We currently don't emit `DW_AT_object_pointer` on function declarations
or definitions. GCC suffers from the same issue:
https://godbolt.org/z/h4jeT54G5
Fixing this will help LLDB in identifying static vs. non-static member
functions (see https://github.com/llvm/llvm-project/issues/120856).
If I interpreted the DWARFv5 spec correctly, it doesn't mandate this
attribute be present *only* for implicit object parameters:
```
If the member function entry describes a non-static member function,
then that entry has a DW_AT_object_pointer attribute whose value is a reference to
the formal parameter entry that corresponds to the object for which the
function is called.
That parameter also has a DW_AT_artificial attribute whose value is true.
```
This patch attaches the `DW_AT_object_pointer` for function
*defintions*. The declarations will be handled in a separate patch.
The part about `DW_AT_artificial` seems overly restrictive, and not true
for explicit object parameters. We probably should relax this part of
the DWARF spec.
Partially fixes https://github.com/llvm/llvm-project/issues/120974
This reverts commit 81fc3add1e627c23b7270fe2739cdacc09063e54.
This breaks some LLDB tests, e.g.
SymbolFile/DWARF/x86/no_unique_address-with-bitfields.cpp:
lldb: ../llvm-project/clang/lib/AST/Decl.cpp:4604: unsigned int clang::FieldDecl::getBitWidthValue() const: Assertion `isa<ConstantExpr>(getBitWidth())' failed.
Save the bitwidth value as a `ConstantExpr` with the value set. Remove
the `ASTContext` parameter from `getBitWidthValue()`, so the latter
simply returns the value from the `ConstantExpr` instead of
constant-evaluating the bitwidth expression every time it is called.
This PR fix the debug infor generation for RWBuffer types.
- This implements the [same fix as
DXC](https://github.com/microsoft/DirectXShaderCompiler/pull/6296).
- Adds the HLSLAttributedResource debug info generation
Closes#118523
---------
Co-authored-by: Joao Saffran <jderezende@microsoft.com>
Co-authored-by: joaosaffran <joao.saffran@microsoft.com>
This is PR is updating the debug info generation for `this`. This is
required to fix the generation of debug information for HLSL RWBuffer
type. This was required from another PR:
https://github.com/llvm/llvm-project/pull/119041/files
Co-authored-by: Joao Saffran <jderezende@microsoft.com>
In Objective-C, forward declarations are currently represented as:
```
DW_TAG_structure_type
DW_AT_name ("Foo")
DW_AT_declaration (true)
DW_AT_APPLE_runtime_class (DW_LANG_ObjC)
```
However, when compiling with `-gmodules`, when a class definition is
turned into a forward declaration within a `DW_TAG_module`, the DIE for
the forward declaration looks as follows:
```
DW_TAG_structure_type
DW_AT_name ("Foo")
DW_AT_declaration (true)
```
Note the absence of `DW_AT_APPLE_runtime_class`. With recent changes in
LLDB, not being able to differentiate between C++ and Objective-C
forward declarations has become problematic (see attached test-case and
explanation in https://github.com/llvm/llvm-project/pull/119860).
ARM ACLE PR#323[1] adds new modal types for 8-bit floating point
intrinsic.
From the PR#323:
```
ACLE defines the `__mfp8` type, which can be used for the E5M2 and E4M3
8-bit floating-point formats. It is a storage and interchange only type
with no arithmetic operations other than intrinsic calls.
````
The type should be an opaque type and its format in undefined in Clang.
Only defined in the backend by a status/format register, for AArch64 the
FPMR.
This patch is an attempt to the add the mfloat8_t scalar type. It has a
parser and codegen for the new scalar type.
The patch it is lowering to and 8bit unsigned as it has no format. But
maybe we should add another opaque type.
[1] https://github.com/ARM-software/acle/pull/323
Add a new enumeration `SuppressInlineNamespaceMode` to `PrintingPolicy` that
is explicit about how to handle inline namespaces. `SuppressInlineNamespace`
uses that enumeration now instead of a Boolean value.
Specializing a template from an inline namespace should be transparent.
For instance
```
namespace foo {
inline namespace v1 {
template<typename A>
void function(A&);
}
}
namespace foo {
template<>
void function<int>(int&);
}
```
`hasName` should match both declarations of `foo::function`.
Makes the behavior of `matchesNodeFullSlow` and `matchesNodeFullFast`
consistent, fixing an assert inside `HasNameMatcher::matchesNode`.
