[gcc]
2016-08-23 Michael Meissner <meissner@linux.vnet.ibm.com>
* config/rs6000/rs6000.c (rs6000_expand_vector_init): Set
initialization of all 0's to the 0 constant, instead of directly
generating XOR. Add support for V4SImode vector initialization on
64-bit systems with direct move, and rework the ISA 3.0 V4SImode
initialization. Change variables used in V4SFmode vector
intialization. For V4SFmode vector splat on ISA 3.0, make sure
any memory addresses are in index form. Add support for using
VSPLTH/VSPLTB to initialize vector short and vector char vectors
with all of the same element.
(regno_or_subregno): New helper function to return a register
number for either REG or SUBREG.
(rs6000_adjust_vec_address): Do not generate ADDI <reg>,R0,<num>.
Use regno_or_subregno where possible.
(rs6000_split_v4si_init_di_reg): New helper function to build up a
DImode value from two SImode values in order to generate V4SImode
vector initialization on 64-bit systems with direct move.
(rs6000_split_v4si_init): Split up the insns for a V4SImode vector
initialization.
(rtx_is_swappable_p): V4SImode vector initialization insn is not
swappable.
* config/rs6000/rs6000-protos.h (rs6000_split_v4si_init): Add
declaration.
* config/rs6000/vsx.md (VSX_SPLAT_I): New mode iterators and
attributes to initialize V8HImode and V16QImode vectors with the
same element.
(VSX_SPLAT_COUNT): Likewise.
(VSX_SPLAT_SUFFIX): Likewise.
(UNSPEC_VSX_VEC_INIT): New unspec.
(vsx_concat_v2sf): Eliminate using 'preferred' register classes.
Allow SFmode values to come from Altivec registers.
(vsx_init_v4si): New insn/split for V4SImode vector initialization
on 64-bit systems with direct move.
(vsx_splat_<mode>, VSX_W iterator): Rework V4SImode and V4SFmode
vector initializations, to allow V4SImode vector initializations
on 64-bit systems with direct move.
(vsx_splat_v4si): Likewise.
(vsx_splat_v4si_di): Likewise.
(vsx_splat_v4sf): Likewise.
(vsx_splat_v4sf_internal): Likewise.
(vsx_xxspltw_<mode>, VSX_W iterator): Eliminate using 'preferred'
register classes.
(vsx_xxspltw_<mode>_direct, VSX_W iterator): Likewise.
(vsx_vsplt<VSX_SPLAT_SUFFIX>_di): New insns to support
initializing V8HImode and V16QImode vectors with the same
element.
* config/rs6000/rs6000.h (TARGET_DIRECT_MOVE_64BIT): Disallow
optimization if -maltivec=be.
[gcc/testsuite]
2016-08-23 Michael Meissner <meissner@linux.vnet.ibm.com>
* gcc.target/powerpc/vec-init-1.c: Add tests where the vector is
being created from pointers to memory locations.
* gcc.target/powerpc/vec-init-2.c: Likewise.
From-SVN: r239712
Switch to a new method for determining the order in which import init
functions are invoked: build an init fcn dependence DAG and walk the DAG
to rewrite/adjust priorities to account for discrepancies introduced by
"go test".
This patch includes a change to the export data format generated
by gccgo. Older versions of gccgo will not be able to read object files
produced by a newer gccgo, but the new gcc will still be able to read
old object files.
Fixesgolang/go#15738.
Reviewed-on: https://go-review.googlesource.com/25301
From-SVN: r239708
2016-08-23 Richard Biener <rguenther@suse.de>
PR tree-optimization/77286
* tree-vect-loop.c (vect_analyze_loop_form_1): Do not modify
the CFG here.
(vect_transform_loop): Split exit edges of loop and scalar
loop if required and at the appropriate time.
From-SVN: r239700
PR libstdc++/71771
* include/bits/stl_iterator.h
(operator-(reverse_iterator<Iter>, reverse_iterator<Iter>): Only
define for C++98 mode.
(operator-(move_iterator<Iter>, move_iterator<Iter>): Don't define.
* testsuite/24_iterators/headers/iterator/synopsis.cc: Use
-std=gnu++98.
* testsuite/24_iterators/headers/iterator/synopsis_c++11.cc: New test.
* testsuite/24_iterators/headers/iterator/synopsis_c++14.cc: New test.
* testsuite/24_iterators/headers/iterator/synopsis_c++17.cc: New test.
* testsuite/24_iterators/move_iterator/greedy_ops.cc: Don't test
difference operator.
* testsuite/24_iterators/reverse_iterator/greedy_ops.cc: Only test
difference operator for C++98.
* testsuite/24_iterators/reverse_iterator/71771.cc: New test.
