gcc/ChangeLog:
2018-01-22 Carl Love <cel@us.ibm.com>
* config/rs6000/rs6000-builtin.def (ST_ELEMREV_V1TI, LD_ELEMREV_V1TI,
LVX_V1TI): Add macro expansion.
* config/rs6000/rs6000-c.c (altivec_builtin_types): Add argument
definitions for VSX_BUILTIN_VEC_XST_BE, VSX_BUILTIN_VEC_ST,
VSX_BUILTIN_VEC_XL, LD_ELEMREV_V1TI builtins.
* config/rs6000/rs6000-p8swap.c (insn_is_swappable_p);
Change check to determine if the instruction is a byte reversing
entry. Fix typo in comment.
* config/rs6000/rs6000.c (altivec_expand_builtin): Add case entry
for VSX_BUILTIN_ST_ELEMREV_V1TI and VSX_BUILTIN_LD_ELEMREV_V1TI.
Add def_builtin calls for new builtins.
* config/rs6000/vsx.md (vsx_st_elemrev_v1ti, vsx_ld_elemrev_v1ti):
Add define_insn expansion.
gcc/testsuite/ChangeLog:
2018-01-22 Carl Love <cel@us.ibm.com>
* gcc.target/powerpc/powerpc.exp: Add torture tests for
builtins-4-runnable.c, builtins-6-runnable.c,
builtins-5-p9-runnable.c, builtins-6-p9-runnable.c.
* gcc.target/powerpc/builtins-6-runnable.c: New test file.
* gcc.target/powerpc/builtins-4-runnable.c: Add additional tests
for signed/unsigned 128-bit and long long int loads.
From-SVN: r256952
PR c++/81933
* typeck2.c (split_nonconstant_init_1): Return false if we didn't
split out anything.
* g++.dg/cpp1y/constexpr-empty4.C: New test.
From-SVN: r256951
This patch fixes various parts of the code to use a larger type than
int for the character length. Depending on the situation,
HOST_WIDE_INT, size_t, or gfc_charlen_t is appropriate.
Regtested on x86_64-pc-linux-gnu and i686-pc-linux-gnu.
gcc/fortran/ChangeLog:
2018-01-22 Janne Blomqvist <jb@gcc.gnu.org>
PR 78534
PR 83704
* arith.c (gfc_arith_concat): Use size_t for string length.
(gfc_compare_string): Likewise.
(gfc_compare_with_Cstring): Likewise.
* array.c (gfc_resolve_character_array_constructor): Use
HOST_WIDE_INT, gfc_mpz_get_hwi.
* check.c (gfc_check_fe_runtime_error): Use size_t.
* data.c (create_character_initializer): Use HOST_WIDE_INT,
gfc_extract_hwi.
* decl.c (gfc_set_constant_character_len): Use gfc_charlen_t.
(add_init_expr_to_sym): Use HOST_WIDE_INT.
* expr.c (gfc_build_init_expr): Use HOST_WIDE_INT,
gfc_extract_hwi.
(gfc_apply_init): Likewise.
* match.h (gfc_set_constant_character_len): Update prototype.
* primary.c (match_string_constant): Use size_t.
* resolve.c (resolve_ordinary_assign): Use HOST_WIDE_INT,
gfc_mpz_get_hwi.
* simplify.c (init_result_expr): Likewise.
(gfc_simplify_len_trim): Use size_t.
* target-memory.c (gfc_encode_character): Use size_t.
(gfc_target_encode_expr): Use HOST_WIDE_INT, gfc_mpz_get_hwi.
(interpret_array): Use size_t.
(gfc_interpret_character): Likewise.
* target-memory.h (gfc_encode_character): Update prototype.
(gfc_interpret_character): Likewise.
(gfc_target_interpret_expr): Likewise.
* trans-const.c (gfc_build_string_const): Use size_t for length
argument.
(gfc_build_wide_string_const): Likewise.
* trans-const.h (gfc_build_string_const): Likewise.
(gfc_build_wide_string_const): Likewise.
2018-01-22 Janne Blomqvist <jb@gcc.gnu.org>
PR 78534
PR 83704
* gfortran.dg/string_1.f90: Remove printing the length.
From-SVN: r256944
2018-01-22 Richard Biener <rguenther@suse.de>
PR tree-optimization/83963
* graphite-scop-detection.c (scop_detection::get_sese): Delay
including the loop exit block.
(scop_detection::merge_sese): Likewise.