This was originally added in https://reviews.llvm.org/D142268 to have
LLDB display variable typenames that benefit from suppressing defaulted
template arguments.
We currently represent template aliases as `DW_AT_typedef`s instead of
`DW_TAG_template_alias`. This means for types like:
```
template <class _Tp>
using __remove_cv_t = __remove_cv(_Tp);
template <class _Tp>
using remove_cv_t = __remove_cv_t<_Tp>;
template<typename T>
class optional {
using value_type = T;
remove_cv_t<value_type> __val_;
}
```
we would generate DWARF like:
```
0x0000274f: DW_TAG_typedef
DW_AT_type (0x0000000000002758 "__remove_cv_t<value_type>")
DW_AT_name ("remove_cv_t<value_type>")
```
This is an actual libc++ type layout introduced in
https://github.com/llvm/llvm-project/pull/110355, and uncovered a
shortcoming of LLDB's data-formatter infrastructure, where we cache
formatters on the contents of `DW_AT_name` (which currently wouldn't be
a fully resolved typename for template specializations).
To unblock the libc++ change, I think we can revert this without much
fallout.
Then we have two options for follow-up (or do both):
1. reland this but adjust the LLDB formatter cache so it doesn't cache
formatters for template specializations
2. implement support for `DW_TAG_template_alias` in LLDB (and make Clang
generate them by default).
Introducing `HLSLAttributedResourceType` - a new type that is similar to
`AttributedType` but with additional data specific to HLSL resources.
`AttributeType` currently only stores an attribute kind and no
additional data from the type attribute parameters. This does not really
work for HLSL resources since its type attributes contain non-boolean
values that need to be retained as well.
For example:
```
template <typename T> class RWBuffer {
__hlsl_resource_t [[hlsl::resource_class(uav)]] [[hlsl::is_rov]] handle;
};
```
The data `HLSLAttributedResourceType` needs to eventually store are:
- resource class (SRV, UAV, CBuffer, Sampler)
- texture dimension(1-3)
- flags is_rov, is_array, is_feedback and is_multisample
- contained type
All of these values except contained type will be stored in
`HLSLAttributedResourceType::Attributes` struct and accessed
individually via the fields. There is also `Data` alias that covers all
of these values as a `unsigned` which is used for hashing and the AST
type serialization.
During type attribute processing all HLSL type attributes will be
validated and collected by SemaHLSL (by
`SemaHLSL::handleResourceTypeAttr`) and in the end combined into a
single `HLSLAttributedResourceType` instance (in
`SemaHLSL::ProcessResourceTypeAttributes`). `SemaHLSL` will also need to
short-term store the `TypeLoc` information for the new type that will be
grabbed by `TypeSpecLocFiller` soon after the type is created.
Part 1/2 of #104861
This commit adds an assert to check for a non-null enum definition in
CGDebugInfo::CreateTypeDefinition(const EnumType*), ensuring
precondition validity.
Previous discussion on https://github.com/llvm/llvm-project/pull/97105
This patch is motivated by the debug-info issue in
https://github.com/llvm/llvm-project/issues/48909. Clang is currently
emitting the `DW_AT_artificial` attribute on debug-info entries for
structured bindings whereas GCC does not. GCC's interpretation of the
DWARF spec is more user-friendly in this regard, so we would like to do
the same in Clang. [`CGDebugInfo` uses `isImplicit` to decide which
variables to mark
artificial](0c4023ae3b/clang/lib/CodeGen/CGDebugInfo.cpp (L4783-L4784))
(e.g., `ImplicitParamDecls`, compiler-generated variables, etc.). But
for the purposes of debug-info, when we say "artificial", what we really
mean in many cases is: "not explicitly spelled out in source".
`VarDecl`s that back tuple-like bindings are [technically
compiler-generated](https://github.com/llvm/llvm-project/issues/48909#issuecomment-2238976579),
but that is a confusing notion for debug-info, since these bindings
*are* spelled out in source. The [documentation for
`isImplicit`](68a0d0c762/clang/include/clang/AST/DeclBase.h (L596-L600))
does to some extent imply that implicit variables aren't written in
source.
This patch adds another condition to deciding whether a `VarDecl` should
be marked artificial. Specifically, don't treat structured bindings as
artificial.
