Comparing to the cc0 version of the CRIS port, I ran a few
microbenchmarks, for example gcc.c-torture/execute/arith-rand.c,
where there's sometimes an addition between an operation of
interest and the test on the result.
Unfortunately this patch doesn't remedy all the performance
regression for that program. But, this patch by itself helps
and makes sense to commit separately: lots of addi.b in
previously empty delay-slots, with functions shortened by one or
a few insns, in libgcc. I had an experience with the
reload-related caveat of % on constraints, which is "fixed"
documentationwise since long (soon 15 years ago;
be3914df4cc8/r105517). I removed an even older related FIXME.
gcc:
PR target/93372
* config/cris/cris.md ("*add<mode>3_addi"): New splitter.
("*addi_b_<mode>"): New pattern.
("*addsi3<setnz>"): Remove stale %-related comment.
gcc/testsuite:
PR target/93372
* gcc.target/cris/pr93372-45.c: New test.
Whoops. This little gem had the effect of making the output
operand (0) constraints disappear but not the input operand (1)
constraints for define_subst:ed patterns, probably because
there's another (match_dup 1) in the output template (not
investigated).
That went surprisingly unnoticed until I added a pass leaning
just a little bit harder on the define_subst:ed patterns and
then only by the libgfortran library generating assembly with
nominally incorrect syntax. (There was a move to a special
register from a general register, and it incorrectly matched a
pattern affecting condition codes.)
gcc:
* config/cris/cris.md ("setnz_subst", "setnz_subst", "setcc_subst"):
Use match_dup in output template, not match_operand.
This eliminates the visited bitmap and makes whether a to be processed
block goes to the next or the current iteration only depend on its
position in RPO order rather than on whether it was visited in the
current iteration. As optimization single-BB iteration is processed
immediately.
2020-07-10 Richard Biener <rguenther@suse.de>
* var-tracking.c (bb_heap_node_t): Remove unused typedef.
(vt_find_locations): Eliminate visited bitmap in favor of
RPO order check. Dump statistics about the number of
local BB dataflow computes.
The store to the whole of each volatile object was picked apart
like there had been an individual assignment to each of the
fields. Reads were added as part of that; see PR for details.
The reads from volatile memory were a clear bug; individual
stores questionable. A separate patch clarifies the docs.
gcc:
2020-07-09 Richard Biener <rguenther@suse.de>
PR middle-end/94600
* expr.c (expand_constructor): Make a temporary also if we're
storing to volatile memory.
gcc/testsuite:
2020-07-09 Hans-Peter Nilsson <hp@axis.com>
PR middle-end/94600
* gcc.dg/pr94600-1.c, gcc.dg/pr94600-2.c, gcc.dg/pr94600-3.c,
gcc.dg/pr94600-4.c, gcc.dg/pr94600-5.c, gcc.dg/pr94600-6.c,
gcc.dg/pr94600-7.c, gcc.dg/pr94600-8.c: New tests.
Combine pass could recognize the pattern defined and split it in split1,
this patch could optimize:
21: r130:DI=r133:DI<<0x20
11: {r129:DI=zero_extend(unspec[[r145:DI]] 87);clobber scratch;}
22: r134:DI=r130:DI|r129:DI
to
21: {r149:DI=zero_extend(unspec[[r145:DI]] 87);clobber scratch;}
22: r134:DI=r149:DI&0xffffffff|r133:DI<<0x20
rldimi is generated instead of sldi+or.
gcc/ChangeLog:
2020-07-13 Xionghu Luo <luoxhu@linux.ibm.com>
* config/rs6000/rs6000.md (rotl_unspec): New
define_insn_and_split.
gcc/testsuite/ChangeLog:
2020-07-13 Xionghu Luo <luoxhu@linux.ibm.com>
* gcc.target/powerpc/vector_float.c: New test.
Move V4SF to V4SI, init vector like V4SI and move to V4SF back.
Better instruction sequence could be generated on Power9:
lfs + xxpermdi + xvcvdpsp + vmrgew
=>
lwz + (sldi + or) + mtvsrdd
With the patch followed, it could be continue optimized to:
lwz + rldimi + mtvsrdd
The point is to use lwz to avoid converting the single-precision to
double-precision upon load, pack four 32-bit data into one 128-bit
register directly.
gcc/ChangeLog:
2020-07-13 Xionghu Luo <luoxhu@linux.ibm.com>
* config/rs6000/rs6000.c (rs6000_expand_vector_init):
Move V4SF to V4SI, init vector like V4SI and move to V4SF back.
