llvm-project/llvm
Matt Arsenault eebe841a47 RegAlloc: Allow targets to split register allocation
AMDGPU normally spills SGPRs to VGPRs. Previously, since all register
classes are handled at the same time, this was problematic. We don't
know ahead of time how many registers will be needed to be reserved to
handle the spilling. If no VGPRs were left for spilling, we would have
to try to spill to memory. If the spilled SGPRs were required for exec
mask manipulation, it is highly problematic because the lanes active
at the point of spill are not necessarily the same as at the restore
point.

Avoid this problem by fully allocating SGPRs in a separate regalloc
run from VGPRs. This way we know the exact number of VGPRs needed, and
can reserve them for a second run.  This fixes the most serious
issues, but it is still possible using inline asm to make all VGPRs
unavailable. Start erroring in the case where we ever would require
memory for an SGPR spill.

This is implemented by giving each regalloc pass a callback which
reports if a register class should be handled or not. A few passes
need some small changes to deal with leftover virtual registers.

In the AMDGPU implementation, a new pass is introduced to take the
place of PrologEpilogInserter for SGPR spills emitted during the first
run.

One disadvantage of this is currently StackSlotColoring is no longer
used for SGPR spills. It would need to be run again, which will
require more work.

Error if the standard -regalloc option is used. Introduce new separate
-sgpr-regalloc and -vgpr-regalloc flags, so the two runs can be
controlled individually. PBQB is not currently supported, so this also
prevents using the unhandled allocator.
2021-07-13 18:49:29 -04:00
..
benchmarks
bindings
cmake [CMake] Disable -fno-semantic-interposition for GCC<10.3 on SystemZ 2021-07-08 22:09:21 -07:00
docs [docs/llvm-cov] Document -compilation-dir 2021-07-13 13:10:02 -07:00
examples [Orc] At CBindings for LazyRexports 2021-07-01 21:52:05 +02:00
include RegAlloc: Allow targets to split register allocation 2021-07-13 18:49:29 -04:00
lib RegAlloc: Allow targets to split register allocation 2021-07-13 18:49:29 -04:00
projects [RFC][debuginfo-test] Rename debug-info lit tests for general purposes 2021-06-28 11:31:40 +01:00
resources
runtimes [runtimes] Fix umbrella component targets 2021-06-12 19:49:44 -07:00
test RegAlloc: Allow targets to split register allocation 2021-07-13 18:49:29 -04:00
tools [LLD] Adding support for RELA for CG Profile. 2021-07-13 13:56:30 -07:00
unittests Revert "[llvm] Add enum iteration to Sequence" 2021-07-13 16:44:42 +00:00
utils [gn build] (manually) port 303ddb60a2 2021-07-13 15:15:38 -04:00
.clang-format
.clang-tidy NFC: .clang-tidy: Inherit configs from parents to improve maintainability 2021-06-08 08:25:59 -07:00
.gitattributes
.gitignore
CMakeLists.txt [RFC][debuginfo-test] Rename debug-info lit tests for general purposes 2021-06-28 11:31:40 +01:00
CODE_OWNERS.TXT
CREDITS.TXT
LICENSE.TXT
README.txt
RELEASE_TESTERS.TXT
configure
llvm.spec.in

README.txt

The LLVM Compiler Infrastructure
================================

This directory and its subdirectories contain source code for LLVM,
a toolkit for the construction of highly optimized compilers,
optimizers, and runtime environments.

LLVM is open source software. You may freely distribute it under the terms of
the license agreement found in LICENSE.txt.

Please see the documentation provided in docs/ for further
assistance with LLVM, and in particular docs/GettingStarted.rst for getting
started with LLVM and docs/README.txt for an overview of LLVM's
documentation setup.

If you are writing a package for LLVM, see docs/Packaging.rst for our
suggestions.