,%[R_I],32(%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " ahi %[R_I],-16\n\t" \
|
|
|
00db10 |
+ " jl 11f\n\t" \
|
|
|
00db10 |
+ " vupllh %%v22,%%v18\n\t" \
|
|
|
00db10 |
+ " vstl %%v22,%[R_I],48(%[R_OUT])\n\t" \
|
|
|
00db10 |
+ "11: \n\t" \
|
|
|
00db10 |
+ " la %[R_OUT],1(%[R_LI],%[R_OUT])\n\t" \
|
|
|
00db10 |
+ "20: \n\t" \
|
|
|
00db10 |
+ ".machine pop" \
|
|
|
00db10 |
+ : /* outputs */ [R_IN] "+a" (inptr) \
|
|
|
00db10 |
+ , [R_OUT] "+a" (outptr), [R_I] "=&a" (index) \
|
|
|
00db10 |
+ , [R_LI] "+a" (loop_count) \
|
|
|
00db10 |
+ : /* inputs */ [R_BUF] "a" (buf) \
|
|
|
00db10 |
+ , [R_TBL] "a" (to_ucs1) \
|
|
|
00db10 |
+ : /* clobber list*/ "memory", "cc" \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v0") ASM_CLOBBER_VR ("v16") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v17") ASM_CLOBBER_VR ("v18") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v19") ASM_CLOBBER_VR ("v20") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v21") ASM_CLOBBER_VR ("v22") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v23") ASM_CLOBBER_VR ("v24") \
|
|
|
00db10 |
+ ); \
|
|
|
00db10 |
+ /* Error occured? */ \
|
|
|
00db10 |
+ if (loop_count != 0) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* Found an invalid character! */ \
|
|
|
00db10 |
+ STANDARD_FROM_LOOP_ERR_HANDLER (1); \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ }
|
|
|
00db10 |
+
|
|
|
00db10 |
+# define LOOP_NEED_FLAGS
|
|
|
00db10 |
+# include <iconv/loop.c>
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Next, define the other direction - from UCS4 to 8bit charset. */
|
|
|
00db10 |
+# define MIN_NEEDED_INPUT MIN_NEEDED_TO
|
|
|
00db10 |
+# define MIN_NEEDED_OUTPUT MIN_NEEDED_FROM
|
|
|
00db10 |
+# define LOOPFCT TO_LOOP
|
|
|
00db10 |
+# define BODY_TO_ORIG \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ uint32_t ch = get32 (inptr); \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ if (__builtin_expect (ch >= sizeof (from_ucs4) / sizeof (from_ucs4[0]), 0)\
|
|
|
00db10 |
+ || (__builtin_expect (from_ucs4[ch], '\1') == '\0' && ch != 0)) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ UNICODE_TAG_HANDLER (ch, 4); \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ /* This is an illegal character. */ \
|
|
|
00db10 |
+ STANDARD_TO_LOOP_ERR_HANDLER (4); \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ *outptr++ = from_ucs4[ch]; \
|
|
|
00db10 |
+ inptr += 4; \
|
|
|
00db10 |
+ }
|
|
|
00db10 |
+# define BODY \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ if (__builtin_expect (inend - inptr < 64, 1) \
|
|
|
00db10 |
+ || outend - outptr < 16) \
|
|
|
00db10 |
+ /* Convert remaining bytes with c code. */ \
|
|
|
00db10 |
+ BODY_TO_ORIG \
|
|
|
00db10 |
+ else \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* Convert 64 ... 1024 bytes at once with tr-instruction. */ \
|
|
|
00db10 |
+ size_t index, tmp; \
|
|
|
00db10 |
+ char buf[256]; \
|
|
|
00db10 |
+ size_t loop_count = (inend - inptr) / 64; \
