There are two main issues preventing us from generating immediate form shifts:
1) We have partial SelectionDAG imported support for G_ASHR and G_LSHR shift
immediate forms, but they currently don't work because the amount type is
expected to be an s64 constant, but we only legalize them to have homogenous
types.
To deal with this, first we introduce a custom legalizer to *only* custom legalize
s32 shifts which have a constant operand into a s64.
There is also an additional artifact combiner to fold zexts(g_constant) to a
larger G_CONSTANT if it's legal, a counterpart to the anyext version committed
in an earlier patch.
2) For G_SHL the importer can't cope with the pattern. For this I introduced an
early selection phase in the arm64 selector to select these forms manually
before the tablegen selector pessimizes it to a register-register variant.
Differential Revision: https://reviews.llvm.org/D63910
llvm-svn: 364994
The BUNDLE itself should not have side effects, and this is a property
of instructions inside the bundle. The hasProperty check already
searches for any member instructions, which was pointless since it was
overridden by this bit.
Allows me to distinguish bundles that have side effects vs. do not in
a future patch. Also fixes an unnecessary scheduling barrier in the
bundle AMDGPU uses to get PC relative addresses.
llvm-svn: 364984
Ordinarily it is lowered as a build_vector of each extract_vector_elt,
which in turn get lowered to bitcasts and bit shifts. Very little
understand the lowered extract pattern, resulting in much worse
code. We treat concat_vectors of v2i16 as legal, so prefer that.
llvm-svn: 364959
The register bank for the destination of the sample argument copy was
wrong. We shouldn't be constraining each source to the result register
bank. Allow constraining the original register to the right size.
llvm-svn: 364928
If the requested source type an be used as a merge source type, create
a merge of merges. This avoids creating large, illegal extensions and
bit-ops directly to the result type.
llvm-svn: 364841
Tests don't cover the masked input path since non-kernel arguments
aren't lowered yet.
Test is copied directly from the existing test, with 2 additions.
llvm-svn: 364833
Replace the brcond for the 2 cases that act as branches. For now
follow how the current system works, although I think we can
eventually get rid of the pseudos.
llvm-svn: 364832
This needs to be extended to s32, and expanded into cmp+select. This
is relying on the fact that widenScalar happens to leave the
instruction in place, but this isn't a guaranteed property of
LegalizerHelper.
llvm-svn: 364831
The condition register bank must be scc or vcc so that a copy will be
inserted, which will be lowered to a compare.
Currently greedy unnecessarily forces using a VCC select.
llvm-svn: 364825
Summary:
ds_ordered_count can now simultaneously operate on up to 4 dwords
in a single instruction, which are taken from (and returned to)
lanes 0..3 of a single VGPR.
Change-Id: I19b6e7b0732b617c10a779a7f9c0303eec7dd276
Reviewers: mareko, arsenm, rampitec
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63716
llvm-svn: 364815
Also works around tablegen defect in selecting add with unused carry,
but if we have to manually select GEP, might as well handle add
manually.
llvm-svn: 364806
There are several things broken, but at least emit the right thing for
gfx9.
The import of the pattern with the unused carry out seems to not
work. Needs a special class for clamp, because OperandWithDefaultOps
doesn't really work.
llvm-svn: 364804
Summary:
The stride should depend on the wave size, not the hardware generation.
Also, the 32_FLOAT format is 0x16, not 16; though that shouldn't be
relevant.
Change-Id: I088f93bf6708974d085d1c50967f119061da6dc6
Reviewers: arsenm, rampitec, mareko
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63808
llvm-svn: 364788
This was checking the size of the register with the value of the size,
which happens to be exec. Also fix assuming VCC is 64-bit to fix
wave32.
Also remove some untested handling for physical registers which is
skipped. This doesn't insert the V_CNDMASK_B32 if SCC is the physical
copy source. I'm not sure if this should be trying to handle this
special case instead of dealing with this in copyPhysReg.
llvm-svn: 364761
Summary:
The new test case led to incorrect code.
Change-Id: Ief48b227e97aa662dd3535c9bafb27d4a184efca
Reviewers: arsenm, david-salinas
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63871
llvm-svn: 364566
Change the interface of CallLowering::lowerFormalArguments to accept
several virtual registers for each formal argument, instead of just one.
This is a follow-up to D46018.
CallLowering::lowerReturn was similarly refactored in D49660. lowerCall
will be refactored in the same way in follow-up patches.
With this change, we forward the virtual registers generated for
aggregates to CallLowering. Therefore, the target can decide itself
whether it wants to handle them as separate pieces or use one big
register. We also copy the pack/unpackRegs helpers to CallLowering to
facilitate this.
ARM and AArch64 have been updated to use the passed in virtual registers
directly, which means we no longer need to generate so many
merge/extract instructions.
