[llvm] 6082263 - Restore "Implement convergence control in MIR using SelectionDAG (#71785)"
Sameer Sahasrabuddhe via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 5 22:49:13 PST 2024
Author: Sameer Sahasrabuddhe
Date: 2024-03-06T12:19:32+05:30
New Revision: 60822637bf007cbaf7401a6ec25cdf2ea7b7edbd
URL: https://github.com/llvm/llvm-project/commit/60822637bf007cbaf7401a6ec25cdf2ea7b7edbd
DIFF: https://github.com/llvm/llvm-project/commit/60822637bf007cbaf7401a6ec25cdf2ea7b7edbd.diff
LOG: Restore "Implement convergence control in MIR using SelectionDAG (#71785)"
This restores commit c7fdd8c11e54585dc9d15d63de9742067e0506b9.
Previously reverted in f010b1bef4dda2c7082cbb41dbabf1f149cce306.
LLVM function calls carry convergence control tokens as operand bundles, where
the tokens themselves are produced by convergence control intrinsics. This patch
implements convergence control tokens in MIR as follows:
1. Introduce target-independent ISD opcodes and MIR opcodes for convergence
control intrinsics.
2. Model token values as untyped virtual registers in MIR.
The change also introduces an additional ISD opcode CONVERGENCECTRL_GLUE and a
corresponding machine opcode with the same spelling. This glues the convergence
control token to SDNodes that represent calls to intrinsics. The glued token is
later translated to an implicit argument in the MIR.
The lowering of calls to user-defined functions is target-specific. On AMDGPU,
the convergence control operand bundle at a non-intrinsic call is translated to
an explicit argument to the SI_CALL_ISEL instruction. Post-selection adjustment
converts this explicit argument to an implicit argument on the SI_CALL
instruction.
Added:
llvm/include/llvm/CodeGen/MachineConvergenceVerifier.h
llvm/lib/CodeGen/MachineConvergenceVerifier.cpp
llvm/test/CodeGen/AMDGPU/convergence-tokens.ll
llvm/test/MachineVerifier/convergencectrl/AMDGPU/basic.mir
llvm/test/MachineVerifier/convergencectrl/AMDGPU/cycles.mir
llvm/test/MachineVerifier/convergencectrl/AMDGPU/lit.local.cfg
llvm/test/MachineVerifier/convergencectrl/AMDGPU/mixed2.mir
llvm/test/MachineVerifier/convergencectrl/AMDGPU/region-nesting.mir
Modified:
llvm/include/llvm/ADT/GenericConvergenceVerifier.h
llvm/include/llvm/CodeGen/FunctionLoweringInfo.h
llvm/include/llvm/CodeGen/ISDOpcodes.h
llvm/include/llvm/CodeGen/SelectionDAGISel.h
llvm/include/llvm/CodeGen/TargetLowering.h
llvm/include/llvm/IR/GenericConvergenceVerifierImpl.h
llvm/include/llvm/Support/TargetOpcodes.def
llvm/include/llvm/Target/Target.td
llvm/include/llvm/Target/TargetSelectionDAG.td
llvm/lib/CodeGen/CMakeLists.txt
llvm/lib/CodeGen/MachineVerifier.cpp
llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
llvm/lib/CodeGen/ValueTypes.cpp
llvm/lib/IR/ConvergenceVerifier.cpp
llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
llvm/lib/Target/AMDGPU/SIISelLowering.cpp
llvm/lib/Target/AMDGPU/SIInstructions.td
llvm/test/CodeGen/AMDGPU/isel-amdgpu-cs-chain-cc.ll
llvm/test/CodeGen/AMDGPU/kernel-vgpr-spill-mubuf-with-voffset.ll
llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll
llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll
llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll
llvm/test/CodeGen/AMDGPU/stacksave_stackrestore.ll
llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
llvm/test/CodeGen/AMDGPU/vgpr_constant_to_sgpr.ll
llvm/test/CodeGen/AMDGPU/whole-wave-register-spill.ll
llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
llvm/test/CodeGen/PowerPC/fmf-propagation.ll
llvm/test/TableGen/GlobalISelCombinerEmitter/builtins/match-table-replacerreg.td
llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-intrinsics.td
llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-patfrag-root.td
llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-variadics.td
llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
Removed:
################################################################################
diff --git a/llvm/include/llvm/ADT/GenericConvergenceVerifier.h b/llvm/include/llvm/ADT/GenericConvergenceVerifier.h
index 0810a0701322907..dc9495ce57bfd66 100644
--- a/llvm/include/llvm/ADT/GenericConvergenceVerifier.h
+++ b/llvm/include/llvm/ADT/GenericConvergenceVerifier.h
@@ -60,6 +60,10 @@ template <typename ContextT> class GenericConvergenceVerifier {
NoConvergence
} ConvergenceKind = NoConvergence;
+ /// The control token operation performed by a convergence control Intrinsic
+ /// in LLVM IR, or by a CONVERGENCECTRL* instruction in MIR
+ enum ConvOpKind { CONV_ANCHOR, CONV_ENTRY, CONV_LOOP, CONV_NONE };
+
// Cache token uses found so far. Note that we track the unique definitions
// and not the token values.
