In upstream dmd, the compiler front-end and run-time have been merged
together into one repository. Both dmd and libdruntime now track that.
D front-end changes:
- Deprecated `scope(failure)' blocks that contain `return' statements.
- Deprecated using integers for `version' or `debug' conditions.
- Deprecated returning a discarded void value from a function.
- `new' can now allocate an associative array.
D runtime changes:
- Added avx512f detection to core.cpuid module.
Phobos changes:
- Changed std.experimental.logger.core.sharedLog to return
shared(Logger).
gcc/d/ChangeLog:
* dmd/MERGE: Merge upstream dmd d7772a2369.
* dmd/VERSION: Bump version to v2.100.1.
* d-codegen.cc (get_frameinfo): Check whether decision to generate
closure changed since semantic finished.
* d-lang.cc (d_handle_option): Remove handling of -fdebug=level and
-fversion=level.
* decl.cc (DeclVisitor::visit (VarDeclaration *)): Generate evaluation
of noreturn variable initializers before throw.
* expr.cc (ExprVisitor::visit (AssignExp *)): Don't generate
assignment for noreturn types, only evaluate for side effects.
* lang.opt (fdebug=): Undocument -fdebug=level.
(fversion=): Undocument -fversion=level.
libphobos/ChangeLog:
* configure: Regenerate.
* configure.ac (libtool_VERSION): Update to 4:0:0.
* libdruntime/MERGE: Merge upstream druntime d7772a2369.
* libdruntime/Makefile.am (DRUNTIME_DSOURCES): Add
core/internal/array/duplication.d.
* libdruntime/Makefile.in: Regenerate.
* src/MERGE: Merge upstream phobos 5748ca43f.
* testsuite/libphobos.gc/nocollect.d:
A SET operation that writes memory may have the same value as an
earlier store but if the alias sets of the new and earlier store do
not conflict then the set is not truly redundant. This can happen,
for example, if objects of different types share a stack slot.
To fix this we define a new function in cselib that first checks for
equality and if that is successful then finds the earlier store in the
value history and checks the alias sets.
The routine is used in two places elsewhere in the compiler:
cfgcleanup and postreload.
gcc/ChangeLog:
PR rtl-optimization/106187
* alias.h (mems_same_for_tbaa_p): Declare.
* alias.cc (mems_same_for_tbaa_p): New function.
* dse.cc (record_store): Use it instead of open-coding
alias check.
* cselib.h (cselib_redundant_set_p): Declare.
* cselib.cc: Include alias.h
(cselib_redundant_set_p): New function.
* cfgcleanup.cc: (mark_effect): Use cselib_redundant_set_p instead
of rtx_equal_for_cselib_p.
* postreload.cc (reload_cse_simplify): Use cselib_redundant_set_p.
(reload_cse_noop_set_p): Delete.
The option prints TOP N counters in a stable format
usage for comparison (diff).
gcc/ChangeLog:
* doc/gcov-dump.texi: Document the new option.
* gcov-dump.cc (main): Parse the new option.
(print_usage): Show the option.
(tag_counters): Sort key:value pairs of TOP N counter.
This patch adds a peephole2 to i386.md to implement the suggestion in
PR target/47949, of using xchg instead of mov for moving values to/from
the %rax/%eax register, controlled by -Oz, as the xchg instruction is
one byte shorter than the move it is replacing.
The new test case is taken from the PR:
int foo(int x) { return x; }
where previously we'd generate:
foo: mov %edi,%eax // 2 bytes
ret
but with this patch, using -Oz, we generate:
foo: xchg %eax,%edi // 1 byte
ret
On the CSiBE benchmark, this saves a total of 10238 bytes (reducing
the -Oz total from 3661796 bytes to 3651558 bytes, a 0.28% saving).
Interestingly, some modern architectures (such as Zen 3) implement
xchg using zero latency register renaming (just like mov), so in theory
this transformation could be enabled when optimizing for speed, if
benchmarking shows the improved code density produces consistently
better performance. However, this is architecture dependent, and
there may be interactions using xchg (instead a single_set) in the
late RTL passes (such as cprop_hardreg), so for now I've restricted
this to -Oz.