AArch64 seems to have had a bug when lowering e.g. [1 x i8*], which was
put into a s64 instead of a p0. Added a test-case which illustrates the
problem more clearly (it crashes without this patch) and fixed the
existing test-case to expect p0.
AMDGPU has been updated to unpack into the virtual registers for
kernels. I think the other code paths fall back for aggregates, so this
should be NFC.
Mips doesn't support aggregates yet, so it's also NFC.
x86 seems to have code for dealing with aggregates, but I couldn't find
the tests for it, so I just added a fallback to DAGISel if we get more
than one virtual register for an argument.
Differential Revision: https://reviews.llvm.org/D63549
llvm-svn: 364510
Summary:
The +DumpCode attribute is a horrible hack in AMDGPU to embed the
disassembly of the generated code into the elf file. It is used by LLPC
to implement an extension that allows the application to read back the
disassembly of the code.
It tries to print an entry label at the start of every function, but
that didn't work for the first function in the module because
DumpCodeInstEmitter wasn't initialised until EmitFunctionBodyStart
which is too late.
Change-Id: I790d73ddf4f51fd02ab32529380c7cb7c607c4ee
Reviewers: arsenm, tpr, kzhuravl
Reviewed By: arsenm
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63712
llvm-svn: 364508
The LivePhysRegs calculated in order to find a scratch register in the
epilogue code wrongly uses 'LiveIns'. Instead, it should use the
'Liveout' sets. For the liveness, also considering the operands of
the terminator (return) instruction which is the insertion point for
the scratch-exec-copy instruction.
Patch by Christudasan Devadasan
llvm-svn: 364470
Original patch https://reviews.llvm.org/D63659 from
Steven Perron <stevenperron@google.com>
The pass AMDGPUUnifyDivergentExitNodes does not update the phi nodes in
the successors of blocks that is splits. This is fixed by calling
BasicBlock::splitBasicBlock to split the block instead of doing it
manually. This does extra work because a new conditional branch is
created in BB which is immediately replaced, but I think the simplicity
is worth it. It also helps make the code more future proof in case other
things need to be updated.
llvm-svn: 364342
Summary:
The symbols use the processor-specific SHN_AMDGPU_LDS section index
introduced with a previous change. The linker is then expected to resolve
relocations, which are also emitted.
Initially disabled for HSA and PAL environments until they have caught up
in terms of linker and runtime loader.
Some notes:
- The llvm.amdgcn.groupstaticsize intrinsics can no longer be lowered
to a constant at compile times, which means some tests can no longer
be applied.
The current "solution" is a terrible hack, but the intrinsic isn't
used by Mesa, so we can keep it for now.
- We no longer know the full LDS size per kernel at compile time, which
means that we can no longer generate a relevant error message at
compile time. It would be possible to add a check for the size of
individual variables, but ultimately the linker will have to perform
the final check.
Change-Id: If66dbf33fccfbf3609aefefa2558ac0850d42275
Reviewers: arsenm, rampitec, t-tye, b-sumner, jsjodin
Subscribers: qcolombet, kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61494
llvm-svn: 364297
Scalar extends to s64 can use S_BFE_{I64|U64}, but vector extends need
to extend to the 32-bit half, and then to 64.
I'm not sure what the line should be between what RegBankSelect
handles, and what instruction select does, but for now I'm erring on
the side of RegBankSelect for future post-RBS combines.
llvm-svn: 364212
This needs different handling if the source is known to be a valid
condition or not. Handle turning it into shifts or a select during
regbankselect.
llvm-svn: 364186
This matters for byval uses outside of the entry block, which appear
as copies.
Previously, the only folding done was during selection, which could
not see the underlying frame index. For any uses outside the entry
block, the frame index was materialized in the entry block relative to
the global scratch wave offset.
This may produce worse code in cases where the offset ends up not
fitting in the MUBUF offset field. A better heuristic would be helpfu
for extreme frames.
llvm-svn: 364185
We sometimes get poor code size because constants of types < 32b are legalized
as 32 bit G_CONSTANTs with a truncate to fit. This works but means that the
localizer can no longer sink them (although it's possible to extend it to do so).
On AArch64 however s8 and s16 constants can be selected in the same way as s32
constants, with a mov pseudo into a W register. If we make s8 and s16 constants
legal then we can avoid unnecessary truncates, they can be CSE'd, and the
localizer can sink them as normal.
There is a caveat: if the user of a smaller constant has to widen the sources,
we end up with an anyext of the smaller typed G_CONSTANT. This can cause
regressions because of the additional extend and missed pattern matching. To
remedy this, there's a new artifact combiner to generate the wider G_CONSTANT
if it's legal for the target.
