public inbox for gcc-prs@sourceware.org help / color / mirror / Atom feed
From: Daniel Villeneuve <Daniel.Villeneuve@gerad.ca> To: nobody@gcc.gnu.org Cc: gcc-prs@gcc.gnu.org, Subject: Re: optimization/6065: gcc seems to reorder IEEE fpsetsticky and sqrt functions Date: Wed, 27 Mar 2002 09:56:00 -0000 [thread overview] Message-ID: <20020327175601.13098.qmail@sources.redhat.com> (raw) The following reply was made to PR optimization/6065; it has been noted by GNATS. From: Daniel Villeneuve <Daniel.Villeneuve@gerad.ca> To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, gcc-bugs@gcc.gnu.org, nobody@gcc.gnu.org, Daniel.Villeneuve@gerad.ca Cc: Subject: Re: optimization/6065: gcc seems to reorder IEEE fpsetsticky and sqrt functions Date: Wed, 27 Mar 2002 12:47:38 -0500 This is a multi-part message in MIME format. --------------3B0C834164807D03C8631D29 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Here is a trimmed down version of the bug report optimization/6065. The compilation command is shorter, the C code is simpler. >Submitter-Id: net >Originator: Daniel Villeneuve >Organization: >Confidential: no >Synopsis: gcc seems to reorder IEEE fpsetsticky and sqrt functions >Severity: serious >Priority: medium >Category: optimization >Class: wrong-code >Release: 3.0.3 >Environment: System: SunOS mckinley 5.7 Generic_106541-05 sun4u sparc SUNW,Ultra-5_10 Architecture: sun4 host: sparc-sun-solaris2.7 build: sparc-sun-solaris2.7 target: sparc-sun-solaris2.7 configured with: ./configure --disable-shared --prefix=/home/sparc-sun-solaris2/gcc-3.0.3 >Description: When compiling with -O2 or higher, gcc seems to move the call to sqrt() outside the bracketing function pair fpsetsticky()/fpgetsticky(), making the result of fpgetsticky incorrect. >How-To-Repeat: Compile the file t_sqrt.c into an executable named t_sqrt, then try: ./t_sqrt 2 whose output ought to be res = 1.41421, inexact flag = 1 but is res = 1.41421, inexact flag = 0 . When compiling with -DVOLATILE, the local double variable is declared as volatile and the code works correctly, even with -O2 or -O3. >Fix: Declare as volatile all floating-point variables used in computations that might affect IEEE status flags (when we want to access those flags, of course). I am not sure this is always correct: it happened to work on my restricted test case. I am aware that C99 provides the FENV_ACCESS pragma and the fe(clear|test)except functions, but they are still not widely available and I would like a portable solution if possible. Is there a better way (compiler options, pragmas, attributes) than using `volatile' to tell gcc not to `over'-optimize the code in such specific circumstances? -- Daniel Villeneuve GERAD/University of Montreal --------------3B0C834164807D03C8631D29 Content-Type: text/plain; charset=us-ascii; name="t_sqrt.c" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="t_sqrt.c" #include <ieeefp.h> #include <math.h> #include <stdlib.h> #include <stdio.h> typedef struct S { double d; int i; } S; static void Sqrt(S *ps) { #ifdef VOLATILE double volatile d; #else double d; #endif int i; d = ps->d, i = ps->i; fpsetsticky(0); d = sqrt(d); if( fpgetsticky() & FP_X_IMP ) i = 1; ps->d = d, ps->i = i; } int main(int argc, char **argv) { S ps[1]; ps->d = atof(argv[1]); ps->i = 0; Sqrt(ps); printf("res = %g, inexact flag = %d\n", ps->d, ps->i); return 0; } --------------3B0C834164807D03C8631D29 Content-Type: text/plain; charset=us-ascii; name="t_sqrt.i" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="t_sqrt.i" # 