DenseMap<const InstructionT *, const InstructionT *> Tokens;
@@ -68,6 +72,8 @@ template <typename ContextT> class GenericConvergenceVerifier {
static bool isInsideConvergentFunction(const InstructionT &I);
static bool isConvergent(const InstructionT &I);
+ static ConvOpKind getConvOp(const InstructionT &I);
+ void checkConvergenceTokenProduced(const InstructionT &I);
const InstructionT *findAndCheckConvergenceTokenUsed(const InstructionT &I);
void reportFailure(const Twine &Message, ArrayRef<Printable> Values);
diff --git a/llvm/include/llvm/CodeGen/FunctionLoweringInfo.h b/llvm/include/llvm/CodeGen/FunctionLoweringInfo.h
index cde7247aeb15116..31af3014afe4e7b 100644
--- a/llvm/include/llvm/CodeGen/FunctionLoweringInfo.h
+++ b/llvm/include/llvm/CodeGen/FunctionLoweringInfo.h
@@ -215,15 +215,7 @@ class FunctionLoweringInfo {
Register CreateRegs(Type *Ty, bool isDivergent = false);
- Register InitializeRegForValue(const Value *V) {
- // Tokens never live in vregs.
- if (V->getType()->isTokenTy())
- return 0;
- Register &R = ValueMap[V];
- assert(R == 0 && "Already initialized this value register!");
- assert(VirtReg2Value.empty());
- return R = CreateRegs(V);
- }
+ Register InitializeRegForValue(const Value *V);
/// GetLiveOutRegInfo - Gets LiveOutInfo for a register, returning NULL if the
/// register is a PHI destination and the PHI's LiveOutInfo is not valid.
diff --git a/llvm/include/llvm/CodeGen/ISDOpcodes.h b/llvm/include/llvm/CodeGen/ISDOpcodes.h
index ef0fec270a43c44..49d51a27e3c0f68 100644
--- a/llvm/include/llvm/CodeGen/ISDOpcodes.h
+++ b/llvm/include/llvm/CodeGen/ISDOpcodes.h
@@ -1386,6 +1386,15 @@ enum NodeType {
#define BEGIN_REGISTER_VP_SDNODE(VPSDID, ...) VPSDID,
#include "llvm/IR/VPIntrinsics.def"
+ // The `llvm.experimental.convergence.*` intrinsics.
+ CONVERGENCECTRL_ANCHOR,
+ CONVERGENCECTRL_ENTRY,
+ CONVERGENCECTRL_LOOP,
+ // This does not correspond to any convergence control intrinsic. It is used
+ // to glue a convergence control token to a convergent operation in the DAG,
+ // which is later translated to an implicit use in the MIR.
+ CONVERGENCECTRL_GLUE,
+
/// BUILTIN_OP_END - This must be the last enum value in this list.
/// The target-specific pre-isel opcode values start here.
BUILTIN_OP_END
diff --git a/llvm/include/llvm/CodeGen/MachineConvergenceVerifier.h b/llvm/include/llvm/CodeGen/MachineConvergenceVerifier.h
new file mode 100644
index 000000000000000..86aa8ce1766b890
--- /dev/null
+++ b/llvm/include/llvm/CodeGen/MachineConvergenceVerifier.h
@@ -0,0 +1,28 @@
+//===- MachineConvergenceVerifier.h - Verify convergencectrl ----*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+/// \file
+///
+/// This file declares the MIR specialization of the GenericConvergenceVerifier
+/// template.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CODEGEN_MACHINECONVERGENCEVERIFIER_H
+#define LLVM_CODEGEN_MACHINECONVERGENCEVERIFIER_H
+
+#include "llvm/ADT/GenericConvergenceVerifier.h"
+#include "llvm/CodeGen/MachineSSAContext.h"
+
+namespace llvm {
+
+using MachineConvergenceVerifier =
+ GenericConvergenceVerifier<MachineSSAContext>;
+
+} // namespace llvm
+
+#endif // LLVM_CODEGEN_MACHINECONVERGENCEVERIFIER_H
diff --git a/llvm/include/llvm/CodeGen/SelectionDAGISel.h b/llvm/include/llvm/CodeGen/SelectionDAGISel.h
index dbd9b391f4a431f..837f8bf7263ea98 100644
--- a/llvm/include/llvm/CodeGen/SelectionDAGISel.h
+++ b/llvm/include/llvm/CodeGen/SelectionDAGISel.h
@@ -459,6 +459,10 @@ class SelectionDAGISel : public MachineFunctionPass {
void Select_ARITH_FENCE(SDNode *N);
void Select_MEMBARRIER(SDNode *N);
+ void Select_CONVERGENCECTRL_ANCHOR(SDNode *N);
+ void Select_CONVERGENCECTRL_ENTRY(SDNode *N);
+ void Select_CONVERGENCECTRL_LOOP(SDNode *N);
+
void pushStackMapLiveVariable(SmallVectorImpl<SDValue> &Ops, SDValue Operand,
SDLoc DL);
void Select_STACKMAP(SDNode *N);
diff --git a/llvm/include/llvm/CodeGen/TargetLowering.h b/llvm/include/llvm/CodeGen/TargetLowering.h
index 7be5b5e8dd2f2f8..2f164a460db8432 100644
--- a/llvm/include/llvm/CodeGen/TargetLowering.h