Differential Revision: https://reviews.llvm.org/D63587
llvm-svn: 364075
Every called function could possibly need this to calculate the
absolute address of stack objectst, and this avoids inserting a copy
around every call site in the kernel. It's also somewhat cleaner to
keep this in a callee saved SGPR.
llvm-svn: 363990
The attribute can specify elimination for leaf or non-leaf, so it
should always be considered. I copied this bug from AArch64, which
probably should also be fixed.
llvm-svn: 363949
Introducing VCC defs during SIFixSGPRCopies is generally
problematic. Avoid it by starting with the VOP3 form with the general
condition register. This is the easiest to fix instance, but doesn't
solve any specific problems I'm looking at.
llvm-svn: 363904
The def instruction for the vreg may not match, because it may be
folding through a reg_sequence. The assert was overly conservative and
not necessary. It's not actually important if DefMI really defined the
register, because the fold that will be done cares about the def of
the value that will be folded.
For some reason copies aren't making it through the reg_sequence,
although they should.
llvm-svn: 363876
This reapplies r363678, using the correct chain for the CopyToReg for
v0. glueCopyToM0 counterintuitively changes the operands of the
original node.
llvm-svn: 363870
This allows targets to make more decisions about reserved registers
after isel. For example, now it should be certain there are calls or
stack objects in the frame or not, which could have been introduced by
legalization.
Patch by Matthias Braun
llvm-svn: 363757
This was ignoring the flag on fneg, and using the source instruction's
flags. Also fixes tests missing from r358702.
Note the expansion itself isn't correct without nnan, but that should
be fixed separately.
llvm-svn: 363637
AMDGPUPropagateAttributes will not work on function bitcatsts,
so move AMDGPUFixFunctionBitcasts before it.
Differential Revision: https://reviews.llvm.org/D63455
llvm-svn: 363614
Summary:
The purpose of the padding is to guard against stale code being
fetched into the instruction cache by the lowest level prefetching.
We're generating relocatable ELF here, and so the padding should
arguably be added by the linker. This is in fact what Mesa does.
This also fixes multi-part shaders for Mesa.
Change-Id: I6bfede58f20e9f337762ccf39ef9e0e263e69e82
Reviewers: arsenm, rampitec, t-tye
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63427
llvm-svn: 363602
Summary:
This is related to the changes to the groupstaticsize intrinsic in
D61494 which would otherwise make the related tests in these files
fail or much less useful.
Note that for some reason, SOPK generation is less effective in the
amdhsa OS, which is why I chose PAL. I haven't investigated this
deeper.
Change-Id: I6bb99569338f7a433c28b4c9eb1e3e036b00d166
Reviewers: arsenm
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D63392
llvm-svn: 363600
The pass works in two modes:
Mode 1: Just set attributes starting from kernels. This can work at
the very beginning of opt and llc pipeline, but cannot clone functions
because it must be a function pass.
Mode 2: Actually clone functions for new attributes. This can only work
after all function passes in the opt pipeline because it has to be a
module pass.
Differential Revision: https://reviews.llvm.org/D63208
llvm-svn: 363586
A target intrinsic may be defined as possibly reading memory, but the
call site may have additional knowledge that it doesn't read
memory. The intrinsic lowering will expect the pessimistic assumption
of the intrinsic definition, so the chain should still be used.
I fixed the same bug in SelectionDAG in r287593.
llvm-svn: 363580
I keep using the wrong instruction when manually writing tests. This
really needs to check the number of operands, but I don't see an easy
way to do that right now.
llvm-svn: 363579
Some GEPs were not being split, presumably because that split would just be
undone by the DAGCombiner. Not performing those splits can prevent important
optimizations, such as preventing the element indices / member offsets from
being (partially) folded into load/store instruction immediates. This patch:
- Makes the splits also occur in the cases where the base address and the GEP
are in the same BB.
- Ensures that the DAGCombiner doesn't reassociate them back again.
Differential Revision: https://reviews.llvm.org/D60294
llvm-svn: 363544
This patch changes MIR stack-id from an integer to an enum,
and adds printing/parsing support for this in MIR files. The default
stack-id '0' is now renamed to 'default'.
This should make MIR tests that have stack objects with different stack-ids
more descriptive. It also clarifies code operating on StackID.
Reviewers: arsenm, thegameg, qcolombet
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D60137
llvm-svn: 363533
Summary:
Instead of encoding a high-word of 0 using a fake TargetGlobalAddress,
just use a literal target constant. This simplifies some subsequent changes.
The generated assembly is now more explicit about the kind of relocation
that is to be used.
Change-Id: I066835202d23b5941fa7a358eb4b89e9b71ab6f8
Reviewers: arsenm, rampitec
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61491
llvm-svn: 363516
Current findBestLoopTop can find and move one kind of block to top, a latch block has one successor. Another common case is:
* a latch block
* it has two successors, one is loop header, another is exit
* it has more than one predecessors
If it is below one of its predecessors P, only P can fall through to it, all other predecessors need a jump to it, and another conditional jump to loop header. If it is moved before loop header, all its predecessors jump to it, then fall through to loop header. So all its predecessors except P can reduce one taken branch.
