RFA: integer subregs of float vectors / Re: c/7225: ice on generic vectors (floats)

RFA: integer subregs of float vectors / Re: c/7225: ice on generic vectors (floats)

[Joern Rennecke]

The following reply was made to PR c/7225; it has been noted by GNATS.

From: Joern Rennecke <joern.rennecke@superh.com>
To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, pinskia@physics.uc.edu,
   gcc-bugs@gcc.gnu.org, nobody@gcc.gnu.org, gcc-patches@gcc.gnu.org
Cc:  
Subject: RFA: integer subregs of float vectors / Re: c/7225: ice on generic 
 vectors (floats)
Date: Thu, 11 Jul 2002 15:59:31 +0100

 This is a multi-part message in MIME format.
 --------------6B10FD973BDC80FBAEB516C0
 Content-Type: text/plain; charset=us-ascii
 Content-Transfer-Encoding: 7bit
 
 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7225
 
 Regression tested on i686-linux-gnu.  The new testcases also succeed.
 
 -- 
 --------------------------
 SuperH
 2430 Aztec West / Almondsbury / BRISTOL / BS32 4AQ
 T:+44 1454 462330
 --------------6B10FD973BDC80FBAEB516C0
 Content-Type: text/plain; charset=us-ascii;
  name="simd-fix-10"
 Content-Transfer-Encoding: 7bit
 Content-Disposition: inline;
  filename="simd-fix-10"
 
 The second hunk is not covered by the original testcase, therefore
 I made a variant that uses double instead of float.
 We still don't conver integer CONST_DOUBLES, but I think we should
 have a testcase first to see that we get this right.
 
 gcc:
 Thu Jul 11 15:39:21 2002  J"orn Rennecke <joern.rennecke@superh.com>
 
 	* simplify-rtx.c (simplify_subreg): Handle floating point
 	CONST_DOUBLEs.  When an integer subreg of a smaller mode than
 	the element mode is requested, compute a subreg with an
 	integer mode of the same size as the element mode first.
 
 testsuite:
 Thu Jul 11 15:39:21 2002  J"orn Rennecke <joern.rennecke@superh.com>
                           Andrew Pinski  <pinskia@physics.uc.edu>
 
 	gcc.c-torture/compile/simd-2.c: New testcase.
 	gcc.c-torture/compile/simd-3.c: Likewise.
 
 Index: simplify-rtx.c
 ===================================================================
 RCS file: /cvs/gcc/gcc/gcc/simplify-rtx.c,v
 retrieving revision 1.109
 diff -p -r1.109 simplify-rtx.c
 *** simplify-rtx.c	4 Jul 2002 06:38:54 -0000	1.109
 --- simplify-rtx.c	11 Jul 2002 14:35:17 -0000
 *************** simplify_subreg (outermode, op, innermod
 *** 2307,2312 ****
 --- 2307,2320 ----
   	  for (; n_elts--; i += step)
   	    {
   	      elt = CONST_VECTOR_ELT (op, i);
 + 	      if (GET_CODE (elt) == CONST_DOUBLE
 + 		  && GET_MODE_CLASS (GET_MODE (elt)) == MODE_FLOAT)
 + 		{
 + 		  elt = gen_lowpart_common (int_mode_for_mode (GET_MODE (elt)),
 + 					    elt);
 + 		  if (! elt)
 + 		    return NULL_RTX;
 + 		}
   	      if (GET_CODE (elt) != CONST_INT)
   		return NULL_RTX;
   	      high = high << shift | sum >> (HOST_BITS_PER_WIDE_INT - shift);
 *************** simplify_subreg (outermode, op, innermod
 *** 2319,2324 ****
 --- 2327,2344 ----
   	  else
   	    return NULL_RTX;
   	}
 +       else if (GET_MODE_CLASS (outermode) == MODE_INT
 + 	       && (elt_size % GET_MODE_SIZE (outermode) == 0))
 + 	{
 + 	  enum machine_mode new_mode
 + 	    = int_mode_for_mode (GET_MODE_INNER (innermode));
 + 	  int subbyte = byte % elt_size;
 + 
 + 	  op = simplify_subreg (new_mode, op, innermode, byte - subbyte);
 + 	    if (! op)
 + 	      return NULL_RTX;
 + 	  return simplify_subreg (outermode, op, new_mode, subbyte);
 + 	}
         else if (GET_MODE_CLASS (outermode) != MODE_VECTOR_INT
   	       && GET_MODE_CLASS (outermode) != MODE_VECTOR_FLOAT)
           /* This shouldn't happen, but let's not do anything stupid.  */
 *** /dev/null	Thu Aug 30 21:30:55 2001
 --- testsuite/gcc.c-torture/compile/simd-2.c	Thu Jul 11 15:38:21 2002
 ***************
 *** 0 ****
 --- 1,17 ----
 + typedef float floatvect2 __attribute__((mode(V2SF)));
 + 
 + typedef union
 + {
 +     floatvect2 vector;
 +     float f[2];
 + }resfloatvect2;
 + 
 + void tempf(float *x, float *y)
 + {
 +         floatvect2 temp={x[0],x[1]};
 +         floatvect2 temp1={y[0],y[1]};
 +         resfloatvect2 temp2;
 +         temp2.vector=temp+temp1;
 +         x[0]=temp2.f[0];
 +         x[1]=temp2.f[1];
 + }
 *** /dev/null	Thu Aug 30 21:30:55 2001
 --- testsuite/gcc.c-torture/compile/simd-3.c	Thu Jul 11 15:38:29 2002
 ***************
 *** 0 ****
 --- 1,17 ----
 + typedef float floatvect2 __attribute__((mode(V2DF)));
 + 
 + typedef union
 + {
 +     floatvect2 vector;
 +     double f[2];
 + }resfloatvect2;
 + 
 + void tempf(double *x, double *y)
 + {
 +         floatvect2 temp={x[0],x[1]};
 +         floatvect2 temp1={y[0],y[1]};
 +         resfloatvect2 temp2;
 +         temp2.vector=temp+temp1;
 +         x[0]=temp2.f[0];
 +         x[1]=temp2.f[1];
 + }
 
 --------------6B10FD973BDC80FBAEB516C0--
 

Re: RFA: integer subregs of float vectors / Re: c/7225: ice on generic vectors (floats)

[Richard Henderson]

The following reply was made to PR c/7225; it has been noted by GNATS.

From: Richard Henderson <rth@redhat.com>
To: Joern Rennecke <joern.rennecke@superh.com>
Cc: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, pinskia@physics.uc.edu,
   gcc-bugs@gcc.gnu.org, nobody@gcc.gnu.org, gcc-patches@gcc.gnu.org
Subject: Re: RFA: integer subregs of float vectors / Re: c/7225: ice on generic vectors (floats)
Date: Thu, 11 Jul 2002 13:59:32 -0700

 On Thu, Jul 11, 2002 at 03:59:31PM +0100, Joern Rennecke wrote:
 > 	* simplify-rtx.c (simplify_subreg): Handle floating point
 > 	CONST_DOUBLEs.  When an integer subreg of a smaller mode than
 > 	the element mode is requested, compute a subreg with an
 > 	integer mode of the same size as the element mode first.
 > 	gcc.c-torture/compile/simd-2.c: New testcase.
 > 	gcc.c-torture/compile/simd-3.c: Likewise.
 
 Ok.
 
 
 r~