* tree-vect-transform.c (vect_get_slp_defs): Don't build a vector
for oprnd1 if not required.
(vectorizable_operation): Use scalar operand in SLP in case of
shift with scalar argument.
From-SVN: r128412
2007-09-11 James E. Wilson <wilson@specifix.com>
* defaults.h (DWARF2_UNWIND_INFO): Don't define if
TARGET_UNWIND_INFO is defined.
* config/ia64/ia64.h (INCOMING_RETURN_ADDR_RTX): Delete undef
after definition.
From-SVN: r128409
* config/spu/spu.c (spu_emit_branch_or_set): Handle NaN values as
operands to DFmode GE or LE compares.
testsuite/
* gcc.target/spu/dfcgt-nan.c: New test.
From-SVN: r128404
* tree-ssa-structalias.c (push_fields_onto_fieldstack): Deal with
TYPE_NONALIASED_COMPONENT like with DECL_NONADDRESSABLE_P.
ada/
* decl.c (array_type_has_nonaliased_component): New predicate.
(gnat_to_gnu_field) <E_Array_Type>: Invoke the above predicate to
set the TYPE_NONALIASED_COMPONENT flag on the type.
<E_Array_Subtype>: Likewise.
* gigi.h (type_for_nonaliased_component_p): Declare.
* utils.c (type_for_nonaliased_component_p): New predicate.
(create_field_decl): Invoke the above predicate to set the
DECL_NONADDRESSABLE_P flag on the field.
From-SVN: r128391
2007-09-11 Christopher D. Rickett <crickett@lanl.gov>
PR fortran/33040
* trans-expr.c (gfc_trans_structure_assign): Convert component
C_NULL_PTR and C_NULL_FUNPTR component initializers to (void *).
* trans-types.c (gfc_get_derived_type): Create a backend_decl for
the c_address field of C_PTR and C_FUNPTR and ensure initializer
is of proper type/kind for (void *).
2007-09-11 Christopher D. Rickett <crickett@lanl.gov>
PR fortran/33040
* gfortran.dg/c_ptr_tests_11.f03: New test case.
From-SVN: r128385
* builtins.def (BUILT_IN_VA_ARG_PACK_LEN): New builtin.
* builtins.c (expand_builtin) <case BUILT_IN_VA_ARG_PACK_LEN>: Issue
error if __builtin_va_arg_pack_len () wasn't optimized out during
inlining.
* tree-inline.c (copy_bb): Replace __builtin_va_arg_pack_len ()
with the number of inline's anonymous arguments.
* doc/extend.texi: Document __builtin_va_arg_pack_len ().
* gcc.dg/va-arg-pack-len-1.c: New test.
* g++.dg/va-arg-pack-len-1.C: New test.
From-SVN: r128376
2007-09-11 Javier Miranda <miranda@adacore.com>
* einfo.ads, einfo.adb (Dispatch_Table_Wrapper): New attribute. Present
in library level record type entities if we are generating statically
allocated dispatch tables.
* exp_disp.adb (Make_Tags/Make_DT): Replace previous code
importing/exporting the _tag declaration by new code
importing/exporting the dispatch table wrapper. This change allows us
to statically allocate of the TSD.
(Make_DT.Export_DT): New procedure.
(Build_Static_DT): New function.
(Has_DT): New function.
* freeze.adb (Freeze_Static_Object): Code cleanup: Do not reset flags
True_Constant and Current_Value. Required to statically
allocate the dispatch tables.
(Check_Allocator): Make function iterative instead of recursive.
Also return inner allocator node, when present, so that we do not have
to look for that node again in the caller.
From-SVN: r128368
gcc/
* config/mips/mips.c (mips_set_mips16_mode): Use separate anchor
settings for MIPS16.
(mips_use_anchors_for_symbol_p): Use default_use_anchors_for_symbol_p.
From-SVN: r128366
gcc/
* config/mips/mips.c (mips_symbol_insns_1): Allow LEAs of
SYMBOL_FORCE_TO_MEM constants.
(mips_rtx_costs): Give a cost of 1 to force_to_mem_operands.
(mips16_rewrite_pool_refs_info): New structure.
(mips16_rewrite_pool_constant): New function, split out from...
(mips16_rewrite_pool_refs): ...here. Take a pointer to a
mips16_rewrite_pool_refs_info structure rather than a pointer
to a constant pool. Force force_to_mem_operands into memory.
(mips16_lay_out_constants): Update call to mips16_rewrite_pool_refs.
* config/mips/predicates.md (force_to_mem_operand): New predicate.
* config/mips/constraints.md (kf): New constraint.
* config/mips/mips.md (*movdi_64bit_mips16): Add a d <- kf alternative.
