This cleans up a couple of places in which the previous patch had:
call_used_or_fixed_regs & ~fixed_reg_set
In that context, regs_invalidated_by_call is IMO more obvious.
2019-09-10 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/frv/frv.c (frv_ifcvt_modify_tests): Use
regs_invalidated_by_call & ~fixed_reg_set instead of
call_used_or_fixed_regs & ~fixed_reg_set.
* config/sh/sh.c (output_stack_adjust): Likewise.
From-SVN: r275601
CALL_USED_REGISTERS and call_used_regs infamously contain all fixed
registers (hence the need for CALL_REALLY_USED_REGISTERS etc.).
We try to recover from this to some extent with:
/* Contains 1 for registers that are set or clobbered by calls. */
/* ??? Ideally, this would be just call_used_regs plus global_regs, but
for someone's bright idea to have call_used_regs strictly include
fixed_regs. Which leaves us guessing as to the set of fixed_regs
that are actually preserved. We know for sure that those associated
with the local stack frame are safe, but scant others. */
HARD_REG_SET x_regs_invalidated_by_call;
Since global registers are added to fixed_reg_set and call_used_reg_set
too, it's always the case that:
call_used_reg_set == regs_invalidated_by_call | fixed_reg_set
This patch replaces all uses of call_used_reg_set with a new macro
call_used_or_fixed_regs to make this clearer.
This is part of a series that allows call_used_regs to be what is
now call_really_used_regs. It's a purely mechanical replacement;
later patches clean up obvious oddities like
"call_used_or_fixed_regs & ~fixed_regs".
2019-09-10 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* hard-reg-set.h (target_hard_regs::x_call_used_reg_set): Delete.
(call_used_reg_set): Delete.
(call_used_or_fixed_regs): New macro.
* reginfo.c (init_reg_sets_1, globalize_reg): Remove initialization
of call_used_reg_set.
* caller-save.c (setup_save_areas): Use call_used_or_fixed_regs
instead of call_used_regs.
(save_call_clobbered_regs): Likewise.
* cfgcleanup.c (old_insns_match_p): Likewise.
* config/c6x/c6x.c (c6x_call_saved_register_used): Likewise.
* config/epiphany/epiphany.c (epiphany_conditional_register_usage):
Likewise.
* config/frv/frv.c (frv_ifcvt_modify_tests): Likewise.
* config/sh/sh.c (output_stack_adjust): Likewise.
* final.c (collect_fn_hard_reg_usage): Likewise.
* ira-build.c (ira_build): Likewise.
* ira-color.c (calculate_saved_nregs): Likewise.
(allocno_reload_assign, calculate_spill_cost): Likewise.
* ira-conflicts.c (ira_build_conflicts): Likewise.
* ira-costs.c (ira_tune_allocno_costs): Likewise.
* ira-lives.c (process_bb_node_lives): Likewise.
* ira.c (setup_reg_renumber): Likewise.
* lra-assigns.c (find_hard_regno_for_1, lra_assign): Likewise.
* lra-constraints.c (need_for_call_save_p): Likewise.
(need_for_split_p, inherit_in_ebb): Likewise.
* lra-lives.c (process_bb_lives): Likewise.
* lra-remat.c (call_used_input_regno_present_p): Likewise.
* postreload.c (reload_combine): Likewise.
* regrename.c (find_rename_reg): Likewise.
* reload1.c (reload_as_needed): Likewise.
* rtlanal.c (find_all_hard_reg_sets): Likewise.
* sel-sched.c (mark_unavailable_hard_regs): Likewise.
* shrink-wrap.c (requires_stack_frame_p): Likewise.
From-SVN: r275600
Reusing the reasoning from the call_fixed_reg_set patch:
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
for (j = 1; j <= MOVE_MAX_WORDS; j++)
if (reg_save_code (i,regno_save_mode[i][j]) == -1)
should be true whenever regno_save_mode[i][j] == VOIDmode,
which it is for:
~call_used_reg_set | ~have_save_mode
So this condition amounts to:
~call_used_reg_set | ~have_save_mode | ~have_save_insn
== ~call_used_reg_set | ~savable_regs
no_caller_save_reg_set is then set if call_used_regs[i], so
no_caller_save_reg_set is:
call_used_reg_set & (~call_used_reg_set | ~savable_regs)
== call_used_reg_set & ~savable_regs
This patch expands its single user accordingly.
Note that ~savable_regs is always empty on LRA targets.
2019-09-10 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* hard-reg-set.h (target_hard_regs::x_no_caller_save_reg_set): Delete.
(no_caller_save_reg_set): Delete.