Since -fpatchable-function-entry is only supported on Linux and used by
Linux kernel, require Linux target for PR target/93492 tests.
PR target/93492
* gcc.target/i386/pr93492-1.c: Require Linux target.
* gcc.target/i386/pr93492-2.c: Likewise.
* gcc.target/i386/pr93492-3.c: Likewise.
* gcc.target/i386/pr93492-4.c: Likewise.
* gcc.target/i386/pr93492-5.c: Likewise.
This patch improves the RTL that the middle-end generates for testing
signed overflow following a widening multiplication. During this
expansion the middle-end generates a truncation which can get used
multiple times. Placing this intermediate value in a pseudo register
reduces the amount of code generated on platforms where this truncation
requires an explicit instruction.
2020-07-11 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog:
* internal-fn.c (expand_mul_overflow): When checking for signed
overflow from a widening multiplication, we access the truncated
lowpart RES twice, so keep this value in a pseudo register.
In the test case, there was a warning about INTENT where an EXTERNAL
masked an interface in an outer scope, when the location of the
symbol was not set, leading to an ICE.
Two problems, two-part solution: It makes no sense to warn about
INTENT for artificially generated formal argument lists, and the
location should be set.
gcc/fortran/ChangeLog:
PR fortran/96073
* frontend-passes.c (check_externals_procedure): Add locus
information for new_sym.
* interface.c (gfc_check_dummy_characteristics): Do not warn
about INTENT for artificially generated variables.
gcc/testsuite/ChangeLog:
PR fortran/96073
* gfortran.dg/interface_48.f90: New test.
The range infrastructure has code to decompose POLY_INT_CST ranges
to worst-case integer bounds. However, it had the fundamental flaw
(obvious in hindsight) that it applied to anti-ranges too, meaning
that a range 2+2X would end up with a range of ~[2, +INF], i.e.
[-INF, 1]. This patch decays to varying in that case instead.
I'm still a bit uneasy about this. ISTM that in terms of
generality:
SSA_NAME => POLY_INT_CST => INTEGER_CST
=> ADDR_EXPR
I.e. an SSA_NAME could store a POLY_INT_CST and a POLY_INT_CST
could store an INTEGER_CST (before canonicalisation). POLY_INT_CST
is also “as constant as” ADDR_EXPR (well, OK, only some ADDR_EXPRs
are run-time rather than link-time constants, whereas all POLY_INT_CSTs
are, but still). So it seems like we should at least be able to treat
POLY_INT_CST as symbolic. On the other hand, I don't have any examples
in which that would be useful.
gcc/
PR tree-optimization/96146
* value-range.cc (value_range::set): Only decompose POLY_INT_CST
bounds to integers for VR_RANGE. Decay to VR_VARYING for anti-ranges
involving POLY_INT_CSTs.
gcc/testsuite/
PR tree-optimization/96146
* gcc.target/aarch64/sve/acle/general/pr96146.c: New test.
Some square brackets were missing escape characters, causing DejaGnu to
try and call a proc with the name "at".
gcc/testsuite/ChangeLog:
* gcc.target/riscv/read-thread-pointer.c: Fix escaping on
regular expression.
get_constant_section() can be passed constant-like non-DECLs, such as
CONSTRUCTOR or STRING_CST, which make DECL_SECTION_NAME unhappy
(asserted in symtab_node::get). This patch ensures that xcoff select
section only invokes resolve_unique_section() for DECLs.
gcc/ChangeLog
2020-07-10 David Edelsohn <dje.gcc@gmail.com>
* config/rs6000/rs6000.c (rs6000_xcoff_select_section): Only
create named section for VAR_DECL or FUNCTION_DECL.
C2X adds BOOL_MAX and BOOL_WIDTH macros to <limits.h>. As GCC only
supports values 0 and 1 for _Bool (regardless of the number of bits in
the representation, other bits are padding bits and if any of them are
nonzero, the representation is a trap representation), the values of
those macros can just be hardcoded directly in <limits.h> rather than
needing corresponding predefined macros.
Bootstrapped with no regressions on x86_64-pc-linux-gnu.
gcc/
* glimits.h [__STDC_VERSION__ > 201710L] (BOOL_MAX, BOOL_WIDTH):
New macros.
gcc/testsuite/
* gcc.dg/c11-bool-limits-1.c, gcc.dg/c2x-bool-limits-1.c: New
tests.
This patch adds execution tests that use the MMA builtins and
check for the right answer, and new tests that checks whether
__builtin_cpu_supports and __builtin_cpu_is return sane
answers for power10.