(scop_detection::add_scop): Do it here instead.
* gcc.dg/graphite/pr83963.c: New testcase.
From-SVN: r256943
This patch fixes my earlier test case that fails for arm-none-eabi
with explicit user option for -mfloat-abi which conflict with
the test case options. I have added a guard to skip the test
on those cases.
ChangeLog entries:
*** gcc/testsuite/ChangeLog ***
2018-01-22 Sudakshina Das <sudi.das@arm.com>
* gcc.c-torture/compile/pr82096.c: Add dg-skip-if and
dg-require-effective-target directives.
From-SVN: r256941
This test has needlessly restrictive requirements. It tries to force a soft-float target and tries to run.
This makes it unsupportable for any non-soft-float variant.
In fact, the test can be a run-time test for any target, and only the scan-assembler tests are specific to
-mfloat-abi=soft. So this patch makes the test always runnable and makes the scan-assembler checks predicable
on the the new arm_sotftfloat effective target check.
* doc/sourcebuild.texi (arm_softfloat): Document.
* lib/target-supports.exp (check_effective_target_arm_softfloat):
New procedure.
* gcc.target/arm/copysign_softfloat_1.c: Allow running everywhere.
Adjust scan-assembler checks for soft-float.
From-SVN: r256940
PR gcc/77734
* config/pa/pa.c (pa_function_ok_for_sibcall): Use
targetm.binds_local_p instead of TREE_PUBLIC to check local binding.
Move TARGET_PORTABLE_RUNTIME check after TARGET_64BIT check.
From-SVN: r256934
PR target/80870
* config/sh/rx/rx.c (config/sh/sh_optimize_sett_clrt.cc):
Use INCLUDE_ALGORITHM and INCLUDE_VECTOR instead of direct includes.
From-SVN: r256926
2018-01-20 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/83900
* simplify.c (gfc_simplify_matmul): Set return type correctly.
2018-01-20 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/83900
* gfortran.dg/matmul_18.f90: New test.
From-SVN: r256919
As Jakub says in the PR, the problem here was that the x86/built-in
version of the scatter support was using a bogus scatter_src_dt
when calling vect_get_vec_def_for_stmt_copy (and had since it
was added). The patch uses the vect_def_type from the original
call to vect_is_simple_use instead.
However, Jakub also pointed out that other parts of the load and store
code passed the vector operand rather than the scalar operand to
vect_is_simple_use. That probably works most of the time since
a constant scalar operand should give a constant vector operand,
and likewise for external and internal definitions. But it
definitely seems more robust to pass the scalar operand.
The patch avoids the issue for gather and scatter offsets by
using the cached gs_info.offset_dt. This is safe because gathers
and scatters are never grouped, so there's only one statement operand
to consider. The patch also caches the vect_def_type for mask operands,
which is safe because grouped masked operations share the same mask.
That just leaves the store rhs. We still need to recalculate the
vect_def_type there since different store values in the group can
have different definition types. But since we still have access
to the original scalar operand, it seems better to use that instead.
2018-01-20 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
PR tree-optimization/83940
* tree-vect-stmts.c (vect_truncate_gather_scatter_offset): Set
offset_dt to vect_constant_def rather than vect_unknown_def_type.
(vect_check_load_store_mask): Add a mask_dt_out parameter and
use it to pass back the definition type.
(vect_check_store_rhs): Likewise rhs_dt_out.
(vect_build_gather_load_calls): Add a mask_dt argument and use
it instead of a call to vect_is_simple_use.
(vectorizable_store): Update calls to vect_check_load_store_mask
and vect_check_store_rhs. Use the dt returned by the latter instead
of scatter_src_dt. Use the cached mask_dt and gs_info.offset_dt
instead of calls to vect_is_simple_use. Pass the scalar rather
than the vector operand to vect_is_simple_use when handling
second and subsequent copies of an rhs value.
(vectorizable_load): Update calls to vect_check_load_store_mask
and vect_build_gather_load_calls. Use the cached mask_dt and
gs_info.offset_dt instead of calls to vect_is_simple_use.
gcc/testsuite/
PR tree-optimization/83940
* gcc.dg/torture/pr83940.c: New test.
From-SVN: r256918
PR middle-end/83945
* tree-emutls.c: Include gimplify.h.
(lower_emutls_2): New function.
(lower_emutls_1): If ADDR_EXPR is a gimple invariant and walk_tree
with lower_emutls_2 callback finds some TLS decl in it, unshare_expr
it before further processing.