2022-08-03 Roger Sayle <roger@nextmovesoftware.com>
Uroš Bizjak <ubizjak@gmail.com>
gcc/ChangeLog
PR target/47949
* config/i386/i386.md (peephole2): New peephole2 to convert
SWI48 moves to/from %rax/%eax where the src is dead to xchg,
when optimizing for minimal size with -Oz.
gcc/testsuite/ChangeLog
PR target/47949
* gcc.target/i386/pr47949.c: New test case.
This patch adds an extra optimization to *cmp<dwi>_doubleword to improve
the code generated for comparisons against -1. Hypothetically, if a
comparison against -1 reached this splitter we'd currently generate code
that looks like:
notq %rdx ; 3 bytes
notq %rax ; 3 bytes
orq %rdx, %rax ; 3 bytes
setne %al
With this patch we would instead generate the superior:
andq %rdx, %rax ; 3 bytes
cmpq $-1, %rax ; 4 bytes
setne %al
which is both faster and smaller, and also what's currently generated
thanks to the middle-end splitting double word comparisons against
zero and minus one during RTL expansion. Should that change, this would
become a missed-optimization regression, but this patch also (potentially)
helps suitable comparisons created by CSE and combine.
2022-08-03 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
* config/i386/i386.md (*cmp<dwi>_doubleword): Add a special case
to split comparisons against -1 using AND and CMP -1 instructions.
This patch improves TImode STV by adding support for logical shifts by
integer constants that are multiples of 8. For the test case:
unsigned __int128 a, b;
void foo() { a = b << 16; }
on x86_64, gcc -O2 currently generates:
movq b(%rip), %rax
movq b+8(%rip), %rdx
shldq $16, %rax, %rdx
salq $16, %rax
movq %rax, a(%rip)
movq %rdx, a+8(%rip)
ret
with this patch we now generate:
movdqa b(%rip), %xmm0
pslldq $2, %xmm0
movaps %xmm0, a(%rip)
ret
2022-08-03 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
* config/i386/i386-features.cc (compute_convert_gain): Add gain
for converting suitable TImode shift to a V1TImode shift.
(timode_scalar_chain::convert_insn): Add support for converting
suitable ASHIFT and LSHIFTRT.
(timode_scalar_to_vector_candidate_p): Consider logical shifts
by integer constants that are multiples of 8 to be candidates.
gcc/testsuite/ChangeLog
* gcc.target/i386/sse4_1-stv-7.c: New test case.
This patch implements some additional zero-extension and sign-extension
related optimizations in simplify-rtx.cc. The original motivation comes
from PR rtl-optimization/71775, where in comment #2 Andrew Pinksi sees:
Failed to match this instruction:
(set (reg:DI 88 [ _1 ])
(sign_extend:DI (subreg:SI (ctz:DI (reg/v:DI 86 [ x ])) 0)))
On many platforms the result of DImode CTZ is constrained to be a
small unsigned integer (between 0 and 64), hence the truncation to
32-bits (using a SUBREG) and the following sign extension back to
64-bits are effectively a no-op, so the above should ideally (often)
be simplified to "(set (reg:DI 88) (ctz:DI (reg/v:DI 86 [ x ]))".
To implement this, and some closely related transformations, we build
upon the existing val_signbit_known_clear_p predicate. In the first
chunk, nonzero_bits knows that FFS and ABS can't leave the sign-bit
bit set, so the simplification of of ABS (ABS (x)) and ABS (FFS (x))
can itself be simplified. The second transformation is that we can
canonicalized SIGN_EXTEND to ZERO_EXTEND (as in the PR 71775 case above)
when the operand's sign-bit is known to be clear. The final two chunks
are for SIGN_EXTEND of a truncating SUBREG, and ZERO_EXTEND of a
truncating SUBREG respectively. The nonzero_bits of a truncating
SUBREG pessimistically thinks that the upper bits may have an
arbitrary value (by taking the SUBREG), so we need look deeper at the
SUBREG's operand to confirm that the high bits are known to be zero.
Unfortunately, for PR rtl-optimization/71775, ctz:DI on x86_64 with
default architecture options is undefined at zero, so we can't be sure
the upper bits of reg:DI 88 will be sign extended (all zeros or all ones).
nonzero_bits knows this, so the above transformations don't trigger,
but the transformations themselves are perfectly valid for other
operations such as FFS, POPCOUNT and PARITY, and on other targets/-march
settings where CTZ is defined at zero.