+++ b/llvm/include/llvm/CodeGen/TargetLowering.h
@@ -4438,6 +4438,7 @@ class TargetLowering : public TargetLoweringBase {
SmallVector<ISD::InputArg, 32> Ins;
SmallVector<SDValue, 4> InVals;
const ConstantInt *CFIType = nullptr;
+ SDValue ConvergenceControlToken;
CallLoweringInfo(SelectionDAG &DAG)
: RetSExt(false), RetZExt(false), IsVarArg(false), IsInReg(false),
@@ -4571,6 +4572,11 @@ class TargetLowering : public TargetLoweringBase {
return *this;
}
+ CallLoweringInfo &setConvergenceControlToken(SDValue Token) {
+ ConvergenceControlToken = Token;
+ return *this;
+ }
+
ArgListTy &getArgs() {
return Args;
}
diff --git a/llvm/include/llvm/IR/GenericConvergenceVerifierImpl.h b/llvm/include/llvm/IR/GenericConvergenceVerifierImpl.h
index f6eb5066d5535df..a3ebde709ae6e1c 100644
--- a/llvm/include/llvm/IR/GenericConvergenceVerifierImpl.h
+++ b/llvm/include/llvm/IR/GenericConvergenceVerifierImpl.h
@@ -61,12 +61,11 @@ void GenericConvergenceVerifier<ContextT>::visit(const BlockT &BB) {
template <class ContextT>
void GenericConvergenceVerifier<ContextT>::visit(const InstructionT &I) {
- auto ID = ContextT::getIntrinsicID(I);
- auto *TokenDef = findAndCheckConvergenceTokenUsed(I);
- bool IsCtrlIntrinsic = true;
+ ConvOpKind ConvOp = getConvOp(I);
- switch (ID) {
- case Intrinsic::experimental_convergence_entry:
+ auto *TokenDef = findAndCheckConvergenceTokenUsed(I);
+ switch (ConvOp) {
+ case CONV_ENTRY:
Check(isInsideConvergentFunction(I),
"Entry intrinsic can occur only in a convergent function.",
{Context.print(&I)});
@@ -78,13 +77,13 @@ void GenericConvergenceVerifier<ContextT>::visit(const InstructionT &I) {
"same basic block.",
{Context.print(&I)});
LLVM_FALLTHROUGH;
- case Intrinsic::experimental_convergence_anchor:
+ case CONV_ANCHOR:
Check(!TokenDef,
"Entry or anchor intrinsic cannot have a convergencectrl token "
"operand.",
{Context.print(&I)});
break;
- case Intrinsic::experimental_convergence_loop:
+ case CONV_LOOP:
Check(TokenDef, "Loop intrinsic must have a convergencectrl token operand.",
{Context.print(&I)});
Check(!SeenFirstConvOp,
@@ -93,14 +92,16 @@ void GenericConvergenceVerifier<ContextT>::visit(const InstructionT &I) {
{Context.print(&I)});
break;
default:
- IsCtrlIntrinsic = false;
break;
}
+ if (ConvOp != CONV_NONE)
+ checkConvergenceTokenProduced(I);
+
if (isConvergent(I))
SeenFirstConvOp = true;
- if (TokenDef || IsCtrlIntrinsic) {
+ if (TokenDef || ConvOp != CONV_NONE) {
Check(isConvergent(I),
"Convergence control token can only be used in a convergent call.",
{Context.print(&I)});
@@ -143,6 +144,10 @@ void GenericConvergenceVerifier<ContextT>::verify(const DominatorTreeT &DT) {
auto checkToken = [&](const InstructionT *Token, const InstructionT *User,
SmallVectorImpl<const InstructionT *> &LiveTokens) {
+ Check(DT.dominates(Token->getParent(), User->getParent()),
+ "Convergence control token must dominate all its uses.",
+ {Context.print(Token), Context.print(User)});
+
Check(llvm::is_contained(LiveTokens, Token),
"Convergence region is not well-nested.",
{Context.print(Token), Context.print(User)});
@@ -161,8 +166,7 @@ void GenericConvergenceVerifier<ContextT>::verify(const DominatorTreeT &DT) {
return;
}
- Check(ContextT::getIntrinsicID(*User) ==
- Intrinsic::experimental_convergence_loop,
+ Check(getConvOp(*User) == CONV_LOOP,
"Convergence token used by an instruction other than "
"llvm.experimental.convergence.loop in a cycle that does "
"not contain the token's definition.",
@@ -199,7 +203,7 @@ void GenericConvergenceVerifier<ContextT>::verify(const DominatorTreeT &DT) {
for (auto &I : *BB) {
if (auto *Token = Tokens.lookup(&I))
checkToken(Token, &I, LiveTokens);
- if (isConvergenceControlIntrinsic(ContextT::getIntrinsicID(I)))
+ if (getConvOp(I) != CONV_NONE)
LiveTokens.push_back(&I);
}
diff --git a/llvm/include/llvm/Support/TargetOpcodes.def b/llvm/include/llvm/Support/TargetOpcodes.def
index 42cb854d9505029..6aded2ceebe13a0 100644
--- a/llvm/include/llvm/Support/TargetOpcodes.def
+++ b/llvm/include/llvm/Support/TargetOpcodes.def
@@ -225,6 +225,11 @@ HANDLE_TARGET_OPCODE(MEMBARRIER)
// using.