* gcc.dg/tls/pr83945.c: New test.
From-SVN: r256916
PR c++/81167
* call.c (joust): Use TREE_TYPE (source) if source is
a POINTER_TYPE_P rather than if ! DECL_CONSTRUCTOR_P (w->fn).
* g++.dg/cpp0x/pr81167.C: New test.
From-SVN: r256905
PR debug/81570
PR debug/83728
* dwarf2cfi.c (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define to
INCOMING_FRAME_SP_OFFSET if not defined.
(scan_trace): Add ENTRY argument. If true and
DEFAULT_INCOMING_FRAME_SP_OFFSET != INCOMING_FRAME_SP_OFFSET,
emit a note to adjust the CFA offset.
(create_cfi_notes): Adjust scan_trace callers.
(create_cie_data): Use DEFAULT_INCOMING_FRAME_SP_OFFSET rather than
INCOMING_FRAME_SP_OFFSET in the CIE.
* config/i386/i386.h (DEFAULT_INCOMING_FRAME_SP_OFFSET): Define.
* config/stormy16/stormy16.h (DEFAULT_INCOMING_FRAME_SP_OFFSET):
Likewise.
* doc/tm.texi.in (DEFAULT_INCOMING_FRAME_SP_OFFSET): Document.
* doc/tm.texi: Regenerated.
From-SVN: r256904
* xcoff.c (xcoff_incl_compare): New function.
(xcoff_incl_search): New function.
(xcoff_process_linenos): Use bsearch to find include file.
(xcoff_initialize_fileline): Sort include file information.
From-SVN: r256895
2018-01-19 Tom de Vries <tom@codesourcery.com>
Cesar Philippidis <cesar@codesourcery.com>
PR target/83920
* config/nvptx/nvptx.c (nvptx_single): Fix jit workaround.
* testsuite/libgomp.oacc-c-c++-common/pr83920.c: New test.
* testsuite/libgomp.oacc-fortran/pr83920.f90: New test.
Co-Authored-By: Cesar Philippidis <cesar@codesourcery.com>
From-SVN: r256894
2018-01-19 Martin Liska <mliska@suse.cz>
* predict.def (PRED_LOOP_EXIT): Change from 85 to 89.
(PRED_LOOP_EXIT_WITH_RECURSION): Change from 72 to 78.
(PRED_LOOP_EXTRA_EXIT): Change from 83 to 67.
(PRED_OPCODE_POSITIVE): Change from 64 to 59.
(PRED_TREE_OPCODE_POSITIVE): Change from 64 to 59.
(PRED_CONST_RETURN): Change from 69 to 65.
(PRED_NULL_RETURN): Change from 91 to 71.
(PRED_LOOP_IV_COMPARE_GUESS): Change from 98 to 64.
(PRED_LOOP_GUARD): Change from 66 to 73.
From-SVN: r256888
2018-01-19 Martin Liska <mliska@suse.cz>
* predict.c (predict_insn_def): Add new assert.
(struct branch_predictor): Change type to signed integer.
(test_prediction_value_range): Amend test to cover
PROB_UNINITIALIZED.
* predict.def (PRED_LOOP_ITERATIONS): Use the new constant.
(PRED_LOOP_ITERATIONS_GUESSED): Likewise.
(PRED_LOOP_ITERATIONS_MAX): Likewise.
(PRED_LOOP_IV_COMPARE): Likewise.
* predict.h (PROB_UNINITIALIZED): Define new constant.
From-SVN: r256887
2018-01-19 Martin Liska <mliska@suse.cz>
* analyze_brprob.py: Support new format that can be easily
parsed. Add new column to report.
2018-01-19 Martin Liska <mliska@suse.cz>
* predict.c (dump_prediction): Add new format for
analyze_brprob.py script which is enabled with -details
suboption.
* profile-count.h (precise_p): New function.
From-SVN: r256886
This PR is an odd case in which, due to the low optimisation level,
we enter vectorisation with:
outer1:
x_1 = PHI <x_3(outer2), ...>;
...
inner:
x_2 = 0;
...
outer2:
x_3 = PHI <x_2(inner)>;
These statements are tentatively treated as a double reduction by
vect_force_simple_reduction, but in the end only x_3 and x_2 are marked
as relevant. vect_analyze_loop_operations skips over x_3, leaving the
vectorizable_reduction check to a presumed future test of x_1, which
in this case never happens. We therefore end up vectorising x_2 only
(complete with peeling for niters!) and leave the scalar x_3 in place.