2022-08-03 Roger Sayle <roger@nextmovesoftware.com>
Segher Boessenkool <segher@kernel.crashing.org>
Richard Sandiford <richard.sandiford@arm.com>
gcc/ChangeLog
* simplify-rtx.cc (simplify_unary_operation_1) <ABS>: Add
optimizations for CLRSB, PARITY, POPCOUNT, SS_ABS and LSHIFTRT
that are all positive to complement the existing FFS and
idempotent ABS simplifications.
<SIGN_EXTEND>: Canonicalize SIGN_EXTEND to ZERO_EXTEND when
val_signbit_known_clear_p is true of the operand.
Simplify sign extensions of SUBREG truncations of operands
that are already suitably (zero) extended.
<ZERO_EXTEND>: Simplify zero extensions of SUBREG truncations
of operands that are already suitably zero extended.
Previously, all gimple_cond types were undserstoof, with float values,
this is no longer true. We should gracefully do nothing if the
gcond type is not supported.
PR tree-optimization/106510
gcc/
* gimple-range-fold.cc (fur_source::register_outgoing_edges):
Check for unsupported statements early.
gcc/testsuite
* gcc.dg/pr106510.c: New.
I missed the -details dump flag, plus I wasn't checking the actual folding.
As a bonus I had flipped the dump file name and the count, so the test
was coming out as unresolved, which I missed because I was only checking
for failures and passes.
Whooops.
gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/vrp-float-1.c: Adjust test so it passes.
When propagating on-entry values in the cache, checking if any equivalence
has a known value can improve results. No new calculations are made.
Only queries via dominators which do not populate the cache are checked.
PR tree-optimization/106474
gcc/
* gimple-range-cache.cc (ranger_cache::fill_block_cache): Query
range of equivalences that may contribute to the range.
gcc/testsuite/
* g++.dg/pr106474.C: New.
Contrary to CTF and our previous expectations, as per [1], turns out
that in BTF:
1) The `encoding' field in integer types shall not be treated as a
bitmap, but as an enumerated, i.e. these bits are exclusive to each
other.
2) The CHAR bit in `encoding' shall _not_ be set when emitting types
for char nor `unsigned char'.
Consequently this patch clears the CHAR bit before emitting the
variable part of BTF integral types. It also updates the testsuite
accordingly, expanding it to check for BOOL bits.
[1] https://lore.kernel.org/bpf/a73586ad-f2dc-0401-1eba-2004357b7edf@fb.com/T/#t
gcc/ChangeLog:
* btfout.cc (output_asm_btf_vlen_bytes): Do not use the CHAR
encoding bit in BTF.
gcc/testsuite/ChangeLog:
* gcc.dg/debug/btf/btf-int-1.c: Do not check for char bits in
bti_encoding and check for bool bits.
I am trying to make sense of back_threader_profitability::profitable_path_p
and the first thing I notice is that we do
/* Threading is profitable if the path duplicated is hot but also
in a case we separate cold path from hot path and permit optimization
of the hot path later. Be on the agressive side here. In some testcases,
as in PR 78407 this leads to noticeable improvements. */
if (m_speed_p
&& ((taken_edge && optimize_edge_for_speed_p (taken_edge))
|| contains_hot_bb))
{
if (n_insns >= param_max_fsm_thread_path_insns)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, " FAIL: Jump-thread path not considered: "
"the number of instructions on the path "
"exceeds PARAM_MAX_FSM_THREAD_PATH_INSNS.\n");
return false;
}
...
}
else if (!m_speed_p && n_insns > 1)
{
if (dump_file && (dump_flags & TDF_DETAILS))
fprintf (dump_file, " FAIL: Jump-thread path not considered: "
"duplication of %i insns is needed and optimizing for size.\n",
n_insns);
return false;
}
...
return true;
thus we apply the n_insns >= param_max_fsm_thread_path_insns only
to "hot paths". The comment above this isn't entirely clear whether
this is by design ("Be on the aggressive side here ...") but I think
this is a mistake. In fact the "hot path" check seems entirely
useless since if the path is not hot we simply continue threading it.
This was caused by r12-324-g69e5544210e3c0 and the following simply
reverts the offending change.
* tree-ssa-threadbackward.cc
(back_threader_profitability::profitable_path_p): Apply
size constraints to all paths again.
Without further ado, here is the implementation for floating point
range operators, plus the switch to enable all ranger clients to
handle floats.