* caller-save.c (init_caller_save): Don't initialize it.
* ira-conflicts.c (ira_build_conflicts): Calculate
no_caller_save_reg_set locally from call_used_reg_set and savable_regs.
From-SVN: r275599
On targets that use reload, call_fixed_reg_set is structurally:
fixed_reg_set -- reginfo.c
| (call_used_reg_set & ~have_save_mode) -- first loop in init_caller_save
| ~have_save_insn -- final loop in init_caller_save
(where "have_save_mode" and "have_save_insn" are just my names).
But the final loop in init_caller_save does:
for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
for (j = 1; j <= MOVE_MAX_WORDS; j++)
if (reg_save_code (i,regno_save_mode[i][j]) == -1)
This last condition ought to be true whenever:
regno_save_mode[i][j] == VOIDmode
since either targetm.hard_regno_mode_ok (i, VOIDmode) should be
false or the VOIDmode save & restore shouldn't match any move insn.
And after the first loop, regno_save_mode[i][j] == VOIDmode whenever
!call_used_regs[i]. So the above is actually:
fixed_reg_set
| (call_used_reg_set & ~have_save_mode)
| (~call_used_reg_set | ~have_save_insn)
which simplifies to:
fixed_reg_set -- reginfo.c
| ~have_save_mode -- first loop in init_caller_save
| ~have_save_insn -- final loop in init_caller_save
| ~call_used_reg_set -- final loop in init_caller_save
So:
~call_fixed_reg_set == (~fixed_reg_set
& have_save_mode
& have_save_insn
& call_used_reg_set) [A]
All users have the form:
(call_used_reg_set or some subset) & ~(call_fixed_reg_set | ...)
i.e.:
(call_used_reg_set or some subset) & ~call_fixed_reg_set & ~(...)
We can therefore drop the "& call_used_reg_set" from [A], leaving:
~fixed_reg_set & have_save_mode & have_save_insn
This patch combines have_save_mode & have_save_insn into a single
condition "a save is possible", represented as savable_regs.
We can then substitute:
~call_fixed_reg_set --> ~fixed_reg_set & savable_regs
(registers we can actually save around calls)
The patch also sets regno_save_mode[i][j] for all registers,
in case non-default ABIs require a save when the default ABI
doesn't. This ensures that savable_regs (like fixed_reg_set but
unlike call_fixed_reg_set) isn't affected by the ABI. This only
becomes significant with later patches and at this point is just
a simplification.
Since init_caller_save is only called for reload targets,
the default assumption for LRA is that all registers are savable,
just like the default assumption before the patch was that
(~)call_fixed_reg_set == (~)fixed_reg_set.
2019-09-10 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* hard-reg-set.h (target_hard_regs::x_call_fixed_reg_set): Delete.
(target_hard_regs::x_savable_regs): New field.
(call_fixed_reg_set): Delete.
(savable_regs): New macro,
* reginfo.c (globalize_reg): Don't set call_fixed_reg_set.
(init_reg_sets_1): Likewise. Initialize savable_regs.
* caller-save.c (init_caller_save): Invoke HARD_REGNO_CALLER_SAVE_MODE
for all registers. Set savable_regs instead of call_fixed_reg_set.
(setup_save_areas, save_call_clobbered_regs): Replace uses of
~call_fixed_reg_set with ~fixed_reg_set & savable_regs.
* config/sh/sh.c (output_stack_adjust): Likewise.
From-SVN: r275598
I have a series of patches that hides call_used_regs from target-
independent code, a knock-on effect of which is that (public) target
macros can't use call_used_regs either. This patch fixes the only
case in which that was a problem.
2019-09-10 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* config/c6x/c6x-protos.h (c6x_set_return_address): Declare.
* config/c6x/c6x.h (REGNO_REG_CLASS): Move implementation to
* config/c6x/c6x.c (c6x_regno_reg_class): ...this new function.
From-SVN: r275597
Only one caller (in dwarf2out.c) was preventing get_call_rtx_from
from taking an rtx_insn *. Since that caller just passes a PATTERN,
it's a trivial change to make.
2019-09-10 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* rtl.h (get_call_rtx_from): Take a const rtx_insn * instead of an rtx.
* rtlanal.c (get_call_rtx_from): Likewise.
* dwarf2out.c (dwarf2out_var_location): Pass the insn rather
than the pattern to get_call_rtx_from.
* config/i386/i386-expand.h (ix86_notrack_prefixed_insn_p): Take
an rtx_insn * instead of an rtx.
* config/i386/i386-expand.c (ix86_notrack_prefixed_insn_p): Likewise.