**Main alternatives considered**
1. Don't use `isImplicit` in `CGDebugInfo` when determining whether to
add `DW_AT_artificial`. Instead use some other property of the AST that
would tell us whether a node was explicitly spelled out in source or not
* Considered using `SourceRange` or `SourceLocation` to tell us this,
but didn't find a good way to, e.g., correctly identify that the
implicit `this` parameter wasn't spelled out in source (as opposed to an
unnamed parameter in a function declaration)
2. We could've also added a bit to `VarDeclBitFields` that indicates
that a `VarDecl` is a holding var, but the use-case didn't feel like
good enough justification for this
3. Don't set the `VarDecl` introduced as part of a tuple-like
decomposition as implicit.
* This may affect AST matching/traversal and this specific use-case
wasn't enough to justify such a change
Fixes https://github.com/llvm/llvm-project/issues/48909
HLSL has a set of intangible types which are described in in the
[draft HLSL Specification
(**[Basic.types]**)](https://microsoft.github.io/hlsl-specs/specs/hlsl.pdf):
There are special implementation-defined types such as handle types,
which fall into a category of standard intangible types. Intangible
types are types that have no defined object representation or value
representation, as such the size is unknown at compile time.
A class type T is an intangible class type if it contains an base
classes or members of intangible class type, standard intangible type,
or arrays of such types. Standard intangible types and intangible class
types are collectively called intangible
types([9](https://microsoft.github.io/hlsl-specs/specs/hlsl.html#Intangible)).
This PR implements one standard intangible type `__hlsl_resource_t`
and sets up the infrastructure that will make it easier to add more
in the future, such as samplers or raytracing payload handles. The
HLSL intangible types are declared in
`clang/include/clang/Basic/HLSLIntangibleTypes.def` and this file is
included with related macro definition in most places that require edits
when a new type is added.
The new types are added as keywords and not typedefs to make sure they
cannot be redeclared, and they can only be declared in builtin implicit
headers. The `__hlsl_resource_t` type represents a handle to a memory
resource and it is going to be used in builtin HLSL buffer types like this:
template <typename T>
class RWBuffer {
[[hlsl::contained_type(T)]]
[[hlsl::is_rov(false)]]
[[hlsl::resource_class(uav)]]
__hlsl_resource_t Handle;
};
Part 1/3 of llvm/llvm-project#90631.
---------
Co-authored-by: Justin Bogner <mail@justinbogner.com>
Currently we would unconditionally add an implicit `this` parameter when
creating an instance method type. However, when we have an explicit
'this', we shouldn't generate one. This patch only passes a valid
`ThisPtr` type to `getOrCreateInstanceMethodType` if one wasn't
explicitly specified. There's no way to get a pointer to a member
function with an explicit `this` parameter (those are treated as regular
function pointers instead). So there's no need to account for that case
in `CGDebugInfo::CreateType`.
Fixes https://github.com/llvm/llvm-project/issues/99744
The builtin causes the program to stop its execution abnormally and
shows a human-readable description of the reason for the termination
when a debugger is attached or in a symbolicated crash log.
The motivation for the builtin is explained in the following RFC:
https://discourse.llvm.org/t/rfc-adding-builtin-verbose-trap-string-literal/75845
clang's CodeGen lowers the builtin to `llvm.trap` and emits debugging
information that represents an artificial inline frame whose name
encodes the category and reason strings passed to the builtin.
This patch adds a new builtin type for AMDGPU's buffer rsrc data type,
which is effectively an AS 8 pointer. This is needed because we'd like
to expose certain intrinsics to users via builtins which take buffer
rsrc as argument.
Currently we use DW_OP_plus_uconst to handle the bitfield offset and
handle the bitfield size by choosing a type size that matches, but this
doesn't work if either offset or size aren't byte-aligned. Extracting
the bits using DW_OP_LLVM_extract_bits means we can handle any kind of
offset or size.
This is a different implementation to #94100, which has been reverted.
When -fdebug-info-for-profiling is specified, for any Load expression if
the pointer operand is not a declared variable, clang will emit debug
info describing the type of the pointer operand (which can be an
intermediate expr)
The option is causing the binary output to be different when compiled
under `-O0`, because it introduce dbg.declare on pseudovariables. Going
to change this implementation to use dbg.value instead.
CGDebugInfo::EmitPseudoVariable currently uses detailed logic to exactly
collect llvm.dbg.declare users of an alloca. This patch replaces this
with an LLVM function for finding debug declares intrinsics and also
adds the same for debug records, simplifying the code and adding record
support.
No tests added in this commit because it is NFC for intrinsics, and
there is no way to make clang emit records directly yet - one of the
existing tests however will switch to covering debug records once
https://github.com/llvm/llvm-project/pull/89799 relands.