|
|
|
00db10 |
+ uint32_t max = sizeof (from_ucs4) / sizeof (from_ucs4[0]); \
|
|
|
00db10 |
+ if (loop_count > (outend - outptr) / 16) \
|
|
|
00db10 |
+ loop_count = (outend - outptr) / 16; \
|
|
|
00db10 |
+ if (loop_count > 16) \
|
|
|
00db10 |
+ loop_count = 16; \
|
|
|
00db10 |
+ size_t remaining_loop_count = loop_count; \
|
|
|
00db10 |
+ /* Step 1: Check for ch>=max, ch == 0 and shorten to bytes. \
|
|
|
00db10 |
+ (ch == 0 is no error, but is handled differently) */ \
|
|
|
00db10 |
+ __asm__ volatile (".machine push\n\t" \
|
|
|
00db10 |
+ ".machine \"z13\"\n\t" \
|
|
|
00db10 |
+ ".machinemode \"zarch_nohighgprs\"\n\t" \
|
|
|
00db10 |
+ /* Setup to check for ch >= max. */ \
|
|
|
00db10 |
+ " vzero %%v21\n\t" \
|
|
|
00db10 |
+ " vleih %%v21,-24576,0\n\t" /* element 0: > */ \
|
|
|
00db10 |
+ " vleih %%v21,-8192,2\n\t" /* element 1: =<> */ \
|
|
|
00db10 |
+ " vlvgf %%v20,%[R_MAX],0\n\t" /* element 0: val */ \
|
|
|
00db10 |
+ /* Process in 64byte - 16 characters blocks. */ \
|
|
|
00db10 |
+ " lghi %[R_I],0\n\t" \
|
|
|
00db10 |
+ " lghi %[R_TMP],0\n\t" \
|
|
|
00db10 |
+ "0: \n\t" \
|
|
|
00db10 |
+ " vlm %%v16,%%v19,0(%[R_IN])\n\t" \
|
|
|
00db10 |
+ /* Test for ch >= max and ch == 0. */ \
|
|
|
00db10 |
+ " vstrczfs %%v22,%%v16,%%v20,%%v21\n\t" \
|
|
|
00db10 |
+ " jno 10f\n\t" \
|
|
|
00db10 |
+ " vstrczfs %%v22,%%v17,%%v20,%%v21\n\t" \
|
|
|
00db10 |
+ " jno 11f\n\t" \
|
|
|
00db10 |
+ " vstrczfs %%v22,%%v18,%%v20,%%v21\n\t" \
|
|
|
00db10 |
+ " jno 12f\n\t" \
|
|
|
00db10 |
+ " vstrczfs %%v22,%%v19,%%v20,%%v21\n\t" \
|
|
|
00db10 |
+ " jno 13f\n\t" \
|
|
|
00db10 |
+ /* Shorten to byte values. */ \
|
|
|
00db10 |
+ " vpkf %%v16,%%v16,%%v17\n\t" \
|
|
|
00db10 |
+ " vpkf %%v18,%%v18,%%v19\n\t" \
|
|
|
00db10 |
+ " vpkh %%v16,%%v16,%%v18\n\t" \
|
|
|
00db10 |
+ /* Store 16bytes to buf. */ \
|
|
|
00db10 |
+ " vst %%v16,0(%[R_I],%[R_BUF])\n\t" \
|
|
|
00db10 |
+ /* Loop until all blocks are processed. */ \
|
|
|
00db10 |
+ " la %[R_IN],64(%[R_IN])\n\t" \
|
|
|
00db10 |
+ " aghi %[R_I],16\n\t" \
|
|
|
00db10 |
+ " brct %[R_LI],0b\n\t" \
|
|
|
00db10 |
+ " j 20f\n\t" \
|
|
|
00db10 |
+ /* Found error ch >= max or ch == 0. */ \
|
|
|
00db10 |
+ "13: aghi %[R_TMP],4\n\t" \
|
|
|
00db10 |
+ "12: aghi %[R_TMP],4\n\t" \
|
|
|
00db10 |
+ "11: aghi %[R_TMP],4\n\t" \
|
|
|
00db10 |
+ "10: vlgvb %[R_I],%%v22,7\n\t" \
|
|
|
00db10 |
+ " srlg %[R_I],%[R_I],2\n\t" \
|
|
|
00db10 |
+ " agr %[R_I],%[R_TMP]\n\t" \
|
|
|
00db10 |
+ "20: \n\t" \
|
|
|
00db10 |
+ ".machine pop" \
|
|
|
00db10 |
+ : /* outputs */ [R_IN] "+a" (inptr) \
|
|
|
00db10 |
+ , [R_I] "=&a" (index) \
|
|
|
00db10 |
+ , [R_TMP] "=d" (tmp) \
|
|
|
00db10 |
+ , [R_LI] "+d" (remaining_loop_count) \
|
|
|
00db10 |
+ : /* inputs */ [R_BUF] "a" (buf) \
|
|
|
00db10 |
+ , [R_MAX] "d" (max) \
|
|
|
00db10 |
