[fixed-point] Support constant folding

["Fu, Chao-Ying" <fu at mips dot com>]

Hello,

  Here is the patch to support constant folding.
Note that two constants must have the same fixed-point types
for + - * /, because type-conversion hasn't been updated for fixed-point.
Thanks!

Ex:
_Fract b1 = 0.1r + 0.2r;
_Fract b2 = 0.5r - 0.3r;
_Fract b3 = 0.5r * 0.5r;
_Fract b4 = 0.2r / 0.5r;
_Fract b5 = 0.01r << 4;
_Fract b6 = 0.1r >> 3;
_Fract b7 = +0.3r;
_Fract b8 = -0.3r;
int b9 = !0.0r;
int c1 = 0.1r != 0.1r;
int c2 = 0.1r == 0.2r;
int c3 = 0.1r > 0.2r;
int c4 = 0.1r >= 0.2r;
int c5 = 0.2r < 0.1r;
int c6 = 0.2r <= 0.1r;
int c7 = 0.1r != 0.2r;
int c8 = 0.1r == 0.1r;
int c9 = 0.1r > 0.0r;
int c10 = 0.1r >= 0.0r;
int c11 = 0.2r < 0.3r;
int c12 = 0.2r <= 0.3r;

Regards,
Chao-ying

2006-11-29  Chao-ying Fu  <fu@mips.com>

	* tree.h (fixed_zerop): Declare.
	* tree.c (fixed_zerop): New function.
	* final.c (output_addr_const): Output the lower data for CONST_FIXED.
	* fold-const.c (fold_negate_expr): Support FIXED_CST.
	(const_binop): Likewise.
	(fold_binary): Likewise.
	(fold_negate_const): Likewise.
	(fold_relational_const): Likewise.
	* c-decl.c (declspecs_add_type): Give errors when complex is used with
	_Fract or _Accum.
	* c-typeck.c (build_unary_op): Support FIXED_POINT_TYPE for
	CONVERT_EXPR, NEGATE_EXPR, and TRUTH_NOT_EXPR.
	(build_binary_op): Check fixed_zerop for *_DIV_EXPR.
	Support FIXED_POINT_TYPE in *_DIV_EXPR, RSHIFT_EXPR, LSHIFT_EXPR,
	EQ_EXPR, NE_EXPR, LE_EXPR, GE_EXPR, LT_EXPR, GT_EXPR.
	* emit-rtl.c (fconst0): New array for fixed-point zero.
	(init_emit_once): Init fconst0.
	* fixed_value.h (fconst0): Declare.
	(fixed_arithmetic, fixed_compare, fixed_arithmetic2): Declare.
	(FIXED_VALUE_NEGATE): Define.
	* fixed_value.c (do_fixed_add, do_fixed_multiply, do_fixed_divide,
	do_fixed_shift): Declare.
	(fixed_from_string): Sign or zero extend the value.
	(do_fixed_add, do_fixed_multiply, do_fixed_divide, do_fixed_shift,
	fixed_arithmetic, fixed_arithmetic2, fixed_compare): New functions.
	* c-common.c (shorten_compare): Test fixed_zerop.
	(c_common_truthvalue_conversion): Support FIXED_CST.

Attachment: gcc.diff
Description: gcc.diff