This caused a segfault in the support for fully-masked loops,
since there were no statements that needed masking. Fixed by
checking for that.
But I think this is also a flaw in vect_analyze_loop_operations.
Outer loop vectorisation reduces the number of times that the
inner loop is executed, so it wouldn't necessarily be valid
to leave the scalar x_3 in place for all vectorisable x_2.
There's already code to forbid that when x_1 isn't present:
/* FORNOW: we currently don't support the case that these phis
are not used in the outerloop (unless it is double reduction,
i.e., this phi is vect_reduction_def), cause this case
requires to actually do something here. */
I think we need to do the same if x_1 is present but not relevant.
2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
PR tree-optimization/83922
* tree-vect-loop.c (vect_verify_full_masking): Return false if
there are no statements that need masking.
(vect_active_double_reduction_p): New function.
(vect_analyze_loop_operations): Use it when handling phis that
are not in the loop header.
gcc/testsuite/
PR tree-optimization/83922
* gcc.dg/pr83922.c: New test.
From-SVN: r256885
This testcase ICEd because we converted the initial value of an
induction to the vector element type even for nested inductions.
This isn't necessary because the initial expression is vectorised
normally, and it meant that init_expr was no longer the original
statement operand by the time we called vect_get_vec_def_for_operand.
Also, adding the conversion code here made the existing SLP conversion
redundant.
2018-01-19 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
PR tree-optimization/83914
* tree-vect-loop.c (vectorizable_induction): Don't convert
init_expr or apply the peeling adjustment for inductions
that are nested within the vectorized loop.
gcc/testsuite/
PR tree-optimization/83914
* gcc.dg/vect/pr83914.c: New test.
From-SVN: r256884
These tests are failing for a silly reason. They scan for an occurrence of the NEGS instruction.
NEGS (and NEG in general) is a pre-UAL alias of RSB with an immediate of 0 and we only emit it
in one pattern: *thumb2_negsi2_short in thumb2.md. In all other instances of negation we emit
the modern RSB mnemonic. This causes needless differences in assembly output.
For example, for these testcases we emit NEG when compiling for -march=armv7-a, but for armv7ve
we emit RSB, causing the scan-assembler tests to fail.
This patch updates the *thumb2_negsi2_short pattern to use the RSB mnemonic and
fixes the flaky scan-assembler directives.
These tests now pass for my compiler configured with:
--with-cpu=cortex-a15 --with-fpu=neon-vfpv4 --with-float=hard --with-mode=thumb
Bootstrapped and tested on arm-none-linux-gnueabihf as well.
* config/arm/thumb2.md (*thumb2_negsi2_short): Use RSB mnemonic
instead of NEG.
* gcc.target/arm/negdi-1.c: Remove bogus assembler scan for negs.
* gcc.target/arm/negdi-2.c: Likewise.
* gcc.target/arm/thumb-16bit-ops.c: Replace scan for NEGS with RSBS.
--This line,gand those below, will be ignored--
M gcc/config/arm/thumb2.md
M gcc/ChangeLog
M gcc/testsuite/gcc.target/arm/thumb-16bit-ops.c
M gcc/testsuite/gcc.target/arm/negdi-1.c
M gcc/testsuite/gcc.target/arm/negdi-2.c
M gcc/testsuite/ChangeLog
From-SVN: r256881
The scan-assembler tests here check for MOVS for Thumb1 and MOV for Thumb2,
but in fact there's no reason why we wouldn't generate MOVS for Thumb2 as well,
it really depends on a lot of optimisation decisions.
The only behaviour we want to test is that we move a 0 constant into a register
only once, which can be achieved with either MOV or MOVS.
Simplify the check by always checking for either MOV or MOVS.
* gcc.target/arm/pr40956.c: Adjust scan-assembler pattern.
From-SVN: r256880
This testcase tests 32-bit ARM state functionality, so add the -marm to make it explicit
as well as to avoid Thumb1 hard-float errors for certain toolchain configurations.
* gcc.target/arm/pr79058.c: Add arm_arm_ok check and -marm to options.
From-SVN: r256878
This makes no difference on most systems, because <sys/resource.h>
renames the type appropriately anyhow, but apparently it makes a
difference on AIX.
Reviewed-on: https://go-review.googlesource.com/88076
From-SVN: r256877