From-SVN: r275593
The standard requires overloads of std::to_chars for char and (un)signed
integer types. This means that our constrained function template is
non-conforming, because the difference is observable when using types
that convert to an integer (e.g. wchar_t, which promotes).
As well as defining the deleted bool overload for LWG 3266, replace the
constrained function template with overloads for each type.
* include/std/charconv (to_chars): Rename to __to_chars_i. Define
non-template overloads for each signed and unsigned integer type and
char. Define deleted overload for bool (LWG 3266).
* testsuite/20_util/to_chars/1_neg.cc: Remove.
* testsuite/20_util/to_chars/3.cc: New test.
* testsuite/20_util/to_chars/lwg3266.cc: New test.
From-SVN: r275588
PR middle-end/91680
* match.pd ((A / (1 << B)) -> (A >> B)): Allow widening cast from
the shift type to type.
* gcc.dg/tree-ssa/pr91680.c: New test.
* g++.dg/torture/pr91680.C: New test.
From-SVN: r275587
/cp
2019-09-10 Paolo Carlini <paolo.carlini@oracle.com>
* decl.c (has_designator_problem): Use cp_expr_loc_or_input_loc
in error_at.
(build_enumerator): Likewise.
(cp_finish_decl): Use DECL_SOURCE_LOCATION.
(grokdeclarator): Use id_loc in two error_at; change errror
message about constinit together constexpr to use two ranges.
/testsuite
2019-09-10 Paolo Carlini <paolo.carlini@oracle.com>
* g++.dg/cpp0x/enum29.C: Test location(s) too.
* g++.dg/cpp0x/lambda/lambda-ice10.C: Likewise.
* g++.dg/cpp2a/constinit3.C: Likewise.
* g++.dg/ext/desig4.C: Likewise.
* g++.dg/ext/label10.C: Likewise.
* g++.old-deja/g++.other/dtor3.C: Likewise.
From-SVN: r275585
The ldaddr macro in sjlj.S needs to be updated to support the FDPIC
model.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
libitm/
* config/arm/sjlj.S (ldaddr): Add FDPIC support.
From-SVN: r275584
Since FDPIC does not support -static, skip the related tests.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
gcc/testsuite/
* lib/target-supports.exp (check_effective_target_static): Disable
for ARM FDPIC target.
From-SVN: r275583
The recent stack_protect_combined_set_insn and
stack_protect_combined_test_insn force recomputing of GOT base, but
need to take into account that in FDPIC mode, the PIC register is
fixed by the ABI (r9).
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
gcc/
* config/arm/arm.md (stack_protect_combined_set_insn): Handle
FDPIC mode.
(stack_protect_combined_test_insn): Likewise.
From-SVN: r275582
uclibc defines bswap_32, so use a different name in this test.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
gcc/testsuite/
* gcc.target/arm/pr43698.c (bswap_32): Rename as my_bswap_32.
From-SVN: r275581
Some tests fail on arm*-*-uclinuxfdpiceabi because it generates PIC
code and they don't support it: skip them. They also fail on
arm*-linux* when forcing -fPIC.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
gcc/testsuite/
* gcc.target/arm/eliminate.c: Accept only nonpic targets.
* g++.dg/other/anon5.C: Likewise.
From-SVN: r275578
Without this, when we are unwinding across a signal frame we can jump
to an even address which leads to an exception.
This is needed in __gnu_persnality_sigframe_fdpic() when restoring the
PC from the signal frame since the PC saved by the kernel has the LSB
bit set to zero.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
libgcc/
* config/arm/unwind-arm.c (_Unwind_VRS_Set): Handle thumb-only
architecture.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275575
We call __aeabi_read_tp() to get the thread pointer. Since this is a
function call, we have to restore the FDPIC register afterwards.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
gcc/
* config/arm/arm.c (arm_load_tp): Add FDPIC support.
* config/arm/arm.md (FDPIC_REGNUM): New constant.
(load_tp_soft_fdpic): New pattern.
(load_tp_soft): Disable in FDPIC mode.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275574
In FDPIC mode, the trampoline generated to support pointers to nested
functions looks like:
.word trampoline address
.word trampoline GOT address
ldr r12, [pc, #8]
ldr r9, [pc, #8]
ldr pc, [pc, #8]
.word static chain value
.word GOT address
.word function's address
because in FDPIC function pointers are actually pointers to function
descriptors, we have to actually generate a function descriptor for
the trampoline.
2019--09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
gcc/
* config/arm/arm.c (arm_asm_trampoline_template): Add FDPIC
support.
(arm_trampoline_init): Likewise.
(arm_trampoline_adjust_address): Likewise.