From-SVN: r239691
* get_dynamic_stack_size is passed a SIZE of a data block (which is
allocated elsewhere), the SIZE_ALIGN of the SIZE (i.e. the alignment
of the underlying memory units (e.g. 32 bytes split into 4 times 8
bytes = 64 bit alignment) and the REQUIRED_ALIGN of the data portion
of the allocated memory.
* Assuming the function is called with SIZE = 2, SIZE_ALIGN = 8 and
REQUIRED_ALIGN = 64 it first adds 7 bytes to SIZE -> 9. This is
what is needed to have two bytes 8-byte-aligned at some memory
location without any known alignment.
* Finally round_push is called to round up SIZE to a multiple of the
stack slot size.
The key to understanding this is that the function assumes that
STACK_DYNMAIC_OFFSET is completely unknown at the time its called
and therefore it does not make assumptions about the alignment of
STACKPOINTER + STACK_DYNMAIC_OFFSET. The latest patch simply
hard-codes that SP + SDO is supposed to be aligned to at least
stack slot size (and does that in a very complicated way). Since
there is no guarantee that this is the case on all targets, the
patch is broken. It may miscalculate a SIZE that is too small in
some cases.
However, on many targets there is some guarantee about the
alignment of SP + SDO even if the actual value of SDO is unknown.
On s390x it's always 8-byte-aligned (stack slot size). So the
right fix should be to add knowledge about the target's guaranteed
alignment of SP + SDO to the function. I'm right now testing a
much simpler patch that uses
REGNO_POINTER_ALIGN(VIRTUAL_STACK_DYNAMIC_REGNUM) as the
alignment.
gcc/ChangeLog:
2016-08-23 Dominik Vogt <vogt@linux.vnet.ibm.com>
* explow.c (get_dynamic_stack_size): Take known alignment of stack
pointer + STACK_DYNAMIC_OFFSET into account when calculating the
size needed. Correct a typo in a comment.
From-SVN: r239688
gcc/testsuite/ChangeLog:
2016-08-23 Dominik Vogt <vogt@linux.vnet.ibm.com>
* gcc.target/s390/insv-1.c: Fix test when running with -m31.
From-SVN: r239687
THe attached patch improves checking of teh results of the subtests
"a" and "f". As they share the same "vone" instruction, the duplicate
scan-assembler-times was bogus. Moved "f" to a separate function to
fix this. Also double check that no extra "vgmf" instructions are
used.
gcc/testsuite/ChangeLog:
2016-08-23 Dominik Vogt <vogt@linux.vnet.ibm.com>
* gcc.target/s390/zvector/vec-genmask-1.c: Improve result
verification.
From-SVN: r239686
Split ~b & a to (b & a) ^ a. This is benefitial on z Systems since we
otherwise need a big -1 constant to be loaded for the ~b.
gcc/ChangeLog:
2016-08-23 Dominik Vogt <vogt@linux.vnet.ibm.com>
* config/s390/s390.md ("*andc_split"): New splitter for and with
complement.
gcc/testsuite/ChangeLog:
2016-08-23 Dominik Vogt <vogt@linux.vnet.ibm.com>
* gcc.target/s390/md/andc-splitter-1.c: New test case.
* gcc.target/s390/md/andc-splitter-2.c: Likewise.
From-SVN: r239685
2016-08-23 Richard Biener <rguenther@suse.de>
PR tree-optimization/27336
* tree-vrp.c (infer_value_range): Handle stmts that can throw
by looking for a non-EH edge.
(process_assert_insertions_for): Likewise.
* c-c++-common/pr27336.c: New testcase.
From-SVN: r239684
When lowering method expressions of the form "P.M" where
P is a pointer type (e.g. "type P *T") make sure we examine
the method set of P and not T during method lookup.
Fixesgolang/go#15722.
Reviewed-on: https://go-review.googlesource.com/24843
From-SVN: r239675
2016-08-22 Steven G. Kargl <kargl@gcc.gnu.org>
Bud Davis <jmdavis@link.com>
PR fortran/60774
* parse.c (next_free,next_fixed): Issue error for statement label
without a statement.
2016-08-22 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/60774
* gfortran.dg/empty_label.f: Adjust test for new error message.
* gfortran.dg/empty_label.f90: Ditto.
* gfortran.dg/empty_label_typedecl.f90: Ditto.
* gfortran.dg/label_3.f90: Deleted (redundant with empty_label.f90).
* gfortran.dg/warnings_are_errors_1.f90: Remove invalid statement label.
Co-Authored-By: Bud Davis <jmdavis@link.com>
From-SVN: r239668
2016-08-22 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/61318
* interface.c (compare_parameter): Use better locus for error message.
2016-08-22 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/61318
* gfortran.dg/pr61318.f90: New test.
From-SVN: r239667
2016-08-22 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/77260
* gcc/fortran/trans-decl.c (generate_local_decl): Suppress warning
for unused variable if symbol is entry point.
