PR c/8639
* fold-const.c (extract_muldiv): Don't propagate division unless
both arguments are multiples of C.
* gcc.c-torture/execute/20021119-1.c: New.
From-SVN: r59466
2002-11-05 Andrew Haley <aph@redhat.com>
* fold-const.c (fold): Don't transform (a0 op compound(a1,a2))
to (compound(a1,a0 op a2)) if a0 or a1 have side effects.
From-SVN: r58822
* fold-const.c (fold_binary_op_with_conditional_arg): Improve
handling of cases where one or both branches of the conditional
have void type, i.e. throw an exception or don't return.
(fold): Only apply (and undo) type conversion to the non-void
branches of a COND_EXPR.
* f/com.c (ffecom_subscript_check_): Cast the failure branch
of the bounds check COND_EXPR to void, to indicate noreturn.
(ffe_truthvalue_conversion): Only apply truth value conversion
to the non-void branches of a COND_EXPR.
From-SVN: r58661
PR c++/7209
* fold_const.c (fold_binary_op_with_conditional_arg): Always
build compound_expr if we used save_expr.
testsuite:
* g++.dg/expr/cond1.C: New test.
From-SVN: r58405
* fold-const.c (make_range): Only narrow to signed range if
the signed range is smaller than the unsigned range.
* gcc.c-torture/execute/20020911-1.c: New.
From-SVN: r57050
* fold-const.c (associate_trees): Only optimize NEGATE_EXPR in one
of the operands into MINUS_EXPR if code is PLUS_EXPR.
* gcc.c-torture/execute/20020805-1.c: New test.
From-SVN: r56058
* Makefile.in: rtlanal.o now depends upon real.h.
* flags.h [flag_signaling_nans]: New flag.
[HONOR_SNANS]: New macro.
* toplev.c [flag_signaling_nans]: Initialize to false.
(f_options): Add processing for "-fsignaling-nans".
(set_fast_math_flags): Clear flag_signaling_nans with -ffast-math.
(process_options): flag_signaling_nans implies flag_trapping_math.
* c-common.c (cb_register_builtins): Define __SUPPORT_SNAN__
when -fsignaling-nans. First step to implementing WG14's N965.
* fold-const.c (fold) [MULT_EXPR]: Conditionalize transforming
1.0 * x into x, and -1.0 * x into -x on !HONOR_SNANS.
[RDIV_EXPR]: Conditionalize x/1.0 into x on !HONOR_SNANS.
* simplify-rtx.c (simplify_relational_operation): Conditionalize
transforming abs(x) < 0.0 into false on !HONOR_SNANS.
* rtlanal.c: #include real.c for TARGET_FLOAT_FORMAT definitions
required by HONOR_SNANS. (may_trap_p): Floating point DIV, MOD,
UDIV, UMOD, GE, GT, LE, LT and COMPARE may always trap with
-fsignaling_nans. EQ and NE only trap for flag_signaling_nans
not flag_trapping_math (i.e. HONOR_SNANS but not HONOR_NANS).
* doc/invoke.texi: Document new -fsignaling-nans compiler option.
From-SVN: r55804
PR opt/4046
* fold-const.c (fold) [COND_EXPR]: Simplify A ? 0 : 1 to !A,
A ? B : 0 to A && B and A ? B : 1 into !A || B if both A and
B are truth values.
From-SVN: r55153
* fold-const.c (comparison_to_compcode): New function to convert
an comparison TREE CODE into a bit-based representation.
(compcode_to_comparison): New function to convert from this bit
based representation back to a comparison TREE CODE.
(fold_truthop): Simplify (x<y) && (x==y) and related composite
comparisons.
* gcc.c-tortuture/execute/compare-1.c: New test case.
* gcc.c-tortuture/execute/compare-2.c: New test case.
* gcc.c-tortuture/execute/compare-3.c: New test case.
From-SVN: r54647
* convert.c (convert_to_integer) [LSHIFT_EXPR]: Don't pass
the truncation down when the target type is signed.
[trunc1]: Use unsigned arithmetic for LSHIFT_EXPR.
* fold-const.c (extract_muldiv) [NOP_EXPR]: Don't pass through
the conversion if the target type is a smaller type.
From-SVN: r54627
* fold-const.c (fold) [EQ_EXPR]: Place both integer and real
constants last in comparisons. Optimize (x+1.0)>0.0 into the
equivalent x > -1.0 when -ffast-math.
* gcc.dg/20020607-2.c: New test case.
From-SVN: r54356
* tree.c (real_minus_onep): New function to test for -1.0.
* fold-const.c (fold) [MULT_EXPR]: Optimize -1.0*x into -x.
* gcc.dg/fnegate-1.c: New test case.
From-SVN: r54149
* tree.h: Forward-declare struct realvaluetype.
(struct tree_real_cst): Point to the REAL_VALUE_TYPE, do not
contain it.