(*movsi_mips16): Likewise.
From-SVN: r128365
gcc/
2007-09-11 Richard Sandiford <richard@codesourcery.com>
Nigel Stephens <nigel@mips.com>
David Ung <davidu@mips.com>
* config/mips/mips.h (CONSTANT_POOL_COST): Move to...
* config/mips/mips.c: ...here and set to 4 for TARGET_MIPS16.
(mips16_constant_cost, mips_immediate_operand_p, mips_binary_cost)
(mips_fp_mult_cost, mips_fp_div_cost, mips_sign_extend_cost)
(mips_zero_extend_cost): New functions.
(mips_rtx_costs): Treat COMPARE constants as having zero cost.
Use the new functions. Tweak many cost estimates, both here
and in the new subroutines. Return false when the cost of the
operands has not been calculated. Check for *clear_upper32.
Check for floating-point multiply-add, reciprocal and rsqrt
patterns. Handle comparison and rotation codes.
Co-Authored-By: David Ung <davidu@mips.com>
Co-Authored-By: Nigel Stephens <nigel@mips.com>
From-SVN: r128364
gcc/
PR c/30013
* config/dfp-bit.c: Don't skip TFmode conversions; move strto*
declarations to top.
(DFP_TO_BFP): Use for either XFmode or TFmode.
(BFP_TO_DFP): Use for either XFmode or TFmode; always use cast
of BFP_VIA_TYPE.
* config/dfp-bit.h: Include float.h.
(LONG_DOUBLE_HAS_XF_MODE, LONG_DOUBLE_HAS_TF_MODE): Define if long
double is one of these modes, rather than using LIBGCC_HAS_*F_MODE
which doesn't mean the same thing.
(BFP_KIND): Use 4 to mean TFmode.
(BFP_FMT): Specify the number of decimal digits based on the
number of mantissa digits.
(BFP_VIA_TYPE): Binary float type to use as cast for sprintf.
(BFP_TO_DFP, DFP_TO_BFP): Define names for TFmode variants.
(STR_TO_BFP): Use strtold for XFmode or TFmode.
(TFtype): Define if TFmode is supported.
* doc/libgcc.texi (Decimal float library routines): Document
TF conversion functions.
gcc/testsuite/
* gcc.dg/dfp/convert-bfp.c: Replace SKIP_LONG_DOUBLE with runtime
checks for size of long double.
* gcc.dg/dfp/convert.h: New file.
* gcc.dg/dfp/convert-bfp-2.c: New test.
* gcc.dg/dfp/convert-bfp-3.c: Ditto.
* gcc.dg/dfp/convert-bfp-4.c: Ditto.
* gcc.dg/dfp/convert-bfp-5.c: Ditto.
* gcc.dg/dfp/convert-bfp-6.c: Ditto.
* gcc.dg/dfp/convert-bfp-7.c: Ditto.
* gcc.dg/dfp/convert-bfp-8.c: Ditto.
* gcc.dg/dfp/convert-bfp-9.c: Ditto.
* gcc.dg/dfp/convert-bfp-10.c: Ditto.
* gcc.dg/dfp/convert-bfp-11.c: Ditto.
From-SVN: r128361
* config/mips/mips.c (mips_scalar_mode_supported_p): Declare.
(TARGET_SCALAR_MODE_SUPPORTED_P): Define.
(mips_emit_compare): Process fixed-point modes.
(mips_pad_arg_upward): Support fixed-point types.
(override_options): Allow fixed-point modes in accumulators.
(mips_pass_by_reference): Pass DQ, UDQ, DA, and UDA modes in registers.
(mips_vector_mode_supported_p): Support V2HQmode, V2UHQmode, V2HAmode,
V2UHAmode, V4QQmode, and V4UQQmode when TARGET_DSP.
(mips_scalar_mode_supported_p): New function to accept fixed-point
modes if the width is not greater than two BITS_PER_WORD.
* config/mips/mips.h (SHORT_FRACT_TYPE_SIZE, FRACT_TYPE_SIZE,
LONG_FRACT_TYPE_SIZE, LONG_LONG_FRACT_TYPE_SIZE,
SHORT_ACCUM_TYPE_SIZE, ACCUM_TYPE_SIZE, LONG_ACCUM_TYPE_SIZE,
LONG_LONG_ACCUM_TYPE_SIZE): Define.
* config/mips/mips.md ("d"): Update mode attribute for fixed-point
modes.
("IMODE"): New mode attribute.
(mips-fixed.md): Include.
* config/mips/mips-modes.def: Create VECTOR_MODES for FRACT, UFRACT,
ACCUM, UACCUM.