Differential Revision: https://reviews.llvm.org/D43256
llvm-svn: 363471
Currently you get extra waits, because waits are inserted for the
register dependencies of the call, and the function prolog waits on
everything.
Currently waits are still inserted on returns. It may make sense to
not do this, and wait in the caller instead.
llvm-svn: 363465
The way SelectionDAG treats memory operands is very frustrating, and
by default drops them unless a property is set on the pattern. There
is no pattern for manually selected instructions, so this requires
manually setting them.
llvm-svn: 363455
Avoid producing illegal register bank copies for reg_sequence and
phi. The default implementation assumes it is possible to pick any
operand's bank and use that for the result, introducing a copy for
operands with a different bank. This does not check for illegal
copies. It is not legal to introduce a VGPR->SGPR copy, so any VGPR
operand requires the result to be a VGPR.
The changes in getInstrMappingImpl aren't strictly necessary, since
AMDGPU now just bypasses this for reg_sequence/phi. This could be
replaced with an assert in case other targets run into this. It is
currently responsible for producing the error for unsatisfiable
copies, but this will be better served with a verifier check.
For phis, for now assume any undetermined operands must be
VGPRs. Eventually, this needs to be able to defer mapping these
operations. This also does not yet have a way to check for whether the
block is in a divergent region.
llvm-svn: 363410
This behavior was added in r130928 for both FastISel and SD, and then
disabled in r131156 for FastISel.
This re-enables it for FastISel with the corresponding fix.
This is triggered only when FastISel can't lower the arguments and falls
back to SelectionDAG for it.
FastISel contains a map of "register fixups" where at the end of the
selection phase it replaces all uses of a register with another
register that FastISel sometimes pre-assigned. Code at the end of
SelectionDAGISel::runOnMachineFunction is doing the replacement at the
very end of the function, while other pieces that come in before that
look through the MachineFunction and assume everything is done. In this
case, the real issue is that the code emitting COPY instructions for the
liveins (physreg to vreg) (EmitLiveInCopies) is checking if the vreg
assigned to the physreg is used, and if it's not, it will skip the COPY.
If a register wasn't replaced with its assigned fixup yet, the copy will
be skipped and we'll end up with uses of undefined registers.
This fix moves the replacement of registers before the emission of
copies for the live-ins.
The initial motivation for this fix is to enable tail calls for
swiftself functions, which were blocked because we couldn't prove that
the swiftself argument (which is callee-save) comes from a function
argument (live-in), because there was an extra copy (vreg to vreg).
A few tests are affected by this:
* llvm/test/CodeGen/AArch64/swifterror.ll: we used to spill x21
(callee-save) but never reload it because it's attached to the return.
We now don't even spill it anymore.
* llvm/test/CodeGen/*/swiftself.ll: we tail-call now.
* llvm/test/CodeGen/AMDGPU/mubuf-legalize-operands.ll: I believe this
test was not really testing the right thing, but it worked because the
same registers were re-used.
* llvm/test/CodeGen/ARM/cmpxchg-O0.ll: regalloc changes
* llvm/test/CodeGen/ARM/swifterror.ll: get rid of a copy
* llvm/test/CodeGen/Mips/*: get rid of spills and copies
* llvm/test/CodeGen/SystemZ/swift-return.ll: smaller stack
* llvm/test/CodeGen/X86/atomic-unordered.ll: smaller stack
* llvm/test/CodeGen/X86/swifterror.ll: same as AArch64
* llvm/test/DebugInfo/X86/dbg-declare-arg.ll: stack size changed
Differential Revision: https://reviews.llvm.org/D62361
llvm-svn: 362963
SIInsertSkips really doesn't understand the control flow, and makes
very stupid assumptions about the block layout. This was able to get
away with not skipping return blocks, since usually after
structurization there is only one placed at the end of the
function. Tail duplication can break this assumption.
llvm-svn: 362754
"Divergence driven ISel. Assign register class for cross block values
according to the divergence."
that discovered the design flaw leading to several issues that
required to be solved before.
This change reverts AMDGPU specific changes and keeps common part
unaffected.
llvm-svn: 362749
This forced the caller to be aware of this, which is an ugly ABI
feature.
Partially reverts r295877. The original reasons for doing this are
mostly fixed. Alloca is now in a non-0 address space, so it should be
OK to have 0 as a valid pointer. Since we treat the absolute address
as the pointer value, this part only really needed to apply to
kernels.
Since r357093, we avoid the need to increment/decrement the offset
register in more cases, and since r354816 the scavenger can fail
without spilling, so it's less critical that we try to avoid an offset
that fits in the MUBUF offset.