* config/arm/arm.h (TRAMPOLINE_SIZE): Likewise.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275571
Use local binding rules to decide whether we can use GOTOFFFUNCDESC to
compute the function address.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
gcc/
* config/arm/arm.c (arm_fdpic_local_funcdesc_p): New function.
(legitimize_pic_address): Enforce binding rules on function
pointers in FDPIC mode.
(arm_assemble_integer): Likewise.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275570
The main difference with existing support is that function addresses
are function descriptor addresses instead. This means that all code
dealing with function pointers now has to cope with function
descriptors instead.
For the same reason, Linux kernel helpers can no longer be called by
dereferencing their address, so we implement wrappers that directly
call the kernel helpers.
When restoring a function address, we also have to restore the FDPIC
register value (r9).
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
gcc/
* ginclude/unwind-arm-common.h (unwinder_cache): Add reserved5
field.
libgcc/
* config/arm/linux-atomic.c (__kernel_cmpxchg): Add FDPIC support.
(__kernel_dmb): Likewise.
(__fdpic_cmpxchg): New function.
(__fdpic_dmb): New function.
* config/arm/unwind-arm.h (FDPIC_REGNUM): New define.
(gnu_Unwind_Find_got): New function.
(_Unwind_decode_typeinfo_ptr): Add FDPIC support.
* unwind-arm-common.inc (UCB_PR_GOT): New.
(funcdesc_t): New struct.
(get_eit_entry): Add FDPIC support.
(unwind_phase2): Likewise.
(unwind_phase2_forced): Likewise.
(__gnu_Unwind_RaiseException): Likewise.
(__gnu_Unwind_Resume): Likewise.
(__gnu_Unwind_Backtrace): Likewise.
* unwind-pe.h (read_encoded_value_with_base): Likewise.
libstdc++/
* libsupc++/eh_personality.cc (get_ttype_entry): Add FDPIC
support.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275568
In FDPIC, we need to make sure __do_global_dtors_aux and frame_dummy
are referenced by their address, not by pointers to the function
descriptors.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
libgcc/
* libgcc/crtstuff.c: Add support for FDPIC.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275567
The FDPIC register is hard-coded to r9, as defined in the ABI.
We have to disable tailcall optimizations if we don't know if the
target function is in the same module. If not, we have to set r9 to
the value associated with the target module.
When generating a symbol address, we have to take into account whether
it is a pointer to data or to a function, because different
relocations are needed.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
gcc/
* config/arm/arm-c.c (__FDPIC__): Define new pre-processor macro
in FDPIC mode.
* config/arm/arm-protos.h (arm_load_function_descriptor): Declare
new function.
* config/arm/arm.c (arm_option_override): Define pic register to
FDPIC_REGNUM.
(arm_function_ok_for_sibcall): Disable sibcall optimization if we
have no decl or go through PLT.
(calculate_pic_address_constant): New function.
(legitimize_pic_address): Call calculate_pic_address_constant.
(arm_load_pic_register): Handle TARGET_FDPIC.
(arm_is_segment_info_known): New function.
(arm_pic_static_addr): Add support for FDPIC.
(arm_load_function_descriptor): New function.
(arm_emit_call_insn): Add support for FDPIC.
(arm_assemble_integer): Add support for FDPIC.
* config/arm/arm.h (PIC_OFFSET_TABLE_REG_CALL_CLOBBERED):
Define. (FDPIC_REGNUM): New define.
* config/arm/arm.md (call): Add support for FDPIC.
(call_value): Likewise.
(restore_pic_register_after_call): New pattern.
(untyped_call): Disable if FDPIC.
(untyped_return): Likewise.
* config/arm/unspecs.md (UNSPEC_PIC_RESTORE): New.
gcc/testsuite/
* gcc.target/arm/fp16-aapcs-2.c: Adjust scan-assembler-times.
* gcc.target/arm/fp16-aapcs-4.c: Likewise.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275566
In FDPIC mode, we set -fPIE unless the user provides -fno-PIE, -fpie,
-fPIC or -fpic: indeed FDPIC code is PIC, but we want to generate code
for executables rather than shared libraries by default.
We also make sure to use the --fdpic assembler option, and select the
appropriate linker emulation.
At link time, we also default to -pie, unless we are generating a
shared library or a relocatable file (-r). Note that static link is
not supported as it requires specifying the dynamic linker because the
executable still has to relocate itself at startup.
We also force 'now' binding since lazy binding is not supported.
We should also apply the same behavior for -Wl,-Ur as for -r, but I
couldn't find how to describe that in the specs fragment.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
Mickaël Guêné <mickael.guene@st.com>
gcc/
* config.gcc: Handle arm*-*-uclinuxfdpiceabi.