This is another attempt to land #81545, which was reverted.
Fixed test case by adding a target triple so that clang generates the same IR for all platforms
This reverts commit 02c6845c762dfd0a19d4a2f997990e160f392dae,
reapplying bfabc958c7c0d7ddc15f23383d9da836e8c6093f.
The patch was reverted due to the test failing on MacOS and Windows
where type units aren't supported. This is addressed by limiting type
unit flag/test coverage to Linux.
Complete C++ type information can be quite expensive - and there's
limited value in representing every member function, even those that
can't be called (we don't do similarly for every non-member function
anyway). So add a flag to opt out of this behavior for experimenting
with this more terse behavior.
I think Sony already does this by default, so perhaps with a change to
the defaults, Sony can migrate to this rather than a downstream patch.
This breaks current debuggers in some expected ways - but those
breakages are visible without this feature too. Consider member function
template instantiations - they can't be consistently enumerated in every
translation unit:
a.h:
```
struct t1 {
template <int i>
static int f1() {
return i;
}
};
namespace ns {
template <int i>
int f1() {
return i;
}
} // namespace ns
```
a.cpp:
```
void f1() {
t1::f1<0>();
ns::f1<0>();
}
```
b.cpp:
```
void f1();
int main() {
f1();
t1::f1<1>();
ns::f1<1>();
}
```
```
(gdb) p ns::f1<0>()
$1 = 0
(gdb) p ns::f1<1>()
$2 = 1
(gdb) p t1::f1<0>()
Couldn't find method t1::f1<0>
(gdb) p t1::f1<1>()
$3 = 1
(gdb) s
f1 () at a.cpp:3
3 t1::f1<0>();
(gdb) p t1::f1<0>()
$4 = 0
(gdb) p t1::f1<1>()
Couldn't find method t1::f1<1>
(gdb)
```
(other similar non-canonical features are implicit special members
(copy/move ctor/assignment operator, default ctor) and nested types (eg:
pimpl idiom, where the nested type is declared-but-not-defined in one
TU, and defined in another TU))
lldb can't parse the template expressions above, so I'm not sure how to
test it there, but I'd guess it has similar problems. (
https://stackoverflow.com/questions/64602475/how-to-print-value-returned-by-template-member-function-in-gdb-lldb-debugging
so... I guess that's just totally not supported in lldb, how
unfortunate. And implicit special members are instantiated implicitly by
lldb, so missing those doesn't tickle the same issue)
Some very rudimentary numbers for a clang debug build:
.debug_info section size:
-g: 476MiB
-g -fdebug-types-section: 357MiB
-g -gomit-unreferenced-members: 340MiB
Though it also means a major reduction in .debug_str size,
-fdebug-types-section doesn't reduce string usage (so the first two
examples have the same .debug_str size, 247MiB), down to 175MiB.
So for total clang binary size (I don't have a quick "debug section size
reduction" on-hand): 1.45 (no type units) GiB -> 1.34 -> 1.22, so it
saves about 120MiB of binary size.
Original Differential Revision: https://reviews.llvm.org/D152017
This reverts commit aeccfee348c717165541d8d895b9b0cdfe31415c, and dependents:
Revert "[NFC] Fix PPC buildbot failure https://lab.llvm.org/buildbot/#/builders/230/builds/29066"
This reverts commit 2b1d1c51f6e321267cc86e9db7808298c59caf0e.
Revert "Fix test - remove unnecessary/incorrect `-S`, in favor of `-emit-llvm`"
This reverts commit ea1ecb50fa831583241fc531153bd2c072955d29.
The test is failing on MacOs and Windows
Such expression does not correspond to a variable in the source code
thus does not have a debug location. When the user collects perf data on
the program, if the intermediate memory load instruction is sampled, it
could not be attributed to any variable/class member, which causes the
sampling results to be under-counted.
This patch adds an option `-fdebug_info_for_pointer_type` to generate a
psuedo variable and its debug info for intermediate expression with
pointer dereferencing, so that perf data collected on the instruction of
that expression can be attributed to the correct class member.
This is a prototype so comments are needed.
Complete C++ type information can be quite expensive - and there's
limited value in representing every member function, even those that
can't be called (we don't do similarly for every non-member function
anyway). So add a flag to opt out of this behavior for experimenting
with this more terse behavior.
I think Sony already does this by default, so perhaps with a change to
the defaults, Sony can migrate to this rather than a downstream patch.
