Extend loadSRsrcFromVGPR to allow moving a range of instructions into
the loop. The call instruction is surrounded by copies into physical
registers which should be part of the waterfall loop.
Differential Revision: https://reviews.llvm.org/D88291
At AMD, in an internal audit of our code, we found some corner cases
where we were not quite differentiating targets enough for some old
hardware. This commit is part of fixing that by adding three new
targets:
* The "Oland" and "Hainan" variants of gfx601 are now split out into
gfx602. LLPC (in the GPUOpen driver) and other front-ends could use
that to avoid using the shaderZExport workaround on gfx602.
* One variant of gfx703 is now split out into gfx705. LLPC and other
front-ends could use that to avoid using the
shaderSpiCsRegAllocFragmentation workaround on gfx705.
* The "TongaPro" variant of gfx802 is now split out into gfx805.
TongaPro has a faster 64-bit shift than its former friends in gfx802,
and a subtarget feature could be set up for that to take advantage of
it. This commit does not make that change; it just adds the target.
V2: Add clang changes. Put TargetParser list in order.
V3: AMDGCNGPUs table in TargetParser.cpp needs to be in GPUKind order,
so fix the GPUKind order.
Differential Revision: https://reviews.llvm.org/D88916
Change-Id: Ia901a7157eb2f73ccd9f25dbacec38427312377d
Following on from D88890, this makes the newly added patterns
conditional on NoFP32Denormals. mad/mac f32 instructions always flush
denormals regardless of the MODE register setting, and I believe the
legacy variants do the same.
Differential Revision: https://reviews.llvm.org/D89123
Note that all subtargets up to GFX10.1 have v_mad_legacy_f32, but GFX8/9
lack v_mac_legacy_f32. GFX10.3 has no mad/mac f32 instructions at all.
Differential Revision: https://reviews.llvm.org/D88890
ExpandUnalignedLoad/Store can sometimes produce unnecessary copies to
temporary stack slot. We should prefer splitting vectors if possible.
Differential Revision: https://reviews.llvm.org/D88882
Summary:
This implements a workaround for a hardware bug in gfx8 and gfx9,
where register usage is not estimated correctly for image_store and
image_gather4 instructions when D16 is used.
Change-Id: I4e30744da6796acac53a9b5ad37ac1c2035c8899
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D81172
Previously we wrote multi-byte values out as-is from host memory. Use
the `emitIntN` helpers in `MCStreamer` to produce a valid descriptor
irrespective of the host endianness.
Reviewed By: arsenm, rochauha
Differential Revision: https://reviews.llvm.org/D88858
Introduce a utility function to make it more
convenient to write code that is the same on
the GFX9 and GFX10 subtargets.
Use isGFX9Plus in the AsmParser for AMDGPU.
Authored By: Joe_Nash
Differential Revision: https://reviews.llvm.org/D88908
uint8_t types are implicitly promoted to int, leading to a
unsigned-signed comparison.
Thanks for the heads-up @uabelho.
Differential Revision: https://reviews.llvm.org/D88876
Refactor exit block creation to a single call ensureEarlyExitBlock.
Add support for generating an early exit block which clears the
exec mask, but only add this instruction when required.
These changes are to facilitate adding more forms of early
termination for PS shaders in the near future.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D88775
Use tablegen generic tables to get the index of image intrinsic
arguments.
Before, the computation of which image intrinsic argument is at which
index was scattered in a few places, tablegen, the SDag instruction
selection and GlobalISel. This patch changes that, so only tablegen
contains code to compute indices and the ImageDimIntrinsicInfo table
provides these information.
Differential Revision: https://reviews.llvm.org/D86270
This tends to increase code size but more importantly it reduces vgpr
usage, and could avoid costly readfirstlanes if the result needs to be
in an sgpr.
Differential Revision: https://reviews.llvm.org/D88580
Convert to use new MachineBasicBlock splitAt function.
Place code in splitBlock function for reuse in future changes.
