* real.c (c4x_single_format, c4x_extended_format): Provide values
for signbit_ro for c4x's single and extended floating point formats.
* optabs.c (expand_copysign): Use the floating point format's
signbit_ro for expanding via expand_copysign_absneg, and it's
signbit_rw field for expanding via expand_copysign_bit.
Co-Authored-By: Richard Henderson <rth@redhat.com>
From-SVN: r96530
* real.h (struct real_format): Split the signbit field into two
two fields, signbit_ro and signbit_rw.
* real.c (ieee_single_format, mips_single_format, ieee_double_format,
mips_double_format, ieee_extended_motorola_format,
ieee_extended_intel_96_format, ieee_extended_intel_96_round_53_format,
ieee_extended_intel_128_format, ibm_extended_format,
mips_extended_format, ieee_quad_format, mips_quad_format,
vax_f_format, vax_d_format, vax_g_format, i370_single_format,
i370_double_format, c4x_single_format, c4x_extended_format,
real_internal_format): Update initializers for new field.
* builtins.c (expand_builtin_signbit): Use signbit_ro field of the
real_format structure.
* optabs.c (expand_absneg_bit, expand_copysign): Use signbit_rw
field of the real_format structure.
From-SVN: r96471
PR rtl-optimization/576
* real.c (real_arithmetic): Change return type from void to bool
to return an indication that the result may be inexact.
* real.h (real_arithmeric): Update prototype.
* fold-const.c (const_binop): Don't constant fold floating
point expressions when the user specifies -frounding-math and
the result may depend upon the run-time rounding mode.
(fold_convert_const_real_from_real): Clean-up.
(fold_initializer): Ignore flag_rounding_math for initializers.
* simplify-rtx.c (simplify_binary_operation): Likewise, don't
constant fold FP operations with flag_rounding_math if the
result may depend upon the run-time rounding mode.
From-SVN: r94020
* real.c (real_floor): Don't leave the result uninitialized when
mode is VOIDMode, but return the unrounded intermediate value.
(real_ceil): Likewise.
From-SVN: r93988
* real.c (real_copysign): New function to implement libm's copysign.
* real.h (real_copysign): Prototype here.
* fold-const.c (tree_expr_nonnegative_p): The result of sqrt, sqrtf
and sqrtl can be negative, as sqrt(-0.0) = -0.0. Correct whitespace.
* builtins.c (fold_builtin_isascii, fold_builtin_toascii,
fold_builtin_isdigit): Add function prototypes.
(fold_builtin_copysign): New function to fold copysign, copysignf
and copysignl. Optimize copysign(x,x) as x. Evaluate copysign of
constant arguments at compile-time using real_copysign. Fold
copysign(X,Y) as fabs(X) if Y is always non-negative.
(fold_builtin_1): Correct minor whitespace/style issues. Call
fold_builtin_copysign for BUILT_IN_COPYSIGN{,F,L}.
* gcc.dg/builtins-41.c: New test case.
* gcc.dg/builtins-42.c: New test case.
From-SVN: r82721
gcc/ChangeLog:
2004-05-27 Paolo Bonzini <bonzini@gnu.org>
Roger Sayle <roger@eyesopen.com>
PR rtl-optimization/15649
Add LTGT_EXPR and improve pretty-printing of unordered
comparisons.
* c-common.c (c_common_truthvalue_conversion):
Handle LTGT_EXPR.
* c-typeck.c (build_binary_op): Likewise.
* dojump.c (do_jump): Likewise.
* expr.c (expand_expr_real_1, do_store_flag): Likewise.
* predict.c (tree_predict_by_opcode): Likewise.
* real.c (real_compare): Likewise.
* tree-cfg.c (verify_expr): Likewise.
* tree-inline.c (estimate_num_insns_1): Likewise.
* tree-pretty-print.c (dump_generic_node): Likewise.
Handle ORDERED_EXPR, UNORDERED_EXPR.
(op_symbol): Print unordered comparisons differently
than ordered ones.
* tree.def (LTGT_EXPR): New '<' tree code.
* doc/c-tree.texi (Expressions): Document floating-point
comparison nodes.
Fold comparisons between floating point values.
* fold-const.c (enum comparison_code): New, from
#define'd constants. Define compcodes for unordered
comparisons and for invalid transformations.
(invert_tree_comparison): Add "honor_nans" parameter.
(fold_truthop): Revamp to work on floating-point types too.
(comparison_to_compcode): Support unordered comparisons.
Use new enum comparison_code.
(compcode_to_comparison): Likewise.
