The peephole optimizer tries to replace
```
%n:sgpr_32 = S_MOV_B32 x
$scc = COPY %n
```
with a `S_MOV_B32` directly into `$scc`.
This crashes because `S_MOV_B32` cannot take `$scc` as input.
We currently generate code like this from GlobalISel when lowering a
G_BRCOND with a constant condition. We should probably look into
removing this kind of branch altogether, but until then we should at
least not crash.
This patch fixes the issue by making sure we don't apply the peephole
optimization when trying to move into a physical register that
doesn't belong to the correct register class.
Differential Revision: https://reviews.llvm.org/D148117
An INLINEASM can have an implicit def of vcc. It is not appropriate for
fixImplicitOperands to change this to vcc_lo on wave32.
Differential Revision: https://reviews.llvm.org/D147157
If more registers are needed for VAddr then the NSA format allows then the
final register can act as a contigous set of remaining addresses. Update
legalizer to pack register for this new format and allow instruction
selection to use NSA encoding when number of addresses exceeds max size.
Also update SIShrinkInstructions to handle partial NSA.
Differential Revision: https://reviews.llvm.org/D144034
Moving this out of AMDGPUBaseInfo enforces that AMDGPUBaseInfo should
not be calling into GCNSubtarget.
Differential Revision: https://reviews.llvm.org/D144564
IsNeverUniform can be set to 1 to mark instructions which are
inherently never-uniform/divergent. Enabling this bit in
Writelane instruction for now. To be extended to all required
instructions.
Reviewed By: arsenm, sameerds, #amdgpu
Differential Revision: https://reviews.llvm.org/D143154
Right now we do opcode wise matching to identify uniform/non-divergent
AMDGPU intrinsics. It is duplicated at 2 places once at IR level uniformity analysis
and at MIR level. Moving them to single tablegen table for consistency and adding
and API rapper to access them.
Reviewed By: arsenm, #amdgpu
Differential Revision: https://reviews.llvm.org/D142961
This was broken for the common case of instructions which are uniform
if their inputs are uniform. This is broken for control flow intrinsics
since the API currently does not express which result operand is in question.
This generates failures in just about every intrinsic test when uniformity
analysis is performed without this.
This was assuming virtual registers only, and asserting on physical.
This was also ignoring AGPRs, and only considering VGPRs.
Reporting the instruction as uniform or not is conceptually wrong,
this should be reported per-operand. An inline asm statement could
include uniform and non-uniform components. This should report
purely for the register defs and ignore the uses.
Fixes asserting on most of the inline asm tests when uniformity
analysis is used.
This was failing when it couldn't find an allocatable class
for special physical register inputs (like $mode), which are all
scalars.
This avoids numerous test failures when regbankselect is updated
to use uniformity analysis.
The new methods return a range for easier iteration. Use them everywhere
instead of getImplicitUses, getNumImplicitUses, getImplicitDefs and
getNumImplicitDefs. A future patch will remove the old methods.
In some use cases the new methods are less efficient because they always
have to scan the whole uses/defs array to count its length, but that
will be fixed in a future patch by storing the number of implicit
uses/defs explicitly in MCInstrDesc. At that point there will be no need
to 0-terminate the arrays.
Differential Revision: https://reviews.llvm.org/D142215
Avoid copying MCInstrDesc instances because a future patch will change
them to find their implicit operands and operand info array based on
their own "this" pointer, so it will only work for MCInstrDescs in the
TargetInsts table, not for a copy of an MCInstrDesc at a different
address.
Differential Revision: https://reviews.llvm.org/D142214
Change MCInstrDesc::operands to return an ArrayRef so we can easily use
it everywhere instead of the (IMHO ugly) opInfo_begin and opInfo_end.
A future patch will remove opInfo_begin and opInfo_end.
Also use it instead of raw access to the OpInfo pointer. A future patch
will remove this pointer.
Differential Revision: https://reviews.llvm.org/D142213
Extension of https://reviews.llvm.org/D141101 to even further reduce the amount of implicit operands we attach. The main benefit is to improve cability of post-ra scheduler, and reduce unneeded dependency resolution (e.g. inserting snops).
Unfortunately, we run into regressions if we completely minimize the amount implicit operands (naively), we run into some regressions (e.g. dual_movs are replaced with multiple calls to v_mov). This is even more reason to switch to LiveRegUnits.
Nonetheless, this patch removes the operands which we can for free (more or less).
Change-Id: Ib4f409202b36bdbc59eed615bc2d19fa8bd8c057
Differential Revision: https://reviews.llvm.org/D141557
Change-Id: I8b039e3c0d39436b384083f8beb947ee1b1730b2
Previously we considered this field to be either N-bit unsigned or
N+1-bit signed, depending on the instruction. I think it's conceptually
simpler to say that the field is always N+1-bit signed, but some
instructions do not allow negative values.