Should yield no functional change.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D88537
Allows the creation of real SOP1 instructions with
assembler mnemonics that differ from their
pseudo-instruction mnemonics. The default behavior
keeps the mnemonics matching.
Corrects a subtarget label typo in a comment.
Authored By: Joe_Nash
Differential Revision: https://reviews.llvm.org/D88708
Check if operand of mul is constant value of one for certain atomic
instructions in order to avoid making unnecessary instructions when
-amdgpu-atomic-optimizer is present.
Differential Revision: https://reviews.llvm.org/D88315
This tends to increase code size but more importantly it reduces vgpr
usage, and could avoid costly readfirstlanes if the result needs to be
in an sgpr.
Differential Revision: https://reviews.llvm.org/D88245
This fixes the declaration of AMDGPULegalizerInfo::legalizeBufferLoad to
match the definition. It is still confusing that that parameter order is
different from legalizeBufferStore.
https://bugs.llvm.org/show_bug.cgi?id=47535
v3f32 should not be expanded to v4f32. getresinfo with a dmask of 7
created an image sample with a v3f32 return value, which was bitcasted
to a v4f32 in constructRetValue.
Differential Revision: https://reviews.llvm.org/D88206
Fix the verifier so that overlapping SGPR operands are counted
independently. We cannot assume that overlapping SGPR accesses
only count as a single constant bus use.
The exception is implicit uses which do not add to constant bus
usage (only) when overlapping.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D87748
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
Fix incorrect merges of m0 inits in loops.
It was assumed that if a clobbering instruction appears in
the same block as an init and the clobbering instruction
does not dominate the init then it does not interfere with
init.
This does not work in the presence of loops, where in this
scenario, the clobbering instruction does interfere with
the init in another iteration.
To fix this, do not check for block equality and defer the
decision to the predecessor check.
Differential Revision: https://reviews.llvm.org/D87882
It's simpler to do this at codegen time than to do ad-hoc constant
folding of machine instructions in SIFoldOperands.
Differential Revision: https://reviews.llvm.org/D88028
This reverts commit 0345d88de6.
Google internal backend uses EntrySU, we are looking into removing
dependency on it.
Differential Revision: https://reviews.llvm.org/D88018
Since 6524a7a2b9, this would sometimes
not emit the or to exec at the beginning of the block, where it really
has to be. If there is an instruction that defines one of the source
operands, split the block and turn the si_end_cf into a terminator.
This avoids regressions when regalloc fast is switched to inserting
reloads at the beginning of the block, instead of spills at the end of
the block.
In a future change, this should always split the block.
This reverts commit c3492a1aa1.
I think this is the wrong strategy and wrong place to do this
transform anyway. Also reverts follow up commit
7d593d0d69.
Alignment requirements for ds_read/write_b96/b128 for gfx9 and onward are
now the same as for other GCN subtargets. This way we can avoid any
unintentional use of these instructions on systems that do not support dword
alignment and instead require natural alignment.
This also makes 'SH_MEM_CONFIG.alignment_mode == STRICT' the default.
Differential Revision: https://reviews.llvm.org/D87821
- Need to lower COPY from SGPR to VGPR to a real instruction as the
standard COPY is used where the source and destination are from the
same register bank so that we potentially coalesc them together and
save one COPY. Considering that, backend optimizations, such as CSE,
won't handle them. However, the copy from SGPR to VGPR always needs
materializing to a native instruction, it should be lowered into a
real one before other backend optimizations.
Differential Revision: https://reviews.llvm.org/D87556
Instruction combining pass turns library rotl implementation to llvm.fshl.i16.
In the selection dag the intrinsic is turned to ISD::ROTL node that cannot be selected.
Need to expand it to shifts again.
Reviewed By: rampitec, arsenm
Differential Revision: https://reviews.llvm.org/D87618
eliminateFrameIndex won't fix up the offset register when the direct
frame index reference is moved to a separate move instruction. Switch
the offset to a base 0 (which it probably should be to begin with).