2020-06-30 Rajalakshmi Srinivasaraghavan <rajis@linux.vnet.ibm.com>
Aaron Sawdey <acsawdey@linux.ibm.com>
gcc/testsuite/
* gcc.target/powerpc/p10-identify.c: New file.
* gcc.target/powerpc/p10-arch31.c: New file.
* gcc.target/powerpc/mma-single-test.c: New file.
* gcc.target/powerpc/mma-double-test.c: New file.
Currently if requires_stack_frame_p() returns true for some insn, the
shrink-wrapping debug output contains only the number of a block containing
that insn.
But it is very useful to see the particular insn that requires the prologue.
Let's call print_rtl_single to display that insn in the following pass dump.
gcc/
* shrink-wrap.c (try_shrink_wrapping): Improve debug output.
In some cases, expand_expr_real_2 prefers to use the mode of the
caller-suggested target instead of the mode of the expression when
passing values to reduce_to_bit_field_precision. E.g.:
else if (target == 0)
op0 = convert_to_mode (mode, op0,
TYPE_UNSIGNED (TREE_TYPE
(treeop0)));
else
{
convert_move (target, op0,
TYPE_UNSIGNED (TREE_TYPE (treeop0)));
op0 = target;
}
where “op0” might not have “mode” for the “else” branch,
but does for all the others.
reduce_to_bit_field_precision discards the suggested target if it
has the wrong mode. This patch moves that to expand_expr_real_2
instead (conditional on reduce_bit_field).
gcc/
PR middle-end/96151
* expr.c (expand_expr_real_2): When reducing bit fields,
clear the target if it has a different mode from the expression.
(reduce_to_bit_field_precision): Don't do that here. Instead
assert that the target already has the correct mode.
This is an arm version of aarch64 patch r11-1741. The approach
is essentially identical, not much more than s/aarch64/arm/.
To recap, PR95726 is about template look-up for things like:
foo<float vecf __attribute__((vector_size(16)))>
foo<float32x4_t>
The immediate cause of the problem is that the hash function usually
returns different hashes for these types, yet the equality function
thinks they are equal. This then raises the question of how the types
are supposed to be treated.
The answer we chose for AArch64 was that the GNU vector type should
be treated as distinct from float32x4_t, but that each type should
implicitly convert to the other.
This would mean that, as far as the PR is concerned, the hashing
function is right to (sometimes) treat the types differently and
the equality function is wrong to treat them as the same.
The most obvious way to enforce the type difference is to use a
target-specific type attribute. That on its own is enough to fix
the PR. The difficulty is deciding whether the knock-on effects
are acceptable.
One obvious effect is that GCC then rejects:
typedef float vecf __attribute__((vector_size(16)));
vecf x;
float32x4_t &z = x;
on the basis that the types are no longer reference-compatible.
For AArch64 we took the approach that this was the correct behaviour.
It is also consistent with current Clang.
A trickier question is whether:
vecf x;
float32x4_t y;
… c ? x : y …
should be valid, and if so, what its type should be [PR92789].
As explained in the comment in the testcase, GCC and Clang both
accepted this, but GCC chose the “then” type while Clang chose
the “else” type. This can lead to different mangling for (probably
artificial) corner cases, as seen for “sel1” and “sel2” in the
testcase.
Adding the attribute makes GCC reject the conditional expression
as ambiguous. For AArch64 we took the approach that this too is
the correct behaviour, for the reasons described in the testcase.
However, it does seem to have the potential to break existing code.
gcc/
PR target/92789
PR target/95726
* config/arm/arm.c (arm_attribute_table): Add
"Advanced SIMD type".
(arm_comp_type_attributes): Check that the "Advanced SIMD type"
attributes are equal.
* config/arm/arm-builtins.c: Include stringpool.h and
attribs.h.
(arm_mangle_builtin_vector_type): Use the mangling recorded
in the "Advanced SIMD type" attribute.
(arm_init_simd_builtin_types): Add an "Advanced SIMD type"
attribute to each Advanced SIMD type, using the mangled type
as the attribute's single argument.
gcc/testsuite/
PR target/92789
PR target/95726
* g++.target/arm/pr95726.C: New test.
This patch adjusts how dynamic reference counts work so that they match
the semantics of the source program more closely, instead of representing
"excess" reference counts beyond those that represent pointers in the
internal libgomp splay-tree data structure. This allows some corner
cases to be handled more gracefully.
