For targets that support xnack replay feature (gfx8+), the
multi-dword scalar loads shouldn't clobber any register that
holds the src address. The constrained version of the scalar
loads have the early clobber flag attached to the dst operand
to restrict RA from re-allocating any of the src regs for its
dst operand.
For unbuffered smem loads, it is illegal for the immediate offset to be
negative if the resulting IOFFSET + (SGPR[Offset] or M0 or zero) is
negative.
New PR of https://github.com/llvm/llvm-project/pull/79553.
Similar to 806761a7629df268c8aed49657aeccffa6bca449.
For IR files without a target triple, -mtriple= specifies the full
target triple while -march= merely sets the architecture part of the
default target triple, leaving a target triple which may not make sense,
e.g. amdgpu-apple-darwin.
Therefore, -march= is error-prone and not recommended for tests without
a target triple. The issue has been benign as we recognize
$unknown-apple-darwin as ELF instead of rejecting it outrightly.
This patch changes AMDGPU tests to not rely on the default
OS/environment components. Tests that need fixes are not changed:
```
LLVM :: CodeGen/AMDGPU/fabs.f64.ll
LLVM :: CodeGen/AMDGPU/fabs.ll
LLVM :: CodeGen/AMDGPU/floor.ll
LLVM :: CodeGen/AMDGPU/fneg-fabs.f64.ll
LLVM :: CodeGen/AMDGPU/fneg-fabs.ll
LLVM :: CodeGen/AMDGPU/r600-infinite-loop-bug-while-reorganizing-vector.ll
LLVM :: CodeGen/AMDGPU/schedule-if-2.ll
```
UniformityAnalysis offers all of the same features and much more, there is no reason left to use the legacy DAs.
See RFC: https://discourse.llvm.org/t/rfc-deprecate-divergenceanalysis-legacydivergenceanalysis/69538
- Remove LegacyDivergenceAnalysis.h/.cpp
- Remove DivergenceAnalysis.h/.cpp + Unit tests
- Remove SyncDependenceAnalysis - it was not a real registered analysis and was only used by DAs
- Remove/adjust references to the passes in the docs where applicable
- Remove TTI hook associated with those passes.
- Move tests to UniformityAnalysis folder.
- Remove RUN lines for the DA, leave only the UA ones.
- Some tests had to be adjusted/removed depending on how they used the legacy DAs.
Reviewed By: foad, sameerds
Differential Revision: https://reviews.llvm.org/D148116
Support for XNACK and SRAMECC is not static on some GPUs. We must be able
to differentiate between different scenarios for these dynamic subtarget
features.
The possible settings are:
- Unsupported: The GPU has no support for XNACK/SRAMECC.
- Any: Preference is unspecified. Use conservative settings that can run anywhere.
- Off: Request support for XNACK/SRAMECC Off
- On: Request support for XNACK/SRAMECC On
GCNSubtarget will track the four options based on the following criteria. If
the subtarget does not support XNACK/SRAMECC we say the setting is
"Unsupported". If no subtarget features for XNACK/SRAMECC are requested we
must support "Any" mode. If the subtarget features XNACK/SRAMECC exist in the
feature string when initializing the subtarget, the settings are "On/Off".
The defaults are updated to be conservatively correct, meaning if no setting
for XNACK or SRAMECC is explicitly requested, defaults will be used which
generate code that can be run anywhere. This corresponds to the "Any" setting.
Differential Revision: https://reviews.llvm.org/D85882
This reverts commit ca907bfb57d8ad3ec3bcc2cff2abab7b1b933af6.
According to michel.daenzer,
> This completely broke the Mesa radeonsi driver on Navi 14. Xorg +
> xterm come up with major corruption & psychedelic colours.
When memory operations are outstanding on function calls, either the
caller or the callee can insert a waitcnt to ensure that all reads are
finished.
Calls need some time to be executed, so if the callee inserts the
waitcnt, filling the instruction buffer and waiting for memory will be
interleaved, hiding some latency. This comes at the cost of having a
waitcnt inside functions that may not be needed as no memory operations
are outstanding.
For function calls, this is already implemented. The same principal
applies to returns: If the caller inserts a waitcnt after the call, the
callee does not have to wait and the return and memory operation can be
run in parallel.
This commit implements waiting in the caller after returning from a
function call.
Differential Revision: https://reviews.llvm.org/D87674
Enable clausing of memory loads on gfx10 by adding a new pass to insert
the s_clause instructions that mark the start of each hard clause.
Differential Revision: https://reviews.llvm.org/D79792
Summary:
This fixes a few issues related to SMRD offsets. On gfx9 and gfx10 we have a
signed byte offset immediate, however we can overflow into a negative since we
treat it as unsigned.
Also, the SMRD SOFFSET sgpr is an unsigned offset on all subtargets. We
sometimes tried to use negative values here.
Third, S_BUFFER instructions should never use a signed offset immediate.
Differential Revision: https://reviews.llvm.org/D77082
"Divergence driven ISel. Assign register class for cross block values
according to the divergence."
that discovered the design flaw leading to several issues that
required to be solved before.
This change reverts AMDGPU specific changes and keeps common part
unaffected.
llvm-svn: 362749
Details: To make instruction selection really divergence driven it is necessary to assign
the correct register classes to the cross block values beforehand. For the divergent targets
same value type requires different register classes dependent on the value divergence.
