This reverts commit ca907bfb57.
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
Clustering loads has caching benefits, but as far as I know there is no
advantage to clustering stores on any AMDGPU subtargets.
The disadvantage is that it tends to increase register pressure and
restricts scheduling freedom.
Differential Revision: https://reviews.llvm.org/D85530
If it is load cluster, we don't need to create the dependency edges(SUb->reg) from SUb to SUa
as they both depend on the base register "reg"
+-------+
+----> reg |
| +---+---+
| ^
| |
| |
| |
| +---+---+
| | SUa | Load 0(reg)
| +---+---+
| ^
| |
| |
| +---+---+
+----+ SUb | Load 4(reg)
+-------+
But if it is store cluster, we need to create it as follow shows to avoid the instruction store
depend on scheduled in-between SUb and SUa.
+-------+
+----> reg |
| +---+---+
| ^
| | Missing +-------+
| | +-------------------->+ y |
| | | +---+---+
| +---+-+-+ ^
| | SUa | Store x 0(reg) |
| +---+---+ |
| ^ |
| | +------------------------+
| | |
| +---+--++
+----+ SUb | Store y 4(reg)
+-------+
Reviewed By: evandro, arsenm, rampitec, foad, fhahn
Differential Revision: https://reviews.llvm.org/D72031
Get rid of all fixmes and base heuristic on `num-clustered-dwords`. The main intuition behind this is as
follows. The existing heuristic roughly summarizes as below:
* Assume, all the mem ops instructions participating in the clustering process, loads/stores same num bytes
* If num bytes loaded by each mem op is 4 bytes, then cluster at max 5 mem ops, that is at max 20 bytes
* If num bytes loaded by each mem op is 8 bytes, then cluster at max 3 mem ops, that is at max 24 bytes
* If num bytes loaded by each mem op is 16 bytes, then cluster at max 2 mem ops, that is at max 32 bytes
So, we need to make sure that the new heuristic do not completey deviate away from the above one, and it
properly handles both the sub-word loads and the wide loads.
Reviewed By: arsenm, rampitec
Differential Revision: https://reviews.llvm.org/D84354
Summary:
Make use of both the - (1) clustered bytes and (2) cluster length, to decide on
the max number of mem ops that can be clustered. On an average, when loads
are dword or smaller, consider `5` as max threshold, otherwise `4`. This
heuristic is purely based on different experimentation conducted, and there is
no analytical logic here.
Reviewers: foad, rampitec, arsenm, vpykhtin
Reviewed By: rampitec
Subscribers: llvm-commits, kerbowa, hiraditya, t-tye, Anastasia, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl, thakis
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82393
Summary:
- AssertAlign node records the guaranteed alignment on its source node,
where these alignments are retrieved from alignment attributes in LLVM
IR. These tracked alignments could help DAG combining and lowering
generating efficient code.
- In this patch, the basic support of AssertAlign node is added. So far,
we only generate AssertAlign nodes on return values from intrinsic
calls.
- Addressing selection in AMDGPU is revised accordingly to capture the
new (base + offset) patterns.
Reviewers: arsenm, bogner
Subscribers: jvesely, wdng, nhaehnle, tpr, hiraditya, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81711
Summary:
Make use of both the - (1) clustered bytes and (2) cluster length, to decide on
the max number of mem ops that can be clustered. On an average, when loads
are dword or smaller, consider `5` as max threshold, otherwise `4`. This heuristic
is purely based on different experimentation conducted, and there is no analytical
logic here.
Reviewers: foad, rampitec, arsenm, vpykhtin
Reviewed By: foad, rampitec
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, Anastasia, t-tye, hiraditya, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81085
Summary:
pickNodeBidirectional tried to compare the best top candidate and the
best bottom candidate by examining TopCand.Reason and BotCand.Reason.
This is unsound because, after calling pickNodeFromQueue, Cand.Reason
does not reflect the most important reason why Cand was chosen. Rather
it reflects the most recent reason why it beat some other potential
candidate, which could have been for some low priority tie breaker
reason.
I have seen this cause problems where TopCand is a good candidate, but
because TopCand.Reason is ORDER (which is very low priority) it is
repeatedly ignored in favour of a mediocre BotCand. This is not how
bidirectional scheduling is supposed to work.