This breaks current debuggers in some expected ways - but those
breakages are visible without this feature too. Consider member function
template instantiations - they can't be consistently enumerated in every
translation unit:
a.h:
```
struct t1 {
template <int i>
static int f1() {
return i;
}
};
namespace ns {
template <int i>
int f1() {
return i;
}
} // namespace ns
```
a.cpp:
```
void f1() {
t1::f1<0>();
ns::f1<0>();
}
```
b.cpp:
```
void f1();
int main() {
f1();
t1::f1<1>();
ns::f1<1>();
}
```
```
(gdb) p ns::f1<0>()
$1 = 0
(gdb) p ns::f1<1>()
$2 = 1
(gdb) p t1::f1<0>()
Couldn't find method t1::f1<0>
(gdb) p t1::f1<1>()
$3 = 1
(gdb) s
f1 () at a.cpp:3
3 t1::f1<0>();
(gdb) p t1::f1<0>()
$4 = 0
(gdb) p t1::f1<1>()
Couldn't find method t1::f1<1>
(gdb)
```
(other similar non-canonical features are implicit special members
(copy/move ctor/assignment operator, default ctor) and nested types (eg:
pimpl idiom, where the nested type is declared-but-not-defined in one
TU, and defined in another TU))
lldb can't parse the template expressions above, so I'm not sure how to
test it there, but I'd guess it has similar problems. (
https://stackoverflow.com/questions/64602475/how-to-print-value-returned-by-template-member-function-in-gdb-lldb-debugging
so... I guess that's just totally not supported in lldb, how
unfortunate. And implicit special members are instantiated implicitly by
lldb, so missing those doesn't tickle the same issue)
Some very rudimentary numbers for a clang debug build:
.debug_info section size:
-g: 476MiB
-g -fdebug-types-section: 357MiB
-g -gomit-unreferenced-members: 340MiB
Though it also means a major reduction in .debug_str size,
-fdebug-types-section doesn't reduce string usage (so the first two
examples have the same .debug_str size, 247MiB), down to 175MiB.
So for total clang binary size (I don't have a quick "debug section size
reduction" on-hand): 1.45 (no type units) GiB -> 1.34 -> 1.22, so it
saves about 120MiB of binary size.
Also open to any riffing on the flag name for sure.
@probinson - would this be an accurate upstreaming of your internal
handling/would you use this functionality? If it wouldn't be useful to
you, it's maybe not worth adding upstream yet - not sure we'll use it at
Google, but if it was useful to you folks and meant other folks could
test with it it seemed maybe useful.
Original Differential Revision: https://reviews.llvm.org/D152017
Workaround for issue #89774 until it can be properly fixed.
When `-gtemplate-alias` is specified Clang emits a DW_TAG_template_alias
for template aliases. This patch avoids an assertion failure by falling
back to the `-gno-template-alias` (default) behaviour, emitting a
DW_TAG_typedef, if the alias is instantiation dependent.
Fix issue https://github.com/llvm/llvm-project/issues/54624
Add front end flags -gtemplate-alias (also a cc1 flag) and -gno-template-alias
to enable/disable usage of the feature. It's enabled by default in the front
end for SCE debugger tuning only.
GCC emits DW_TAG_typedef for template aliases (as does Clang with this feature
disabled), and GDB and LLDB appear not to support DW_TAG_template_alias.
The -Xclang option -gsimple-template-names=mangled is treated the same as
=full, which is not a regression from current behaviour for template
aliases.
The current implementation omits defaulted arguments and as a consequence
also omits empty parameter packs that come after defaulted arguments. Again,
this isn't a regression as the DW_TAG_typedef name doesn't contain defaulted
arguments.
LLVM support added in https://github.com/llvm/llvm-project/pull/88943
Added template-alias.cpp - Check the metadata construction & interaction with
-gsimple-template-names.
Added variadic-template-alias.cpp - Check template parameter packs work.
Added defaulted-template-alias.cpp - Check defaulted args (don't) work.
Modified debug-options.c - Check Clang generates correct cc1 flags.
HLSL constant sized array function parameters do not decay to pointers.
Instead constant sized array types are preserved as unique types for
overload resolution, template instantiation and name mangling.
This implements the change by adding a new `ArrayParameterType` which
represents a non-decaying `ConstantArrayType`. The new type behaves the
same as `ConstantArrayType` except that it does not decay to a pointer.