2016-08-22 Steven G. Kargl <kargl@gcc.gnu.org>
PR fortran/77260
* gfortran.dg/pr77260_1.f90: New test.
* gfortran.dg/pr77260_2.f90: Ditto.
From-SVN: r239666
The __builtin_isinf_sign folding uses a type-specific signbit built-in
function, meaning it only works for the types float, double and long
double, not for types such as _FloatN, _FloatNx, __float128. Since
the signbit built-in function is now type-generic, that can be used
unconditionally, much as the code uses the type-generic isinf built-in
function unconditionally, and this patch makes it do so, thereby
enabling __builtin_isinf_sign (which glibc uses to expand the isinf
macro since that macro in glibc traditionally provided the stronger
guarantees about the return value given by __builtin_isinf_sign) to
work for all floating-point types.
The test gcc.dg/torture/builtin-isinf_sign-1.c needs updating because
it tests that comparisons of calls to __builtin_isinf_sign to
conditional expressions involving __builtin_isinf and
__builtin_signbit* get optimized away, and with a change of what
particular built-in function for signbit is used, GCC doesn't notice
the expressions with type-generic and non-type-generic built-in
functions are equivalent at -O0 or -O1 (it does optimize away the
original test at -O2).
Bootstrapped with no regressions on x86_64-pc-linux-gnu.
PR middle-end/77269
gcc:
* builtins.c (fold_builtin_classify): Use builtin_decl_explicit
(BUILT_IN_SIGNBIT) to expand __builtin_isinf_sign.
gcc/testsuite:
* gcc.dg/torture/builtin-isinf_sign-1.c: Use __builtin_signbit not
__builtin_signbitf and __builtin_signbitl in expected generic
expansion.
* gcc.dg/torture/float128-tg-2.c, gcc.dg/torture/float128x-tg-2.c,
gcc.dg/torture/float16-tg-2.c, gcc.dg/torture/float32-tg-2.c,
gcc.dg/torture/float32x-tg-2.c, gcc.dg/torture/float64-tg-2.c,
gcc.dg/torture/float64x-tg-2.c, gcc.dg/torture/floatn-tg-2.h: New
tests.
From-SVN: r239665
gcc/ChangeLog:
* print-tree.c (print_node) [VECTOR_CST]: Coalesce the output of
identical consecutive elements.
[SSA_NAME]: Print the name's def stmt on its own line. When printing
the node's def stmt, avoid printing an unwanted trailing newline by
replacing the call to print_gimple_stmt() with its inlined body and
adjusting it to not set pp_needs_newline and to call pp_flush()
instead of pp_newline_and_flush().
From-SVN: r239661
This patch adds a minimal set of built-in functions for the new
_FloatN and _FloatNx types.
The functions added are __builtin_fabs*, __builtin_copysign*,
__builtin_huge_val*, __builtin_inf*, __builtin_nan* and
__builtin_nans* (where * = fN or fNx). That is, 42 new entries are
added to the enum of built-in functions and the associated array of
decls, where not all of them are actually supported on any one target.
These functions are believed to be sufficient for libgcc (complex
multiplication and division use __builtin_huge_val*,
__builtin_copysign* and __builtin_fabs*) and for glibc (which also
depends on complex multiplication from libgcc, as well as using such
functions itself). The basic target-independent support for folding /
expanding calls to these built-in functions is wired up, so those for
constants can be used in static initializers, and the fabs and
copysign built-ins can always be expanded to bit-manipulation inline
(for any format setting signbit_ro and signbit_rw, which covers all
formats supported for _FloatN and _FloatNx), although insn patterns
for fabs (abs<mode>2) and copysign (copysign<mode>3) will be used when
available and may result in more optimal code.
The complex multiplication and division functions in libgcc rely on
predefined macros (defined with -fbuilding-libgcc) to say what the
built-in function suffixes to use with a particular mode are. This
patch updates that code accordingly, where previously it involved a
hack supposing that machine-specific suffixes for constants were also
suffixes for built-in functions.
As with the main _FloatN / _FloatNx patch, this patch does not update
code dealing only with optimizations that currently has cases only
covering float, double and long double, though some such cases are
straightforward and may be covered in a followup patch.
The functions are defined with DEF_GCC_BUILTIN, so calls to the TS
18661-3 functions such as fabsf128 and copysignf128, without the
__builtin_, will not be optimized. As noted in the original _FloatN /
_FloatNx patch submission, in principle the bulk of the libm functions
that have built-in versions should have those versions extended to
cover the new types, but that would require more consideration of the
effects of increasing the size of the enum and initializing many more
functions at startup.
I don't know whether target-specific built-in functions can readily be
made into aliases for target-independent functions, but if they can,
it would make sense to do so for the x86, ia64 and rs6000 *q functions
corresponding to these, so that they can benefit from the
architecture-independent folding logic and from any optimizations
enabled for these functions in future, and so that less
target-specific code is needed to support them.