These are bare bone implementations good enough for relation operators
to work, while keeping the NAN bits up to date in the frange. There
is also minimal support for keeping track of +-INF when it is obvious.
Tested on x86-64 Linux.
gcc/ChangeLog:
* range-op-float.cc (finite_operands_p): New.
(frelop_early_resolve): New.
(default_frelop_fold_range): New.
(class foperator_equal): New.
(class foperator_not_equal): New.
(class foperator_lt): New.
(class foperator_le): New.
(class foperator_gt): New.
(class foperator_ge): New.
(class foperator_unordered): New.
(class foperator_ordered): New.
(class foperator_relop_unknown): New.
(floating_op_table::floating_op_table): Add above classes to
floating op table.
* value-range.h (frange::supports_p): Enable.
gcc/testsuite/ChangeLog:
* g++.dg/opt/pr94589-2.C: XFAIL.
* gcc.dg/tree-ssa/vrp-float-1.c: New test.
* gcc.dg/tree-ssa/vrp-float-11.c: New test.
* gcc.dg/tree-ssa/vrp-float-3.c: New test.
* gcc.dg/tree-ssa/vrp-float-4.c: New test.
* gcc.dg/tree-ssa/vrp-float-6.c: New test.
* gcc.dg/tree-ssa/vrp-float-7.c: New test.
* gcc.dg/tree-ssa/vrp-float-8.c: New test.
This patch Allows us to export floating point ranges into the SSA name
(SSA_NAME_RANGE_INFO).
[Richi, in PR24021 you suggested that match.pd could use global float
ranges, because it would generally not invoke ranger. This patch
implements the boiler plate to save the frange globally.]
[Jeff, we've also been talking in parallel of using NAN knowledge
during expansion to RTL. This patch will provide the NAN bits in the
SSA name.]
Since frange's currently implementation is just a shell, with no
actual endpoints, frange_storage_slot only contains frange_props which
fits inside a byte. When we have endpoints, y'all can decide if it's
worth saving them, or if the NAN/etc bits are good enough.
gcc/ChangeLog:
* tree-core.h (struct tree_ssa_name): Add frange_info and
reshuffle the rest.
* value-range-storage.cc (vrange_storage::alloc_slot): Add case
for frange.
(vrange_storage::set_vrange): Same.
(vrange_storage::get_vrange): Same.
(vrange_storage::fits_p): Same.
(frange_storage_slot::alloc_slot): New.
(frange_storage_slot::set_frange): New.
(frange_storage_slot::get_frange): New.
(frange_storage_slot::fits_p): New.
* value-range-storage.h (class frange_storage_slot): New.
ipa-* still works on legacy value_range's which only support
integrals. This patch limits the query to integrals, as to not get a
floating point range that can't exist in an irange.
gcc/ChangeLog:
* ipa-prop.cc (ipa_compute_jump_functions_for_edge): Limit ranger
query to integrals.
This adds EDGE_COPY_SRC_JOINER_BLOCK sources to the set of blocks
we need to check we can duplicate.
PR tree-optimization/106497
* tree-ssa-threadupdate.cc (fwd_jt_path_registry::update_cfg):
Also verify we can copy EDGE_COPY_SRC_JOINER_BLOCK.
* gcc.dg/torture/pr106497.c: New testcase.
gcc/ChangeLog:
* profile.cc (compute_branch_probabilities): Dump details only
if TDF_DETAILS.
* symtab.cc (symtab_node::dump_base): Do not dump pointer unless
TDF_ADDRESS is used, it makes comparison harder.
The following reduces the number of SSA updates done during autopar
OMP expansion, specifically avoiding the cases that just add virtual
operands (where maybe none have been before) in dead regions of the CFG.
Instead virtual SSA update is delayed until after the pass. There's
much more TLC needed here, but test coverage makes it really difficult.
PR tree-optimization/106498
* omp-expand.cc (expand_omp_taskreg): Do not perform virtual
SSA update here.
(expand_omp_for): Or here.
(execute_expand_omp): Instead schedule it here together
with CFG cleanup via TODO.
This adjusts the assert guard to include -flto-partition=none which
behaves as WPA.
PR lto/106334
* dwarf2out.cc (dwarf2out_register_external_die): Adjust
assert.
The following builds upon the logic of the PR105679 fix by avoiding
to thread to a known edge that is predicted as probably never executed.
