This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter.
Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible.
To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future.
To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin.
To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute.
To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins.
There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch.
Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue.
Differential Revision: https://reviews.llvm.org/D48617
llvm-svn: 336583
As reported here: https://bugs.llvm.org/show_bug.cgi?id=37033
Any usage of a builtin function that uses a va_list by reference
will cause an assertion when redeclaring it.
After discussion in the review, it was concluded that the correct
way of accomplishing this fix is to make attempts to redeclare certain
builtins an error. Unfortunately, doing this limitation for all builtins
is likely a breaking change, so this commit simply limits it to
types with custom type checking and those that take a reference.
Two tests needed to be updated to make this work.
Differential Revision: https://reviews.llvm.org/D45383
llvm-svn: 330160
Currently we only have OpenCL 2.0 Builtins i.e. pipes or address space conversions.
They have to be added only in the version 2.0 compilation mode to make the identifiers
available for use in the other versions.
Review: http://reviews.llvm.org/D20249
llvm-svn: 274509
Fix arc patch fuzz error.
Summary:
Support for the pipe built-in functions for OpenCL 2.0.
The pipe builtin functions may have infinite kinds of element types, one approach
would be to just generate calls that would always use generic types such as void*.
This patch is based on bader's opencl support patch on SPIR-V branch.
Reviewers: Anastasia, pekka.jaaskelainen
Subscribers: keryell, bader, cfe-commits
Differential Revision: http://reviews.llvm.org/D15914
llvm-svn: 258782
Summary:
Support for the pipe built-in functions for OpenCL 2.0.
The pipe builtin functions may have infinite kinds of element types, one approach
would be to just generate calls that would always use generic types such as void*.
This patch is based on bader's opencl support patch on SPIR-V branch.
Reviewers: Anastasia, pekka.jaaskelainen
Subscribers: keryell, bader, cfe-commits
Differential Revision: http://reviews.llvm.org/D15914
llvm-svn: 258773
* adds -aux-triple option to specify target triple
* propagates aux target info to AST context and Preprocessor
* pulls in target specific preprocessor macros.
* pulls in target-specific builtins from aux target.
* sets appropriate host or device attribute on builtins.
Differential Revision: http://reviews.llvm.org/D12917
llvm-svn: 248299
This extends the target builtin support to allow language specific annotations
(i.e. LANGBUILTIN). This is to allow MSVC compatibility whilst retaining the
ability to have EABI targets use a __builtin_ prefix. This is merely to allow
uniformity in the EABI case where the unprefixed name is provided as an alias in
the header.
llvm-svn: 212196
We already have builtins that are only available in GNU mode, so this
mirrors that.
Reviewers: rsmith
Differential Revision: http://llvm-reviews.chandlerc.com/D2128
llvm-svn: 194615
uncovered.
This required manually correcting all of the incorrect main-module
headers I could find, and running the new llvm/utils/sort_includes.py
script over the files.
I also manually added quite a few missing headers that were uncovered by
shuffling the order or moving headers up to be main-module-headers.
llvm-svn: 169237
and language-specific initialization. Use this to allow ASTUnit to
create a preprocessor object *before* loading the AST file. No actual
functionality change.
llvm-svn: 138983
inconsistent with the type that the builtin *should* have, forget
about the builtin altogether: we don't want subsequence analyses,
CodeGen, etc., to think that we have a proper builtin function.
C is protected from errors here because it allows one to use a
library builtin without having a declaration, and detects inconsistent
(re-)declarations of builtins during declaration merging. C++ was
unprotected, and therefore would crash.
Fixes PR8839.
llvm-svn: 122351
a specific language. We are adding such language info. by
extensing Builtins.def and via a language flag added
to LIBBUILTIN/BUILTIN and check for that when deciding
a name is builtin or not. Implements //rdar://8689273.
llvm-svn: 120429
represent builtins that have the "scanf" attribution (via the format attribute) just
like we do with printf functions. Follow-up work is needed to add similar support
for fscanf et al.
This is to support format-string checking for scanf functions.
llvm-svn: 108499
registered when PCH wasn't being used. We should always install (in BuiltinInfo)
information about target-specific builtins, but we shouldn't register any builtin
identifier infos. This fixes the build of apps that use PCH and target specific
builtins together.
llvm-svn: 73492
preprocessor and initialize it early in clang-cc. This
ensures that __has_builtin works in all modes, not just
when ASTContext is around.
llvm-svn: 73319
Craig Topper