Currently the dbg_value ends up in the relaxed branch block. A future
commit will push the dbg_value out of this block, and I'm not sure how
to coax the IR into producing the same MIR at the relevant point.
For an instruction in the basic block BB, SinkingPass enumerates basic blocks
dominated by BB and BB's successors. For each enumerated basic block,
SinkingPass uses `AllUsesDominatedByBlock` to check whether the basic
block dominates all of the instruction's users. This is inefficient.
Use the nearest common dominator of all users to avoid enumerating the
candidate. The nearest common dominator may be in a parent loop which is
not beneficial. In that case, find the ancestors in the dominator tree.
In the case that the instruction has no user, with this change we will
not perform unnecessary move. This causes some amdgpu test changes.
A stage-2 x86-64 clang is a byte identical with this change.
The implicit def of the super register would appear to kill any live
uses of components before the spill, and would be deleted by
MachineCopyPropagation. We need to add implicit uses of the super
register, similarly to what copyPhysReg does. VGPR tuples appear to be
correctly handled already. I need to double check the SGPR->memory
path.
https://reviews.llvm.org/D83833
Patch adds two new GICombinerRules for G_SELECT. The rules include:
combining selects with undef comparisons into their first selectee value,
and to combine away selects with constant comparisons. Patch additionally
adds a new combiner test for the AArch64 target to test these new G_SELECT
combiner rules and the existing select_same_val combiner rule.
Patch by mkitzan
This is the first of a set of DAGCombiner changes enabling strictfp
optimizations. I want to test to waters with this to make sure changes
like these are acceptable for the strictfp case- this particular change
should preserve exception ordering and result precision perfectly, and
many other possible changes appear to be able to as well.
Copied from regular fadd combines but modified to preserve ordering via
the chain, this change allows strict_fadd x, (fneg y) to become
struct_fsub x, y and strict_fadd (fneg x), y to become strict_fsub y, x.
Differential Revision: https://reviews.llvm.org/D85548
There is no justification for changing vcc_lo to vcc
when shrinking V_CNDMASK, and such a change could
later confuse live variable analysis.
Make sure the original register is preserved.
Differential Revision: https://reviews.llvm.org/D86541
This would assert with unaligned DS access enabled. The offset may not
be aligned. Theoretically the pattern predicate should check the
memory alignment, although it is possible to have the memory be
aligned but not the immediate offset.
In this case I would expect it to use ds_{read|write}_b64 with
unaligned access, but am not clear if there's a reason it doesn't.
If the condition output is negated, swap the branch targets. This is
similar to what SelectionDAG does for when SelectionDAGBuilder
decides to invert the condition and swap the branches.
This is leaving behind a dead constant def for some reason.
This produces less work for addressing mode matching. I think this is
safe since I don't think machine IR is supposed to give the same
aliasing properties as getelementptr in the IR.
If a workgroup size is known to be not greater than wavefront size
the s_barrier instruction is not needed since all threads are guaranteed
to come to the same point at the same time.
This is the same optimization that was implemented for SelectionDAG in
D31731.
Differential Revision: https://reviews.llvm.org/D86609
Before calling target hook to determine if two loads/stores are clusterable,
we put them into different groups to avoid fake cluster due to dependency.
For now, we are putting the loads/stores into the same group if they have
the same predecessor. We assume that, if two loads/stores have the same
predecessor, it is likely that, they didn't have dependency for each other.
However, one SUnit might have several predecessors and for now, we just
pick up the first predecessor that has non-data/non-artificial dependency,
which is too arbitrary. And we are struggling to fix it.
So, I am proposing some better implementation.
1. Collect all the loads/stores that has memory info first to reduce the complexity.
2. Sort these loads/stores so that we can stop the seeking as early as possible.
3. For each load/store, seeking for the first non-dependency instruction with the
sorted order, and check if they can cluster or not.
Reviewed By: Jay Foad
Differential Revision: https://reviews.llvm.org/D85517
Most notably, we were incorrectly reporting <3 x s16> as a legal type
for these. Make sure these aren't legal to help make progress on
fixing the artifact combiner and vector legalizer
rules. Unfortunately, this means spreading the -global-isel-abort=0
hack, although this doesn't change the legalizer result in any
situation.
Implicit uses of non-register value types places impossible to satisfy
constraints on the legalizer / artifact combiner. These prevent
writing sensible legalize rules for the artifacts without triggering
infinite loops in the legalizer.
The verifier really needs to enforce this, but I'm not sure what the
exact conditions would look like yet.