Bootstrapped with no regressions on x86_64-pc-linux-gnu.
gcc:
* tree.h (CASE_FLT_FN_FLOATN_NX, float16_type_node)
(float32_type_node, float64_type_node, float32x_type_node)
(float128x_type_node): New macros.
* builtin-types.def (BT_FLOAT16, BT_FLOAT32, BT_FLOAT64)
(BT_FLOAT128, BT_FLOAT32X, BT_FLOAT64X, BT_FLOAT128X)
(BT_FN_FLOAT16, BT_FN_FLOAT32, BT_FN_FLOAT64, BT_FN_FLOAT128)
(BT_FN_FLOAT32X, BT_FN_FLOAT64X, BT_FN_FLOAT128X)
(BT_FN_FLOAT16_FLOAT16, BT_FN_FLOAT32_FLOAT32)
(BT_FN_FLOAT64_FLOAT64, BT_FN_FLOAT128_FLOAT128)
(BT_FN_FLOAT32X_FLOAT32X, BT_FN_FLOAT64X_FLOAT64X)
(BT_FN_FLOAT128X_FLOAT128X, BT_FN_FLOAT16_CONST_STRING)
(BT_FN_FLOAT32_CONST_STRING, BT_FN_FLOAT64_CONST_STRING)
(BT_FN_FLOAT128_CONST_STRING, BT_FN_FLOAT32X_CONST_STRING)
(BT_FN_FLOAT64X_CONST_STRING, BT_FN_FLOAT128X_CONST_STRING)
(BT_FN_FLOAT16_FLOAT16_FLOAT16, BT_FN_FLOAT32_FLOAT32_FLOAT32)
(BT_FN_FLOAT64_FLOAT64_FLOAT64, BT_FN_FLOAT128_FLOAT128_FLOAT128)
(BT_FN_FLOAT32X_FLOAT32X_FLOAT32X)
(BT_FN_FLOAT64X_FLOAT64X_FLOAT64X)
(BT_FN_FLOAT128X_FLOAT128X_FLOAT128X): New type definitions.
* builtins.def (DEF_GCC_FLOATN_NX_BUILTINS): New macro.
(copysign, fabs, huge_val, inf, nan, nans): Use it.
* builtins.c (expand_builtin): Use CASE_FLT_FN_FLOATN_NX for fabs
and copysign.
(fold_builtin_0): Use CASE_FLT_FN_FLOATN_NX for inf and huge_val.
(fold_builtin_1): Use CASE_FLT_FN_FLOATN_NX for fabs.
* doc/extend.texi (Other Builtins): Document these built-in
functions.
* fold-const-call.c (fold_const_call): Use CASE_FLT_FN_FLOATN_NX
for nan and nans.
gcc/c-family:
* c-family/c-cppbuiltin.c (c_cpp_builtins): Check _FloatN and
_FloatNx types for suffixes for built-in functions.
gcc/testsuite:
* gcc.dg/torture/float128-builtin.c,
gcc.dg/torture/float128-ieee-nan.c,
gcc.dg/torture/float128x-builtin.c,
gcc.dg/torture/float128x-nan.c, gcc.dg/torture/float16-builtin.c,
gcc.dg/torture/float16-nan.c, gcc.dg/torture/float32-builtin.c,
gcc.dg/torture/float32-nan.c, gcc.dg/torture/float32x-builtin.c,
gcc.dg/torture/float32x-nan.c, gcc.dg/torture/float64-builtin.c,
gcc.dg/torture/float64-nan.c, gcc.dg/torture/float64x-builtin.c,
gcc.dg/torture/float64x-nan.c, gcc.dg/torture/floatn-builtin.h,
gcc.dg/torture/floatn-nan.h: New tests.
From-SVN: r239658
gcc/testsuite:
* gcc.dg/torture/fp-int-convert-float128x-timode.c,
gcc.dg/torture/fp-int-convert-float32x-timode.c,
gcc.dg/torture/fp-int-convert-float64x-timode.c: Correct type
names in calls to TEST_I_F.
From-SVN: r239655
gcc/ChangeLog:
2016-08-20 Kugan Vivekanandarajah <kuganv@linaro.org>
* tree-vrp.c (vrp_visit_assignment_or_call): Changed to Return VR.
(vrp_visit_cond_stmt): Just sets TAKEN_EDGE_P.
(vrp_visit_switch_stmt): Likewise.
(extract_range_from_stmt): Factored out from vrp_visit_stmt.
(extract_range_from_phi_node): Factored out from vrp_visit_phi_stmt.
(vrp_visit_stmt): Use extract_range_from_stmt.
(vrp_visit_phi_node): Use extract_range_from_phi_node.
From-SVN: r239639