PA patch for conversion.c testsuite failure

PA patch for conversion.c testsuite failure

[Steve Ellcey]

Here is a patch to fix the PA failure on conversion.c.  It uses a signed
floating pt quad to fixed pt quad conversion to implement a long double to
unsigned long long conversion.  Tested on HP-UX 11.* PA32 and PA64.  I
did not test on HP-UX 10.20 but I did verify that the libc function I am
using (_U_Qfcnvfxt_quad_to_quad) exists on 10.20.

Steve Ellcey
sje@cup.hp.com


2002-07-16  Steve Ellcey  <sje@cup.hp.com>
	* gcc/config/pa/long_double.h (FIXUNS_TRUNCTFDI2_LIBCALL): New define.
	(fixunstfdi_libfunc): Change to use FIXUNS_TRUNCTFDI2_LIBCALL.
	* gcc/config/pa/quadlib.c (_U_Qfcnvfxt_quad_to_udbl): New function.


*** gcc.orig/gcc/config/pa/long_double.h	Tue Jul 16 10:19:38 2002
--- gcc/gcc/config/pa/long_double.h	Tue Jul 16 10:20:07 2002
*************** Boston, MA 02111-1307, USA.  */
*** 42,47 ****
--- 42,48 ----
  #define FIX_TRUNCTFSI2_LIBCALL \
    (TARGET_64BIT ? "__U_Qfcnvfxt_quad_to_sgl" : "_U_Qfcnvfxt_quad_to_sgl")
  #define FIXUNS_TRUNCTFSI2_LIBCALL "_U_Qfcnvfxt_quad_to_usgl"
+ #define FIXUNS_TRUNCTFDI2_LIBCALL "_U_Qfcnvfxt_quad_to_udbl"
  #define FIX_TRUNCTFDI2_LIBCALL "_U_Qfcnvfxt_quad_to_dbl"
  #define EQTF2_LIBCALL "_U_Qfeq"
  #define NETF2_LIBCALL "_U_Qfne"
*************** Boston, MA 02111-1307, USA.  */
*** 80,86 ****
      fixtfsi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIX_TRUNCTFSI2_LIBCALL);\
      fixtfdi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIX_TRUNCTFDI2_LIBCALL);\
      fixunstfsi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIXUNS_TRUNCTFSI2_LIBCALL);\
!     fixunstfdi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIX_TRUNCTFDI2_LIBCALL);  \
      eqtf2_libfunc = gen_rtx_SYMBOL_REF (Pmode, EQTF2_LIBCALL);		\
      netf2_libfunc = gen_rtx_SYMBOL_REF (Pmode, NETF2_LIBCALL);		\
      gttf2_libfunc = gen_rtx_SYMBOL_REF (Pmode, GTTF2_LIBCALL);		\
--- 81,87 ----
      fixtfsi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIX_TRUNCTFSI2_LIBCALL);\
      fixtfdi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIX_TRUNCTFDI2_LIBCALL);\
      fixunstfsi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIXUNS_TRUNCTFSI2_LIBCALL);\
!     fixunstfdi_libfunc = gen_rtx_SYMBOL_REF (Pmode, FIXUNS_TRUNCTFDI2_LIBCALL);\
      eqtf2_libfunc = gen_rtx_SYMBOL_REF (Pmode, EQTF2_LIBCALL);		\
      netf2_libfunc = gen_rtx_SYMBOL_REF (Pmode, NETF2_LIBCALL);		\
      gttf2_libfunc = gen_rtx_SYMBOL_REF (Pmode, GTTF2_LIBCALL);		\
*** gcc.orig/gcc/config/pa/quadlib.c	Tue Jul 16 10:19:43 2002
--- gcc/gcc/config/pa/quadlib.c	Tue Jul 16 10:22:58 2002
*************** long double _U_Qfneg (long double);
*** 42,47 ****
--- 42,48 ----
  int __U_Qfcnvfxt_quad_to_sgl (long double);
  #endif
  unsigned int _U_Qfcnvfxt_quad_to_usgl(long double);
+ unsigned long long _U_Qfcnvfxt_quad_to_udbl(long double);
  
  int
  _U_Qfeq (long double a, long double b)
*************** _U_Qfcnvfxt_quad_to_usgl(long double a)
*** 113,116 ****
--- 114,133 ----
  {
    extern long long _U_Qfcnvfxt_quad_to_dbl (long double a);
    return (unsigned int) _U_Qfcnvfxt_quad_to_dbl (a);
+ }
+ 
+ /* HP only has signed conversion in library, so need to synthesize an
+    unsigned version */
+ typedef union {
+     long long unsigned int u[2];
+     long double d[1];
+ } quad_type;
+ 
+ unsigned long long
+ _U_Qfcnvfxt_quad_to_udbl(long double a)
+ {
+   extern quad_type _U_Qfcnvfxt_quad_to_quad (long double a);
+   quad_type u;
+   u = _U_Qfcnvfxt_quad_to_quad(a);
+   return u.u[1];
  }

Re: PA patch for conversion.c testsuite failure

[Mark Mitchell]

--On Tuesday, July 16, 2002 02:53:25 PM -0400 John David Anglin 
<dave@hiauly1.hia.nrc.ca> wrote:

> Re: <http://gcc.gnu.org/ml/gcc-patches/2002-07/msg00795.html>.
>
> 2002-07-16  Steve Ellcey  <sje@cup.hp.com>
>         * gcc/config/pa/long_double.h (FIXUNS_TRUNCTFDI2_LIBCALL): New
> define. 	(fixunstfdi_libfunc): Change to use FIXUNS_TRUNCTFDI2_LIBCALL.
> 	* gcc/config/pa/quadlib.c (_U_Qfcnvfxt_quad_to_udbl): New function.
>
> OK for main.
>
> Mark, I believe this is safe although it is not a regression.  Is this
> OK for 3.1?

We are less than a week from 3.1.1 so now is not the time
for non-regression patches.

Yours,

--
Mark Mitchell                   mark@codesourcery.com
CodeSourcery, LLC               http://www.codesourcery.com

Re: PA patch for conversion.c testsuite failure

[John David Anglin]

Re: <http://gcc.gnu.org/ml/gcc-patches/2002-07/msg00795.html>.

2002-07-16  Steve Ellcey  <sje@cup.hp.com>
        * gcc/config/pa/long_double.h (FIXUNS_TRUNCTFDI2_LIBCALL): New define.
	(fixunstfdi_libfunc): Change to use FIXUNS_TRUNCTFDI2_LIBCALL.
	* gcc/config/pa/quadlib.c (_U_Qfcnvfxt_quad_to_udbl): New function.

OK for main.

Mark, I believe this is safe although it is not a regression.  Is this
OK for 3.1?

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)