PR tree-optimization/106495
* tree-ssa-threadbackward.cc
(back_threader_profitability::profitable_path_p): If known_edge
is probably never executed avoid threading.
This adds a clarifying "note" to address space mismatch diagnostics.
For example, it improves the diagnostic for
gcc.target/i386/addr-space-typeck-2.c from:
addr-space-typeck-2.c: In function 'test_bad_call':
addr-space-typeck-2.c:12:22: error: passing argument 2 of 'expects_seg_gs'
from pointer to non-enclosed address space
12 | expects_seg_gs (0, ptr, 1);
| ^~~
to:
addr-space-typeck-2.c: In function 'test_bad_call':
addr-space-typeck-2.c:12:22: error: passing argument 2 of 'expects_seg_gs'
from pointer to non-enclosed address space
12 | expects_seg_gs (0, ptr, 1);
| ^~~
addr-space-typeck-2.c:7:51: note: expected '__seg_gs void *' but argument
is of type 'void *'
7 | extern void expects_seg_gs (int i, void __seg_gs *param, int j);
| ~~~~~~~~~~~~~~~^~~~~
I took the liberty of adding the test coverage to i386 since we need
a specific target to test this on.
gcc/c/ChangeLog:
* c-typeck.cc (build_c_cast): Quote names of address spaces in
diagnostics.
(convert_for_assignment): Add a note to address space mismatch
diagnostics, specifying the expected and actual types.
gcc/testsuite/ChangeLog:
* gcc.target/i386/addr-space-typeck-1.c: New test.
* gcc.target/i386/addr-space-typeck-2.c: New test.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
This patch resolves PR target/106481, and is an oversight in my recent
battles with REG_EQUAL notes during TImode STV (see PR target/106278
https://gcc.gnu.org/pipermail/gcc-patches/2022-July/598416.html).
The patch above's/current behaviour is that we check that the mode of
the REG_EQUAL note is TImode before using PUT_MODE to set it to V1TImode.
However, the new test case reveals that this doesn't consider REG_EQUAL
notes that are CONST_INT or CONST_WIDE_INT, i.e. that are VOIDmode,
and so STV produces:
(insn 85 84 86 2 (set (reg:V1TI 113)
(reg:V1TI 84)) "pr106481.c":13:3 1766 {movv1ti_internal}
(expr_list:REG_EQUAL (const_wide_int 0x0ffffffff00000004)
(nil)))
which causes problems as the const_wide_int isn't a valid immediate
constant for V1TImode. With this patch, we now generate the correct:
(insn 85 84 86 2 (set (reg:V1TI 113)
(reg:V1TI 84)) "pr106481.c":13:3 1766 {movv1ti_internal}
(expr_list:REG_EQUAL (const_vector:V1TI [
(const_wide_int 0x0ffffffff00000004)
])
(nil)))
2022-08-01 Roger Sayle <roger@nextmovesoftware.com>
Uroš Bizjak <ubizjak@gmail.com>
gcc/ChangeLog
PR target/106481
* config/i386/i386-features.cc (timode_scalar_chain::convert_insn):
Convert a CONST_SCALAR_INT_P in a REG_EQUAL note into a V1TImode
CONST_VECTOR.
gcc/testsuite/ChangeLog
PR target/106481
* gcc.target/i386/pr106481.c: New test case.
We can't always use the PLT entry as the function address for local IFUNC
functions. When the PIC register is needed for PLT call, indirect call
via the PLT entry will fail since the PIC register may not be set up
properly for indirect call. Add ix86_ifunc_ref_local_ok to return false
when the PLT entry can't be used as local IFUNC function pointers.
gcc/
PR target/83782
* config/i386/i386.cc (ix86_ifunc_ref_local_ok): New.
(TARGET_IFUNC_REF_LOCAL_OK): Use it.
gcc/testsuite/
PR target/83782
* gcc.target/i386/pr83782-1.c: Require non-ia32.
* gcc.target/i386/pr83782-2.c: Likewise.
* gcc.target/i386/pr83782-3.c: New test.
The kernel bpftool expects BTF_KIND_FUNC entries in BTF to include an
annotation reflecting the linkage of functions (static, global). For
whatever reason they abuse the `vlen' field of the BTF_KIND_FUNC entry
instead of adding a variable-part to the record like it is done with
other entry kinds.