WeakRefDirective should specify a directive to declare "a global as being a weak undefined symbol".
The directive used by AMDGPU was incorrect - ".weakref" was intended for other purposes.
The correct directive is ".weak" and it is already defined as default for ELF.
So the redefinition was removed.
Reviewers: arsenm, rampitec
Differential Revision: https://reviews.llvm.org/D87762
Also renamed the fields to follow style guidelines.
Accessors help with readability - weight mutation, in particular,
is easier to follow this way.
Differential Revision: https://reviews.llvm.org/D87725
Fix lowering and instruction selection for v3x16 types
and enable InstCombine to emit them.
This patch only implements it for the selection dag.
GlobalISel tests in GlobalISel/llvm.amdgcn.image.load.1d.d16.ll and
GlobalISel/llvm.amdgcn.image.store.2d.d16.ll still don't work.
Differential Revision: https://reviews.llvm.org/D84420
Pre-gfx10 all MODE-setting instructions were S_SETREG_B32 which is
marked as having unmodeled side effects, which makes the machine
scheduler treat it as a barrier. Now that we have proper implicit $mode
operands we can use a no-side-effects S_SETREG_B32_mode pseudo instead
for setregs that only touch the FP MODE bits, to give the scheduler more
freedom.
Differential Revision: https://reviews.llvm.org/D87446
We have a single noret intrinsic an a lot of special handling
around it. Declare it just as any other but do not define rtn
instructions itself instead.
Differential Revision: https://reviews.llvm.org/D87719
GetElementPtrInst::Create returns a GetElementPtrInst* so we don't need to cast. Similarly IntegerType inherits from the Type base class.
Also, I've used auto* in a few places to cleanup the code.
Helps fix some clang-tidy warnings which saw the dyn_casts and warned that these can return null.
The versions that take 'unsigned' will be removed in the future.
I tried to use getOriginalAlign instead of getAlign in some
places. getAlign factors in the minimum alignment implied by
the offset in the pointer info. Since we're also passing the
pointer info we can use the original alignment.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D87592
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
Check for NoNaNsFPMath function attribute in isKnownNeverSNaN.
Function attributes are in held in 'TargetMachine.Options'.
Among other things, this allows selection of some patterns imported
in D87351 since G_FCANONICALIZE is not generated when isKnownNeverSNaN
returns true in lowerFMinNumMaxNum.
However we notice some incorrect results since function attributes are
not correctly written in TargetMachine.Options when next function is
processed. Take a look at @v_test_no_global_nnans_med3_f32_pat0_srcmod0,
it has "no-nans-fp-math"="false" but TargetMachine.Options still has it
set to true since first function in test file had this attribute set to
true. This will be fixed in D87511.
Differential Revision: https://reviews.llvm.org/D87456
Predicates with 'let PredicateCodeUsesOperands = 1' want to examine
matched operands. When we encounter predicate code that uses operands,
analyze its named operand arguments and create a map between argument
index and name. Later, when leaf node with name is encountered, emit
GIM_RecordNamedOperand that will store that operand at its argument
index in operand list. This operand list will be an argument to c++
code of the predicate.
Differential Revision: https://reviews.llvm.org/D87285
It was found some packed immediate operands (e.g. `<half 1.0, half 2.0>`) are
incorrectly processed so one of two packed values were lost.
Introduced new function to check immediate 32-bit operand can be folded.
Converted condition about current op_sel flags value to fall-through.
Fixes: SWDEV-247595
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D87158
We weren't using this before, so none of the MachineFunction CFG edges had the
branch probability information added. As a result, block placement later in the
pipeline was flying blind.
This is enabled only with optimizations enabled like SelectionDAG.
Differential Revision: https://reviews.llvm.org/D86824
This combine previously tried to take sequences like:
%cond = G_ICMP pred, a, b
G_BRCOND %cond, %truebb
G_BR %falsebb
%truebb:
...