Differential Revision: https://reviews.llvm.org/D140883
This bit is not set/reserved in td file. Let's remove it for now,
we can always add it back if we need it.
Reviewed by: foad
Differential Revision: https://reviews.llvm.org/D141223
Implicit def operands are needed when we spill partially undef super registers by each individual subregister. The implicit-def operands will allow us to lower spills without the verifier complaining. Currently, we are overzeously attaching implicit operands, when we really only need them on the first sub reg spill op. By more selectively attached the implicit ops, we will free up some unneeded dependencies for the post-ra scheduler.
Moreover, this enables a previously incorrect optimization / resolves a correctness issue in indirectCopyToAGPR. When lowering AGPR copies on GFX908, we can improve CodeGen by reusing accvgpr_writes. However, we could not reliably determine which agprs accvgpr_writes actually define due to implicit-defs.
Differential Revision: https://reviews.llvm.org/D141101
They do not seem to be GFX10-specific anymore. Also renames the
corresponding feature.
Reviewed By: dp
Differential Revision: https://reviews.llvm.org/D141069
Use deduction guides instead of helper functions.
The only non-automatic changes have been:
1. ArrayRef(some_uint8_pointer, 0) needs to be changed into ArrayRef(some_uint8_pointer, (size_t)0) to avoid an ambiguous call with ArrayRef((uint8_t*), (uint8_t*))
2. CVSymbol sym(makeArrayRef(symStorage)); needed to be rewritten as CVSymbol sym{ArrayRef(symStorage)}; otherwise the compiler is confused and thinks we have a (bad) function prototype. There was a few similar situation across the codebase.
3. ADL doesn't seem to work the same for deduction-guides and functions, so at some point the llvm namespace must be explicitly stated.
4. The "reference mode" of makeArrayRef(ArrayRef<T> &) that acts as no-op is not supported (a constructor cannot achieve that).
Per reviewers' comment, some useless makeArrayRef have been removed in the process.
This is a follow-up to https://reviews.llvm.org/D140896 that introduced
the deduction guides.
Differential Revision: https://reviews.llvm.org/D140955
Accelerate finding the base class for a physical register by
building a statically mapping table from physical registers
to base classes using TableGen.
Replace uses of SIRegisterInfo::getPhysRegClass with
TargetRegisterInfo::getPhysRegBaseClass in order to use
the computed table.
Reviewed By: arsenm, foad
Differential Revision: https://reviews.llvm.org/D139422
Uniformity analysis is a generalization of divergence analysis to
include irreducible control flow:
1. The proposed spec presents a notion of "maximal convergence" that
captures the existing convention of converging threads at the
headers of natual loops.
2. Maximal convergence is then extended to irreducible cycles. The
identity of irreducible cycles is determined by the choices made
in a depth-first traversal of the control flow graph. Uniformity
analysis uses criteria that depend only on closed paths and not
cycles, to determine maximal convergence. This makes it a
conservative analysis that is independent of the effect of DFS on
CycleInfo.
3. The analysis is implemented as a template that can be
instantiated for both LLVM IR and Machine IR.
Validation:
- passes existing tests for divergence analysis
- passes new tests with irreducible control flow
- passes equivalent tests in MIR and GMIR
Based on concepts originally outlined by
Nicolai Haehnle <nicolai.haehnle@amd.com>
With contributions from Ruiling Song <ruiling.song@amd.com> and
Jay Foad <jay.foad@amd.com>.
Support for GMIR and lit tests for GMIR/MIR added by
Yashwant Singh <yashwant.singh@amd.com>.
Differential Revision: https://reviews.llvm.org/D130746
Currently, the custom SGPR spill lowering pass spills
SGPRs into physical VGPR lanes and the remaining VGPRs
are used by regalloc for vector regclass allocation.
This imposes many restrictions that we ended up with
unsuccessful SGPR spilling when there won't be enough
VGPRs and we are forced to spill the leftover into
memory during PEI. The custom spill handling during PEI
has many edge cases and often breaks the compiler time
to time.
This patch implements spilling SGPRs into virtual VGPR
lanes. Since we now split the register allocation for
SGPRs and VGPRs, the virtual registers introduced for
the spill lanes would get allocated automatically in
the subsequent regalloc invocation for VGPRs.
Spill to virtual registers will always be successful,
even in the high-pressure situations, and hence it avoids
most of the edge cases during PEI. We are now left with
only the custom SGPR spills during PEI for special registers
like the frame pointer which isn an unproblematic case.
This patch also implements the whole wave spills which
might occur if RA spills any live range of virtual registers
involved in the whole wave operations. Earlier, we had
been hand-picking registers for such machine operands.