Restrict to callable functions for now to split this into 2 steps to
limit thte number of test updates and in case anything breaks.
llvm-svn: 362665
Since the beginning, the offset of a frame index has been consistently
interpreted backwards. It was treating it as an offset from the
scratch wave offset register as a frame register. The correct
interpretation is the offset from the SP on entry to the function,
before the prolog. Frame index elimination then should select either
SP or another register as an FP.
Treat the scratch wave offset on kernel entry as the pre-incremented
SP. Rely more heavily on the standard hasFP and frame pointer
elimination logic, and clean up the private reservation code. This
saves a copy in most callee functions.
The kernel prolog emission code is still kind of a mess relying on
checking the uses of physical registers, which I would prefer to
eliminate.
Currently selection directly emits MUBUF instructions, which require
using a reference to some register. Use the register chosen for SP,
and then ignore this later. This should probably be cleaned up to use
pseudos that don't refer to any specific base register until frame
index elimination.
Add a workaround for shaders using large numbers of SGPRs. I'm not
sure these cases were ever working correctly, since as far as I can
tell the logic for figuring out which SGPR is the scratch wave offset
doesn't match up with the shader input initialization in the shader
programming guide.
llvm-svn: 362661
This is something of a workaround, and the state of stack realignment
controls is kind of a mess. Ideally, we would be able to specify the
stack is infinitely aligned on entry to a kernel.
TargetFrameLowering provides multiple controls which apply at
different points. The StackRealignable field is used during
SelectionDAG, and for some reason distinct from this
hook. StackAlignment is a single field not dependent on the
function. It would probably be better to make that dependent on the
calling convention, and the maximum value for kernels.
Currently this doesn't really change anything, since the frame
lowering mostly does its own thing. This helps avoid regressions in a
future change which will rely more heavily on hasFP.
llvm-svn: 362447
Summary:
This prevents regressions in next patch,
and somewhat recovers from the regression to AMDGPU test in D62223.
It is indeed not great that we leave vector decrement,
don't transform it into vector add all-ones..
https://rise4fun.com/Alive/ZRl
This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs, and then reverted in
rL362109 to fix missing constant folds that were causing
endless combine loops.
Reviewers: RKSimon, craig.topper, spatel, arsenm
Reviewed By: RKSimon, arsenm
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62263
llvm-svn: 362144
Summary:
The main motivation is shown by all these `neg` instructions that are now created.
In particular, the `@reg32_lshr_by_negated_unfolded_sub_b` test.
AArch64 test changes all look good (`neg` created), or neutral.
X86 changes look neutral (vectors), or good (`neg` / `xor eax, eax` created).
I'm not sure about `X86/ragreedy-hoist-spill.ll`, it looks like the spill
is now hoisted into preheader (which should still be good?),
2 4-byte reloads become 1 8-byte reload, and are elsewhere,
but i'm not sure how that affects that loop.
I'm unable to interpret AMDGPU change, looks neutral-ish?
This is hopefully a step towards solving [[ https://bugs.llvm.org/show_bug.cgi?id=41952 | PR41952 ]].
https://rise4fun.com/Alive/pkdq (we are missing more patterns, i'll submit them later)
This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs, and then reverted in
rL362109 to fix missing constant folds that were causing
endless combine loops.
Reviewers: craig.topper, RKSimon, spatel, arsenm
Reviewed By: RKSimon
Subscribers: bjope, qcolombet, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62223
llvm-svn: 362142
With LLPC, previous investigation has suggested that si-scheduler
interacts badly with SiFormMemoryClauses on an XNACK target in some
games.
That needs further investigation in the future. In the meantime, this
commit adds a target-specific attribute to allow us to disable
SIFormMemoryClauses by setting it to 1 on a per-function basis for LLPC
to use.
Differential Revision: https://reviews.llvm.org/D62572
Change-Id: Ia0ca12ce79093cbbe86caded723ffb13384ede92
llvm-svn: 362127
I was looking into an endless combine loop the uncommitted follow-up patch
was causing, and it appears even these patches can exibit such an
endless loop. The root cause is that we try to hoist one binop (add/sub) with
constant operand, and if we get two such binops both of which are
eligible for this hoisting, we get stuck.
Some cases may highlight missing constant-folds.
Reverts r361871,r361872,r361873,r361874.
llvm-svn: 362109
Summary:
This prevents regressions in next patch,
and somewhat recovers from the regression to AMDGPU test in D62223.
It is indeed not great that we leave vector decrement,
don't transform it into vector add all-ones..
https://rise4fun.com/Alive/ZRl
This is a recommit, originally committed in rL361855, but reverted
to investigate test-suite compile-time hangs.
Reviewers: RKSimon, craig.topper, spatel, arsenm
Reviewed By: RKSimon, arsenm
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62263
llvm-svn: 361873
Summary:
The main motivation is shown by all these `neg` instructions that are now created.
In particular, the `@reg32_lshr_by_negated_unfolded_sub_b` test.
AArch64 test changes all look good (`neg` created), or neutral.