%falsebb:
...
and by inverting the compare predicate and swapping branch targets, delete the
G_BR and instead have a single conditional branch to the falsebb. Since in an
earlier patch we have a combine to fold not(icmp) into just an inverted icmp,
we don't need this combine to do as much. This patch instead generalizes the
combine by just looking for:
G_BRCOND %cond, %truebb
G_BR %falsebb
%truebb:
...
%falsebb:
...
and then inverting the condition using a not (xor). The xor can be folded away
in a separate combine. This change also lets us avoid some optimization code
in the IRTranslator.
I also think that deleting G_BRs in the combiner is unnecessary. That's
something that targets can decide to do at selection time and could simplify
generic code in future.
Differential Revision: https://reviews.llvm.org/D86664
Add subtarget feature check to avoid using ds_read/write_b96/128 with too
low alignment if a bug is present on that specific hardware.
Add this "feature" to GFX 10.1.1 as it is also affected.
Add global-isel test.
optimizeEndCF removes EXEC restoring instruction case this instruction is the only one except the branch to the single successor and that successor contains EXEC mask restoring instruction that was lowered from END_CF belonging to IF_ELSE.
As a result of such optimization we get the basic block with the only one instruction that is a branch to the single successor.
In case the control flow can reach such an empty block from S_CBRANCH_EXEZ/EXECNZ it might happen that spill/reload instructions that were inserted later by register allocator are placed under exec == 0 condition and never execute.
Removing empty block solves the problem.
This change require further work to re-implement LIS updates. Recently, LIS is always nullptr in this pass. To enable it we need another patch to fix many places across the codegen.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D86634
During the PEI pass, the dead TargetStackID::SGPRSpill spill slots
are not being removed while spilling the FP/BP to memory.
Fixes: SWDEV-250393
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D87032
Previously SDNodeFlags::instersectWith(Flags) would do nothing if Flags was
in an undefined state, which is very bad given that this is the default when
getNode() is called without passing an explicit SDNodeFlags argument.
This meant that if an already existing and reused node had a flag which the
second caller to getNode() did not set, that flag would remain uncleared.
This was exposed by https://bugs.llvm.org/show_bug.cgi?id=47092, where an NSW
flag was incorrectly set on an add instruction (which did in fact overflow in
one of the two original contexts), so when SystemZElimCompare removed the
compare with 0 trusting that flag, wrong-code resulted.
There is more that needs to be done in this area as discussed here:
Differential Revision: https://reviews.llvm.org/D86871
Review: Ulrich Weigand, Sanjay Patel
- When an operand is changed into an immediate value or like, ensure their
target flags being cleared or set properly.
Differential Revision: https://reviews.llvm.org/D87109
Use forward declarations and move the include down to dependent files that actually use it.
This also exposes a number of implicit dependencies on KnownBits.h
Summary of changes:
- Changed parser to eliminate generation of excessive error messages;
- Corrected lit tests to match all expected error messages;
- Corrected lit tests to guard against unwanted extra messages (added option "--implicit-check-not=error:");
- Added missing checks and fixed some typos in tests.
See bug 46907: https://bugs.llvm.org/show_bug.cgi?id=46907
Reviewers: arsenm, rampitec
Differential Revision: https://reviews.llvm.org/D86940
The addend in a REL32 reloc needs to be adjusted to account for the
offset from the PC value returned by the s_getpc instruction to the
point where the reloc is applied. This was being done correctly for
(GOTPC)REL32_LO but not for (GOTPC)REL32_HI. This will only make a
difference if the target symbol happens to get loaded almost exactly
a multiple of 4G away from the relocated instructions.
Differential Revision: https://reviews.llvm.org/D86938
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.
There's a special case in hasAttribute for None when pImpl is null. If pImpl is not null we dispatch to pImpl->hasAttribute which will always return false for Attribute::None.
So if we just want to check for None its sufficient to just check that pImpl is null. Which can even be done inline.
This patch adds a helper for that case which I hope will speed up our getSubtargetImpl implementations.