SelectionDAG and GlobalISel take different failure paths for these and
end up producing different failure errors. Check both so the test
passes when the default is switched.
This interferes with GlobalISel's much better handling of the
situation.
This should really be disable for GlobalISel. However, the fallback
only re-runs the selection passes, and doesn't go back and rerun any
codegen IR passes. I haven't come up with a good solution to this
problem.
This is to initially handleg immAllOnesV, which should match
G_BUILD_VECTOR or G_BUILD_VECTOR_TRUNC. In the future, it could be
used for other patterns cases that map to multiple G_* instructions,
such as G_ADD and G_PTR_ADD.
D77152 tried to do this but got it wrong in the shift-by-zero case.
D86430 reverted the wrong code. Reimplement the optimization with
different code depending on whether the shift amount is known to be
non-zero (modulo bitwidth).
This improves code quality for fshl tests on AMDGPU, which only has an
fshr instruction.
Differential Revision: https://reviews.llvm.org/D86438
Handle workitem intrinsics. There isn't really away to adequately test
this right now, since none of the known bits users are fine grained
enough to test the edge conditions. This triggers a number of
instances of the new 64-bit to 32-bit shift combine in the existing
tests.
shl ([sza]ext x, y) => zext (shl x, y).
Turns expensive 64 bit shifts into 32 bit if it does not overflow the
source type:
This is a port of an AMDGPU DAG combine added in
5fa289f0d8. InstCombine does this
already, but we need to do it again here to apply it to shifts
introduced for lowered getelementptrs. This will help matching
addressing modes that use 32-bit offsets in a future patch.
TableGen annoyingly assumes only a single match data operand, so
introduce a reusable struct. However, this still requires defining a
separate GIMatchData for every combine which is still annoying.
Adds a morally equivalent function to the existing
getShiftAmountTy. Without this, we would have to do try to repeatedly
query the legalizer info and guess at what type to use for the shift.
This is a fixup of commit 0819a6416f (D77152) which could
result in miscompiles. The miscompile could only happen for targets
where isOperationLegalOrCustom could return different values for
FSHL and FSHR.
The commit mentioned above added logic in expandFunnelShift to
convert between FSHL and FSHR by swapping direction of the
funnel shift. However, that transform is only legal if we know
that the shift count (modulo bitwidth) isn't zero.
Basically, since fshr(-1,0,0)==0 and fshl(-1,0,0)==-1 then doing a
rewrite such as fshr(X,Y,Z) => fshl(X,Y,0-Z) would be incorrect if
Z modulo bitwidth, could be zero.
```
$ ./alive-tv /tmp/test.ll
----------------------------------------
define i32 @src(i32 %x, i32 %y, i32 %z) {
%0:
%t0 = fshl i32 %x, i32 %y, i32 %z
ret i32 %t0
}
=>
define i32 @tgt(i32 %x, i32 %y, i32 %z) {
%0:
%t0 = sub i32 32, %z
%t1 = fshr i32 %x, i32 %y, i32 %t0
ret i32 %t1
}
Transformation doesn't verify!
ERROR: Value mismatch
Example:
i32 %x = #x00000000 (0)
i32 %y = #x00000400 (1024)
i32 %z = #x00000000 (0)
Source:
i32 %t0 = #x00000000 (0)
Target:
i32 %t0 = #x00000020 (32)
i32 %t1 = #x00000400 (1024)
Source value: #x00000000 (0)
Target value: #x00000400 (1024)
```
It could be possible to add back the transform, given that logic
is added to check that (Z % BW) can't be zero. Since there were
no test cases proving that such a transform actually would be useful
I decided to simply remove the faulty code in this patch.
Reviewed By: foad, lebedev.ri
Differential Revision: https://reviews.llvm.org/D86430
This is the slowest operation in the already slow pass.
Instead of sorting just put a stall list into an ordered
map.
Differential Revision: https://reviews.llvm.org/D86253
Do not break down local loads and stores so ds_read/write_b96/b128 in
ISelLowering can be selected on subtargets that support them and if align
requirements allow them.
Differential Revision: https://reviews.llvm.org/D84403
Fix local ds_read/write_b96/b128 so they can be selected if the alignment
allows. Otherwise, either pick appropriate ds_read2/write2 instructions or break
them down.
Differential Revision: https://reviews.llvm.org/D81638
Features UnalignedBufferAccess and UnalignedDSAccess are now used to determine
whether hardware supports such access.
UnalignedAccessMode should be used to enable them.
hasUnalignedBufferAccessEnabled() and hasUnalignedDSAccessEnabled() can be
now used to quickly check both.
