From patchwork Fri Jul 2 17:11:31 2010 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Ken Werner X-Patchwork-Id: 57696 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) by ozlabs.org (Postfix) with SMTP id A7C0DB70A0 for ; Sat, 3 Jul 2010 03:11:51 +1000 (EST) Received: (qmail 28493 invoked by alias); 2 Jul 2010 17:11:48 -0000 Received: (qmail 28481 invoked by uid 22791); 2 Jul 2010 17:11:46 -0000 X-SWARE-Spam-Status: No, hits=-1.9 required=5.0 tests=AWL, BAYES_00, TW_BJ, T_RP_MATCHES_RCVD X-Spam-Check-By: sourceware.org Received: from mtagate6.uk.ibm.com (HELO mtagate6.uk.ibm.com) (194.196.100.166) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Fri, 02 Jul 2010 17:11:40 +0000 Received: from d06nrmr1707.portsmouth.uk.ibm.com (d06nrmr1707.portsmouth.uk.ibm.com [9.149.39.225]) by mtagate6.uk.ibm.com (8.13.1/8.13.1) with ESMTP id o62HBbra023449 for ; Fri, 2 Jul 2010 17:11:37 GMT Received: from d06av01.portsmouth.uk.ibm.com (d06av01.portsmouth.uk.ibm.com [9.149.37.212]) by d06nrmr1707.portsmouth.uk.ibm.com (8.13.8/8.13.8/NCO v10.0) with ESMTP id o62HBbbx1470496 for ; Fri, 2 Jul 2010 18:11:37 +0100 Received: from d06av01.portsmouth.uk.ibm.com (loopback [127.0.0.1]) by d06av01.portsmouth.uk.ibm.com (8.12.11.20060308/8.13.3) with ESMTP id o62HBbcX027462 for ; Fri, 2 Jul 2010 18:11:37 +0100 Received: from leonard.localnet (dyn-9-152-224-29.boeblingen.de.ibm.com [9.152.224.29]) by d06av01.portsmouth.uk.ibm.com (8.12.11.20060308/8.12.11) with ESMTP id o62HBVlo026874 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NO); Fri, 2 Jul 2010 18:11:36 +0100 From: Ken Werner To: gdb-patches@sourceware.org Subject: Re: [patch, libiberty, gdb] IEEE 754 half floating-point format Date: Fri, 2 Jul 2010 19:11:31 +0200 User-Agent: KMail/1.13.2 (Linux/2.6.32-23-generic-pae; KDE/4.4.2; i686; ; ) References: <201007021859.05191.ken@linux.vnet.ibm.com> In-Reply-To: <201007021859.05191.ken@linux.vnet.ibm.com> Cc: gcc-patches@gcc.gnu.org MIME-Version: 1.0 Message-Id: <201007021911.31134.ken@linux.vnet.ibm.com> X-IsSubscribed: yes Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org On Friday, July 02, 2010 06:59:05 pm Ken Werner wrote: > Hi, > > The attached patch adds the half precision floating-point format as > described in IEEE 754-2008 as binary16. Languages such as OpenCL and > architectures like ARM support this representation. > Any suggestions are welcome. > > Regards > -ken Whoops, I intended to crosspost this patch on gcc-patches as well because it touches files within the libiberty and include directory. Sorry. Regards -ken include/ChangeLog: 2010-07-02 Ken Werner * gfloatformat.h (floatformat_ieee_half_big): Add declaration. (floatformat_ieee_half_little): Likewise. libiberty/ChangeLog: 2010-07-02 Ken Werner * floatformat.c (floatformat_ieee_half_big): New variable. (floatformat_ieee_half_little): Likewise. ChangeLog: 2010-07-02 Ken Werner * gdbtypes.h (floatformats_ieee_half): Add declaration. * gdbtypes.c (floatformats_ieee_half): New variable. * doublest.c (floatformat_from_length): Set format to gdbarch_half_format if length matches. * gdbarch.sh (half_bit): New architecture method. (half_format): Likewise. * gdbarch.h: Regenerate. * gdbarch.c: Likewise. Index: include/floatformat.h =================================================================== RCS file: /cvs/src/src/include/floatformat.h,v retrieving revision 1.14 diff -u -p -r1.14 floatformat.h --- include/floatformat.h 8 Nov 2007 00:08:48 -0000 1.14 +++ include/floatformat.h 2 Jul 2010 15:35:08 -0000 @@ -96,6 +96,8 @@ struct floatformat /* floatformats for IEEE single and double, big and little endian. */ +extern const struct floatformat floatformat_ieee_half_big; +extern const struct floatformat floatformat_ieee_half_little; extern const struct floatformat floatformat_ieee_single_big; extern const struct floatformat floatformat_ieee_single_little; extern const struct floatformat floatformat_ieee_double_big; Index: libiberty/floatformat.c =================================================================== RCS file: /cvs/src/src/libiberty/floatformat.c,v retrieving revision 1.25 diff -u -p -r1.25 floatformat.c --- libiberty/floatformat.c 13 Nov 2007 15:12:42 -0000 1.25 +++ libiberty/floatformat.c 2 Jul 2010 15:35:08 -0000 @@ -77,7 +77,23 @@ floatformat_always_valid (const struct f a system header, what we do if not, etc. */ #define FLOATFORMAT_CHAR_BIT 8 -/* floatformats for IEEE single and double, big and little endian. */ +/* floatformats for IEEE half, single and double, big and little endian. */ +const struct floatformat floatformat_ieee_half_big = +{ + floatformat_big, 16, 0, 1, 5, 15, 31, 6, 10, + floatformat_intbit_no, + "floatformat_ieee_half_big", + floatformat_always_valid, + NULL +}; +const struct floatformat floatformat_ieee_half_little = +{ + floatformat_little, 16, 0, 1, 5, 15, 31, 6, 10, + floatformat_intbit_no, + "floatformat_ieee_half_little", + floatformat_always_valid, + NULL +}; const struct floatformat floatformat_ieee_single_big = { floatformat_big, 32, 0, 1, 8, 127, 255, 9, 23, Index: gdb/gdbtypes.h =================================================================== RCS file: /cvs/src/src/gdb/gdbtypes.h,v retrieving revision 1.132 diff -u -p -r1.132 gdbtypes.h --- gdb/gdbtypes.h 29 Jun 2010 16:17:58 -0000 1.132 +++ gdb/gdbtypes.h 2 Jul 2010 15:35:08 -0000 @@ -1191,6 +1191,7 @@ extern const struct objfile_type *objfil /* Explicit floating-point formats. See "floatformat.h". */ +extern const struct floatformat *floatformats_ieee_half[BFD_ENDIAN_UNKNOWN]; extern const struct floatformat *floatformats_ieee_single[BFD_ENDIAN_UNKNOWN]; extern const struct floatformat *floatformats_ieee_double[BFD_ENDIAN_UNKNOWN]; extern const struct floatformat *floatformats_ieee_double_littlebyte_bigword[BFD_ENDIAN_UNKNOWN]; Index: gdb/gdbtypes.c =================================================================== RCS file: /cvs/src/src/gdb/gdbtypes.c,v retrieving revision 1.194 diff -u -p -r1.194 gdbtypes.c --- gdb/gdbtypes.c 29 Jun 2010 16:53:09 -0000 1.194 +++ gdb/gdbtypes.c 2 Jul 2010 15:35:07 -0000 @@ -42,6 +42,10 @@ /* Floatformat pairs. */ +const struct floatformat *floatformats_ieee_half[BFD_ENDIAN_UNKNOWN] = { + &floatformat_ieee_half_big, + &floatformat_ieee_half_little +}; const struct floatformat *floatformats_ieee_single[BFD_ENDIAN_UNKNOWN] = { &floatformat_ieee_single_big, &floatformat_ieee_single_little Index: gdb/doublest.c =================================================================== RCS