1 "t_sqrt.c" # 1 "/usr/include/ieeefp.h" 1 3 # 19 "/usr/include/ieeefp.h" 3 #pragma ident "@(#)ieeefp.h 1.15 97/11/22 SMI" # 64 "/usr/include/ieeefp.h" 3 typedef enum fpclass_t { FP_SNAN = 0, FP_QNAN = 1, FP_NINF = 2, FP_PINF = 3, FP_NDENORM = 4, FP_PDENORM = 5, FP_NZERO = 6, FP_PZERO = 7, FP_NNORM = 8, FP_PNORM = 9 } fpclass_t; extern fpclass_t fpclass(double); extern int finite(double); extern int unordered(double, double); # 113 "/usr/include/ieeefp.h" 3 typedef enum fp_rnd { FP_RN = 0, FP_RZ = 1, FP_RP = 2, FP_RM = 3 } fp_rnd; extern fp_rnd fpsetround(fp_rnd); extern fp_rnd fpgetround(void); # 191 "/usr/include/ieeefp.h" 3 extern int fpgetmask(void); extern int fpsetmask(int); extern int fpgetsticky(void); extern int fpsetsticky(int); # 209 "/usr/include/ieeefp.h" 3 extern int isnanf(float); extern int isnand(double); # 2 "t_sqrt.c" 2 # 1 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/math.h" 1 3 # 27 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/math.h" 3 #pragma ident "@(#)math.h 2.7 98/01/27" # 46 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/math.h" 3 typedef union _h_val { unsigned long _i[sizeof(double) / sizeof(unsigned long)]; double _d; } _h_val; extern const _h_val __huge_val; # 134 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/math.h" 3 extern double acos (double); extern double asin (double); extern double atan (double); extern double atan2 (double, double); extern double cos (double); extern double sin (double); extern double tan (double); extern double cosh (double); extern double sinh (double); extern double tanh (double); extern double exp (double); extern double frexp (double, int *); extern double ldexp (double, int); extern double log (double); extern double log10 (double); extern double modf (double, double *); extern double pow (double, double); extern double sqrt (double); extern double ceil (double); extern double fabs (double); extern double floor (double); extern double fmod (double, double); # 3 "t_sqrt.c" 2 # 1 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 1 3 # 25 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 #pragma ident "@(#)stdlib.h 1.44 98/01/22 SMI" # 1 "/usr/include/sys/feature_tests.h" 1 3 # 13 "/usr/include/sys/feature_tests.h" 3 #pragma ident "@(#)feature_tests.h 1.17 97/12/04 SMI" # 1 "/usr/include/sys/isa_defs.h" 1 3 # 9 "/usr/include/sys/isa_defs.h" 3 #pragma ident "@(#)isa_defs.h 1.15 97/11/22 SMI" # 16 "/usr/include/sys/feature_tests.h" 2 3 # 28 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 2 3 # 38 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 typedef struct { int quot; int rem; } div_t; typedef struct { long quot; long rem; } ldiv_t; # 66 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 typedef unsigned int size_t; # 77 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 typedef long uid_t; # 107 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 typedef long int wchar_t; # 148 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 extern unsigned char __ctype[]; extern double atof(const char *); extern int atoi(const char *); extern long int atol(const char *); extern double strtod(const char *, char **); extern long int strtol(const char *, char **, int); extern unsigned long int strtoul(const char *, char **, int); extern int rand(void); extern void srand(unsigned int); extern void *calloc(size_t, size_t); extern void free(void *); extern void *malloc(size_t); extern void *realloc(void *, size_t); extern