Differential Revision: https://reviews.llvm.org/D84522
Adjust alignment requirements for ds_read/write_b96/b128.
GFX9 and onwards allow misaligned access for reads and writes but only if
SH_MEM_CONFIG.alignment_mode allows it.
UnalignedDSAccess is set on GCN subtargets from GFX9 onward to let us know if we
can relax alignment requirements.
UnalignedAccessMode acts similary to UnalignedBufferAccess for DS instructions
but only from GFX9 onward and is supposed to match alignment_mode. By default
alignment of 4 is required.
Differential Revision: https://reviews.llvm.org/D82788
In SelectionDAGBuilder always translate the fshl and fshr intrinsics to
FSHL and FSHR (or ROTL and ROTR) instead of lowering them to shifts and
ORs. Improve the legalization of FSHL and FSHR to avoid code quality
regressions.
Differential Revision: https://reviews.llvm.org/D77152
Summary:
- HIP uses an unsized extern array `extern __shared__ T s[]` to declare
the dynamic shared memory, which size is not known at the
compile time.
Reviewers: arsenm, yaxunl, kpyzhov, b-sumner
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D82496
Assuming this is used to split a memory access into smaller pieces,
the new access should still have the same aliasing properties as the
original memory access. As far as I can tell, this wasn't
intentionally dropped. It may be necessary to drop this if you are
moving the operand outside of the bounds of the original object in
such a way that it may alias another IR object, but I don't think any
of the existing users are doing this. Some of the uses widen into
unused alignment padding, which I think is OK.
Custom lower and widen odd sized loads up to the alignment. The
default set of legalization actions doesn't have a way to represent
this. This fixes naturally aligned <3 x s8> and <3 x s16> loads.
This also starts moving towards eliminating the buggy and
overcomplicated legalization rules for narrowing. All the memory size
changes should be done in the lower or custom action, not NarrowScalar
/ FewerElements. These currently have redundant and ambiguous code
with the lower action.
Summary:
When the resource descriptor is of vgpr, we need a waterfall loop
to read into a sgpr. In this patchm we generalized the implementation
to work for any regster class sizes, and extend the work to MIMG
instructions.
Fixes: SWDEV-223405
Reviewers:
arsenm, nhaehnle
Differential Revision:
https://reviews.llvm.org/D82603
By detecting this sign extend pattern early, we can uncover opportunities for
more optimizations.
Differential Revision: https://reviews.llvm.org/D85965
We weren't looking through the parameters on calls at all.
E.g., say you had
```
declare i32 @zext(i32 zeroext %x)
...
%y = call i32 @zext(i32 %something)
...
```
At the point of the call, we wouldn't know that the %something should have the
zeroext attribute.
This sets flags in about the same way as
TargetLoweringBase::ArgListEntry::setAttributes.
Differential Revision: https://reviews.llvm.org/D86125
Previously, it would successfully select and assert if not HSA or PAL
when expanding the pseudoinstruction. We don't need the
pseudoinstruction anymore since we know the total size after
legalization.
The code to determine the value size was overcomplicated and only
correct in the case where the result register already had a register
class assigned. We can always take the size directly from the
register's type.
The previous implementation was incorrect, and based off incorrect
instruction definitions. Unfortunately we can't match natural
addressing in a lot of cases due to the shift/scale applied in
getelementptrs. This relies on reducing the 64-bit shift to 32-bits.
The artifact combiner searches for the uses of G_MERGE_VALUES for
unmerge/trunc that need further combining. This also needs to handle
the vector merge opcodes the same way. This fixes leaving behind some
pairs I expected to be removed, that were if the legalizer is run a
second time.
We may have an SGPR->VGPR copy if a totally uniform pointer
calculation is used for a VGPR pointer operand.
Also hack around a bug in MUBUF matching which would incorrectly use
MUBUF for global when flat was requested. This should really be a
predicate on the parent pattern, but the DAG always checked this
manually inside the complex pattern.
The VGPR component is a 32-bit offset, not 64-bits.
I'm not sure what the correct syntax is for this. This maintains the
vaddr position and leaves saddr in the end "off" position. This is
particularly terrible for stores, since the operand order is now <vgpr
offset>, <data>, <sgpr base>, splitting the pointer operands. I
suppose this is a logical consequence from the mistake of not putting
the data operand first. I'm not sure what sp3 does.
This was only used for matching the saddr addressing mode of global
instructions, but this was not implemented correctly. The instruction
definitions aren't even correct, and are defined as using a 64-bit
VGPR component. Eliminate this pass to enable correcting the
instruction definitions. A new matching implementation can work in
GlobalISel or relying on DAG divergence information for the base
address.
