There comes a time in the life of any amateur code generator when dumb string
concatenation just won't cut it any more. For NeonEmitter.cpp, that time has
come.
There were a bunch of magic type codes which meant different things depending on
the context. There were a bunch of special cases that really had no reason to be
there but the whole thing was so creaky that removing them would cause something
weird to fall over. There was a 1000 line switch statement for code generation
involving string concatenation, which actually did lexical scoping to an extent
(!!) with a bunch of semi-repeated cases.
I tried to refactor this three times in three different ways without
success. The only way forward was to rewrite the entire thing. Luckily the
testing coverage on this stuff is absolutely massive, both with regression tests
and the "emperor" random test case generator.
The main change is that previously, in arm_neon.td a bunch of "Operation"s were
defined with special names. NeonEmitter.cpp knew about these Operations and
would emit code based on a huge switch. Actually this doesn't make much sense -
the type information was held as strings, so type checking was impossible. Also
TableGen's DAG type actually suits this sort of code generation very well
(surprising that...)
So now every operation is defined in terms of TableGen DAGs. There are a bunch
of operators to use, including "op" (a generic unary or binary operator), "call"
(to call other intrinsics) and "shuffle" (take a guess...). One of the main
advantages of this apart from making it more obvious what is going on, is that
we have proper type inference. This has two obvious advantages:
1) TableGen can error on bad intrinsic definitions easier, instead of just
generating wrong code.
2) Calls to other intrinsics are typechecked too. So
we no longer need to work out whether the thing we call needs to be the Q-lane
version or the D-lane version - TableGen knows that itself!
Here's an example: before:
case OpAbdl: {
std::string abd = MangleName("vabd", typestr, ClassS) + "(__a, __b)";
if (typestr[0] != 'U') {
// vabd results are always unsigned and must be zero-extended.
std::string utype = "U" + typestr.str();
s += "(" + TypeString(proto[0], typestr) + ")";
abd = "(" + TypeString('d', utype) + ")" + abd;
s += Extend(utype, abd) + ";";
} else {
s += Extend(typestr, abd) + ";";
}
break;
}
after:
def OP_ABDL : Op<(cast "R", (call "vmovl", (cast $p0, "U",
(call "vabd", $p0, $p1))))>;
As an example of what happens if you do something wrong now, here's what happens
if you make $p0 unsigned before the call to "vabd" - that is, $p0 -> (cast "U",
$p0):
arm_neon.td:574:1: error: No compatible intrinsic found - looking up intrinsic 'vabd(uint8x8_t, int8x8_t)'
Available overloads:
- float64x2_t vabdq_v(float64x2_t, float64x2_t)
- float64x1_t vabd_v(float64x1_t, float64x1_t)
- float64_t vabdd_f64(float64_t, float64_t)
- float32_t vabds_f32(float32_t, float32_t)
... snip ...
This makes it seriously easy to work out what you've done wrong in fairly nasty
intrinsics.
As part of this I've massively beefed up the documentation in arm_neon.td too.
Things still to do / on the radar:
- Testcase generation. This was implemented in the previous version and not in
the new one, because
- Autogenerated tests are not being run. The testcase in test/ differs from
the autogenerated version.
- There were a whole slew of special cases in the testcase generation that just
felt (and looked) like hacks.
If someone really feels strongly about this, I can try and reimplement it too.
- Big endian. That's coming soon and should be a very small diff on top of this one.
llvm-svn: 211101
Clean up the __has_attribute implementation without modifying its behavior.
Replaces the tablegen-driven AttrSpellings.inc, which lived in the lexing layer with AttrHasAttributeImpl.inc, which lives in the basic layer. Updates the preprocessor to call through to this new functionality which can take additional information into account (such as scopes and syntaxes).
Expose the ability for parts of the compiler to ask whether an attribute is supported for a given spelling (including scope), syntax, triple and language options.
llvm-svn: 205181
Replaces the tablegen-driven AttrSpellings.inc, which lived in the lexing layer with AttrHasAttributeImpl.inc, which lives in the basic layer. Updates the preprocessor to call through to this new functionality which can take additional information into account (such as scopes and syntaxes).
Expose the ability for parts of the compiler to ask whether an attribute is supported for a given spelling (including scope), syntax, triple and language options.
llvm-svn: 204952
This patch adds some very, very sparse initial documentation for some attributes. Additional effort from attribute authors is greatly appreciated.
llvm-svn: 201515
important for thread safety attributes, which contain expressions that were
not being visited, and were thus invisible to various tools. There are now
Visit*Attr methods that can be overridden for every attribute.
llvm-svn: 198224
which we don't think can't have one, only allow it in the tiny number of
attributes which opts into this weird parse rule.