Reviewers: rampitec, nhaehnle
Differential Revision: https://reviews.llvm.org/D59990
This commit was reverted because of the build failure.
The reason was mlformed patch.
Build failure fixed.
llvm-svn: 361741
Details: To make instruction selection really divergence driven it is necessary to assign
the correct register classes to the cross block values beforehand. For the divergent targets
same value type requires different register classes dependent on the value divergence.
Reviewers: rampitec, nhaehnle
Differential Revision: https://reviews.llvm.org/D59990
llvm-svn: 361644
Summary:
Moving SMRD to VMEM in SIFixSGPRCopies is rather bad for performance if
the load is really uniform. So select the scalar load intrinsics directly
to either VMEM or SMRD buffer loads based on divergence analysis.
If an offset happens to end up in a VGPR -- either because a floating
point calculation was involved, or due to other remaining deficiencies
in SIFixSGPRCopies -- we use v_readfirstlane.
There is some unrelated churn in tests since we now select MUBUF offsets
in a unified way with non-scalar buffer loads.
Change-Id: I170e6816323beb1348677b358c9d380865cd1a19
Reviewers: arsenm, alex-t, rampitec, tpr
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D53283
llvm-svn: 348050
An extractelement with non-constant index will be lowered either to
scratch or movrel loop in most cases. This patch converts such
instruction into a set of selects if vector size is not too big.
Differential Revision: https://reviews.llvm.org/D54351
llvm-svn: 346800
Summary:
Moving SMRD to VMEM in SIFixSGPRCopies is rather bad for performance if
the load is really uniform. So select the scalar load intrinsics directly
to either VMEM or SMRD buffer loads based on divergence analysis.
If an offset happens to end up in a VGPR -- either because a floating
point calculation was involved, or due to other remaining deficiencies
in SIFixSGPRCopies -- we use v_readfirstlane.
There is some unrelated churn in tests since we now select MUBUF offsets
in a unified way with non-scalar buffer loads.
Change-Id: I170e6816323beb1348677b358c9d380865cd1a19
Reviewers: arsenm, alex-t, rampitec, tpr
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D53283
llvm-svn: 344696
Summary:
Patch by Marek Olsak and David Stuttard, both of AMD.
This adds a new amdgcn intrinsic supporting s.buffer.load, in particular
multiple dword variants. These are convenient to use from some front-end
implementations.
Also modified the existing llvm.SI.load.const intrinsic to common up the
underlying implementation.
This modification also requires that we can lower to non-uniform loads correctly
by splitting larger dword variants into sizes supported by the non-uniform
versions of the load.
V2: Addressed minor review comments.
V3: i1 glc is now i32 cachepolicy for consistency with buffer and
tbuffer intrinsics, plus fixed formatting issue.
V4: Added glc test.
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D51098
Change-Id: I83a6e00681158bb243591a94a51c7baa445f169b
llvm-svn: 340684
This allows to hoist code portion to compute reciprocal of loop
invariant denominator in integer division after codegen prepare
expansion.
Differential Revision: https://reviews.llvm.org/D48604
llvm-svn: 335988
This replaces most argument uses with loads, but for
now not all.
The code in SelectionDAG for calling convention lowering
is actively harmful for amdgpu_kernel. It attempts to
split the argument types into register legal types, which
results in low quality code for arbitary types. Since
all kernel arguments are passed in memory, we just want the
raw types.
I've tried a couple of methods of mitigating this in SelectionDAG,
but it's easier to just bypass this problem alltogether. It's
possible to hack around the problem in the initial lowering,
but the real problem is the DAG then expects to be able to use
CopyToReg/CopyFromReg for uses of the arguments outside the block.
Exposing the argument loads in the IR also has the advantage
that the LoadStoreVectorizer can merge them.
I'm not sure the best approach to dealing with the IR
argument list is. The patch as-is just leaves the IR arguments
in place, so all the existing code will still compute the same
kernarg size and pointlessly lowers the arguments.
Arguably the frontend should emit kernels with an empty argument
list in the first place. Alternatively a dummy array could be
inserted as a single argument just to reserve space.
This does have some disadvantages. Local pointer kernel arguments can
no longer have AssertZext placed on them as the equivalent !range
metadata is not valid on pointer typed loads. This is mostly bad
for SI which needs to know about the known bits in order to use the
DS instruction offset, so in this case this is not done.
More importantly, this skips noalias arguments since this pass
does not yet convert this to the equivalent !alias.scope and !noalias
metadata. Producing this metadata correctly seems to be tricky,
although this logically is the same as inlining into a function which
doesn't exist. Additionally, exposing these loads to the vectorizer
may result in degraded aliasing information if a pointer load is
merged with another argument load.
I'm also not entirely sure this is preserving the current clover
ABI, although I would greatly prefer if it would stop widening
arguments and match the HSA ABI. As-is I think it is extending
< 4-byte arguments to 4-bytes but doesn't align them to 4-bytes.
llvm-svn: 335650
Summary:
This handles def-after-use of physregs, and allows us to merge loads and
stores even across some physreg defs (typically M0 defs).
Change-Id: I076484b2bda27c2cf46013c845a0380c5b89b67b
Reviewers: arsenm, mareko, rampitec
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D42647
llvm-svn: 325882