It did not process hazard for ds_permute because it does not
load or store even though it is DS.
Differential Revision: https://reviews.llvm.org/D86003
These should really match either G_BUILD_VECTOR or
G_BUILD_VECTOR_TRUNC, but there doesn't seem to be an existing
mechanism for matching alternative opcodes. There is GIM_SwitchOpcode,
but it seems to assume it's oly only used for matcher optimization.
I could also omit any opcode check and rely on the matcher directly
checking the opcode, but the table optimizer currently assumes there
has to be an opcode check.
Also doesn't try to handle undef elements like the DAG version.
Unfortunately this ends up not working as expected on targets with
16-bit operations due to AMDGPUCodeGenPrepare's promotion of uniform
16-bit ops to i32.
The vector case annoyingly requires switching the checked opcode,
since constants for vectors aren't directly handled.
I also need to think more carefully about whether this is valid for i1.
PAL recently got support for multiple ELF sections and relocations,
therefore we can now use .rodata sections instead of forcing constants
into .text.
Differential Revision: https://reviews.llvm.org/D85895
If we need a scratch register for the spill don't use the same scratch
register that is being used for the MBUF offset.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D85772
SIPreEmitPeephole does not process all terminators, which means
it can fail to handle SI_RETURN_TO_EPILOG if immediately preceeded
by a branch to the early exit block.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D85872
From the code after the 'break', they are processing 64bit scalar and
vector bitcast. So I think the break-condition should be (cond1 || cond2)
This means we only execute following code if (64bit and dest-is-vector).
Also remove a previous fix which is not needed with this new fix.
(introduced in: 1349a04ef5)
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D85804
This mirrors the support for the equivalent extracts. This also
creates a huge mess that would be greatly improved if we had any bit
operation combines.
ISD::ATOMIC_STORE arbitrarily has the operands in the opposite order
from regular ISD::STORE, which always introduced an annoying
duplication of patterns to handle both cases. Since in GlobalISel
there's just the one G_STORE, we need to swap the operands to
correctly emit the type check for the pointer operand.
Some work started in 20aafa3156 to
migrate SelectionDAG to use ISD::STORE for atomics, but that work
seems to have stalled. Since this is the pretty much the last
operation which matters which isn't supported for AMDGPU, use this
compatibility hack to unblock declaring it functionally complete.
Not sure what's going on with the pending_phis AArch64 test. It seems
it didn't always use atomics, and I'm not sure what it was originally
testing matters anymore.
SplitKit forms invalid COPY subreg bundles without a leading
BUNDLE instruction. That manifests itself in post-RA scheduler
counting instruction and asserting on "Instruction count mismatch".
The bundle shall be undone by VirtRegRewriter::expandCopyBundle(),
but it does not because VirtRegRewriter::handleIdentityCopy() can
turn COPY bundle into a KILL bundle.
Process KILLs as well.
Differential Revision: https://reviews.llvm.org/D85484
This can fold the immediate into the physical destination, but this
should not look for further users of the register. Fixes regression
introduced by 766cb615a3.
Fix 64-bit copy to SCC by restricting the pattern resulting
in such a copy to subtargets supporting 64-bit scalar compare,
and mapping the copy to S_CMP_LG_U64.
Before introducing the S_CSELECT pattern with explicit SCC
(0045786f14), there was no need
for handling 64-bit copy to SCC ($scc = COPY sreg_64).
The proposed handling to read only the low bits was however
based on a false premise that it is only one bit that matters,
while in fact the copy source might be a vector of booleans and
all bits need to be considered.
The practical problem of mapping the 64-bit copy to SCC is that
the natural instruction to use (S_CMP_LG_U64) is not available
on old hardware. Fix it by restricting the problematic pattern
to subtargets supporting the instruction (hasScalarCompareEq64).
Differential Revision: https://reviews.llvm.org/D85207
Regions are sometimes skipped which should be rescheduled without memory op
clustering. RegionIdx is not incremented when iterating over regions that
are flagged to be skipped, causing the index to be incorrect.
Thanks to Vang Thao for discovering this bug!
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D85498
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
Use the same basic strategy as LegalizeVectorTypes. Try to index into
smaller pieces if there's a constant index, and otherwise fall back to
a stack temporary.
If we were to have an operation with an s16 def that needs to be
executed in a waterfall loop, not having s16 legal would place an
avoidable burden on RegBankSelect to widen it.