2020-07-10 Julian Brown <julian@codesourcery.com>
Thomas Schwinge <thomas@codesourcery.com>
libgomp/
* libgomp.h (struct splay_tree_key_s): Change virtual_refcount to
dynamic_refcount.
(struct gomp_device_descr): Remove GOMP_MAP_VARS_OPENACC_ENTER_DATA.
* oacc-mem.c (acc_map_data): Substitute virtual_refcount for
dynamic_refcount.
(acc_unmap_data): Update comment.
(goacc_map_var_existing, goacc_enter_datum): Adjust for
dynamic_refcount semantics.
(goacc_exit_datum_1, goacc_exit_datum): Re-add some error checking.
Adjust for dynamic_refcount semantics.
(goacc_enter_data_internal): Implement "present" case of dynamic
memory-map handling here. Update "non-present" case for
dynamic_refcount semantics.
(goacc_exit_data_internal): Use goacc_exit_datum_1.
* target.c (gomp_map_vars_internal): Remove
GOMP_MAP_VARS_OPENACC_ENTER_DATA handling. Update for dynamic_refcount
handling.
(gomp_unmap_vars_internal): Remove virtual_refcount handling.
(gomp_load_image_to_device): Substitute dynamic_refcount for
virtual_refcount.
* testsuite/libgomp.oacc-c-c++-common/pr92843-1.c: Remove XFAILs.
* testsuite/libgomp.oacc-c-c++-common/refcounting-1.c: New test.
* testsuite/libgomp.oacc-c-c++-common/refcounting-2.c: New test.
* testsuite/libgomp.oacc-c-c++-common/struct-3-1-1.c: New test.
* testsuite/libgomp.oacc-fortran/deep-copy-6.f90: Remove XFAILs and
trace output.
* testsuite/libgomp.oacc-fortran/deep-copy-6-no_finalize.F90: Remove
trace output.
* testsuite/libgomp.oacc-fortran/dynamic-incr-structural-1.f90: New
test.
* testsuite/libgomp.oacc-c-c++-common/structured-dynamic-lifetimes-4.c:
Remove stale comment.
* testsuite/libgomp.oacc-fortran/mdc-refcount-1-1-1.f90: Remove XFAILs.
* testsuite/libgomp.oacc-fortran/mdc-refcount-1-1-2.F90: Likewise.
* testsuite/libgomp.oacc-fortran/mdc-refcount-1-2-1.f90: Likewise.
* testsuite/libgomp.oacc-fortran/mdc-refcount-1-2-2.f90: Likewise.
* testsuite/libgomp.oacc-fortran/mdc-refcount-1-3-1.f90: Likewise.
* testsuite/libgomp.oacc-fortran/mdc-refcount-1-4-1.f90: Adjust XFAIL.
Co-Authored-By: Thomas Schwinge <thomas@codesourcery.com>
This patch factors out the parts of goacc_enter_datum and
goacc_exit_datum that can be shared with goacc_enter_data_internal
and goacc_exit_data_internal respectively (in the next patch),
without overloading function return values or complicating code paths
unnecessarily.
2020-07-10 Julian Brown <julian@codesourcery.com>
Thomas Schwinge <thomas@codesourcery.com>
libgomp/
* oacc-mem.c (goacc_map_var_existing): New function.
(goacc_enter_datum): Use above function.
(goacc_exit_datum_1): New function.
(goacc_exit_datum): Use above function.
Co-Authored-By: Thomas Schwinge <thomas@codesourcery.com>
gcc/ChangeLog:
PR target/95581
* config/rs6000/rs6000-call.c: Add new type v16qi_ftype_pcvoid.
(altivec_init_builtins) Change __builtin_altivec_mask_for_load to use
v16qi_ftype_pcvoid with correct number of parameters.
Another thing newly allowed by P1907R1. The ABI group has discussed
representing unions with designated initializers, and has separately
specified how to represent designators; this patch implements both.
gcc/cp/ChangeLog:
* tree.c (structural_type_p): Allow unions.
* mangle.c (write_expression): Express unions with a designator.
libiberty/ChangeLog:
* cp-demangle.c (cplus_demangle_operators): Add di, dx, dX.
(d_expression_1): Handle di and dX.
(is_designated_init, d_maybe_print_designated_init): New.
(d_print_comp_inner): Use d_maybe_print_designated_init.
* testsuite/demangle-expected: Add designator tests.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/nontype-class-union1.C: New test.
