The current rs6000 rtx_cost for comparisons against 0 is very high if
TARGET_ISEL && !TARGET_MFCRF, much higher than for reg-reg comparisons,
much higher than a load of 0 and such a reg-reg-comparison. This leads
to infinite recursion in CSE (see PR81288).
This patch removes the too-high cost, also simplifying this code.
PR 81288/target
* config/rs6000/rs6000.c (rs6000_rtx_costs): Do not handle
TARGET_ISEL && !TARGET_MFCRF differently. Simplify code.
From-SVN: r255188
If we have a negative condition we can use a literal 0 in the isel,
instead of having to load it into a register. If the condition is from
a comparison with an immediate we can change e.g. LT to LE and adjust
the immediate, saving a li instruction.
* config/rs6000/rs6000.md (<code><GPR:mode><GPR2:mode>2_isel): Change
LT/GT/LTU/GTU to LE/GE/LEU/GEU where possible.
From-SVN: r255186
PR c++/81888
* parser.c (cp_parser_decomposition_declaration): Reject just
BRACE_ENCLOSED_INITIALIZER_P initializers with nelts != 1 rather
than all such CONSTRUCTORs, and only if is_direct_init is true.
* g++.dg/cpp1z/decomp30.C: Add a test for structured binding with
= {} and = { a, a } initializers.
* g++.dg/cpp1z/decomp31.C: New test.
From-SVN: r255180
2017-11-27 Michael Meissner <meissner@linux.vnet.ibm.com>
PR middle_end/82333
* varasm.c (compare_constant): Take the mode of the constants into
account when comparing floating point constants.
[gcc/testsuite]
2017-11-27 Michael Meissner <meissner@linux.vnet.ibm.com>
PR middle_end/82333
* gcc.target/powerpc/pr82333.c: New test.
From-SVN: r255177
PR c++/81675
* cp-gimplify.c (cp_fold) <case COND_EXPR>: Don't return immediately
for VOID_TYPE_P COND_EXPRs, instead fold the operands and if op0 is
INTEGER_CST, ensure that both op1 and op2 are non-NULL and fall
through into normal folding, otherwise just rebuild x if any op
changed.
* g++.dg/warn/pr81675.C: New test.
From-SVN: r255167
bootstrap-ubsan shows:
gcc/hash-map.h:277:19: runtime error: member access within null pointer of type 'struct hash_map'
Fix the issue by returning early.
From-SVN: r255166
According to the description of inssp instruction from Intel CET it
adusts the shadow stack pointer (ssp) only by value in the range of
[0..255]. As a number of adjustment could be greater than 255 there
should be a loop generated to adjust ssp.
gcc/
* config/i386/i386.md: Add a loop with incssp.
* testsuite/gcc.target/i386/cet-sjlj-1.c: Fix test.
* testsuite/gcc.target/i386/cet-sjlj-4.c: Likewise.
From-SVN: r255164
PR debug/81307
* dbxout.c (lastlineno): New variable.
(dbx_debug_hooks): Use dbxout_switch_text_section as
switch_text_section debug hook.
(dbxout_function_end): Switch to current_function_section
rather than function_section. If crtl->has_bb_partition,
output just one N_FUN, depending on in_cold_section_p.
(dbxout_source_line): Remember last lineno in lastlineno.
(dbxout_switch_text_section): New function.
(dbxout_function_decl): Adjust dbxout_block caller.
(dbx_block_with_cold_children): New function.
(dbxout_block): Return true if any LBRAC/RBRAC have been
emitted. Use dbx_block_with_cold_children at depth == 0
in second partition. Add PARENT_BLOCKNUM argument, pass
it optionally adjusted to children. Output LBRAC/RBRAC
around recursive call only if the block is in the current
partition, if not and anything was output, emit empty
range LBRAC/RBRAC.
* final.c (final_scan_insn): Compute cold_function_name
before calling switch_text_section debug hook. Call
that hook even if dwarf2out_do_frame if not emitting
dwarf debug info.
* g++.dg/debug/debug9.C: Remove -fno-reorder-blocks-and-partition
workaround.
From-SVN: r255161
PR rtl-optimization/81553
* combine.c (simplify_if_then_else): In (if_then_else COND (OP Z C1) Z)
to (OP Z (mult COND (C1 * STORE_FLAG_VALUE))) optimization, if OP
is a shift where C1 has different mode than the whole shift, use C1's
mode for MULT rather than the shift's mode.
* gcc.c-torture/compile/pr81553.c: New test.
From-SVN: r255150
PR target/82848
* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin): Don't fold
builtins not enabled in the currently selected ISA.
* gcc.target/powerpc/pr82848.c: New test.
From-SVN: r255148
PR fortran/81304
* trans-openmp.c (gfc_trans_omp_array_reduction_or_udr): Set
attr.implicit_type in intrinsic_sym to avoid undesirable warning.
* testsuite/libgomp.fortran/pr81304.f90: New test.
From-SVN: r255144
This patch implements the some of the division optimizations discussed in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71026.
The division reciprocal optimization now handles divisions by squares:
x / (y * y) -> x * (1 / y) * (1 / y)
This requires at least one more division by y before it triggers - the
3 divisions of (1/ y) are then CSEd into a single division. Overall
this changes 1 division into 1 multiply, which is generally much faster.
2017-11-24 Jackson Woodruff <jackson.woodruff@arm.com>
gcc/
PR tree-optimization/71026
* tree-ssa-math-opts (is_division_by_square, is_square_of): New.
(insert_reciprocals): Change to insert reciprocals before a division
by a square and to insert the square of a reciprocal.
(execute_cse_reciprocals_1): Change to consider division by a square.
(register_division_in): Add importance parameter.
testsuite/
PR tree-optimization/71026
* gfortran.dg/extract_recip_1.f: New test.
* gcc.dg/extract_recip_3.c: New test.
* gcc.dg/extract_recip_4.c: New test.
From-SVN: r255141
* tree-object-size.c (pass_through_call): Use gimple_call_return_flags
ERF_RETURN*ARG* for builtins other than BUILT_IN_ASSUME_ALIGNED,
check for the latter with gimple_call_builtin_p. Do not handle
BUILT_IN_STPNCPY_CHK which is not a pass through call.
* gcc.dg/builtin-object-size-18.c: New test.
From-SVN: r255133