This was trying to constrain a physical register. By the verifier's
understanding, it's impossible to have a 1-bit copy to vcc/vcc_lo so
don't try to handle physregs.
The functionality is used when calling imageAtomicExhange() on float
type imageBuffer in Graphics shaders.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D85187
This reverts commit ac70b37a00
which reverted commit 8aeb2fe13a
because codegen tests got broken and i needed time to investigate.
This shows some regressions in tests, but they are all around GEP's,
so i'm not really sure how important those are.
https://rise4fun.com/Alive/1Gn
Get the argument register and ensure there's a copy to the virtual
register. AMDGPU and AArch64 have similarish code to get the livein
value, and I also want to use this in multiple places.
This is a bit more aggressive about setting the register class than
the original function, but that's probably OK.
I think we're missing a few verifier checks for function live ins. I
noticed AArch64's calling convention code is not actually adding
liveins to functions, only the entry block (which apparently might not
matter that much?). There should probably be a verifier check that
entry block live ins are also live into the function. We also might
need a verifier check that the copy to the livein virtual register is
in the entry block.
There seems to be an unrelated CSEMIRBuilder bug that was causing
expensive checks failures in this case. Hack the test to avoid this
problem for now until that's fixed.
When scavenging consider the sub-register of the source operand
to determine the bank of a candidate register (not just sub0).
Without this it is possible to introduce an infinite loop,
e.g. $sgpr15_sgpr16_sgpr17 can be assigned for a conflict between
$sgpr0 and SGPR_96:sub1.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D84910
This patch stops unconditionally transforming FSUB(-0,X) into an FNEG(X) while building the DAG. There is also one small change to handle the new FSUB(-0,X) similarly to FNEG(X) in the AMDGPU backend.
Differential Revision: https://reviews.llvm.org/D84056
There were various hacks used to try to avoid making s1 SGPR vs. s1
VCC ambiguous after constraining the register before we had a strategy
to deal with this. This also attempted to handle undef operands, which
are now illegal gMIR.
Use pad with undef and unmerge with unused results. This is annoyingly
similar to several other places in LegalizerHelper, but they're all
slightly different.
For AMDGPU, vectors with elements < 32 bits should be indexed in
32-bit elements and the desired bits extracted from there. For
elements > 64-bits, these should be reduce to 64/32 elements to enable
the normal dynamic indexing paths.
In the dynamic index cases, this produces shorter code most of the
time. This does immediately regress the constant index cases, but this
should be fixed once we have the most basic of shift combines.
The element size > 64 case is pretty much ported from the exisiting
DAG implementation for extract element promote. The increasing element
size case is new.
Avoid recursively calling copyPhysReg for AGPR handling. This was
dropping the necessary super register implicit defs to avoid liveness
verifier errors.
We are using undef on the indirect move source subreg and then
using implicit super-reg. This creates a problem in RA when
Greedy decides to split the register. It reassigns the implicit
super-reg but does not bother to change undef source because
it is really does not matter. The fix is to stop lying to RA and
drop undef flag.
This has also hit a problem in SIFoldOperands as it can fold
immediate into an indirect move since there is no undef flag
anymore. That results in multiple test failures, so added the
check for this case.
Differential Revision: https://reviews.llvm.org/D84899
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
I still think it's highly questionable that we have two intrinsics
with identical behavior and only vary by the name of the libcall used
if it happens to be lowered that way, but try to reduce the feature
delta between SDAG and GlobalISel for recently added intrinsics. I'm
not sure which opcode should be considered the canonical one, but
lower roundeven back to round.
MFMA instructions shall not be scheduled back to back
to avoid MAI SIMD stall. Tell post-RA schedule we would
prefer some other instruction instead.
Differential Revision: https://reviews.llvm.org/D84883
I never completed the work on the patches referenced by
f8bf7d7f42, but this was intended to
avoid folding immediate writes into m0 which the coalescer doesn't
understand very well. Relax this to allow simple SGPR immediates to
fold directly into VGPR copies. This pattern shows up routinely in
current GlobalISel code since nothing is smart enough to emit VGPR
constants yet.
expressions.
Also "fix" the longstanding bug where the computed size depends on the
order of the visitation. We could try to predict the allocation order
used by legalization, but it would never be 100% perfect. Until we
start fixing the addresses somehow (or have a more reliable allocation
scheme later), just try to compute the size based on the worst case
padding.
This introduces the same bug llvm.amdgcn.s.setreg has where if the
user specified an immediate outside of the valid 16-bit range, it will
select into a verifier error.