HANDLE_TARGET_OPCODE(JUMP_TABLE_DEBUG_INFO)
+HANDLE_TARGET_OPCODE(CONVERGENCECTRL_ENTRY)
+HANDLE_TARGET_OPCODE(CONVERGENCECTRL_ANCHOR)
+HANDLE_TARGET_OPCODE(CONVERGENCECTRL_LOOP)
+HANDLE_TARGET_OPCODE(CONVERGENCECTRL_GLUE)
+
/// The following generic opcodes are not supposed to appear after ISel.
/// This is something we might want to relax, but for now, this is convenient
/// to produce diagnostics.
diff --git a/llvm/include/llvm/Target/Target.td b/llvm/include/llvm/Target/Target.td
index 0d97a47190b1969..0577c58f8da2df8 100644
--- a/llvm/include/llvm/Target/Target.td
+++ b/llvm/include/llvm/Target/Target.td
@@ -1483,6 +1483,25 @@ def JUMP_TABLE_DEBUG_INFO : StandardPseudoInstruction {
let isMeta = true;
}
+let hasSideEffects = false, isMeta = true, isConvergent = true in {
+def CONVERGENCECTRL_ANCHOR : StandardPseudoInstruction {
+ let OutOperandList = (outs unknown:$dst);
+ let InOperandList = (ins);
+}
+def CONVERGENCECTRL_ENTRY : StandardPseudoInstruction {
+ let OutOperandList = (outs unknown:$dst);
+ let InOperandList = (ins);
+}
+def CONVERGENCECTRL_LOOP : StandardPseudoInstruction {
+ let OutOperandList = (outs unknown:$dst);
+ let InOperandList = (ins unknown:$src);
+}
+def CONVERGENCECTRL_GLUE : StandardPseudoInstruction {
+ let OutOperandList = (outs);
+ let InOperandList = (ins unknown:$src);
+}
+}
+
// Generic opcodes used in GlobalISel.
include "llvm/Target/GenericOpcodes.td"
diff --git a/llvm/include/llvm/Target/TargetSelectionDAG.td b/llvm/include/llvm/Target/TargetSelectionDAG.td
index d84c2d30e447262..d7bf8c35ee10721 100644
--- a/llvm/include/llvm/Target/TargetSelectionDAG.td
+++ b/llvm/include/llvm/Target/TargetSelectionDAG.td
@@ -789,6 +789,16 @@ def assertsext : SDNode<"ISD::AssertSext", SDT_assert>;
def assertzext : SDNode<"ISD::AssertZext", SDT_assert>;
def assertalign : SDNode<"ISD::AssertAlign", SDT_assert>;
+def convergencectrl_anchor : SDNode<"ISD::CONVERGENCECTRL_ANCHOR",
+ SDTypeProfile<1, 0, [SDTCisVT<0,untyped>]>>;
+def convergencectrl_entry : SDNode<"ISD::CONVERGENCECTRL_ENTRY",
+ SDTypeProfile<1, 0, [SDTCisVT<0,untyped>]>>;
+def convergencectrl_loop : SDNode<"ISD::CONVERGENCECTRL_LOOP",
+ SDTypeProfile<1, 1,
+ [SDTCisVT<0,untyped>, SDTCisVT<1,untyped>]>>;
+def convergencectrl_glue : SDNode<"ISD::CONVERGENCECTRL_GLUE",
+ SDTypeProfile<0, 1, [SDTCisVT<0, untyped>]>>;
+
//===----------------------------------------------------------------------===//
// Selection DAG Condition Codes
diff --git a/llvm/lib/CodeGen/CMakeLists.txt b/llvm/lib/CodeGen/CMakeLists.txt
index e02c1d6417e077b..2c24de60edd43e7 100644
--- a/llvm/lib/CodeGen/CMakeLists.txt
+++ b/llvm/lib/CodeGen/CMakeLists.txt
@@ -110,6 +110,7 @@ add_llvm_component_library(LLVMCodeGen
MachineBranchProbabilityInfo.cpp
MachineCFGPrinter.cpp
MachineCombiner.cpp
+ MachineConvergenceVerifier.cpp
MachineCopyPropagation.cpp
MachineCSE.cpp
MachineCheckDebugify.cpp
diff --git a/llvm/lib/CodeGen/MachineConvergenceVerifier.cpp b/llvm/lib/CodeGen/MachineConvergenceVerifier.cpp
new file mode 100644
index 000000000000000..8ec85bdf48d45a7
--- /dev/null
+++ b/llvm/lib/CodeGen/MachineConvergenceVerifier.cpp
@@ -0,0 +1,100 @@
+//===- MachineConvergenceVerifier.cpp - Verify convergencectrl ------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CodeGen/MachineConvergenceVerifier.h"
+#include "llvm/CodeGen/GlobalISel/GenericMachineInstrs.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/MachineSSAContext.h"
+#include "llvm/IR/GenericConvergenceVerifierImpl.h"
+
+using namespace llvm;
+
+template <>
+auto GenericConvergenceVerifier<MachineSSAContext>::getConvOp(
+ const MachineInstr &MI) -> ConvOpKind {
+ switch (MI.getOpcode()) {
+ default:
+ return CONV_NONE;
+ case TargetOpcode::CONVERGENCECTRL_ENTRY:
+ return CONV_ENTRY;
+ case TargetOpcode::CONVERGENCECTRL_ANCHOR:
+ return CONV_ANCHOR;
+ case TargetOpcode::CONVERGENCECTRL_LOOP:
+ return CONV_LOOP;
+ }
+}
+
+template <>
+void GenericConvergenceVerifier<
+ MachineSSAContext>::checkConvergenceTokenProduced(const MachineInstr &MI) {
+ Check(!MI.hasImplicitDef(),
+ "Convergence control tokens are defined explicitly.",