X86 changes look neutral (vectors), or good (`neg` / `xor eax, eax` created).
I'm not sure about `X86/ragreedy-hoist-spill.ll`, it looks like the spill
is now hoisted into preheader (which should still be good?),
2 4-byte reloads become 1 8-byte reload, and are elsewhere,
but i'm not sure how that affects that loop.
I'm unable to interpret AMDGPU change, looks neutral-ish?
This is hopefully a step towards solving [[ https://bugs.llvm.org/show_bug.cgi?id=41952 | PR41952 ]].
https://rise4fun.com/Alive/pkdq (we are missing more patterns, i'll submit them later)
This is a recommit, originally committed in rL361852, but reverted
to investigate test-suite compile-time hangs.
Reviewers: craig.topper, RKSimon, spatel, arsenm
Reviewed By: RKSimon
Subscribers: bjope, qcolombet, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62223
llvm-svn: 361871
Summary:
- There's a regression due to the cross-block RC assignment. Use the
proper way to derive the output register RC in inline asm.
Reviewers: rampitec, alex-t
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, dstuttard, tpr, t-tye, eraman, hiraditya, llvm-commits, yaxunl
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62537
llvm-svn: 361868
Summary:
This prevents regressions in next patch,
and somewhat recovers from the regression to AMDGPU test in D62223.
It is indeed not great that we leave vector decrement,
don't transform it into vector add all-ones..
https://rise4fun.com/Alive/ZRl
Reviewers: RKSimon, craig.topper, spatel, arsenm
Reviewed By: RKSimon, arsenm
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62263
llvm-svn: 361855
Summary:
The main motivation is shown by all these `neg` instructions that are now created.
In particular, the `@reg32_lshr_by_negated_unfolded_sub_b` test.
AArch64 test changes all look good (`neg` created), or neutral.
X86 changes look neutral (vectors), or good (`neg` / `xor eax, eax` created).
I'm not sure about `X86/ragreedy-hoist-spill.ll`, it looks like the spill
is now hoisted into preheader (which should still be good?),
2 4-byte reloads become 1 8-byte reload, and are elsewhere,
but i'm not sure how that affects that loop.
I'm unable to interpret AMDGPU change, looks neutral-ish?
This is hopefully a step towards solving [[ https://bugs.llvm.org/show_bug.cgi?id=41952 | PR41952 ]].
https://rise4fun.com/Alive/pkdq (we are missing more patterns, i'll submit them later)
Reviewers: craig.topper, RKSimon, spatel, arsenm
Reviewed By: RKSimon
Subscribers: bjope, qcolombet, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, javed.absar, dstuttard, tpr, t-tye, kristof.beyls, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62223
llvm-svn: 361852
If the only VGPRs used for SGPR spilling were not CSRs, this was
enabling all laness and immediately restoring exec. This is the usual
situation in leaf functions.
llvm-svn: 361848
Summary:
- Don't treat the use of a scalar register as `vreg_1` an VGPR usage.
Otherwise, that promotes that scalar register into vector one, which
breaks the assumption that scalar register holds the lane mask.
- The issue is triggered in a complicated case, where if the uses of
that (lane mask) scalar register is legalized firstly before its
definition, e.g., due to the mismatch block placement and its
topological order or loop. In that cases, the legalization of PHI
introduces the use of that scalar register as `vreg_1`.
Reviewers: rampitec, nhaehnle, arsenm, alex-t
Subscribers: kzhuravl, jvesely, wdng, dstuttard, tpr, t-tye, hiraditya, llvm-commits, yaxunl
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62492
llvm-svn: 361847
Summary:
- The current implementation simplifies the case where the source of
`copyto` is `implicit-def`ed. However, it only works when that
`implicit-def` is single-used since it detects that from
`implicit-def` and cannot determine which destination vreg should be
used if there are multiple uses.
- This patch changes that detection when `copyto` is being emitted. If
that `copyto`'s source is defined from `implicit-def`, it simplifies
it. Hence, it works even that `implicit-def` is multi-used.
- Except it simplifies the internal IR, it won't improve the quality of
code generation. However, it helps to detect 'implicit-def` in a
straight-forward manner in some passes, such as `si-i1-copies`. A test
case is added.
Reviewers: sunfish, nhaehnle
Subscribers: jvesely, hiraditya, asbirlea, llvm-commits, yaxunl
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62342
llvm-svn: 361777
Details: To make instruction selection really divergence driven it is necessary to assign
the correct register classes to the cross block values beforehand. For the divergent targets
same value type requires different register classes dependent on the value divergence.
Reviewers: rampitec, nhaehnle
Differential Revision: https://reviews.llvm.org/D59990
This commit was reverted because of the build failure.
The reason was mlformed patch.