GlobalISel let through a call to null, which would then fold into the
source operand like any other inline immediate. The SelectionDAG
lowering deletes calls to null and undef as a workaround from before
calls were supported. We should probably drop the special handling
case in the DAG lowering now, since the middle end optimizers delete
null calls anyway.
This needs an implicit def of the super-register in case one of the
lanes isn't defined, similar to copyPhysReg (or the not-VGPR spill
case below). This showed up in GlobalISel testing since it currently
doesn't fold out many undef instructions.
We can't fold the masked compare value through the select if the
select condition is re-defed after the and instruction. Fixes a
verifier error and trying to use the outgoing value defined in the
block.
I'm not sure why this pass is bothering to handle physregs. It's
making this more complex and forces extra liveness computation.
Update logic for reserving VGPR for SGPR spills. A CSR VGPR being reserved for
SGPR spills could be clobbered if there were no free lower VGPR's available.
Create a stack object so that it will be spilled in the prologue. Also
adds more tests.
Differential Revision: https://reviews.llvm.org/D83730
The DAG behavior allows matchching input patterns with a single result
to the first result of an output instruction that defines multiple
results. The remaining defs are implicitly dead.
This starts to fix using manual selection for AMDGPU add/sub (although
it's still needed, mostly because it's also still needed for
G_PTR_ADD).
Summary:
D78800 skipped generating cache invalidating instrucions altogether
on AMDPAL. However, this is sometimes too restrictive - we want a
more flexible option to be able to toggle this behaviour on and off
while we work towards developing a correct implementation of the
alternative memory model.
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, dexonsmith, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D84448
We don't really need these asserts. The LegalizerInfo is also
overly-aggressivly constructed, even when not in use. It needs to not
assert on dummy targets that have manually specified, unrelated
features.
Summary:
In parallelizeChainedStores, a TokenFactor was created with the size greater than 3000.
We found that DAGCombiner::visitTokenFactor will consume a huge amount of time on
such nodes. Since the number of operands already exceeds TokenFactorInlineLimit, we propose
to give up simplification with the consideration of compile time.
Reviewers:
@spatel, @arsenm
Differential Revision:
https://reviews.llvm.org/D84204
PHIElimination/createPHISourceCopy inserts non-branch terminators
after the control flow pseudo if a successor phi reads a register
defined by the control flow pseudo. If this happens, we need to split
the expansion of the control flow pseudo to ensure all the branches
are after all of the other mask management instructions.
GlobalISel hit this in testscases that happened to be tail
duplicated. The original testcase still does not work, since the same
problem appears to be present in a later pass.
Currently supported LLVM MTBUF syntax is shown below. It is not compatible with SP3.
op dst, addr, rsrc, FORMAT, soffset
This change adds support for SP3 syntax:
op dst, addr, rsrc, soffset SP3FORMAT
In addition to being compatible with SP3, this syntax allows using symbolic names for data, numeric and unified formats. Below is a list of added syntax variants.
format:<expression>
format:[<numeric-format-name>,<data-format-name>]
format:[<data-format-name>,<numeric-format-name>]
format:[<data-format-name>]
format:[<numeric-format-name>]
format:[<unified-format-name>]
The last syntax variant is supported for GFX10 only.
See llvm bug 37738
Reviewers: arsenm, rampitec, vpykhtin
Differential Revision: https://reviews.llvm.org/D84026
Widen or narrow a type to a type with the same scalar size as
another. This can be used to force G_PTR_ADD/G_PTRMASK's scalar
operand to match the bitwidth of the pointer type. Use this to
disallow narrower types for G_PTRMASK.
If the operand index exceeded the limit of unsigned char, it wrapped
and would point to the wrong operand. Increase the size of the operand
index field to avoid this, and also don't bother trying to fold into
implicit operands.
This adds the llvm.abs(), llvm.umin(), llvm.umax(), llvm.smin(),
and llvm.smax() intrinsics specified in D81829. For SelectionDAG,
the ISD opcodes and all the legalization and lowering already exist,
so this just wires them up to the intrinsic in the SDAG builder and
adds rudimentary tests. For GlobalISel only the min/max intrinsics
are wired up, as llvm.abs() will require the addition of a G_ABS op,
and corresponding legalization support.
Differential Revision: https://reviews.llvm.org/D84125
Add support in LegalizerHelper for lowering G_SADDSAT etc. either
using add/subtract-with-overflow or using max/min instructions.
Enable this lowering for AMDGPU so it can be tested. The legalization
rules are still approximate and skips out on using the clamp bit to
treat these as legal, which has never been used before. This also
doesn't yet try to deal with expanding SALU cases.