(combine_compcodes): New function.
(invert_truthvalue): Let invert_tree_comparison decide
whether it is valid to fold the comparison. Fold ORDERED
and UNORDERED even if flag_unsafe_math_optimizations is off,
and the remaining even if flag_unsafe_math_optimizations
is off but we are under -fno-trapping-math.
(fold_relational_const): Integer modes do not honor NaNs.
gcc/testsuite/ChangeLog:
2004-05-27 Paolo Bonzini <bonzini@gnu.org>
* gcc.c-torture/compare-fp-1.c, gcc.c-torture/compare-fp-2.c,
gcc.c-torture/compare-fp-3.c, gcc.c-torture/compare-fp-4.c,
gcc.c-torture/compare-fp-3.x, gcc.c-torture/compare-fp-4.x,
gcc.c-torture/pr15649-1.c: New.
Co-Authored-By: Roger Sayle <roger@eyesopen.com>
From-SVN: r82365
* real.h (struct real_value): Use the same type for all
bitfields. Rename exp to uexp.
(REAL_EXP, SET_REAL_EXP): New accessor macros for uexp.
Adjust all uses of exp...
* builtins.c: ... here, ...
* emit-rtl.c: ... here, and ...
* real.c: ... and here.
From-SVN: r79802
* real.c (encode_ibm_extended): Normalize the input value before
converting it to a double. Handle the case where a normal value
rounds to infinity.
From-SVN: r77498
* real.c (real_floor, real_ceil): Tweak to allow input and output
arguments to overlap.
(real_round): New function to implement round(3m) semantics.
* real.h (real_round): Prototype here.
* builtins.c (fold_builtin_round): New function to constant fold
round, roundf and roundl.
(fold_builtin): Call fold_builtin_round for BUILT_IN_ROUND{,F,L}.
* gcc.dg/builtins-29.c: New test case.
From-SVN: r76428
* config/rs6000/darwin-ldouble.c: Add big comment explaining
exactly what is expected as a 'long double'.
(_xlqadd): When a value to be returned is representable as a
'double', just return it directly, do not construct it using a union.
Also, correct final fixup.
(_xlqmul): Likewise.
(_xlqdiv): Likewise.
* real.c (encode_ibm_extended): Make consistent with darwin-ldouble.c.
From-SVN: r75629
* Makefile.in (BUILD_RTL): Replace $(BUILD_PREFIX)insn-modes.o
with min-insn-modes.o.
(STAGESTUFF): Add min-insn-modes.c.
(genobjs): Add genmodes.o.
(print-rtl.o, print-rtl1.o): Depend on $(TM_P_H).
(insn-modes.o): Depend on $(TM_H) not $(GTM_H); also real.h.
(min-insn-modes.c, min-insn-modes.o): New rules.
(s-modes): Also generate min-insn-modes.c.
($(BUILD_PREFIX_1)insn-modes.o): Kill.
* genmodes.c (struct mode_data): Add format field.
(blank_mode, validate_mode, complete_mode): Update to match.
(make_scalar_mode): Separate into make_int_mode and make_float_mode.
(_SCALAR_MODE): Kill.
(FLOAT_MODE, FRACTIONAL_FLOAT_MODE): Add format argument.
(emit_insn_modes_c_header): Adjust.
(emit_min_insn_modes_c_header, emit_real_format_for_mode)
(emit_min_insn_modes_c): New functions.
(emit_insn_modes_c): Call emit_real_format_for_mode.
(main): Add -m option to generate min-insn-modes.c.
* machmode.h: Update documentation. Add format argument to
all uses of FLOAT_MODE.
* real.c: Don't define real_format_for_mode here.
* dwarfout.c: Move default definition of PRINT_REG...
* defaults.h: ...here.
* print-rtl.c: Include tm_p.h.
(DEBUG_PRINT_REG, DEBUG_REGISTER_NAMES, debug_reg_names, reg_names):
Kill.
(print_rtx): Use PRINT_REG, not DEBUG_PRINT_REG. But surround
this entire block with #ifndef GENERATOR_FILE.
* regclass.c: Unconditionally define reg_names.
* config/mips/mips.h, config/rs6000/rs6000.h, config/sh/sh.h
Don't define DEBUG_REGISTER_NAMES.
* config/rs6000/darwin.h: Don't use DEBUG_REGISTER_NAMES in
redefinition of REGISTER_NAMES.
* config/i386/i386.h: Don't define DEBUG_PRINT_REG.
* combine.c: Change all preprocessor conditionals on
EXTRA_CC_MODES to use SELECT_CC_MODE instead; rearrange a bit
for clarity.