+ : /* clobber list*/ "memory", "cc" \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v16") ASM_CLOBBER_VR ("v17") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v18") ASM_CLOBBER_VR ("v19") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v20") ASM_CLOBBER_VR ("v21") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v22") \
|
|
|
00db10 |
+ ); \
|
|
|
00db10 |
+ /* Error occured in step 1? An error (ch >= max || ch == 0) \
|
|
|
00db10 |
+ occured, if remaining_loop_count > 0. The error occured \
|
|
|
00db10 |
+ at character-index (index) after already processed blocks. */ \
|
|
|
00db10 |
+ loop_count -= remaining_loop_count; \
|
|
|
00db10 |
+ if (loop_count > 0) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* Step 2: Translate already processed blocks in buf and \
|
|
|
00db10 |
+ check for errors (from_ucs4[ch] == 0). */ \
|
|
|
00db10 |
+ __asm__ volatile (".machine push\n\t" \
|
|
|
00db10 |
+ ".machine \"z13\"\n\t" \
|
|
|
00db10 |
+ ".machinemode \"zarch_nohighgprs\"\n\t" \
|
|
|
00db10 |
+ " sllk %[R_I],%[R_LI],4\n\t" \
|
|
|
00db10 |
+ " ahi %[R_I],-1\n\t" \
|
|
|
00db10 |
+ /* Execute tr with correct len. */ \
|
|
|
00db10 |
+ " exrl %[R_I],21f\n\t" \
|
|
|
00db10 |
+ /* Post-processing. */ \
|
|
|
00db10 |
+ " lghi %[R_I],0\n\t" \
|
|
|
00db10 |
+ "0: \n\t" \
|
|
|
00db10 |
+ /* Find invalid character - value == 0. */ \
|
|
|
00db10 |
+ " vl %%v16,0(%[R_I],%[R_BUF])\n\t" \
|
|
|
00db10 |
+ " vfenezbs %%v17,%%v16,%%v16\n\t" \
|
|
|
00db10 |
+ " je 10f\n\t" \
|
|
|
00db10 |
+ /* Store 16bytes to buf_out. */ \
|
|
|
00db10 |
+ " vst %%v16,0(%[R_I],%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " aghi %[R_I],16\n\t" \
|
|
|
00db10 |
+ " brct %[R_LI],0b\n\t" \
|
|
|
00db10 |
+ " la %[R_OUT],0(%[R_I],%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " j 20f\n\t" \
|
|
|
00db10 |
+ "21: tr 0(1,%[R_BUF]),0(%[R_TBL])\n\t" \
|
|
|
00db10 |
+ /* Found an error: from_ucs4[ch] == 0. */ \
|
|
|
00db10 |
+ "10: la %[R_OUT],0(%[R_I],%[R_OUT])\n\t" \
|
|
|
00db10 |
+ " vlgvb %[R_I],%%v17,7\n\t" \
|
|
|
00db10 |
+ "20: \n\t" \
|
|
|
00db10 |
+ ".machine pop" \
|
|
|
00db10 |
+ : /* outputs */ [R_OUT] "+a" (outptr) \
|
|
|
00db10 |
+ , [R_I] "=&a" (tmp) \
|
|
|
00db10 |
+ , [R_LI] "+d" (loop_count) \
|
|
|
00db10 |
+ : /* inputs */ [R_BUF] "a" (buf) \
|
|
|
00db10 |
+ , [R_TBL] "a" (from_ucs4) \
|
|
|
00db10 |
+ : /* clobber list*/ "memory", "cc" \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v16") \
|
|
|
00db10 |
+ ASM_CLOBBER_VR ("v17") \
|
|
|
00db10 |
+ ); \
|
|
|
00db10 |
+ /* Error occured in processed bytes of step 2? \
|
|
|
00db10 |
+ Thus possible error in step 1 is obselete.*/ \
|
|
|
00db10 |
+ if (tmp < 16) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ index = tmp; \
|
|
|
00db10 |
+ inptr -= loop_count * 64; \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ /* Error occured in step 1/2? */ \
|
|
|
00db10 |
+ if (index < 16) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* Found an invalid character (see step 2) or zero \
|
|
|
00db10 |
+ (see step 1) at index! Convert the chars before index \
|
|
|
00db10 |