(TREE_REAL_CST_PTR): New accessor.
(TREE_REAL_CST): Update.
* real.h: Include machmode.h.
(realvaluetype): Make it struct realvaluetype, not a typedef.
(build_real): Prototype here.
* tree.c: Include real.h.
(build_real): Allocate the REAL_VALUE_TYPE as a separate
object in GC memory, set TREE_REAL_CST_PTR to point to it.
(build_real_from_int_cst): Use build_real.
* ggc-common.c (ggc_mark_trees): Mark TREE_REAL_CST_PTR of a
REAL_CST.
* builtins.c, c-common.c, c-lex.c, dwarf2out.c, expr.c,
fold-const.c, print-tree.c, real.c, cp/mangle.c, cp/tree.c,
f/bld.c, f/com.c, f/expr.c, f/target.c, java/decl.c,
java/jcf-parse.c, java/parse.y, java/typeck.c: Include real.h.
* Makefile.in, cp/Make-lang.in, f/Make-lang.in,
java/Make-lang.in: Update dependency lists.
From-SVN: r53959
2002-05-15 Jakub Jelinek <jakub@redhat.com>
* fold-const.c (fold): Fix a typo.
2002-05-15 Eric Botcazou <ebotcazou@multimania.com>
* fold-const.c (fold) [LT_EXPR]: Move the transformation of a
comparison against the highest or lowest integer value before
the 'X >= CST to X > (CST - 1)' and 'X < CST to X <= (CST - 1)'
transformation and that of an unsigned comparison against 0
right after.
From-SVN: r53493
PR opt/3995
* fold-const.c (sign_bit_p): New function.
(fold) [EQ_EXPR]: Use this to convert (A & C) == 0 into A >= 0 and
(A & C) != 0 into A < 0, when constant C is the sign bit of A's type.
Reapply fold when converting (A & C) == C into (A & C) != 0.
(fold_binary_op_with_conditional_arg): Fix typo in comment.
testsuite/
* gcc.c-torture/execute/20020506-1.c: New test case.
From-SVN: r53241
* fold-const.c (lshift-double): Cast the high word to an unsigned
HOST_WIDE_INT when extracting sign bit to avoid compiler warning.
(div_and_round_double): Cast carry to a signed HOST_WIDE_INT to
avoid compiler warning. (fold): Remove redundant code from
BIT_AND_EXPR as integer operands are canonicalized to be arg1.
From-SVN: r53228
2002-04-18 David S. Miller <davem@redhat.com>
* config/sparc/sparc.h (BRANCH_COST): Define.
* fold-const.c (BRANCH_COST): Don't provide default here, expr.h
does it.
From-SVN: r52500
* fold-const.c (fold) [NOP_EXPR]: Convert (T)(x&c) into ((T)x&(T)c)
for integer constant c (if x has unsigned type or sign bit is not
set in c). This folds the zero/sign extension into the bit-wise and
operation.
* gcc.c-torture/compile/20020415-1.c: New.
Co-Authored-By: Jakub Jelinek <jakub@redhat.com>
From-SVN: r52465
* toplev.c: Don't include setjmp.h. Kill float_handler_set,
float_handled, float_handler, float_signal, set_float_handler,
and do_float_handler. Set handler for SIGFPE to crash_signal.
* toplev.h: Don't prototype do_float_handler.
* c-lex.c: Fold parse_float into lex_number. Make warning
about portability of hex float constants more informative, and
don't issue it on top of a syntax error.
* fold-const.c: Fold const_binop_1 and fold_convert_1 into
their callers.
* real.h: Define REAL_VALUE_ABS here...
* simplify-rtx.c: ... not here. Fold check_fold_consts,
simplify_unary_real, simplify_binary_real, and
simplify_binary_is2orm1 into their callers.
* tree.c: Fold build_real_from_int_cst_1 into caller.
* doc/tm.texi: Document REAL_VALUE_ABS and REAL_VALUE_NEGATIVE.
* tsystem.h: Include float.h here...
* libgcc2.c: ... not here.
java:
* lex.c: Change java_perform_atof to take normal parameters
instead of a pointer to a parameter block. Call it directly
from java_lex.
testsuite:
* gcc.dg/c90-hexfloat-1.c: Adjust error regexps.
From-SVN: r51336
* real.h: Don't define REAL_INFINITY or REAL_IS_NOT_DOUBLE.
Always make REAL_VALUE_TYPE a struct containing an array of
HOST_WIDE_INT, not a double. Tidy up the code deciding how
big it is. Don't declare or use union real_extract.
* emit-rtl.c (init_emit_once), varasm.c (immed_real_const_1,
decode_rtx_const, output_constant_pool), config/a29k/a29k.c
(print_operand), config/arm/arm.c (output_move_double),
config/arm/arm.md (consttable_4, consttable_8),
config/romp/romp.c (output_fpops), config/s390/s390.h
(ASM_OUTPUT_SPECIAL_POOL_ENTRY), config/xtensa/xtensa.c
(xtensa_output_literal): Don't use union real_extract.