file: /cvs/src/src/gdb/doublest.c,v retrieving revision 1.44 diff -u -p -r1.44 doublest.c --- gdb/doublest.c 14 May 2010 17:53:16 -0000 1.44 +++ gdb/doublest.c 2 Jul 2010 15:35:05 -0000 @@ -777,7 +777,10 @@ floatformat_from_length (struct gdbarch { const struct floatformat *format; - if (len * TARGET_CHAR_BIT == gdbarch_float_bit (gdbarch)) + if (len * TARGET_CHAR_BIT == gdbarch_half_bit (gdbarch)) + format = gdbarch_half_format (gdbarch) + [gdbarch_byte_order (gdbarch)]; + else if (len * TARGET_CHAR_BIT == gdbarch_float_bit (gdbarch)) format = gdbarch_float_format (gdbarch) [gdbarch_byte_order (gdbarch)]; else if (len * TARGET_CHAR_BIT == gdbarch_double_bit (gdbarch)) Index: gdb/gdbarch.sh =================================================================== RCS file: /cvs/src/src/gdb/gdbarch.sh,v retrieving revision 1.512 diff -u -p -r1.512 gdbarch.sh --- gdb/gdbarch.sh 26 May 2010 18:19:27 -0000 1.512 +++ gdb/gdbarch.sh 2 Jul 2010 15:35:07 -0000 @@ -363,12 +363,14 @@ v:int:long_bit:::8 * sizeof (long):4*TAR # machine. v:int:long_long_bit:::8 * sizeof (LONGEST):2*gdbarch->long_bit::0 -# The ABI default bit-size and format for "float", "double", and "long -# double". These bit/format pairs should eventually be combined into -# a single object. For the moment, just initialize them as a pair. +# The ABI default bit-size and format for "half", "float", "double", and +# "long double". These bit/format pairs should eventually be combined +# into a single object. For the moment, just initialize them as a pair. # Each format describes both the big and little endian layouts (if # useful). +v:int:half_bit:::16:2*TARGET_CHAR_BIT::0 +v:const struct floatformat **:half_format:::::floatformats_ieee_half::pformat (gdbarch->half_format) v:int:float_bit:::8 * sizeof (float):4*TARGET_CHAR_BIT::0 v:const struct floatformat **:float_format:::::floatformats_ieee_single::pformat (gdbarch->float_format) v:int:double_bit:::8 * sizeof (double):8*TARGET_CHAR_BIT::0 Index: gdb/gdbarch.h =================================================================== RCS file: /cvs/src/src/gdb/gdbarch.h,v retrieving revision 1.416 diff -u -p -r1.416 gdbarch.h --- gdb/gdbarch.h 26 May 2010 18:19:26 -0000 1.416 +++ gdb/gdbarch.h 2 Jul 2010 15:35:06 -0000 @@ -118,12 +118,18 @@ extern void set_gdbarch_long_bit (struct extern int gdbarch_long_long_bit (struct gdbarch *gdbarch); extern void set_gdbarch_long_long_bit (struct gdbarch *gdbarch, int long_long_bit); -/* The ABI default bit-size and format for "float", "double", and "long - double". These bit/format pairs should eventually be combined into - a single object. For the moment, just initialize them as a pair. +/* The ABI default bit-size and format for "half", "float", "double", and + "long double". These bit/format pairs should eventually be combined + into a single object. For the moment, just initialize them as a pair. Each format describes both the big and little endian layouts (if useful). */ +extern int gdbarch_half_bit (struct gdbarch *gdbarch); +extern void set_gdbarch_half_bit (struct gdbarch *gdbarch, int half_bit); + +extern const struct floatformat ** gdbarch_half_format (struct gdbarch *gdbarch); +extern void