P1907R1 made various adjustments to non-type template parameters, notably
introducing the notion of "structural type". I implemented an early version
of that specification in r10-4426, but it was adjusted in the final paper to
allow more. This patch implements allowing template parameters of
floating-point type; still to be implemented are unions and subobjects.
gcc/cp/ChangeLog:
* pt.c (convert_nontype_argument): Handle REAL_TYPE.
(invalid_nontype_parm_type_p): Allow all structural types.
* tree.c (structural_type_p): Use SCALAR_TYPE_P.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/pr81246.C: No error in C++20.
* g++.dg/cpp0x/variadic74.C: No error in C++20.
* g++.dg/cpp1z/nontype-auto3.C: No error in C++20.
* g++.dg/template/crash106.C: No error in C++20.
* g++.dg/template/crash119.C: No error in C++20.
* g++.dg/template/nontype12.C: No error in C++20.
* g++.dg/template/void3.C: Don't require follow-on message.
* g++.dg/template/void7.C: Don't require follow-on message.
* g++.dg/template/void9.C: Don't require follow-on message.
We were wrongly checking is_empty_class on the result of strip_array_types
rather than the actual field type. We weren't considering the alignment of
the data member. We needed to handle unions the same way as
layout_nonempty_base_or_field.
gcc/cp/ChangeLog:
PR c++/96105
PR c++/96052
PR c++/95976
* class.c (check_field_decls): An array of empty classes is not an
empty data member.
(layout_empty_base_or_field): Handle explicit alignment.
Fix union handling.
gcc/testsuite/ChangeLog:
PR c++/96105
PR c++/96052
PR c++/95976
* g++.dg/cpp2a/no_unique_address4.C: New test.
* g++.dg/cpp2a/no_unique_address5.C: New test.
* g++.dg/cpp2a/no_unique_address6.C: New test.
gcc/ada/
* sem_ch13.adb (Analyze_Attribute_Definition_Clause) <Address>:
Issue an unconditional warning for an overlay that changes the
scalar storage order.
gcc/ada/
* sem_res.adb (Resolve_Expression_With_Actions): Check the rules
of AI12-0368, and mark the declare expression as static or known
at compile time as appropriate.
* sem_ch4.adb: Minor reformatting.
* libgnat/a-stoufo.ads, libgnat/a-stoufo.adb: Allow up to 9
replacement parameters. I'm planning to use this in the test
case for this ticket.
gcc/ada/
* exp_ch3.adb (Build_Access_Subprogram_Wrapper_Body): Create a
proper signature when the access type denotes a parameterless
subprogram.
* exp_ch6.adb (Expand_Call): Handle properly a parameterless
indirect call when the corresponding access type has contracts.
gcc/ada/
* exp_aggr.adb
(Convert_To_Positional): Add Dims local variable
and pass it in calls to Is_Flat and Flatten.
(Check_Static_Components): Pass Dims in call to
Is_Static_Element.
(Nonflattenable_Next_Aggr): New predicate.
(Flatten): Add Dims parameter and Expr local variable. Call
Nonflattenable_Next_Aggr in a couple of places. In the case
when an Others choice is present, check that the element is
either static or a nested aggregate that can be flattened,
before disregarding the replication limit for elaboration
purposes. Check that a nested array is flattenable in the case
of a multidimensional array in any position. Remove redundant
check in the Others case and pass Dims in call to
Is_Static_Element. Use Expr variable.
(Is_Flat): Change type of Dims parameter from Int to Nat.
(Is_Static_Element): Add Dims parameter. Replace tests on
literals with call to Compile_Time_Known_Value. If everything
else failed and the dimension is 1, preanalyze the expression
before calling again Compile_Time_Known_Value on it. Return
true for null.
(Late_Expansion): Do not expand further if the assignment to the
target can be done directly by the back end.
gcc/ada/
* libgnat/g-socket.adb (Wait_On_Socket): Fix memory leaks and
file descriptor leaks. A memory leak was created each time the
routine was called without a selector (Selector = Null). Also,
in case of exception in the routine a memory leak and descriptor
leak was created as the created file selector was not closed.
gcc/ada/
* sem_util.adb
(Immediate_Context_Implies_Is_Potentially_Unevaluated): New
subprogram.
(Is_Potentially_Unevaluated): Do not stop climbing the tree on
the first candidate subexpression; required to handle nested
expressions.
gcc/ada/
* sem_ch3.adb (Check_Completion): Refactor chained
if-then-elsif-... statement to be more like a case
statement (note: we can't simply use case statement because of
Is_Intrinsic_Subprogram in the first condition).