+ {Context.print(&MI)});
+ const MachineOperand &Def = MI.getOperand(0);
+ const MachineRegisterInfo &MRI = Context.getFunction()->getRegInfo();
+ Check(MRI.getUniqueVRegDef(Def.getReg()),
+ "Convergence control tokens must have unique definitions.",
+ {Context.print(&MI)});
+}
+
+template <>
+const MachineInstr *
+GenericConvergenceVerifier<MachineSSAContext>::findAndCheckConvergenceTokenUsed(
+ const MachineInstr &MI) {
+ const MachineRegisterInfo &MRI = Context.getFunction()->getRegInfo();
+ const MachineInstr *TokenDef = nullptr;
+
+ for (unsigned I = 0, E = MI.getNumOperands(); I != E; ++I) {
+ const MachineOperand &MO = MI.getOperand(I);
+ if (!MO.isReg() || !MO.isUse())
+ continue;
+ Register OpReg = MO.getReg();
+ if (!OpReg.isVirtual())
+ continue;
+
+ const MachineInstr *Def = MRI.getUniqueVRegDef(OpReg);
+ if (!Def)
+ continue;
+ if (getConvOp(*Def) == CONV_NONE)
+ continue;
+
+ CheckOrNull(
+ MI.isConvergent(),
+ "Convergence control tokens can only be used by convergent operations.",
+ {Context.print(OpReg), Context.print(&MI)});
+
+ CheckOrNull(!TokenDef,
+ "An operation can use at most one convergence control token.",
+ {Context.print(OpReg), Context.print(&MI)});
+
+ TokenDef = Def;
+ }
+
+ if (TokenDef)
+ Tokens[&MI] = TokenDef;
+
+ return TokenDef;
+}
+
+template <>
+bool GenericConvergenceVerifier<MachineSSAContext>::isInsideConvergentFunction(
+ const MachineInstr &MI) {
+ // The class MachineFunction does not have any property to indicate whether it
+ // is convergent. Trivially return true so that the check always passes.
+ return true;
+}
+
+template <>
+bool GenericConvergenceVerifier<MachineSSAContext>::isConvergent(
+ const MachineInstr &MI) {
+ return MI.isConvergent();
+}
+
+template class llvm::GenericConvergenceVerifier<MachineSSAContext>;
diff --git a/llvm/lib/CodeGen/MachineVerifier.cpp b/llvm/lib/CodeGen/MachineVerifier.cpp
index 2632b5b9feac9d6..1d0757c5d7f5f59 100644
--- a/llvm/lib/CodeGen/MachineVerifier.cpp
+++ b/llvm/lib/CodeGen/MachineVerifier.cpp
@@ -39,6 +39,8 @@
#include "llvm/CodeGen/LiveStacks.h"
#include "llvm/CodeGen/LiveVariables.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineConvergenceVerifier.h"
+#include "llvm/CodeGen/MachineDominators.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineFunctionPass.h"
@@ -220,6 +222,11 @@ namespace {
LiveStacks *LiveStks = nullptr;
SlotIndexes *Indexes = nullptr;
+ // This is calculated only when trying to verify convergence control tokens.
+ // Similar to the LLVM IR verifier, we calculate this locally instead of
+ // relying on the pass manager.
+ MachineDomTree DT;
+
void visitMachineFunctionBefore();
void visitMachineBasicBlockBefore(const MachineBasicBlock *MBB);
void visitMachineBundleBefore(const MachineInstr *MI);
@@ -2955,7 +2962,30 @@ void MachineVerifier::checkPHIOps(const MachineBasicBlock &MBB) {
}
}
+static void
+verifyConvergenceControl(const MachineFunction &MF, MachineDomTree &DT,
+ std::function<void(const Twine &Message)> FailureCB) {
+ MachineConvergenceVerifier CV;
+ CV.initialize(&errs(), FailureCB, MF);
+
+ for (const auto &MBB : MF) {
+ CV.visit(MBB);
+ for (const auto &MI : MBB.instrs())
+ CV.visit(MI);
+ }
+
+ if (CV.sawTokens()) {
+ DT.recalculate(const_cast<MachineFunction &>(MF));
+ CV.verify(DT);
+ }
+}
+
void MachineVerifier::visitMachineFunctionAfter() {
+ auto FailureCB = [this](const Twine &Message) {
+ report(Message.str().c_str(), MF);
+ };
+ verifyConvergenceControl(*MF, DT, FailureCB);
+
calcRegsPassed();
for (const MachineBasicBlock &MBB : *MF)
diff --git a/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp b/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
index 4172fbc96d1e5dd..e01cd8cbf925a20 100644
--- a/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/FunctionLoweringInfo.cpp
@@ -395,6 +395,16 @@ Register FunctionLoweringInfo::CreateRegs(const Value *V) {
!TLI->requiresUniformRegister(*MF, V));
}
+Register FunctionLoweringInfo::InitializeRegForValue(const Value *V) {
+ // Tokens live in vregs only when used for convergence control.