This patch makes GCC to include this linkage info in BTF_KIND_FUNC
entries.
Tested in bpf-unknown-none target.
gcc/ChangeLog:
PR debug/106263
* ctfc.h (struct ctf_dtdef): Add field linkage.
* ctfc.cc (ctf_add_function): Set ctti_linkage.
* dwarf2ctf.cc (gen_ctf_function_type): Pass a linkage for
function types and subprograms.
* btfout.cc (btf_asm_func_type): Emit linkage information for the
function.
(btf_dtd_emit_preprocess_cb): Propagate the linkage information
for functions.
gcc/testsuite/ChangeLog:
PR debug/106263
* gcc.dg/debug/btf/btf-function-4.c: New test.
* gcc.dg/debug/btf/btf-function-5.c: Likewise.
Ensure that both parameters to vector shifts use the same mode. This is most
important for amdgcn where the masks are DImode.
gcc/ChangeLog:
* omp-simd-clone.cc (simd_clone_adjust): Convert shift_cnt to match
the mask type.
Co-authored-by: Jakub Jelinek <jakub@redhat.com>
This patch fixes a missed optimization in match.pd. It takes the pattern,
x / y * y == x, and optimizes it to x % y == 0. This produces fewer
instructions. This simplification does not happen for complex types.
This patch also adds tests for the optimization rule.
Bootstrapped/regtested on x86_64-pc-linux-gnu.
PR tree-optimization/104992
gcc/ChangeLog:
* match.pd (x / y * y == x): New simplification.
gcc/testsuite/ChangeLog:
* g++.dg/pr104992-1.C: New test.
* gcc.dg/pr104992.c: New test.
GCC fails to bootstrap when configured with --enable-languages=all on
machines that have older versions of GNAT installed as the system Ada
compiler. In configure, it's not sufficient to check whether gnat is
available, but whether a sufficiently recent version of GNAT is
installed. This patch tweaks config/acx.m4 so that conftest.adb also
contains a reference to System.CRTL.int64 as required by the current
version of gcc/ada/osint.adb. This fixes the build when the system
Ada is GNAT v4.8.5 (on Redhat 7) by disabling ada, but continues to
work fine when the system Ada is GNAT v11.3.1.
2022-08-01 Roger Sayle <roger@nextmovesoftware.com>
Arnaud Charlet <charlet@adacore.com>
config/ChangeLog
* acx.m4 (AC_PROG_GNAT): Update conftest.adb to include
features required of the host gnat compiler.
ChangeLog
* configure: Regenerate.
The boz_15.f90 test FAILs on powerpc64le-linux when -mabi=ieeelongdouble
is used (either default through --with-long-double-format=ieee or
when used explicitly).
The problem is that the read/write transfer routines are called with
BT_REAL (or BT_COMPLEX) type and kind 17 which is magic we use to say
it is the IEEE quad real(kind=16) rather than the IBM double double
real(kind=16). For the floating point input/output we then handle kind
17 specially, but for B/O/Z we just treat the bytes of the floating point
value as binary blob and using 17 in that case results in unexpected
behavior, for write it means we don't estimate right how many chars we'll
need and print ******************** etc. rather than what we should, and
even with explicit size we'd print one further byte than intended.
For read it would even mean overwriting some unrelated byte after the
floating point object.
Fixed by using 16 instead of 17 in the read_radix and write_{b,o,z} calls.
2022-08-01 Jakub Jelinek <jakub@redhat.com>
PR libfortran/106079
* io/transfer.c (formatted_transfer_scalar_read,
formatted_transfer_scalar_write): For type BT_REAL with kind 17
change kind to 16 before calling read_radix or write_{b,o,z}.
These are some assorted cleanups to the frange class to make it easier
to drop in an implementation with FP endpoints:
* frange::set() had some asserts limiting the type of arguments
passed. There's no reason why we can't handle all the variants.
Worse comes to worse, we can always return a VARYING which is
conservative and correct.
* frange::normalize_kind() now returns a boolean that can be used in
union and intersection to indicate that the range changed.
* Implement vrp_val_max and vrp_val_min for floats. Also, move them
earlier in the header file so frange can use them.
Tested on x86-64 Linux.
gcc/ChangeLog:
* value-range.cc (tree_compare): New.
(frange::set): Make more general.