* genopinit.c: Remove mention of EXTRA_CC_MODES in comment.
* configure.in: Don't define EXTRA_CC_MODES.
* configure, config.in: Regenerate.
* doc/tm.texi: Remove documentation of EXTRA_CC_MODES.
* config/arc/arc.c, config/m32r/m32r.c, config/sparc/sparc.c:
May assume that GET_MODE_CLASS is accurate for extra CC modes
at all times.
* config/i860/i860.h (INIT_CUMULATIVE_ARGS): Pass correct
number of arguments to aggregate_value_p.
* genmodes.c (RESET_FLOAT_FORMAT, reset_float_format): New.
* machmode.def: Explain ARCH-modes.def. Document
RESET_FLOAT_FORMAT. Improve commentary on various mode
clusters. Do not define OI, PQI, PHI, PSI, PDI, QF, HF, TQF,
XF, or TF modes here. Remove backward-compatibility
definition of CC.
* config/alpha/alpha-modes.def: New file; define TF mode.
* config/arc/arm-modes.def: Define XF mode.
* config/c4x/c4x-modes.def: Define QF and HF modes. Unset
float format for SF and DF modes.
* config/dsp16xx/dsp16xx-modes.def: New file; define HF mode.
* config/i386/i386-modes.def: Define XF and TF modes.
* config/i960/i960-modes.def: Define TF mode.
* config/ia64/ia64-modes.def: Define TF and OI modes.
* config/m68k/m68k-modes.def: New file; define XF mode.
* config/mips/mips-modes.def: New file; define TF mode, reset
formats for SF and DF modes.
* config/pa/pa-modes.def: Define TF mode.
* config/rs6000/rs6000.c: Define TF and PSI modes.
* config/s390/s390-modes.def: Define OI mode.
* config/sh/sh-modes.def: New file; define PSI mode.
* config/sparc/sparc-modes.def: Define TF mode.
* config/vax/vax-modes.def: New file; reset formats for SF and
DF modes.
* config/c4x/c4x.c (c4x_override_options): No need to mess
with real_format_for_mode or set REAL_MODE_FORMATs.
(c4x_immed_int_constant): Don't apply GET_MODE_CLASS to rtx
variable.
* config/i386/i386.c (override_options): No need to set
REAL_MODE_FORMATs here.
* config/i960/i960.c (i960_initialize): Likewise.
* config/m68k/m68k.c (m68k_override_options): Likewise.
* config/ia64/ia64.c (ia64_override_options): Set REAL_MODE_FORMAT
for TFmode only if not the default.
* config/mips/mips.c (override_options): Likewise.
* config/vax/vax.c (override_optionms): Set REAL_MODE_FORMAT for
DFmode only if not the default.
* config/i370/i370.h (RET_REG): Don't consider TFmode.
* config/m68hc11/m68hc11.c (print_operand): Don't consider XFmode.
* config/dsp16xx/dsp16xx.c (hard_regno_mode_ok): #if 0 out use
of modes that don't appear anywhere in the machine description.
* config/arc/arc-modes.def, config/arm/arm-modes.def
* config/c4x/c4x-modes.def, config/frv/frv-modes.def
* config/i386/i386-modes.def, config/i960/i960-modes.def
* config/ia64/ia64-modes.def, config/mmix/mmix-modes.def
* config/pa/pa-modes.def, config/pdp11/pdp11-modes.def
* config/rs6000/rs6000-modes.def, config/s390/s390-modes.def
* config/sparc/sparc-modes.def: Convert to new style for
declaring extra CC modes.
From-SVN: r72440
* real.c (real_trunc, real_floor, real_ceil): New functions
to implement trunc, floor and ceil respectively.
* real.h (real_trunc, real_floor, real_ceil): Prototype here.
* builtins.c (integer_valued_real_p): New function to test if
a floating point expression has an integer valued result.
(fold_trunc_transparent_mathfn): Optimize foo(foo(x)) as
foo(x) where foo is an integer rounding function. Similarly,
optimize foo(bar(x)) as bar(x), and foo((double)(int)x) as
(double)(int)x when both foo and bar are integer rounding
functions and we don't need to honor errno.
(fold_builtin_trunc, fold_builtin_floor, fold_builtin_ceil):
New functions to fold trunc, floor and ceil.
(fold_builtin): Use fold_builtin_trunc to fold BUILT_IN_TRUNC*,
fold_builtin_floor to fold BUILT_IN_FLOOR* and fold_builtin_ceil
to fold BUILT_IN_CEIL*.