To fix this I changed the code to always compare TopCand and BotCand
directly, like the generic implementation of pickNodeBidirectional does.
This removes some uncommented AMDGPU-specific logic; if this logic turns
out to be important then perhaps it could be moved into an override of
tryCandidate instead.
Graphics shader benchmarking on gfx10 shows a lot more positive than
negative effects from this change.
Reviewers: arsenm, tstellar, rampitec, kzhuravl, vpykhtin, dstuttard, tpr, atrick, MatzeB
Subscribers: jvesely, wdng, nhaehnle, yaxunl, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68338
Similar to what we already do with SimplifyDemandedVectorElts, call SimplifyDemandedBits across all the extracted elements of the source vector, treating it as single use.
There's a minor regression in store-weird-sizes.ll which will be addressed in an upcoming SimplifyDemandedBits patch.
We are relying on atrificial DAG edges inserted by the
MemOpClusterMutation to keep loads and stores together in the
post-RA scheduler. This does not work all the time since it
allows to schedule a completely independent instruction in the
middle of the cluster.
Removed the DAG mutation and added pass to bundle already
clustered instructions. These bundles are unpacked before the
memory legalizer because it does not work with bundles but also
because it allows to insert waitcounts in the middle of a store
cluster.
Removing artificial edges also allows a more relaxed scheduling.
Differential Revision: https://reviews.llvm.org/D72737
This was committed in rL358887 but reverted in rL360066 due to a x86 regression, really it should be have been pre-committed instead of being part of the SimplifyDemandedBits bitcast patch.
llvm-svn: 360263
Reverts "[X86] Remove (V)MOV64toSDrr/m and (V)MOVDI2SSrr/m. Use 128-bit result MOVD/MOVQ and COPY_TO_REGCLASS instead"
Reverts "[TargetLowering][AMDGPU][X86] Improve SimplifyDemandedBits bitcast handling"
Eric Christopher and Jorge Gorbe Moya reported some issues with these patches to me off list.
Removing the CodeGenOnly instructions has changed how fneg is handled during fast-isel with sse/sse2. We're now emitting fsub -0.0, x instead
moving to the integer domain(in a GPR), xoring the sign bit, and then moving back to xmm. This is because the fast isel table no longer
contains an entry for (f32/f64 bitcast (i32/i64)) so the target independent fneg code fails. The use of fsub changes the behavior of nan with
respect to -O2 codegen which will always use a pxor. NOTE: We still have a difference with double with -m32 since the move to GPR doesn't work
there. I'll file a separate PR for that and add test cases.
Since removing the CodeGenOnly instructions was fixing PR41619, I'm reverting r358887 which exposed that PR. Though I wouldn't be surprised
if that bug can still be hit independent of that.
This should hopefully get Google back to green. I'll work with Simon and other X86 folks to figure out how to move forward again.
llvm-svn: 360066
This patch adds support for BigBitWidth -> SmallBitWidth bitcasts, splitting the DemandedBits/Elts accordingly.
The AMDGPU backend needed an extra (srl (and x, c1 << c2), c2) -> (and (srl(x, c2), c1) combine to encourage BFE creation, I investigated putting this in DAGCombine but it caused a lot of noise on other targets - some improvements, some regressions.
The X86 changes are all definite wins.
Differential Revision: https://reviews.llvm.org/D60462
llvm-svn: 358887
D52935 introduced the ability for SimplifyDemandedBits to call SimplifyDemandedVectorElts through BITCASTs if the demanded bit mask entirely covered the sub element.
This patch relaxes this to demanding an element if we need any bit from it.
Differential Revision: https://reviews.llvm.org/D54761
llvm-svn: 348073
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
If the truncate is only accessing the first element of the vector,
we can use the original source value.
This helps with some combine ordering issues after operations are
lowered to integer operations between bitcasts of build_vector.
In particular it stops unnecessarily materializing the unused
top half of a vector in some cases.
llvm-svn: 331909
GFX9 stopped using m0 for most DS instructions. Select
a different instruction without the use. I think this will
be less error prone than trying to manually maintain m0
uses as needed.
llvm-svn: 319270
Matt Arsenault