Differential Revision: https://reviews.llvm.org/D86744
Intrinsic declarations use the default subtarget, but this should be
using the subtarget for the calling function. I haven't been able to
come up with a case where it matters though.
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.
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
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.
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.
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 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
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.
The SGPR spills happen in SILowerSGPRSpills() and allSGPRSpillsAreDead()
make sure there are no SGPR spills pending during PEI. But the FP/BP
spills happen during PEI and are exceptions.
Use actual frame indices of FP/BP in allSGPRSpillsAreDead() to
accommodate the exceptions.
Differential Revision: https://reviews.llvm.org/D86291
When sampling from images with coordinates that only have 16 bit
accuracy, convert the image intrinsic call to use a16 or g16.
This does only happen if the target hardware supports it.
An alternative would be to always apply this combination, independent of
the target hardware and extend 16 bit arguments to 32 bit arguments
during legalization. To me, this sounds like an unnecessary roundtrip
that could prevent some further InstCombine optimizations.
Differential Revision: https://reviews.llvm.org/D85887
The `UnrollMaxBlockToAnalyze` parameter is used at the stage when we have no
information about a loop body BB cost. In some cases, e.g. for simple loop
```
for(int i=0; i<32; ++i){
D = Arr2[i*8 + C1];
Arr1[i*64 + C2] += C3 * D;
Arr1[i*64 + C2 + 2048] += C4 * D;
}
```
current default parameter value is not enough to run deeper cost analyze so the
loop is not completely unrolled.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D86248
Previously we weren't adding the LegalizerInfo to the post-legalizer
combiner. Since that's fixed, we don't need to try to filter out the
one case that was breaking.
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
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.
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.
If the same stream object is used for multiple compiles, the PAL metadata from eariler compilations will leak into later one. See https://github.com/GPUOpen-Drivers/llpc/issues/882 for how this is happening in LLPC.
No tests were added because multiple compiles will have to happen using the same pass manager, and I do not see a setup for that on the LLVM side. Let me know if there is a good way to test this.
Reviewed By: nhaehnle
Differential Revision: https://reviews.llvm.org/D85667
This was always set to 0. Use a default value of 0 in this context to
satisfy the instruction definition patterns. We can't unconditionally
use SLC with a default value of 0 due to limitations in TableGen's
handling of defaulted operands when followed by non-default operands.
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
This patch implements initial backend support for a -mtune CPU controlled by a "tune-cpu" function attribute. If the attribute is not present X86 will use the resolved CPU from target-cpu attribute or command line.
This patch adds MC layer support a tune CPU. Each CPU now has two sets of features stored in their GenSubtargetInfo.inc tables . These features lists are passed separately to the Processor and ProcessorModel classes in tablegen. The tune list defaults to an empty list to avoid changes to non-X86. This annoyingly increases the size of static tables on all target as we now store 24 more bytes per CPU. I haven't quantified the overall impact, but I can if we're concerned.
One new test is added to X86 to show a few tuning features with mismatched tune-cpu and target-cpu/target-feature attributes to demonstrate independent control. Another new test is added to demonstrate that the scheduler model follows the tune CPU.
I have not added a -mtune to llc/opt or MC layer command line yet. With no attributes we'll just use the -mcpu for both. MC layer tools will always follow the normal CPU for tuning.
Differential Revision: https://reviews.llvm.org/D85165
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.
Changes the Offset arguments to both functions from int64_t to TypeSize
& updates all uses of the functions to create the offset using TypeSize::Fixed()
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D85220
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
Add cases of fused fmul+fadd/fsub with f16 and f64 operands to cost model.
Also added operations with contract attribute.
Fixed line endings in test.
Reviewed By: rampitec
Differential Revision: https://reviews.llvm.org/D84995
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
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.
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.
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.
Currently GlobalISel doesn't force all VGPR phi operands to VGPRs, so
this hit a case where it was queried with a VGPR and SGPR. This could
arguably be a verifier error, but it's currently not.
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.
Sebastian Neubauer