set_gdbarch_half_format (struct gdbarch *gdbarch, const struct floatformat ** half_format); + extern int gdbarch_float_bit (struct gdbarch *gdbarch); extern void set_gdbarch_float_bit (struct gdbarch *gdbarch, int float_bit); Index: gdb/gdbarch.c =================================================================== RCS file: /cvs/src/src/gdb/gdbarch.c,v retrieving revision 1.468 diff -u -p -r1.468 gdbarch.c --- gdb/gdbarch.c 26 May 2010 18:19:27 -0000 1.468 +++ gdb/gdbarch.c 2 Jul 2010 15:35:06 -0000 @@ -145,6 +145,8 @@ struct gdbarch int int_bit; int long_bit; int long_long_bit; + int half_bit; + const struct floatformat ** half_format; int float_bit; const struct floatformat ** float_format; int double_bit; @@ -293,6 +295,8 @@ struct gdbarch startup_gdbarch = 8 * sizeof (int), /* int_bit */ 8 * sizeof (long), /* long_bit */ 8 * sizeof (LONGEST), /* long_long_bit */ + 16, /* half_bit */ + 0, /* half_format */ 8 * sizeof (float), /* float_bit */ 0, /* float_format */ 8 * sizeof (double), /* double_bit */ @@ -451,6 +455,7 @@ gdbarch_alloc (const struct gdbarch_info gdbarch->int_bit = 4*TARGET_CHAR_BIT; gdbarch->long_bit = 4*TARGET_CHAR_BIT; gdbarch->long_long_bit = 2*gdbarch->long_bit; + gdbarch->half_bit = 2*TARGET_CHAR_BIT; gdbarch->float_bit = 4*TARGET_CHAR_BIT; gdbarch->double_bit = 8*TARGET_CHAR_BIT; gdbarch->long_double_bit = 8*TARGET_CHAR_BIT; @@ -562,6 +567,9 @@ verify_gdbarch (struct gdbarch *gdbarch) /* Skip verify of int_bit, invalid_p == 0 */ /* Skip verify of long_bit, invalid_p == 0 */ /* Skip verify of long_long_bit, invalid_p == 0 */ + /* Skip verify of half_bit, invalid_p == 0 */ + if (gdbarch->half_format == 0) + gdbarch->half_format = floatformats_ieee_half; /* Skip verify of float_bit, invalid_p == 0 */ if (gdbarch->float_format == 0) gdbarch->float_format = floatformats_ieee_single; @@ -935,6 +943,12 @@ gdbarch_dump (struct gdbarch *gdbarch, s "gdbarch_dump: get_syscall_number = <%s>\n", host_address_to_string (gdbarch->get_syscall_number)); fprintf_unfiltered (file, + "gdbarch_dump: half_bit = %s\n", + plongest (gdbarch->half_bit)); + fprintf_unfiltered (file, + "gdbarch_dump: half_format = %s\n", + pformat (gdbarch->half_format)); + fprintf_unfiltered (file, "gdbarch_dump: has_dos_based_file_system = %s\n", plongest (gdbarch->has_dos_based_file_system)); fprintf_unfiltered (file, @@ -1375,6 +1389,39 @@ set_gdbarch_long_long_bit (struct gdbarc } int +gdbarch_half_bit (struct gdbarch *gdbarch) +{ + gdb_assert (gdbarch != NULL); + /* Skip verify of half_bit, invalid_p == 0 */ + if (gdbarch_debug >= 2) + fprintf_unfiltered (gdb_stdlog, "gdbarch_half_bit called\n"); + return gdbarch->half_bit; +} + +void +set_gdbarch_half_bit (struct gdbarch *gdbarch, + int half_bit) +{ + gdbarch->half_bit = half_bit; +} + +const struct floatformat ** +gdbarch_half_format (struct gdbarch *gdbarch) +{ + gdb_assert (gdbarch != NULL); + if (gdbarch_debug >= 2) + fprintf_unfiltered (gdb_stdlog, "gdbarch_half_format called\n"); + return gdbarch->half_format; +} + +void +set_gdbarch_half_format (struct gdbarch *gdbarch, + const struct floatformat ** half_format) +{ + gdbarch->half_format = half_format; +} + +int gdbarch_float_bit (struct gdbarch *gdbarch) { gdb_assert (gdbarch != NULL);