This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
[Committed] Bug combining type conversions in fold-const.c
- From: Roger Sayle <roger at eyesopen dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: Mon, 9 Feb 2004 06:27:50 -0700 (MST)
- Subject: [Committed] Bug combining type conversions in fold-const.c
The following patch fixes a latent bug in GCC's constant folding. fold
contains code to combine two consecutive type conversions into a single
conversion when possible. Examples of which include the following
problematic cases:
(FIX_TRUNC_EXPR:int (NOP_EXPR:double (...:float))
(FIX_CEIL_EXPR:int (NOP_EXPR:double (...:float))
(FIX_FLOOR_EXPR:int (NOP_EXPR:double (...:float))
where the conversion to an integer can be performed as a float->int
conversion rather than as a float->double extension followed by a
double->int conversion.
The problem is that instead of using the original FP->int conversion
operator, fold previously called the front-end's "convert" function
which would apply the default integer to fp conversion for that
language, FIX_TRUNC_EXPR for C/C++ or perform rounding in FORTRAN.
This issue was exposed by my recent patch to provide a language
independent "fold_convert" for use by the middle-end (i.e. without
generating warnings, etc...). As theoretically, fold_convert never
needs to suggest a default real->integer conversion it just calls
abort. To my surprise this abort triggered whilst compiling 175.vpr
from SPECint2000, revealing the latent bug fixed below.
The following patch has been tested by a full bootstrap on
i686-pc-linux-gnu, all languages except treelang, and regression
tested with a top-level "make -k check" with no new failures.
Committed to mainline.
2004-02-09 Roger Sayle <roger@eyesopen.com>
* fold-const.c (fold) <NOP_EXPR>: Use the original type conversion
tree code rather than call fold_convert, which doesn't specify a
default floating point to integer conversion.
* gcc.c-torture/compile/20040209-1.c: New test case.
Index: fold-const.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/fold-const.c,v
retrieving revision 1.328
diff -c -3 -p -r1.328 fold-const.c
*** fold-const.c 7 Feb 2004 18:57:29 -0000 1.328
--- fold-const.c 8 Feb 2004 23:54:07 -0000
*************** fold (tree expr)
*** 5641,5648 ****
if (TYPE_MAIN_VARIANT (inside_type) == TYPE_MAIN_VARIANT (final_type)
&& ((inter_int && final_int) || (inter_float && final_float))
&& inter_prec >= final_prec)
! return fold_convert (final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0));
/* Likewise, if the intermediate and final types are either both
float or both integer, we don't need the middle conversion if
--- 5641,5648 ----
if (TYPE_MAIN_VARIANT (inside_type) == TYPE_MAIN_VARIANT (final_type)
&& ((inter_int && final_int) || (inter_float && final_float))
&& inter_prec >= final_prec)
! return fold (build1 (code, final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0)));
/* Likewise, if the intermediate and final types are either both
float or both integer, we don't need the middle conversion if
*************** fold (tree expr)
*** 5657,5672 ****
&& ! (final_prec != GET_MODE_BITSIZE (TYPE_MODE (final_type))
&& TYPE_MODE (final_type) == TYPE_MODE (inter_type))
&& ! final_ptr)
! return fold_convert (final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0));
/* If we have a sign-extension of a zero-extended value, we can
replace that by a single zero-extension. */
if (inside_int && inter_int && final_int
&& inside_prec < inter_prec && inter_prec < final_prec
&& inside_unsignedp && !inter_unsignedp)
! return fold_convert (final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0));
/* Two conversions in a row are not needed unless:
- some conversion is floating-point (overstrict for now), or
--- 5657,5672 ----
&& ! (final_prec != GET_MODE_BITSIZE (TYPE_MODE (final_type))
&& TYPE_MODE (final_type) == TYPE_MODE (inter_type))
&& ! final_ptr)
! return fold (build1 (code, final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0)));
/* If we have a sign-extension of a zero-extended value, we can
replace that by a single zero-extension. */
if (inside_int && inter_int && final_int
&& inside_prec < inter_prec && inter_prec < final_prec
&& inside_unsignedp && !inter_unsignedp)
! return fold (build1 (code, final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0)));
/* Two conversions in a row are not needed unless:
- some conversion is floating-point (overstrict for now), or
*************** fold (tree expr)
*** 5690,5697 ****
&& ! (final_prec != GET_MODE_BITSIZE (TYPE_MODE (final_type))
&& TYPE_MODE (final_type) == TYPE_MODE (inter_type))
&& ! final_ptr)
! return fold_convert (final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0));
}
if (TREE_CODE (TREE_OPERAND (t, 0)) == MODIFY_EXPR
--- 5690,5697 ----
&& ! (final_prec != GET_MODE_BITSIZE (TYPE_MODE (final_type))
&& TYPE_MODE (final_type) == TYPE_MODE (inter_type))
&& ! final_ptr)
! return fold (build1 (code, final_type,
! TREE_OPERAND (TREE_OPERAND (t, 0), 0)));
}
if (TREE_CODE (TREE_OPERAND (t, 0)) == MODIFY_EXPR
/* The following code used to ICE in fold_convert. */
float ceilf(float);
int foo(float x)
{
return (double)ceilf(x);
}
Roger
--
Roger Sayle, E-mail: roger@eyesopen.com
OpenEye Scientific Software, WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road, Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507. Fax: (+1) 505-473-0833