+ if (V->getType()->isTokenTy() && !isa<ConvergenceControlInst>(V))
+ return 0;
+ Register &R = ValueMap[V];
+ assert(R == Register() && "Already initialized this value register!");
+ assert(VirtReg2Value.empty());
+ return R = CreateRegs(V);
+}
+
/// GetLiveOutRegInfo - Gets LiveOutInfo for a register, returning NULL if the
/// register is a PHI destination and the PHI's LiveOutInfo is not valid. If
/// the register's LiveOutInfo is for a smaller bit width, it is extended to
diff --git a/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp b/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
index 032cff416cda93e..54409cbf91f1f7a 100644
--- a/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/InstrEmitter.cpp
@@ -285,6 +285,30 @@ Register InstrEmitter::getVR(SDValue Op,
return I->second;
}
+static bool isConvergenceCtrlMachineOp(SDValue Op) {
+ if (Op->isMachineOpcode()) {
+ switch (Op->getMachineOpcode()) {
+ case TargetOpcode::CONVERGENCECTRL_ANCHOR:
+ case TargetOpcode::CONVERGENCECTRL_ENTRY:
+ case TargetOpcode::CONVERGENCECTRL_LOOP:
+ case TargetOpcode::CONVERGENCECTRL_GLUE:
+ return true;
+ }
+ return false;
+ }
+
+ // We can reach here when CopyFromReg is encountered. But rather than making a
+ // special case for that, we just make sure we don't reach here in some
+ // surprising way.
+ switch (Op->getOpcode()) {
+ case ISD::CONVERGENCECTRL_ANCHOR:
+ case ISD::CONVERGENCECTRL_ENTRY:
+ case ISD::CONVERGENCECTRL_LOOP:
+ case ISD::CONVERGENCECTRL_GLUE:
+ llvm_unreachable("Convergence control should have been selected by now.");
+ }
+ return false;
+}
/// AddRegisterOperand - Add the specified register as an operand to the
/// specified machine instr. Insert register copies if the register is
@@ -346,9 +370,12 @@ InstrEmitter::AddRegisterOperand(MachineInstrBuilder &MIB,
// multiple uses.
// Tied operands are never killed, so we need to check that. And that
// means we need to determine the index of the operand.
- bool isKill = Op.hasOneUse() &&
- Op.getNode()->getOpcode() != ISD::CopyFromReg &&
- !IsDebug &&
+ // Don't kill convergence control tokens. Initially they are only used in glue
+ // nodes, and the InstrEmitter later adds implicit uses on the users of the
+ // glue node. This can sometimes make it seem like there is only one use,
+ // which is the glue node itself.
+ bool isKill = Op.hasOneUse() && !isConvergenceCtrlMachineOp(Op) &&
+ Op.getNode()->getOpcode() != ISD::CopyFromReg && !IsDebug &&
!(IsClone || IsCloned);
if (isKill) {
unsigned Idx = MIB->getNumOperands();
@@ -1191,6 +1218,17 @@ EmitMachineNode(SDNode *Node, bool IsClone, bool IsCloned,
}
}
+ if (SDNode *GluedNode = Node->getGluedNode()) {
+ // FIXME: Possibly iterate over multiple glue nodes?
+ if (GluedNode->getOpcode() ==
+ ~(unsigned)TargetOpcode::CONVERGENCECTRL_GLUE) {
+ Register VReg = getVR(GluedNode->getOperand(0), VRBaseMap);
+ MachineOperand MO = MachineOperand::CreateReg(VReg, /*isDef=*/false,
+ /*isImp=*/true);
+ MIB->addOperand(MO);
+ }
+ }
+
// Run post-isel target hook to adjust this instruction if needed.
if (II.hasPostISelHook())
TLI->AdjustInstrPostInstrSelection(*MIB, Node);
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index d6ab0cf6df8763c..22e57d0d99e9b1b 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -5226,6 +5226,17 @@ void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,
// Create the node.
SDValue Result;
+
+ if (auto Bundle = I.getOperandBundle(LLVMContext::OB_convergencectrl)) {
+ auto *Token = Bundle->Inputs[0].get();
+ SDValue ConvControlToken = getValue(Token);
+ assert(Ops.back().getValueType() != MVT::Glue &&
+ "Did not expected another glue node here.");
+ ConvControlToken =
+ DAG.getNode(ISD::CONVERGENCECTRL_GLUE, {}, MVT::Glue, ConvControlToken);
+ Ops.push_back(ConvControlToken);
+ }
+
// In some cases, custom collection of operands from CallInst I may be needed.
TLI.CollectTargetIntrinsicOperands(I, Ops, DAG);
if (IsTgtIntrinsic) {
@@ -6226,6 +6237,27 @@ bool SelectionDAGBuilder::visitEntryValueDbgValue(
return true;
}
+/// Lower the call to the specified intrinsic function.