* config/arm/bpabi.h (TARGET_FDPIC_ASM_SPEC): New.
(SUBTARGET_EXTRA_ASM_SPEC): Use TARGET_FDPIC_ASM_SPEC.
* config/arm/linux-eabi.h (FDPIC_CC1_SPEC): New.
(CC1_SPEC): Use FDPIC_CC1_SPEC.
(MUSL_DYNAMIC_LINKER): Add -fdpic suffix when needed.
* config/arm/uclinuxfdpiceabi.h: New file.
libsanitizer/
* configure.tgt (arm*-*-*fdpiceabi): Sanitizers are
unsupported in this configuration.
Co-Authored-By: Mickaël Guêné <mickael.guene@st.com>
From-SVN: r275565
The new arm-uclinuxfdpiceabi target behaves pretty much like
arm-linux-gnueabi. In order to enable the same set of features, we
have to update several configure scripts that generally match targets
like *-*-linux*: in most places, we add *-uclinux* where there is
already *-linux*, or uclinux* when there is already linux*.
In gcc/config.gcc and libgcc/config.host we use *-*-uclinuxfdpiceabi
because there is already a different behaviour for *-*uclinux* target.
In libtool.m4, we use uclinuxfdpiceabi in cases where ELF shared
libraries support is required, as uclinux does not guarantee that.
2019-09-10 Christophe Lyon <christophe.lyon@st.com>
config/
* futex.m4: Handle *-uclinux*.
* tls.m4 (GCC_CHECK_TLS): Likewise.
gcc/
* config.gcc: Handle *-*-uclinuxfdpiceabi.
libatomic/
* configure.tgt: Handle arm*-*-uclinux*.
* configure: Regenerate.
libgcc/
* config.host: Handle *-*-uclinuxfdpiceabi.
libitm/
* configure.tgt: Handle *-*-uclinux*.
* configure: Regenerate.
* libtool.m4: Handle uclinuxfdpiceabi.
From-SVN: r275564
Define partial specializations for std::remove_cv so that
std::remove_const and std::remove_volatile don't need to be
instantiated.
* include/std/type_traits (__remove_cv_t): New alias template.
(is_void, is_integral, is_floating_point, is_pointer)
(is_member_object_pointer, is_member_function_pointer, is_null_pointer)
(is_member_point), __is_signed_integer, __is_unsigned_integer)
(__make_unsigned_selector, __make_signed_selector, remove_pointer)
(__decay_selector): Use __remove_cv_t.
(remove_cv): Add partial specializations for cv-qualified types.
(__decay_t): New alias template.
(__decay_and_strip, __common_type_impl, __result_of_impl): Use
__decay_t.
(__enable_if_t): Move earlier in the file.
(_Require): Use __enable_if_t.
(swap(T&, T&)): Use _Require.
(swap(T(&)[N])): Use __enable_if_t.
From-SVN: r275562
This relatively minor change increases the number of inlinable
functions/methods in the standard library from 983 to 2179.
In particular it permits inlining math/bits/RotateLeftNN. This
restores the speed of crypto/sha256 back to what it was before the
update to 1.13beta1.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/194340
From-SVN: r275558
They were lost when the files were moved in the update to Go1.13beta1.
These changes should be made in the master repo for the 1.14 release,
as riscv64 support is added there.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/194343
From-SVN: r275551
The C file has a build tag, but the procedure we use for building C
files ignores build tags.
This should fix the libgo build on non-x86 systems.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/194378
From-SVN: r275544
2019-09-09 Bernd Edlinger <bernd.edlinger@hotmail.de>
* expmed.c (extract_bit_field): Update function comment
regarding alt_rtl.
* expr.c (expand_expr_real): Update function comment
regarding alt_rtl.
(expand_misaligned_mem_ref): New helper function.
(expand_expr_real_2): Use expand_misaligned_mem_ref.
Remove duplicate assignment to "base" at case MEM_REF.
Remove a shadowed variable "unsignedp" at case VCE.
From-SVN: r275541
For the main package we add an implicit import of the runtime package,
to ensure that it is initialized. That import used the predeclared
location, which caused various tests, notably Named_type::is_builtin,
to treat these imported names as builtin. Start using a real location,
so that those tests do the right thing.
By the way, this implicit import is a partial cause of golang/go#19773.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/194337
From-SVN: r275540
We forgot to ever traverse types of constant expressions. This rarely
makes a difference--evidently, since nobody noticed--but it does
matter when we inline constant expressions: we need to ensure that the
type is visible to the importing code.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/194317
From-SVN: r275539