Summary: We do this already for output operands, but missed it for (non-tied) input operands.
Reviewers: arsenm, Petar.Avramovic
Reviewed By: arsenm
Subscribers: jvesely, wdng, nhaehnle, rovka, hiraditya, llvm-commits, kerbowa
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83763
16-bit literals are encoded as 32-bit values. If high 16-bits of the value is 0xFFFF, the decoded instruction cannot be reassembled.
For example, the following code
0xff,0x04,0x04,0x52,0xcd,0xab,0xff,0xff
was decoded as
v_mul_lo_u16_e32 v2, 0xffffabcd, v2
However this literal is actually a 64-bit constant 0x00000000ffffabcd which violates requirements described in the documentation - the truncation is not safe.
This change corrects decoding to make reassembly possible.
Reviewers: arsenm, rampitec
Differential Revision: https://reviews.llvm.org/D84098
Range that f16 can represent fits into i32.
Lower as f16->i32->i64 instead of f16->f32->i64
since f32->i64 has long expansion.
Differential Revision: https://reviews.llvm.org/D84166
These are treated identically to value aggregates placed in the kernel
argument list. A %struct.foo or %struct.foo addrspace(4)*
byref(sizeof(%struct.foo)) align(alignof(%struct.foo)) argument should
produce the same offsets and argument metadata.
This handles all 3 kernel ABI implementations, and the two HSA
metadata emission paths.
The AMDGPU handling of f16 vectors is terrible still since it gets
scalarized even when the vector operation is legal.
The code is is essentially duplicated between the non-strict and
strict case. Apparently no other expansions are currently trying to do
this. This is mostly because I found the behavior of
getStrictFPOperationAction to be confusing. In the ARM case, it would
expand strict_fsub even though it shouldn't due to the later check. At
that point, the logic required to check for legality was more complex
than just duplicating the 2 instruction expansion.
This was failing to add the size of LDS globals that weren't directly
used by an instruction. They could be used by constant expressions
which are transitively used by the function. This requires a better
search, but just abort on this for now for correctness.
Add narrowScalarFor action.
Add narrow scalar for typeIndex == 0 for G_FPTOSI/G_FPTOUI.
Legalize using narrowScalarFor as s16->s32 G_FPTOSI/G_FPTOUI
followed by s32->s64 G_SEXT/G_ZEXT.
Differential Revision: https://reviews.llvm.org/D84010
As explained in the comment:
// For a FLAT instruction the hardware decides whether to access
// global/scratch/shared memory based on the high bits of vaddr,
// ignoring the offset field, so we have to ensure that when we add
// remainder to vaddr it still points into the same underlying object.
// The easiest way to do that is to make sure that we split the offset
// into two pieces that are both >= 0 or both <= 0.
In particular FLAT (as opposed to SCRATCH and GLOBAL) instructions have
an unsigned immediate offset field, so we can't use it to help split a
negative offset.
Differential Revision: https://reviews.llvm.org/D83394
tryLatency compares two sched candidates. For the top zone it prefers
the one with lesser depth, but only if that depth is greater than the
total latency of the instructions we've already scheduled -- otherwise
its latency would be hidden and there would be no stall.
Unfortunately it only tests the depth of one of the candidates. This can
lead to situations where the TopDepthReduce heuristic does not kick in,
but a lower priority heuristic chooses the other candidate, whose depth
*is* greater than the already scheduled latency, which causes a stall.
The fix is to apply the heuristic if the depth of *either* candidate is
greater than the already scheduled latency.
All this also applies to the BotHeightReduce heuristic in the bottom
zone.
Differential Revision: https://reviews.llvm.org/D72392
When SCC is dead, but VCC is required then replace s_and / s_andn2
with s_mov into VCC when mask value is 0 or -1.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D83850
The hardware has created a real mess in the naming for add/sub, which
have been renamed basically every generation. Switch the carry out
pseudos to have the gfx9/gfx10 names. We were using the original SI/CI
v_add_i32/v_sub_i32 names. Later targets reintroduced these names as
carryless instructions with a saturating clamp bit, which we do not
define. Do this rename so we can unambiguously add these missing
instructions.
The carry-in versions should also be renamed, but at least those had a
consistent _u32 name to begin with. The 16-bit instructions were also
renamed, but aren't ambiguous.