I've manually checked that the handlers for all these attributes can in fact
cope with an identifier as the argument. This is still somewhat terrible; we
should move more fully towards picking the parsing rules based on the
attribute, and make the Parse -> Sema interface more type-safe.
llvm-svn: 193295
The CMake build was still using it because I forgot to s/CLANG/LLVM/ in
the tablegen() call. The Makefile build is already using llvm-tblgen.
llvm-svn: 184192
This change partly addresses a heinous problem we have with the
parsing of attribute arguments that are a lone identifier. Previously,
we would end up parsing the 'align' attribute of this as an expression
"(Align)":
template<unsigned Size, unsigned Align>
class my_aligned_storage
{
__attribute__((align((Align)))) char storage[Size];
};
while this would parse as a "parameter name" 'Align':
template<unsigned Size, unsigned Align>
class my_aligned_storage
{
__attribute__((align(Align))) char storage[Size];
};
The code that handles the alignment attribute would completely ignore
the parameter name, so the while the first of these would do what's
expected, the second would silently be equivalent to
template<unsigned Size, unsigned Align>
class my_aligned_storage
{
__attribute__((align)) char storage[Size];
};
i.e., use the maximal alignment rather than the specified alignment.
Address this by sniffing the "Args" provided in the TableGen
description of attributes. If the first argument is "obviously"
something that should be treated as an expression (rather than an
identifier to be matched later), parse it as an expression.
Fixes <rdar://problem/13700933>.
llvm-svn: 180973
This change partly addresses a heinous problem we have with the
parsing of attribute arguments that are a lone identifier. Previously,
we would end up parsing the 'align' attribute of this as an expression
"(Align)":
template<unsigned Size, unsigned Align>
class my_aligned_storage
{
__attribute__((align((Align)))) char storage[Size];
};
while this would parse as a "parameter name" 'Align':
template<unsigned Size, unsigned Align>
class my_aligned_storage
{
__attribute__((align(Align))) char storage[Size];
};
The code that handles the alignment attribute would completely ignore
the parameter name, so the while the first of these would do what's
expected, the second would silently be equivalent to
template<unsigned Size, unsigned Align>
class my_aligned_storage
{
__attribute__((align)) char storage[Size];
};
i.e., use the maximal alignment rather than the specified alignment.
Address this by sniffing the "Args" provided in the TableGen
description of attributes. If the first argument is "obviously"
something that should be treated as an expression (rather than an
identifier to be matched later), parse it as an expression.
Fixes <rdar://problem/13700933>.
llvm-svn: 180970
This change introduces a 'kind' attribute for the <Para> tag, that captures the
kind of the parent block command.
For example:
\todo Meow.
used to be just <Para>Meow.</Para>, but now it is
<Para kind="todo">Meow.</Para>
llvm-svn: 174216
This reimplements r173850 with a better approach:
(1) use a TableGen-generated matcher instead of doing a linear search;
(2) avoid allocations for new strings by converting code points to string
iterals with TableGen.
llvm-svn: 173931
Introduce a spelling index to Attr class, which is an index into the attribute spelling list of an attribute defined in Attr.td.
This index will determine the actual spelling used by an attribute, as it incorporates both the syntax and naming of the attribute.
When constructing an attribute AST node, the spelling index is computed based on attribute kind, scope (if it's a C++11 attribute), and
name, then passed to Attr that will use the index to print itself.
Thanks to Richard Smith for the idea and review.
llvm-svn: 173358
Now we have a list of all commands. This is a good thing in itself, but it
also enables us to easily implement typo correction for command names.
With this change we have objects that contain information about each command,
so it makes sense to resolve command name just once during lexing (currently we
store command names as strings and do a linear search every time some property
value is needed). Thus comment token and AST nodes were changed to contain a
command ID -- index into a tables of builtin and registered commands. Unknown
commands are registered during parsing and thus are also uniformly assigned an
ID. Using an ID instead of a StringRef is also a nice memory optimization
since ID is a small integer that fits into a common bitfield in Comment class.
This change implies that to get any information about a command (even a command
name) we need a CommandTraits object to resolve the command ID to CommandInfo*.
Currently a fresh temporary CommandTraits object is created whenever it is
needed since it does not have any state. But with this change it has state --
new commands can be registered, so a CommandTraits object was added to
ASTContext.
Also, in libclang CXComment has to be expanded to include a CXTranslationUnit
so that all functions working on comment AST nodes can get a CommandTraits
object. This breaks binary compatibility of CXComment APIs.
Now clang_FullComment_getAsXML(CXTranslationUnit TU, CXComment CXC) doesn't
need TU parameter anymore, so it was removed. This is a source-incompatible
change for this C API.
llvm-svn: 163540
Benjamin Kramer