* config/mips/mips-fixed.md: New file.
From-SVN: r128360
* config/dfp-bit.c (dfp_unary_func): Delete.
(dfp_unary_op): Delete.
(dfp_binary_op): Use decFloat functions instead of decNumber
functions for binary operations.
(d32_binary_op): Convert 32-bit operands to 64 bits for evaluation.
(dnn_binary_op): Call dfp_binary_op with decFloat rather than
DFP_C_TYPE.
(dfp_compare_op): Use decFloat functions insteadof decNumber
functions for comparisons.
(d32_compare_op): Convert 32-bit operands to 64 bits for evaluation.
(dnn_binary_op): Call dfp_compare_op with decFloat rather than
DFP_C_TYPE.
(DFP_ADD, DFP_SUB, DFP_MULTIPLE, DFP_DIVIDE): Use macros for
call to dxx_binary_op and decFloat function.
(DFP_EQ, DFP_NE, DFP_LT, DFP_GT, DFP_LE, DFP_GE): Use macros for
calls to dxx_binary_op and decFloat function.
* config/dfp-bit.h: Include decFloat header files.
(decFloat, DFP_BINARY_OP, DFP_COMPARE_OP, DEC_FLOAT_ADD,
DEC_FLOAT_SUBTRACT, DEC_FLOAT_MULTIPLY, DEC_FLOAT_DIVIDE,
DEC_FLOAT_COMPARE, DEC_FLOAT_IS_ZERO, DEC_FLOAT_IS_NAN,
DEC_FLOAT_IS_SIGNED: Define for each of 3 operand widths.
From-SVN: r128358
Jan Sjodin <jan.sjodin@amd.com>
* tree-vect-analyze.c (vect_analyze_operations): Change
comparison of loop iterations with threshold to less than
or equal to instead of less than. Reduce
min_scalar_loop_bound by one.
* tree-vect-transform.c (vect_estimate_min_profitable_iters):
Change prologue and epilogue iterations estimate to vf/2,
when unknown at compile-time. Change versioning guard
cost to taken_branch_cost. If peeling for alignment is
unknown at compile-time, change peel guard costs to one
taken branch and one not-taken branch per peeled loop.
If peeling for alignment is known but number of scalar loop
iterations is unknown at compile-time, change peel guard
costs to one taken branch per peeled loop. Change the cost
model equation to consider vector iterations as the loop
iterations less the prologue and epilogue iterations.
Change outside vector cost check to less than or equal to
zero instead of equal to zero.
(vect_do_peeling_for_loop_bound): Reduce
min_scalar_loop_bound by one.
* tree-vectorizer.h: Add TARG_COND_TAKEN_BRANCH_COST and
TARG_COND_NOT_TAKEN_BRANCH_COST.
* config/i386/i386.h (processor_costs): Add
scalar_stmt_cost, scalar_load_cost, scalar_store_cost,
vec_stmt_cost, vec_to_scalar_cost, scalar_to_vec_cost,
vec_align_load_cost, vect_unalign_load_cost,
vec_store_cost, cond_taken_branch_cost,
cond_not_taken_branch_cost.
Define macros for x86 costs.
* config/i386/i386.c:
(size_cost): Set scalar_stmt_cost, scalar_load_cost,
scalar_store_cost, vec_stmt_cost, vec_to_scalar_cost,
scalar_to_vec_cost, vec_align_load_cost,
vect_unalign_load_cost, vec_store_cost,
cond_taken_branch_cost, cond_not_taken_branch_cost to one.
(i386_cost, i486_cost, pentium_cost, pentiumpro_cost,
geode_cost, k6_cost, athlon_cost, pentium4_cost, nocona_cost,
core2_cost, generic64_cost, generic32_cost): Set to default
untuned costs.
(k8_cost, amdfam10_cost): Costs for vectorization tuned.
(x86_builtin_vectorization_cost): New.
2007-09-10 Harsha Jagasia <harsha.jagasia@amd.com>
* gcc.dg/vect/costmodel/i386/costmodel-vect-31.c:
Change dg-final to expect 1 non-profitable loop and
3 profitable loops.
* gcc.dg/vect/costmodel/x86-64/costmodel-vect-31.c:
Change dg-final to expect 1 non-profitable loop and
3 profitable loops.
* gcc.dg/vect/costmodel/x86-64/costmodel-fast-math-vect-pr29925.c:
Change dg-final to expect 1 profitable loop.
* gcc.dg/vect/costmodel/i386/costmodel-fast-math-vect-pr29925.c:
Change dg-final to expect 1 profitable loop.
Co-Authored-By: Jan Sjodin <jan.sjodin@amd.com>
From-SVN: r128353