* config/dsp16xx/dsp16xx.c (print_operand), config/i860/i860.c
(sfmode_constant_to_ulong), config/ns32k/merlin.h
(PRINT_OPERAND), config/ns32k/ns32k.c (print_operand),
config/pdp11/pdp11.h (PRINT_OPERAND), config/we32k/we32k.h
(PRINT_OPERAND): Don't use local version of union
real_extract.
* config/convex/convex.c (check_float_value), config/vax/vax.c
(vax_float_literal), config/m88k/m88k.md (divdf3),
config/dsp16xx/dsp16xx.md (fixuns_trunchfhi2),
config/pdp11/pdp11.c (output_move_quad): Don't do host
arithmetic on target floating point quantities.
* config/a29k/a29k.md, config/dsp16xx/dsp16xx.c
(output_dsp16xx_float_const): Don't test HOST_FLOAT_FORMAT.
* fold-const.c (fold), simplify-rtx.c (simplify_binary_real):
Use MODE_HAS_INFINITIES rather than #ifdef REAL_INFINITY.
* real.c (earith): Test INFINITY rather than REAL_INFINITY;
NANS implies INFINITY, so can drop #ifdef NANS inside #ifndef
INFINITY.
* print-rtl.c (print_rtx): Disable code which needs
floating-point emulator.
* libgcc2.c: Include float.h and use DBL_MANT_DIG,
FLT_MANT_DIG, to define DF_SIZE and SF_SIZE, rather than
depending on HOST_FLOAT_FORMAT to be defined properly.
* ch/grant.c, cp/error.c: Always use REAL_VALUE_TO_DECIMAL;
don't test REAL_IS_NOT_DOUBLE.
* config/1750a/1750a.c (get_double, float_label): Delete.
(print_operand): Delete huge commented-out chunk. Use
REAL_VALUE_TO_DECIMAL.
* config/1750a/1750a-protos.h: Delete prototypes of deleted
functions.
* config/convex/convex.h: Always set TARGET_FLOAT_FORMAT to
IEEE_FLOAT_FORMAT.
* config/i370/i370.h (PRINT_OPERAND [TARGET_HLASM version]):
Use REAL_VALUE_TO_DECIMAL as ELF version does.
* config/m88k/m88k.c (real_power_of_2_operand,
legitimize_operand): Take the REAL_VALUE_TYPE and/or union
real_extract out of the union; run the input through
REAL_VALUE_TO_TARGET_DOUBLE, then plug the pair of longwords
from that into the union.
* config/pdp11/pdp11.c (output_move_double): Rearrange
parentheses to make automatic indenter happy.
* doc/tm.texi (Cross-compilation): Rename node to "Floating
Point" and rewrite to describe current situation. Also adjust
documentation of REAL_VALUE_TO_TARGET_SINGLE and friends to
match code.
* doc/rtl.texi: Adjust cross reference.
From-SVN: r51210
* defaults.h (MODE_HAS_NANS, MODE_HAS_INFINITIES): New.
(MODE_HAS_SIGNED_ZEROS, MODE_HAS_SIGN_DEPENDENT_ROUNDING): New.
* flags.h (HONOR_NANS, HONOR_INFINITIES, HONOR_SIGNED_ZEROS): New.
(HONOR_SIGN_DEPENDENT_ROUNDING): New.
* builtins.c (expand_builtin_mathfn): Use HONOR_NANS.
* c-common.c (truthvalue_conversion): Reduce x - y != 0 to x != y
unless x and y could be infinite.
(expand_unordered_cmp): New, mostly split from expand_tree_builtin.
Check that the common type of both arguments is a real, even for
targets without unordered comparisons. Allow an integer argument
to be compared against a real.
(expand_tree_builtin): Use expand_unordered_cmp.
* combine.c (combine_simplify_rtx): Use the new HONOR_... macros.
* cse.c (fold_rtx): Likewise. Fix indentation.
* fold-const.c (fold_real_zero_addition_p): New.
(fold): Use it, and the new HONOR_... macros.
* ifcvt.c (noce_try_minmax): Use the new HONOR_... macros.
* jump.c (reversed_comparison_code_parts): After searching for
the true comparison mode, use HONOR_NANS to decide whether it
can be safely reversed.
(reverse_condition_maybe_unordered): Remove IEEE check.
* simplify-rtx.c (simplify_binary_operation): Use the new macros
to decide which simplifications are valid. Allow the following
simplifications for IEEE: (-a + b) to (b - a), (a + -b) to (a - b),
and (a - -b) to (a + b).
(simplify_relational_operation): Use HONOR_NANS.
* doc/tm.texi: Document the MODE_HAS_... macros.
From-SVN: r50401