* fold-const.c (tree_expr_nonnegative_p): Handle FLOAT_EXPR and
the remaining integer rounding functions.
* gcc.dg/builtins-25.c: New testcase.
* gcc.dg/builtins-26.c: New testcase.
From-SVN: r68903
* real.h (ieee_extended_intel_96_round_53_format): New.
* real.c (ieee_extended_intel_96_round_53_format): New.
* config/i386/freebsd.h (SUBTARGET_OVERRIDE_OPTIONS): Use it
for XFmode and TFmode.
From-SVN: r68445
* real.c (real_maxval): New function to return the largest finite
value representable in a given mode (i.e. FLT_MAX and DBL_MAX).
* real.h (real_maxval): Prototype here.
* fold-const.c (fold_inf_compare): Transform comparisons against
+-Infinity into comparisons against DBL_MAX (or equivalent).
* gcc.c-torture/execute/ieee/inf-2.c: New test case.
From-SVN: r67112
* real.c (real_powi): New function to calculate the value of
a real raised to an integer power, i.e. pow(x,n) for int n.
(real_sqrt): Convert to using the faster do_add, do_multiply
and do_divide API for consistency with the rest of real.c.
* real.h (real_powi): Prototype here.
* builtins.c (fold_builtin): Avoid local variable mode when
evaluating sqrt at compile time. Attempt to evaluate pow at
compile-time, by checking for an integral exponent.
* gcc.dg/builtins-14.c: New test case.
From-SVN: r66515
* real.c (do_add): Change to return a bool indicating that the
result of the operation may be inexact due to loss of precision.
(do_multiply): Likewise.
(do_divide): Likewise.
From-SVN: r65980
* real.h (EXP_BITS): Make room for...
(struct real_value): ... added canonical bit.
(struct real_format): Added pnan.
(mips_single_format, mips_double_format, mips_extended_format,
mips_quad_format): New.
* real.c: Copy p to pnan in all formats.
(get_canonical_qnan, get_canonical_snan): Set canonical bit.
(real_nan): Use pnan to compute significand's shift.
(real_identical): Disregard significand in canonical
NaNs.
(real_hash): Likewise. Take signalling into account.
(encode_ieee_single, encode_ieee_double, encode_ieee_quad):
Disregard significand bits in canonical NaNs. Set all bits of
canonical NaN if !qnan_msb_set.
(encode_ibm_extended, decode_ibm_extended): Likewise. Use
qnan_msb_set to tell the base double format.
(ibm_extended_format): Use 53 as pnan.
(mips_single_format, mips_double_format, mips_extended_format,
mips_quad_format): Copied from the corresponding ieee/ibm
formats, with qnan_msb_set false.
* config/mips/iris6.h (MIPS_TFMODE_FORMAT): Use mips_extended_format.
* config/mips/linux64.h (MIPS_TFMODE_FORMAT): Use mips_quad_format.
* config/mips/mips.c (override_options): Use mips_single_format
and mips_double_format. Default TFmode to mips_quad_format.
* config/mips/t-linux64 (tp-bit.c): Define QUIET_NAN_NEGATED.
* config/mips/t-irix6: Likewise.
* config/mips/t-mips (fp-bit.c, dp-bit.c): Likewise.
* config/fp-bit.c (pack_d, unpack_d): Obey it.
From-SVN: r65146
* emit-rtl.c (dconstm2, dconsthalf): New real constants.
(init_emit_once): Initialize dconstm2 and dconsthalf here.
* real.h (dconstm2, dconsthalf): Add prototypes here.
* real.c (real_sqrt): Use dconsthalf rather than local copy.
* builtins.c (fold_builtin): When optimizing sqrt(exp(x)) as
exp(x/2.0) remember to fold the division if possible.
Fold sin(0.0) as 0.0, cos(0.0) as 1.0, pow(x,1.0) as x,
pow(x,-1.0) as 1.0/x, pow(x,2.0) as x*x, pow(x,-2.0) as
1.0/(x*x) and pow(x,0.5) as sqrt(x).
* gcc.dg/builtins-3.c: Add new tests for sin and cos.
* gcc.dg/builtins-7.c: New test case.
* gcc.dg/builtins-8.c: New test case.
From-SVN: r65088
* real.c (real_sqrt): Return a bool result indicating whether
a floating point exception or trap should be raised.
* real.h (real_sqrt): Update function prototype.
* builtins.c (fold_builtin): Only fold non-trapping square
roots unless we're ignoring errno and trapping math.
From-SVN: r61337