Values of `ConstantArrayType` in HLSL decay during overload resolution
via a new `HLSLArrayRValue` cast to `ArrayParameterType`.
`ArrayParamterType` values are passed indirectly by-value to functions
in IR generation resulting in callee generated memcpy instructions.
The behavior of HLSL function calls is documented in the [draft language
specification](https://microsoft.github.io/hlsl-specs/specs/hlsl.pdf)
under the Expr.Post.Call heading.
Additionally the design of this implementation approach is documented in
[Clang's
documentation](https://clang.llvm.org/docs/HLSL/FunctionCalls.html)
Resolves#70123
swifttailcc is a new calling convention in LLVM introduced in
https://reviews.llvm.org/D95443. We add a new DWARF tag to capture
this in debuginfo.
Differential Revision: https://reviews.llvm.org/D95704
[Note: This patch seems to have been forgotten about, landing it now with considerable delay]
[RISCV] RISCV vector calling convention (1/2)
This is the vector calling convention based on
https://github.com/riscv-non-isa/riscv-elf-psabi-doc,
the idea is to split between "scalar" callee-saved registers
and "vector" callee-saved registers. "scalar" ones remain the
original strategy, however, "vector" ones are handled together
with RVV objects.
The stack layout would be:
|--------------------------| <-- FP
| callee-allocated save |
| area for register varargs|
|--------------------------|
| callee-saved registers | <-- scalar callee-saved
| (scalar) |
|--------------------------|
| RVV alignment padding |
|--------------------------|
| callee-saved registers | <-- vector callee-saved
| (vector) |
|--------------------------|
| RVV objects |
|--------------------------|
| padding before RVV |
|--------------------------|
| scalar local variables |
|--------------------------| <-- BP
| variable size objects |
|--------------------------| <-- SP
Note: This patch doesn't contain "tuple" type, e.g. vint32m1x2.
It will be handled in https://github.com/riscv-non-isa/riscv-elf-psabi-doc (2/2).
Differential Revision: https://reviews.llvm.org/D154576
In PR #79382, I need to add a new type that derives from
ConstantArrayType. This means that ConstantArrayType can no longer use
`llvm::TrailingObjects` to store the trailing optional Expr*.
This change refactors ConstantArrayType to store a 60-bit integer and
4-bits for the integer size in bytes. This replaces the APInt field
previously in the type but preserves enough information to recreate it
where needed.
To reduce the number of places where the APInt is re-constructed I've
also added some helper methods to the ConstantArrayType to allow some
common use cases that operate on either the stored small integer or the
APInt as appropriate.
Resolves#85124.
In `-fbounds-safety`, bounds annotations are considered type attributes
rather than declaration attributes. Constructing them as type attributes
allows us to extend the attribute to apply nested pointers, which is
essential to annotate functions that involve out parameters: `void
foo(int *__counted_by(*out_count) *out_buf, int *out_count)`.
We introduce a new sugar type to support bounds annotated types,
`CountAttributedType`. In order to maintain extra data (the bounds
expression and the dependent declaration information) that is not
trackable in `AttributedType` we create a new type dedicate to this
functionality.
This patch also extends the parsing logic to parse the `counted_by`
argument as an expression, which will allow us to extend the model to
support arguments beyond an identifier, e.g., `__counted_by(n + m)` in
the future as specified by `-fbounds-safety`.
This also adjusts `__bdos` and array-bounds sanitizer code that already
uses `CountedByAttr` to check `CountAttributedType` instead to get the
field referred to by the attribute.
The new experimental calling convention preserve_none is the opposite
side of existing preserve_all. It tries to preserve as few general
registers as possible. So all general registers are caller saved
registers. It can also uses more general registers to pass arguments.
This attribute doesn't impact floating-point registers. Floating-point
registers still follow the c calling convention.
Currently preserve_none is supported on X86-64 only. It changes the c
calling convention in following fields:
* RSP and RBP are the only preserved general registers, all other
general registers are caller saved registers.
* We can use [RDI, RSI, RDX, RCX, R8, R9, R11, R12, R13, R14, R15, RAX]
to pass arguments.
It can improve the performance of hot tailcall chain, because many
callee saved registers' save/restore instructions can be removed if the
tail functions are using preserve_none. In my experiment in protocol
buffer, the parsing functions are improved by 3% to 10%.
Implements https://isocpp.org/files/papers/P2662R3.pdf
The feature is exposed as an extension in older language modes.
Mangling is not yet supported and that is something we will have to do before release.