(frange::normalize_kind): Cleanup and return bool.
(frange::union_): Use normalize_kind return value.
(frange::intersect): Same.
(frange::verify_range): Remove unnecessary else.
* value-range.h (vrp_val_max): Move before frange class.
(vrp_val_min): Same.
(frange::frange): Remove set to m_type.
Make all vrange::supports_*_p methods const_tree as they can end up
being called from functions that are const_tree.
Tested on x86-64 Linux.
gcc/ChangeLog:
* value-range.cc (vrange::supports_type_p): Use const_tree.
(irange::supports_type_p): Same.
(frange::supports_type_p): Same.
* value-range.h (Value_Range::supports_type_p): Same.
(irange::supports_p): Same.
Even though ranger is type agnostic, SCEV seems to only work with
integers. This patch removes some FIXME notes making it explicit that
bounds_of_var_in_loop only works with iranges.
Tested on x86-64 Linux.
gcc/ChangeLog:
* gimple-range-fold.cc (fold_using_range::range_of_phi): Only
query SCEV for integers.
(fold_using_range::range_of_ssa_name_with_loop_info): Remove
irange check.
2bfb0fcb51510f22723c8cdfefe [Sanitizer][MIPS] Fix stat struct size for the O32 ABI.
Signed-off-by: Dimitrije Milosevic <dimitrije.milosevic@syrmia.com>.
This patch adds rot[lr]64ti2_doubleword patterns to the x86_64 backend,
to move splitting of 128-bit TImode rotates by 64 bits after reload,
matching what we now do for 64-bit DImode rotations by 32 bits with -m32.
In theory moving when this rotation is split should have little
influence on code generation, but in practice "reload" sometimes
decides to make use of the increased flexibility to reduce the number
of registers used, and the code size, by using xchg.
For example:
__int128 x;
__int128 y;
__int128 a;
__int128 b;
void foo()
{
unsigned __int128 t = x;
t ^= a;
t = (t<<64) | (t>>64);
t ^= b;
y = t;
}
Before:
movq x(%rip), %rsi
movq x+8(%rip), %rdi
xorq a(%rip), %rsi
xorq a+8(%rip), %rdi
movq %rdi, %rax
movq %rsi, %rdx
xorq b(%rip), %rax
xorq b+8(%rip), %rdx
movq %rax, y(%rip)
movq %rdx, y+8(%rip)
ret
After:
movq x(%rip), %rax
movq x+8(%rip), %rdx
xorq a(%rip), %rax
xorq a+8(%rip), %rdx
xchgq %rdx, %rax
xorq b(%rip), %rax
xorq b+8(%rip), %rdx
movq %rax, y(%rip)
movq %rdx, y+8(%rip)
ret
One some modern architectures this is a small win, on some older
architectures this is a small loss. The decision which code to
generate is made in "reload", and could probably be tweaked by
register preferencing. The much bigger win is that (eventually) all
TImode mode shifts and rotates by constants will become potential
candidates for TImode STV.
2022-07-31 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
* config/i386/i386.md (define_expand <any_rotate>ti3): For
rotations by 64 bits use new rot[lr]64ti2_doubleword pattern.
(rot[lr]64ti2_doubleword): New post-reload splitter.
This patch resolves PR target/106450, some more fall-out from more
aggressive TImode scalar-to-vector (STV) optimizations. I continue
to be caught out by how far TImode STV has diverged from DImode/SImode
STV, and therefore requires additional (unexpected) tweaking. Many
thanks to H.J. Lu for pointing out timode_remove_non_convertible_regs
needs to be extended to handle XOR (and other new operations).
Unhelpfully the comment above this function states that it's the TImode
version of "remove_non_convertible_regs", which doesn't exist anymore,
so I've resurrected an explanatory comment from the git history.
By refactoring the checks for hard regs and already "marked" regs
into timode_check_non_convertible_regs itself, all of its callers are
simplified. This patch then FOR_EACH_INSN_USE and FOR_EACH_INSN_DEF
to generically handle arbitrary (non-move) instructions (including
unary and binary operations), calling timode_check_non_convertible_regs
on each TImode register USE and DEF.
2022-07-31 Roger Sayle <roger@nextmovesoftware.com>
H.J. Lu <hjl.tools@gmail.com>
gcc/ChangeLog
PR target/106450
* config/i386/i386-features.cc (timode_check_non_convertible_regs):
Do nothing if REGNO is set in the REGS bitmap, or is a hard reg.