void abort(void); extern int atexit(void (*)(void)); extern void exit(int); extern void _exithandle(void); extern char *getenv(const char *); extern int system(const char *); extern void *bsearch(const void *, const void *, size_t, size_t, int (*)(const void *, const void *)); extern void qsort(void *, size_t, size_t, int (*)(const void *, const void *)); extern int abs(int); extern div_t div(int, int); extern long int labs(long); extern ldiv_t ldiv(long, long); extern int mbtowc(wchar_t *, const char *, size_t); extern int mblen(const char *, size_t); extern int wctomb(char *, wchar_t); extern size_t mbstowcs(wchar_t *, const char *, size_t); extern size_t wcstombs(char *, const wchar_t *, size_t); # 219 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdlib.h" 3 extern int mkstemp(char *); # 4 "t_sqrt.c" 2 # 1 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 1 3 # 14 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 # 1 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdarg.h" 1 3 # 43 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdarg.h" 3 typedef __builtin_va_list __gnuc_va_list; # 15 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 2 3 # 34 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 #pragma ident "@(#)stdio.h 1.69 98/07/13 SMI" # 1 "/usr/include/sys/feature_tests.h" 1 3 # 37 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 2 3 # 1 "/usr/include/sys/va_list.h" 1 3 # 9 "/usr/include/sys/va_list.h" 3 #pragma ident "@(#)va_list.h 1.11 97/11/22 SMI" # 26 "/usr/include/sys/va_list.h" 3 typedef void *__va_list; # 38 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 2 3 # 1 "/usr/include/stdio_tag.h" 1 3 # 9 "/usr/include/stdio_tag.h" 3 #pragma ident "@(#)stdio_tag.h 1.3 98/04/20 SMI" # 21 "/usr/include/stdio_tag.h" 3 typedef struct __FILE __FILE; # 39 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 2 3 # 1 "/usr/include/stdio_impl.h" 1 3 # 9 "/usr/include/stdio_impl.h" 3 #pragma ident "@(#)stdio_impl.h 1.7 98/04/17 SMI" # 1 "/usr/include/sys/isa_defs.h" 1 3 # 12 "/usr/include/stdio_impl.h" 2 3 # 22 "/usr/include/stdio_impl.h" 3 typedef int ssize_t; # 38 "/usr/include/stdio_impl.h" 3 struct __FILE { ssize_t _cnt; unsigned char *_ptr; unsigned char *_base; unsigned char _flag; unsigned char _file; unsigned __orientation:2; unsigned __filler:6; }; # 40 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 2 3 # 55 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 typedef __FILE FILE; # 103 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 typedef long fpos_t; # 203 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 extern FILE __iob[20]; # 215 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 extern unsigned char _sibuf[], _sobuf[]; # 244 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 extern unsigned char *_bufendtab[]; extern FILE *_lastbuf; # 275 "/home/sparc-sun-solaris2/gcc-3.0.3/lib/gcc-lib/sparc-sun-solaris2.7/3.0.3/include/stdio.h" 3 extern int remove(const char *); extern int rename(const char *, const char *); extern FILE *tmpfile(void); extern char *tmpnam(char *); extern int fclose(FILE *); extern int fflush(FILE *); extern FILE *fopen(const char *, const char *); extern FILE *freopen(const char *, const char *, FILE *); extern void setbuf(FILE *, char *); extern int setvbuf(FILE *, char *, int, size_t); extern int fprintf(FILE *, const char *, ...); extern int fscanf(FILE *, const char *, ...); extern int printf(const char *, ...); extern int scanf(const char *, ...); extern int sprintf(char *, const char *, ...); extern int sscanf(const char *, const char *, ...); extern int vfprintf(FILE *, const char *, __gnuc_va_list); extern int vprintf(const char *, __gnuc_va_list); extern int vsprintf(char *, const char *, __gnuc_va_list); extern int fgetc(FILE *); extern char *fgets(char *, int, FILE *); extern int fputc(int, FILE *); extern int fputs(const char *, FILE *); extern int getc(FILE *); extern int getchar(void); extern char *gets(char *); extern int putc(int, FILE *); extern int putchar(int); extern int puts(const char *); extern int ungetc(int, FILE *); extern size_t fread(void *, size_t, size_t, FILE *); extern size_t fwrite(const void *, size_t, size_t, FILE *); extern int fgetpos(FILE *, fpos_t *); extern int fseek(FILE *, long, int); extern int fsetpos(FILE *, const fpos_t *); extern long ftell(FILE *); extern void rewind(FILE *); extern void clearerr(FILE *); extern int feof(FILE *); extern int ferror(FILE *); extern void perror(const char *); extern int __filbuf(FILE *); extern int __flsbuf(int, FILE *); # 5 "t_sqrt.c" 2 typedef struct S { double d; int i; } S; static void Sqrt(S *ps) { double d; int i; d = ps->d, i = ps->i; fpsetsticky(0); d = sqrt(d); if( fpgetsticky() & 0x01 ) i = 1; ps->d = d, ps->i = i; } int main(int argc, char **argv) { S ps[1]; ps->d = atof(argv[1]); ps->i = 0; Sqrt(ps); printf("res = %g, inexact flag = %d\n", ps->d, ps->i); return 0; } --------------3B0C834164807D03C8631D29 Content-Type: text/plain; charset=us-ascii; name="Makefile" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="Makefile" t_sqrt: t_sqrt.c gcc -v -save-temps -ansi -O2 -o t_sqrt t_sqrt.c -lm # gcc -DVOLATILE -v -save-temps -ansi -O2 -o t_sqrt t_sqrt.c -lm clean: rm -f t_sqrt t_sqrt.[ois] --------------3B0C834164807D03C8631D29 Content-Type: text/plain; charset=us-ascii; name="t_sqrt.s" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="t_sqrt.s" .file "t_sqrt.i" .section ".text" .align 4 .type Sqrt,#function .proc 020 Sqrt: !#PROLOGUE# 0 save %sp, -128, %sp !#PROLOGUE# 1 ldd [%i0], %o0 std %o0, [%fp-24] ldd [%fp-24], %f2 mov 0, %o0 fsqrtd %f2, %f2 ld [%i0+8], %l0 call fpsetsticky, 0 std %f2, [%fp-32] ldd [%fp-32], %f4 fmovs %f4, %f2 fmovs %f5, %f3 fcmpd %f4, %f2 nop fbe .LL2 ldd [%fp-24], %o0 call sqrt, 0 nop std %f0, [%fp-32] .LL2: call fpgetsticky, 0 nop andcc %o0, 1, %g0 bne,a .LL3 mov 1, %l0 .LL3: ldd [%fp-32], %o0 st %l0, [%i0+8] std %o0, [%i0] ret restore .LLfe1: .size Sqrt,.LLfe1-Sqrt .section ".rodata" .align 8 .LLC0: .asciz "res = %g, inexact flag = %d\n" .section ".text" .align 4 .global main .type main,#function .proc 04 main: !#PROLOGUE# 0 save %sp, -128, %sp !#PROLOGUE# 1 ld [%i1+4], %o0 call atof, 0 mov 0, %i0 add %fp, -32, %o0 std %f0, [%fp-32] call Sqrt, 0 st %g0, [%fp-24] sethi %hi(.LLC0), %o0 ld [%fp-32], %o1 or %o0, %lo(.LLC0), %o0 ld [%fp-28], %o2 call printf, 0 ld [%fp-24], %o3 ret restore .LLfe2: .size main,.LLfe2-main .ident "GCC: (GNU) 3.0.3" --------------3B0C834164807D03C8631D29--
next reply other threads:[~2002-03-27 17:56 UTC|newest] Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top 2002-03-27 9:56 Daniel Villeneuve [this message] -- strict thread matches above, loose matches on Subject: below -- 2002-03-26 9:46 Daniel Villeneuve
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20020327175601.13098.qmail@sources.redhat.com \ --to=daniel.villeneuve@gerad.ca \ --cc=gcc-prs@gcc.gnu.org \ --cc=nobody@gcc.gnu.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).