+ manually. If there is a zero at index detected by step 1, \
|
|
|
00db10 |
+ there could be invalid characters before this zero. */ \
|
|
|
00db10 |
+ int i; \
|
|
|
00db10 |
+ uint32_t ch; \
|
|
|
00db10 |
+ for (i = 0; i < index; i++) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ ch = get32 (inptr); \
|
|
|
00db10 |
+ if (__builtin_expect (from_ucs4[ch], '\1') == '\0') \
|
|
|
00db10 |
+ break; \
|
|
|
00db10 |
+ *outptr++ = from_ucs4[ch]; \
|
|
|
00db10 |
+ inptr += 4; \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ if (i == index) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ ch = get32 (inptr); \
|
|
|
00db10 |
+ if (ch == 0) \
|
|
|
00db10 |
+ { \
|
|
|
00db10 |
+ /* This is no error, but handled differently. */ \
|
|
|
00db10 |
+ *outptr++ = from_ucs4[ch]; \
|
|
|
00db10 |
+ inptr += 4; \
|
|
|
00db10 |
+ continue; \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ UNICODE_TAG_HANDLER (ch, 4); \
|
|
|
00db10 |
+ \
|
|
|
00db10 |
+ /* This is an illegal character. */ \
|
|
|
00db10 |
+ STANDARD_TO_LOOP_ERR_HANDLER (4); \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ } \
|
|
|
00db10 |
+ }
|
|
|
00db10 |
+
|
|
|
00db10 |
+# define LOOP_NEED_FLAGS
|
|
|
00db10 |
+# include <iconv/loop.c>
|
|
|
00db10 |
+
|
|
|
00db10 |
+
|
|
|
00db10 |
+/* Generate ifunc'ed loop function. */
|
|
|
00db10 |
+__typeof(__from_generic_c)
|
|
|
00db10 |
+__attribute__ ((ifunc ("__from_generic_resolver")))
|
|
|
00db10 |
+__from_generic;
|
|
|
00db10 |
+
|
|
|
00db10 |
+static void *
|
|
|
00db10 |
+__from_generic_resolver (unsigned long int dl_hwcap)
|
|
|
00db10 |
+{
|
|
|
00db10 |
+ if (sizeof (from_ucs4) / sizeof (from_ucs4[0]) <= 256
|
|
|
00db10 |
+ && dl_hwcap & HWCAP_S390_VX)
|
|
|
00db10 |
+ return &__from_generic_vx;
|
|
|
00db10 |
+ else
|
|
|
00db10 |
+ return &__from_generic_c;
|
|
|
00db10 |
+}
|
|
|
00db10 |
+
|
|
|
00db10 |
+__typeof(__to_generic_c)
|
|
|
00db10 |
+__attribute__ ((ifunc ("__to_generic_resolver")))
|
|
|
00db10 |
+__to_generic;
|
|
|
00db10 |
+
|
|
|
00db10 |
+static void *
|
|
|
00db10 |
+__to_generic_resolver (unsigned long int dl_hwcap)
|
|
|
00db10 |
+{
|
|
|
00db10 |
+ if (sizeof (from_ucs4) / sizeof (from_ucs4[0]) <= 256
|
|
|
00db10 |
+ && dl_hwcap & HWCAP_S390_VX)
|
|
|
00db10 |
+ return &__to_generic_vx;
|
|
|
00db10 |
+ else
|
|
|
00db10 |
+ return &__to_generic_c;
|
|
|
00db10 |
+}
|
|
|
00db10 |
+
|
|
|
00db10 |
+strong_alias (__to_generic_c_single, __to_generic_single)
|
|
|
00db10 |
+
|
|
|
00db10 |
+# undef FROM_LOOP
|
|
|
00db10 |
+# undef TO_LOOP
|
|
|
00db10 |
+# define FROM_LOOP __from_generic
|
|
|
00db10 |
+# define TO_LOOP __to_generic
|
|
|
00db10 |
+# include <iconv/skeleton.c>
|
|
|
00db10 |
+
|
|
|
00db10 |
+#else
|
|
|
00db10 |
+/* Generate this module without ifunc if build environment lacks vector
|
|
|
00db10 |
+ support. Instead the common 8bit-generic.c is used. */
|
|
|
00db10 |
+# include "iconvdata/8bit-generic.c"
|
|
|
00db10 |
+#endif /* !defined HAVE_S390_VX_ASM_SUPPORT */