Build failure fixed.
llvm-svn: 361741
Details: To make instruction selection really divergence driven it is necessary to assign
the correct register classes to the cross block values beforehand. For the divergent targets
same value type requires different register classes dependent on the value divergence.
Reviewers: rampitec, nhaehnle
Differential Revision: https://reviews.llvm.org/D59990
llvm-svn: 361644
We were assuming a much larger possible per-wave visible stack
allocation than is possible:
faa3ae5138/src/core/runtime/amd_gpu_agent.cpp (L70)
Based on this, we can assume the high 15 bits of a frame index or sret
are 0. The frame index value is the per-lane offset, so the maximum
frame index value is MAX_WAVE_SCRATCH / wavesize.
Remove the corresponding subtarget feature and option that made
this configurable.
llvm-svn: 361541
Keep it optional in cases this is ever needed in some global
context. Currently it's only used for getting an upper bound inline
asm code size.
For AMDGPU, gfx10 increases the maximum instruction size to
20-bytes. This avoids penalizing older subtargets when estimating code
size, and making some annoying branch relaxation test adjustments.
llvm-svn: 361405
Register coalescer fails for the test in the patch with the assertion in
JoinVals::ConflictResolution `DefMI != nullptr'. It attempts to join
live intervals for two adjacent instructions and erase the copy:
%2:vreg_256 = COPY %1
%3:vreg_256 = COPY killed %1
The LI needs to be adjusted to kill subrange for the erased instruction
and extend the subrange of the original def. That was done for the main
interval only but not for the subrange. As a result subrange had a VNI
pointing to the erased slot resulting in the above failure.
Differential Revision: https://reviews.llvm.org/D62162
llvm-svn: 361293
Unfortunately the way SIInsertSkips works is backwards, and is
required for correctness. r338235 added handling of some special cases
where skipping is mandatory to avoid side effects if no lanes are
active. It conservatively handled asm correctly, but the same logic
needs to apply to calls.
Usually the call sequence code is larger than the skip threshold,
although the way the count is computed is really broken, so I'm not
sure if anything was likely to really hit this.
llvm-svn: 361202
The code did not match the example in the comment, and was checking
the undef flag on the copy dest instead of source. The existing tests
were only hitting the > 2 operands case.
llvm-svn: 361156
Summary:
Avoid introducing hazard mitigation when lgkmcnt is reduced to 0.
Clarify code comments to explain assumptions made for this hazard
mitigation. Expand and correct test cases to cover variants of
s_waitcnt.
Reviewers: nhaehnle, rampitec
Subscribers: arsenm, kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62058
llvm-svn: 361124
Summary:
Was looking into supporting `(srl (shl x, c1), c2)` with c1 != c2 in dagcombiner,
this test changes, but makes `update_llc_test_checks.py` unhappy.
**Many** AMDGPU tests specify `-march`, not `-mtriple`, which results in `update_llc_test_checks.py`
defaulting to x86 asm function detection heuristics, which don't work here.
I propose to fix this by adding an infrastructure to map from `-march` to `-mtriple`,
in the UpdateTestChecks tooling.
Reviewers: RKSimon, MaskRay, arsenm
Reviewed By: arsenm
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D62099
llvm-svn: 361101
This is ported from the custom AMDGPU DAG implementation. I think this
is a better default expansion than what the DAG currently uses, at
least if the target has CTLZ.
This implements the signed version in terms of the unsigned
conversion, which is implemented with bit operations. SelectionDAG has
several other implementations that should eventually be ported
depending on what instructions are legal.
llvm-svn: 361081
Summary:
In order to combine memory operations efficiently, the load/store
optimizer might move some instructions around. It's usually safe
to move instructions down past the merged instruction because the
pass checks if memory operations can be re-ordered.
Though, the current logic doesn't handle Write-after-Write hazards.
This fixes a reflection issue with Monster Hunter World and DXVK.
v2: - rebased on top of master
- clean up the test case
- handle WaW hazards correctly
Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=40130
Original patch by Samuel Pitoiset.
Reviewers: tpr, arsenm, nhaehnle
Reviewed By: nhaehnle
Subscribers: ronlieb, arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye
Differential Revision: https://reviews.llvm.org/D61313
llvm-svn: 361008
The recent introduction of v3i32 etc as an MVT, and its use in AMDGPU
3-dword memory instructions, caused a de-optimization problem for code
with such a load that then bitcasts via vector of i8, because v12i8 is
not an MVT so it legalizes the bitcast by widening it.
This commit adds the ability to widen a bitcast using extract_subvector
on the result, so the value does not need to go via memory.
Differential Revision: https://reviews.llvm.org/D60457
Change-Id: Ie4abb7760547e54a2445961992eafc78e80d4b64
llvm-svn: 360942
This is the conservatively correct default. It is always safe to
assume xnack is enabled, but not the converse.