But now with SGPR spills into virtual VGPR lanes, we are
exposing them to the allocator.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D124196
With D134950, targets get notified when a virtual register is created and/or
cloned. Targets can do the needful with the delegate callback. AMDGPU propagates
the virtual register flags maintained in the target file itself. They are useful
to identify a certain type of machine operands while inserting spill stores and
reloads. Since RegAllocFast spills the physical register itself, there is no way
its virtual register can be mapped back to retrieve the flags. It can be solved
by passing the virtual register as an additional argument. This argument has no
use when the spill interfaces are called during the greedy allocator or even the
PrologEpilogInserter and can pass a null register in such cases.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D138656
C++17 allows us to call constructors pair and tuple instead of helper
functions make_pair and make_tuple.
Differential Revision: https://reviews.llvm.org/D139828
OMod was disabled if OpSel was enabled, but that restriction is more
specific than necessary. Any VOP3 with float operands can use OMod.
On GFX11, FMAC_F16_e64 can use op_sel.
Previously, SIFoldOperands and convertToThreeAddress were accidentally correct when
they reinterpreted the zero OMod operand on V_FMAC_F16_e64 as the OpSel operand on
V_FMA_F16_gfx9_e64. Now we explicitly add op_sel if required.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D139469
Update spill code to account for new vector types with
bit widths: 288, 320, 352, 384.
Related to D138205.
Differential Revision: https://reviews.llvm.org/D139203
isLiteralConstant and isLiteralConstantLike were similar to
!isInlineConstant with slight differences like handling isReg operands.
To avoid a profusion of similar functions with undocumented differences,
this patch removes all the isLiteralConstant* variants. Callers are responsible
for handling the isReg case.
Differential Revision: https://reviews.llvm.org/D125759
In a lot of places, we were just calling `getNamedOperandIdx` to check if the result was != or == to -1.
This is fine in itself, but it's verbose and doesn't make the intention clear, IMHO. I added a `hasNamedOperand` and replaced all cases I could find with regexes and manually.
Reviewed By: arsenm, foad
Differential Revision: https://reviews.llvm.org/D137540
Create new VALU instructions in moveToVALU instead of mutating the
existing SALU instruction. This makes it easier to add extra operands so
we can convert to the VOP3 form of VALU instructions.
NFCI but it does have the minor side effect of removing duplicate
implicit operands that were present on the original SALU if they are
default implicit operands for the VALU.
Differential Revision: https://reviews.llvm.org/D137324
This patch fixes:
llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:7383: warning: enumerated
mismatch in conditional expression:
‘llvm::AMDGPU::UfmtGFX11::UnifiedFormat’ vs
‘llvm::AMDGPU::UfmtGFX10::UnifiedFormat’
Strict WQM does not require a WQM transistion if it occurs within
an existing WQM section.
This occurs heavily in GFX11 pixel shaders with LDS_PARAM_LOAD.
Which leads to unnecessary EXEC mask manipulation.
To avoid these transitions, detect WQM -> Strict WQM -> WQM
and substitute new ENTER_PSEUDO_WM/EXIT_PSEUDO_WM markers instead.
These are treat similarly by WWM register pre-allocation pass,
but do not manipulate EXEC or use registers to save EXEC state.
Reviewed By: piotr
Differential Revision: https://reviews.llvm.org/D136813
Due to the encoding changes in GFX11, we had a hack in place that
disables the use of VGPRs above 128. This patch removes the need for
that hack.
We introduce a new register class VGPR_32_Lo128 which is used for 16-bit
operands of VOP1, VOP2, and VOPC instructions. This register class only has the
low 128 VGPRs, but is otherwise identical to VGPR_32. Therefore, 16-bit VOP1,
VOP2, and VOPC instructions are correctly limited to use the first 128
VGPRs, while the other instructions can freely use all 256.
We introduce new pseduo-instructions used on GFX11 which have the suffix
t16 (True 16) to use the VGPR_32_Lo128 register class.
Reviewed By: foad, rampitec, #amdgpu
Differential Revision: https://reviews.llvm.org/D133723
This is helpful for detecting whether a block ends with divergent branch
in passes before lowering the pseudo control flow instructions.
Differential Revision: https://reviews.llvm.org/D133184
Currently there isn't a generic way to get a smaller register class
that can be produced from a subregister of a larger class. Replaces a
manually implemented version for AMDGPU. This will be used to improve
subregister support in the allocator.
D125803 introduced shrinking of F16 FMA to FMAAK/FMAMK in
SIShrinkInstructions (useful on GFX10+ where VOP3 instructions may have
a literal operand) but failed to handle the V_FMA_F16_gfx9_e64 form of
the opcode which is used on GFX9+.
Differential Revision: https://reviews.llvm.org/D133489