+void SelectionDAGBuilder::visitConvergenceControl(const CallInst &I,
+ unsigned Intrinsic) {
+ SDLoc sdl = getCurSDLoc();
+ switch (Intrinsic) {
+ case Intrinsic::experimental_convergence_anchor:
+ setValue(&I, DAG.getNode(ISD::CONVERGENCECTRL_ANCHOR, sdl, MVT::Untyped));
+ break;
+ case Intrinsic::experimental_convergence_entry:
+ setValue(&I, DAG.getNode(ISD::CONVERGENCECTRL_ENTRY, sdl, MVT::Untyped));
+ break;
+ case Intrinsic::experimental_convergence_loop: {
+ auto Bundle = I.getOperandBundle(LLVMContext::OB_convergencectrl);
+ auto *Token = Bundle->Inputs[0].get();
+ setValue(&I, DAG.getNode(ISD::CONVERGENCECTRL_LOOP, sdl, MVT::Untyped,
+ getValue(Token)));
+ break;
+ }
+ }
+}
+
/// Lower the call to the specified intrinsic function.
void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
unsigned Intrinsic) {
@@ -7885,6 +7917,10 @@ void SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I,
case Intrinsic::experimental_vector_deinterleave2:
visitVectorDeinterleave(I);
return;
+ case Intrinsic::experimental_convergence_anchor:
+ case Intrinsic::experimental_convergence_entry:
+ case Intrinsic::experimental_convergence_loop:
+ visitConvergenceControl(I, Intrinsic);
}
}
@@ -8561,6 +8597,14 @@ void SelectionDAGBuilder::LowerCallTo(const CallBase &CB, SDValue Callee,
}
}
+ SDValue ConvControlToken;
+ if (auto Bundle = CB.getOperandBundle(LLVMContext::OB_convergencectrl)) {
+ auto *Token = Bundle->Inputs[0].get();
+ ConvControlToken = getValue(Token);
+ } else {
+ ConvControlToken = DAG.getUNDEF(MVT::Untyped);
+ }
+
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(getCurSDLoc())
.setChain(getRoot())
@@ -8569,7 +8613,8 @@ void SelectionDAGBuilder::LowerCallTo(const CallBase &CB, SDValue Callee,
.setConvergent(CB.isConvergent())
.setIsPreallocated(
CB.countOperandBundlesOfType(LLVMContext::OB_preallocated) != 0)
- .setCFIType(CFIType);
+ .setCFIType(CFIType)
+ .setConvergenceControlToken(ConvControlToken);
std::pair<SDValue, SDValue> Result = lowerInvokable(CLI, EHPadBB);
if (Result.first.getNode()) {
@@ -9121,7 +9166,8 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
assert(!I.hasOperandBundlesOtherThan(
{LLVMContext::OB_deopt, LLVMContext::OB_funclet,
LLVMContext::OB_cfguardtarget, LLVMContext::OB_preallocated,
- LLVMContext::OB_clang_arc_attachedcall, LLVMContext::OB_kcfi}) &&
+ LLVMContext::OB_clang_arc_attachedcall, LLVMContext::OB_kcfi,
+ LLVMContext::OB_convergencectrl}) &&
"Cannot lower calls with arbitrary operand bundles!");
SDValue Callee = getValue(I.getCalledOperand());
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
index 2084de473b8062e..dcf46e0563ff9de 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h
@@ -623,6 +623,7 @@ class SelectionDAGBuilder {
void visitIntrinsicCall(const CallInst &I, unsigned Intrinsic);
void visitTargetIntrinsic(const CallInst &I, unsigned Intrinsic);
void visitConstrainedFPIntrinsic(const ConstrainedFPIntrinsic &FPI);
+ void visitConvergenceControl(const CallInst &I, unsigned Intrinsic);
void visitVPLoad(const VPIntrinsic &VPIntrin, EVT VT,
const SmallVectorImpl<SDValue> &OpValues);
void visitVPStore(const VPIntrinsic &VPIntrin,
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index 18ca17e53dac380..fa71adc8da3f381 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -165,6 +165,9 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
if (cast<ConstantSDNode>(this)->isOpaque())
return "OpaqueTargetConstant";
return "TargetConstant";
+
+ // clang-format off
+
case ISD::TargetConstantFP: return "TargetConstantFP";
case ISD::TargetGlobalAddress: return "TargetGlobalAddress";
case ISD::TargetGlobalTLSAddress: return "TargetGlobalTLSAddress";
@@ -449,6 +452,11 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::SET_FPMODE: return "set_fpmode";
case ISD::RESET_FPMODE: return "reset_fpmode";
+ // Convergence control instructions
+ case ISD::CONVERGENCECTRL_ANCHOR: return "convergencectrl_anchor";
+ case ISD::CONVERGENCECTRL_ENTRY: return "convergencectrl_entry";
+ case ISD::CONVERGENCECTRL_LOOP: return "convergencectrl_loop";
+
// Bit manipulation
case ISD::ABS: return "abs";