This does regress assembler error message quality in some cases. In
mismatched wave32/wave64 situations, this will switch from
"unsupported instruction" to "invalid operand", with the error
pointing at the wrong position. I couldn't quite follow how the
assembler selects these, but the previous behavior seemed accidental
to me. It looked like there was a partial attempt to handle this which
was never completed (i.e. there is an AMDGPUOperand::isBoolReg but it
isn't used for anything).
Add widenScalar for TypeIdx == 0 for G_SITOFP/G_UITOFP.
Legailize, using widenScalar, as s64->s32 G_SITOFP/G_UITOFP
followed by s32->s16 G_FPTRUNC.
Differential Revision: https://reviews.llvm.org/D83880
Summary:
llc reports `fp32-denormals` is not recognized. I guess it was intended to be
`-denormal-fp-math-f32={preserve-sign|ieee} -mattr=+mad-mac-f32-insts`
Reviewers: rampitec
Reviewed By: rampitec
Subscribers: jvesely, nhaehnle, llvm-commits, kerbowa
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D83883
Add handling of s_andn2 and mask of 0.
This eliminates redundant instructions from uniform control flow.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D83641
Fix two obvious errors in the code and also update the test check.
Also add one test to catch the failure.
Patch by Ruiling Song!
Differential Revision: https://reviews.llvm.org/D83280
The hardware spec require src0 of s_cmpk should be a register. So, we
should not optimize s_cmp to s_cmpk if src0 is not register.
Patch by Ruiling Song!
Preserve SCC dead flags in SIOptimizeExecMaskingPreRA.
This helps with removing redundant s_andn2 instructions later.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D83637
It is possible that LowerSwitch pass leaves certain blocks
unreachable from the entry. If not removed, these dead blocks
can cause undefined behavior in the subsequent passes.
It caused a crash in the AMDGPU backend after the instruction
selection when a PHI node has its incoming values coming from
these unreachable blocks.
In the AMDGPU pass flow, the last invocation of UnreachableBlockElim
precedes where LowerSwitch is currently placed and eventually
missed out on the opportunity to get these blocks eliminated.
This patch ensures that LowerSwitch pass get inserted earlier
to make use of the existing unreachable block elimination pass.
Reviewed By: sameerds, arsenm
Differential Revision: https://reviews.llvm.org/D83584
This doesn't appear used for anything, and is emitted incorrectly
based on the description. This also depends on the IR type, and
pointee element type.
The unfortunate split between finalizeLowering and the selector pass
means there's a point where the verifier fails. The DAG selector pass
skips the verifier, but this seems to not work when using the
GlobalISel fallback.
Updated the AArch64 tests the best I could with my vague, inferred
understanding of AArch64 register banks. As far as I can tell, there
is only one 32-bit/64-bit type which will use the gpr register bank,
so we have to use the fpr bank for the other operand.
GFX10 image instructions use one or more address operands starting at
vaddr0, instead of a single vaddr operand, to allow for NSA forms.
Differential Revision: https://reviews.llvm.org/D81675
Fix the division/remainder algorithm by adding a second quotient
refinement step, which is required in some cases like
0xFFFFFFFFu / 0x11111111u (https://bugs.llvm.org/show_bug.cgi?id=46212).
Also document, rewrite and simplify it by ensuring that we always have a
lower bound on inv(y), which simplifies the UNR step and the quotient
refinement steps.
Differential Revision: https://reviews.llvm.org/D83381
handleAssignments was assuming every argument type is an MVT, and
assignArg would always fail. This fixes one of the hacks in the
current AMDGPU calling convention code that pre-processes the
arguments.
The tests in a5b9ad7e9a actually failed
the verifier, which for some reason is not the default. Also add tests
for 0-sized function arguments, which do not add entries to the
expected register lists.
This was resulting in a missing vreg def in the use select
instruction.
The output of the pseudo doesn't make sense, since it really shouldn't
have the vreg output in the first place, and instead an implicit scc
def to match the real scalar behavior.
We could have easier to understand tests if we selected scalar
versions of the [us]{add|sub}.with.overflow intrinsics.
This does still end up producing vector code in the end, since it gets
moved later.
The default constructor wasn't setting isSet o the ArgDescriptor, so
while these had the value set, they were treated as missing. This only
ended up mattering in the indirect call case (and for regular calls in
GlobalISel, which current doesn't have a way to support the variable
ABI).
Exit early if the exec mask is zero at the end of control flow.
Mark the ends of control flow during control flow lowering and
convert these to exits during the insert skips pass.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D82737
Exit early if the exec mask is zero at the end of control flow.
Mark the ends of control flow during control flow lowering and
convert these to exits during the insert skips pass.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D82737
Matt Arsenault