(timode_remove_non_convertible_regs): Update comment.
Call timode_check_non_convertible_reg on all TImode register
DEFs and USEs in each instruction.
gcc/testsuite/ChangeLog
PR target/106450
* gcc.target/i386/pr106450.c: New test case.
gcc/fortran/ChangeLog:
PR fortran/92805
* match.cc (gfc_match_small_literal_int): Make gobbling of leading
whitespace optional.
(gfc_match_name): Likewise.
(gfc_match_char): Likewise.
* match.h (gfc_match_small_literal_int): Adjust prototype.
(gfc_match_name): Likewise.
(gfc_match_char): Likewise.
* primary.cc (match_kind_param): Match small literal int or name
without gobbling whitespace.
(get_kind): Do not skip over blanks.
(match_string_constant): Likewise.
gcc/testsuite/ChangeLog:
PR fortran/92805
* gfortran.dg/literal_constants.f: New test.
* gfortran.dg/literal_constants.f90: New test.
Co-authored-by: Steven G. Kargl <kargl@gcc.gnu.org>
gcc/fortran/ChangeLog:
PR fortran/77652
* check.cc (gfc_check_associated): Make the rank check of POINTER
vs. TARGET match the allowed forms of pointer assignment for the
selected Fortran standard.
gcc/testsuite/ChangeLog:
PR fortran/77652
* gfortran.dg/associated_target_9a.f90: New test.
* gfortran.dg/associated_target_9b.f90: New test.
In C++, since all tokens are lexed from libcpp up front, diagnostics generated
by libcpp after lexing has completed do not get a valid location from libcpp
(rather, libcpp thinks they all pertain to the end of the file.) This has long
been addressed using the global variable "done_lexing", which the C++ frontend
sets at the appropriate time; when done_lexing is true, then c_cpp_diagnostic(),
which outputs libcpp's diagnostics, uses input_location instead of the wrong
libcpp location. The C++ frontend arranges that input_location will point to the
token it is currently processing, so this generally works fine. However, there
is one exception currently, which is -Wunused-macros. This gets generated at the
end of processing in cpp_finish (), since we need to wait until then to
determine whether a macro was eventually used or not. But the locations it
passes to c_cpp_diagnostic () were remembered from the original lexing and hence
they should not be overridden with input_location, which is now the one
incorrectly pointing to the end of the file.
Fixed by setting done_lexing=false again just prior to calling cpp_finish (). I
also renamed the variable from done_lexing to "override_libcpp_locations", since
it's now not strictly about lexing anymore.
There is no new testcase with this patch, since we already had an xfailed
testcase which is now fixed.
gcc/c-family/ChangeLog:
PR c++/66290
* c-common.h: Rename global done_lexing to
override_libcpp_locations.
* c-common.cc (c_cpp_diagnostic): Likewise.
* c-opts.cc (c_common_finish): Set override_libcpp_locations
(formerly done_lexing) immediately prior to calling cpp_finish ().
gcc/cp/ChangeLog:
PR c++/66290
* parser.cc (cp_lexer_new_main): Rename global done_lexing to
override_libcpp_locations.
gcc/testsuite/ChangeLog:
PR c++/66290
* c-c++-common/pragma-diag-15.c: Remove xfail for C++.
This patch fixes PR bootstrap/106472 by adding a missing dependency
to Makefile.def to allow make bootstrap when configured using
"--enable-languages=go" (and not using make with multiple threads).
2022-07-31 Roger Sayle <roger@nextmovesoftware.com>
ChangeLog
PR bootstrap/106472
* Makefile.def (dependencies): Make configure-target-libgo depend
upon all-target-libbacktrace.
Here the CONSTRUCTOR we were providing for D{} had an entry for the B base
subobject at offset 0 following the entry for the C base, causing
output_constructor_regular_field to ICE due to going backwards. It might be
nice for that function to be more tolerant of empty fields, but it also
seems reasonable for the front end to prune the useless entry.
PR c++/106369
gcc/cp/ChangeLog:
* constexpr.cc (reduced_constant_expression_p): Return false
if a CONSTRUCTOR initializes an empty field.
gcc/testsuite/ChangeLog:
* g++.dg/cpp1z/constexpr-lambda27.C: New test.