Introduce a feature to blacklist targets where xnack can never be
meaningfully enabled. I'm not sure the targets this is applied to is
100% correct.
llvm-svn: 360903
Trace through multiple COPYs when looking for a physreg source. Add
hinting for vregs that will be copied into physregs (we only hinted
for vregs getting copied to a physreg previously). Give hinted a
register a bonus when deciding which value to spill. This is part of
my rewrite regallocfast series. In fact this one doesn't even have an
effect unless you also flip the allocation to happen from back to
front of a basic block. Nonetheless it helps to split this up to ease
review of D52010
Patch by Matthias Braun
llvm-svn: 360887
Summary:
SGPR in CC can be either hw initialized or set by other chained shaders
and so this increases the SGPR count availalbe to CC to 105.
Change-Id: I3dfadc750fe4a3e2bd07117a2899fd13f3e2fef3
Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61261
llvm-svn: 360778
Tighten conditions on SMEM WAR hazard unit tests to ensure rejection
of workaround insertion where a s_waitcnt is present in dependency
chain. The current workaround code already conforms to these revise
tests.
llvm-svn: 360686
This also allows three op patterns to use increased constant bus
limit of GFX10.
Differential Revision: https://reviews.llvm.org/D61763
llvm-svn: 360395
The VOP3 form should always be the preferred selection, to be shrunk
later. This should only be an optimization issue, but this partially
works around a problem from clobbering VCC when SIFixSGPRCopies
rewrites an SCC defining operation directly to VCC.
3 of the testcases are regressions from failing to fold the immediate
in cases it should. These can be avoided by improving the VCC liveness
handling in SIFoldOperands. Simply increasing the threshold to
computeRegisterLiveness works, although this is common enough that VCC
liveness should probably be tracked throughout the pass. The hack of
leaving behind an implicit_def instruction to avoid breaking iterator
wastes instruction count, which inhibits finding the VCC def in long
chains of adds. Doing this however exposes different, worse looking
regressions from poor scheduling behavior. This could probably be
avoided around by forcing the shrink of the addc here, but the
scheduler should probably be fixed.
The r600 add test needs to be split out because it asserts on the
arguments in the new test during the calling convention lowering.
llvm-svn: 360293
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
Summary: GCNHazardRecognizer fails to identify hazards that are in and around bundles. This patch allows the hazard recognizer to consider bundled instructions in both scheduler and hazard recognizer mode. We ignore “bundledness” for the purpose of detecting hazards and examine the instructions individually.
Reviewers: arsenm, msearles, rampitec
Reviewed By: rampitec
Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61564
llvm-svn: 360199
Summary:
No test case because I don't know of a way to trigger this, but I
accidentally caused this to fail while working on a different change.
Change-Id: I8015aa447fe27163cc4e4902205a203bd44bf7e3
Reviewers: arsenm, rampitec
Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61490
llvm-svn: 360123
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
The VOP3 form should always be the preferred selection form to be
shrunk later.
The r600 sub test needs to be split out because it asserts on the
arguments in the new test during the calling convention lowering.
llvm-svn: 359899
This was broken if the original operand was killed. The kill flag
would appear on both instructions, and fail the verifier. Keep the
kill flag, but remove the operands from the old instruction. This has
an added benefit of really reducing the use count for future folds.
Ideally the pass would be structured more like what PeepholeOptimizer
does to avoid this hack to avoid breaking instruction iterators.
llvm-svn: 359891
When a fold of an immediate into a sub/subrev required shrinking the
instruction, the wrong VOP2 opcode was used. This was using the VOP2
equivalent of the original instruction, not the commuted instruction
with the inverted opcode.
llvm-svn: 359883
We use both -long-option and --long-option in tests. Switch to --long-option for consistency.
In the "llvm-readelf" mode, -long-option is discouraged as it conflicts with grouped short options and it is not accepted by GNU readelf.
While updating the tests, change llvm-readobj -s to llvm-readobj -S to reduce confusion ("s" is --section-headers in llvm-readobj but --symbols in llvm-readelf).
llvm-svn: 359649
Summary:
Extract the logic for doing reassociations
from DAGCombiner::reassociateOps into a helper
function DAGCombiner::reassociateOpsCommutative,
and use that helper to trigger reassociation
on the original operand order, or the commuted
operand order.
Codegen is not identical since the operand order will
be different when doing the reassociations for the
commuted case. That causes some unfortunate churn in
some test cases. Apart from that this should be NFC.
Reviewers: spatel, craig.topper, tstellar
Reviewed By: spatel
Subscribers: dmgreen, dschuff, jvesely, nhaehnle, javed.absar, sbc100, jgravelle-google, hiraditya, aheejin, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D61199
llvm-svn: 359476
This reverts commit 7a6ef3004655dd86d722199c471ae78c28e31bb4.
We discovered some internal test failures, so reverting for now.
Differential Revision: https://reviews.llvm.org/D61213
llvm-svn: 359363
Michael Liao