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/Makefile b/sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
index 0805b07..11ad2b9 100644
|
|
|
00db10 |
--- a/sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/Makefile
|
|
|
00db10 |
@@ -42,3 +42,13 @@ sysdep_routines += wcslen wcslen-vx wcslen-c \
|
|
|
00db10 |
wmemset wmemset-vx wmemset-c \
|
|
|
00db10 |
wmemcmp wmemcmp-vx wmemcmp-c
|
|
|
00db10 |
endif
|
|
|
00db10 |
+
|
|
|
00db10 |
+ifeq ($(subdir),iconvdata)
|
|
|
00db10 |
+override define generate-8bit-table
|
|
|
00db10 |
+$(make-target-directory)
|
|
|
00db10 |
+LC_ALL=C $(SHELL) ./gen-8bit.sh $< > $(@:stmp=T)
|
|
|
00db10 |
+LC_ALL=C $(SHELL) ../sysdeps/s390/multiarch/gen-8bit.sh $< >> $(@:stmp=T)
|
|
|
00db10 |
+$(move-if-change) $(@:stmp=T) $(@:stmp=h)
|
|
|
00db10 |
+touch $@
|
|
|
00db10 |
+endef
|
|
|
00db10 |
+endif
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/gen-8bit.sh b/sysdeps/s390/multiarch/gen-8bit.sh
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..6f88c4b
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/gen-8bit.sh
|
|
|
00db10 |
@@ -0,0 +1,6 @@
|
|
|
00db10 |
+#!/bin/sh
|
|
|
00db10 |
+echo "static const uint8_t to_ucs1[256] = {"
|
|
|
00db10 |
+sed -ne '/^[^[:space:]]*[[:space:]]*.x00/d;/^END/q' \
|
|
|
00db10 |
+ -e 's/^<U00\(..\)>[[:space:]]*.x\(..\).*/ [0x\2] = 0x\1,/p' \
|
|
|
00db10 |
+ "$@" | sort -u
|
|
|
00db10 |
+echo "};"
|
|
|
00db10 |
diff --git a/sysdeps/s390/multiarch/iconv/skeleton.c b/sysdeps/s390/multiarch/iconv/skeleton.c
|
|
|
00db10 |
new file mode 100644
|
|
|
00db10 |
index 0000000..3a90031
|
|
|
00db10 |
--- /dev/null
|
|
|
00db10 |
+++ b/sysdeps/s390/multiarch/iconv/skeleton.c
|
|
|
00db10 |
@@ -0,0 +1,21 @@
|
|
|
00db10 |
+/* Skeleton for a conversion module - S390 version.
|
|
|
00db10 |
+ Copyright (C) 2016 Free Software Foundation, Inc.
|
|
|
00db10 |
+ This file is part of the GNU C Library.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is free software; you can redistribute it and/or
|
|
|
00db10 |
+ modify it under the terms of the GNU Lesser General Public
|
|
|
00db10 |
+ License as published by the Free Software Foundation; either
|
|
|
00db10 |
+ version 2.1 of the License, or (at your option) any later version.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ The GNU C Library is distributed in the hope that it will be useful,
|
|
|
00db10 |
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
00db10 |
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
00db10 |
+ Lesser General Public License for more details.
|
|
|
00db10 |
+
|
|
|
00db10 |
+ You should have received a copy of the GNU Lesser General Public
|
|
|
00db10 |
+ License along with the GNU C Library; if not, see
|
|
|
00db10 |
+ <http://www.gnu.org/licenses/>. */
|
|
|
00db10 |
+
|
|
|
00db10 |
+#ifndef IGNORE_ICONV_SKELETON
|
|
|
00db10 |
+# include_next <iconv/skeleton.c>
|
|
|
00db10 |
+#endif
|
|
|
00db10 |
--
|
|
|
00db10 |
1.8.3.1
|
|
|
00db10 |
|