case ISD::BITREVERSE: return "bitreverse";
@@ -464,6 +472,8 @@ std::string SDNode::getOperationName(const SelectionDAG *G) const {
case ISD::INIT_TRAMPOLINE: return "init_trampoline";
case ISD::ADJUST_TRAMPOLINE: return "adjust_trampoline";
+ // clang-format on
+
case ISD::CONDCODE:
switch (cast<CondCodeSDNode>(this)->get()) {
default: llvm_unreachable("Unknown setcc condition!");
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index 9b5ab4267b80e93..1c14e4da8e9d382 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -2370,6 +2370,21 @@ void SelectionDAGISel::Select_MEMBARRIER(SDNode *N) {
N->getOperand(0));
}
+void SelectionDAGISel::Select_CONVERGENCECTRL_ANCHOR(SDNode *N) {
+ CurDAG->SelectNodeTo(N, TargetOpcode::CONVERGENCECTRL_ANCHOR,
+ N->getValueType(0));
+}
+
+void SelectionDAGISel::Select_CONVERGENCECTRL_ENTRY(SDNode *N) {
+ CurDAG->SelectNodeTo(N, TargetOpcode::CONVERGENCECTRL_ENTRY,
+ N->getValueType(0));
+}
+
+void SelectionDAGISel::Select_CONVERGENCECTRL_LOOP(SDNode *N) {
+ CurDAG->SelectNodeTo(N, TargetOpcode::CONVERGENCECTRL_LOOP,
+ N->getValueType(0), N->getOperand(0));
+}
+
void SelectionDAGISel::pushStackMapLiveVariable(SmallVectorImpl<SDValue> &Ops,
SDValue OpVal, SDLoc DL) {
SDNode *OpNode = OpVal.getNode();
@@ -3117,6 +3132,15 @@ void SelectionDAGISel::SelectCodeCommon(SDNode *NodeToMatch,
case ISD::JUMP_TABLE_DEBUG_INFO:
Select_JUMP_TABLE_DEBUG_INFO(NodeToMatch);
return;
+ case ISD::CONVERGENCECTRL_ANCHOR:
+ Select_CONVERGENCECTRL_ANCHOR(NodeToMatch);
+ return;
+ case ISD::CONVERGENCECTRL_ENTRY:
+ Select_CONVERGENCECTRL_ENTRY(NodeToMatch);
+ return;
+ case ISD::CONVERGENCECTRL_LOOP:
+ Select_CONVERGENCECTRL_LOOP(NodeToMatch);
+ return;
}
assert(!NodeToMatch->isMachineOpcode() && "Node already selected!");
diff --git a/llvm/lib/CodeGen/ValueTypes.cpp b/llvm/lib/CodeGen/ValueTypes.cpp
index 5dcabdb6a9e06d6..58db686ec7d578f 100644
--- a/llvm/lib/CodeGen/ValueTypes.cpp
+++ b/llvm/lib/CodeGen/ValueTypes.cpp
@@ -629,6 +629,8 @@ EVT EVT::getEVT(Type *Ty, bool HandleUnknown){
switch (Ty->getTypeID()) {
default:
return MVT::getVT(Ty, HandleUnknown);
+ case Type::TokenTyID:
+ return MVT::Untyped;
case Type::IntegerTyID:
return getIntegerVT(Ty->getContext(), cast<IntegerType>(Ty)->getBitWidth());
case Type::FixedVectorTyID:
diff --git a/llvm/lib/IR/ConvergenceVerifier.cpp b/llvm/lib/IR/ConvergenceVerifier.cpp
index 336c202b6f94cee..e73aeaade5f7149 100644
--- a/llvm/lib/IR/ConvergenceVerifier.cpp
+++ b/llvm/lib/IR/ConvergenceVerifier.cpp
@@ -14,6 +14,30 @@
using namespace llvm;
+template <>
+auto GenericConvergenceVerifier<SSAContext>::getConvOp(const Instruction &I)
+ -> ConvOpKind {
+ const auto *CB = dyn_cast<CallBase>(&I);
+ if (!CB)
+ return CONV_NONE;
+ switch (CB->getIntrinsicID()) {
+ default:
+ return CONV_NONE;
+ case Intrinsic::experimental_convergence_anchor:
+ return CONV_ANCHOR;
+ case Intrinsic::experimental_convergence_entry:
+ return CONV_ENTRY;
+ case Intrinsic::experimental_convergence_loop:
+ return CONV_LOOP;
+ }
+}
+
+template <>
+void GenericConvergenceVerifier<SSAContext>::checkConvergenceTokenProduced(
+ const Instruction &I) {
+ return;
+}
+
template <>
const Instruction *
GenericConvergenceVerifier<SSAContext>::findAndCheckConvergenceTokenUsed(
@@ -38,11 +62,10 @@ GenericConvergenceVerifier<SSAContext>::findAndCheckConvergenceTokenUsed(
auto *Token = Bundle->Inputs[0].get();
auto *Def = dyn_cast<Instruction>(Token);
- CheckOrNull(
- Def && isConvergenceControlIntrinsic(SSAContext::getIntrinsicID(*Def)),
- "Convergence control tokens can only be produced by calls to the "
- "convergence control intrinsics.",
- {Context.print(Token), Context.print(&I)});
+ CheckOrNull(Def && getConvOp(*Def) != CONV_NONE,
+ "Convergence control tokens can only be produced by calls to the "
+ "convergence control intrinsics.",
+ {Context.print(Token), Context.print(&I)});
if (Def)
Tokens[&I] = Def;
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
index 2e5b02fbe856601..d1af74e6e0d352d 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
@@ -2688,7 +2688,18 @@ void AMDGPUDAGToDAGISel::SelectINTRINSIC_W_CHAIN(SDNode *N) {
void AMDGPUDAGToDAGISel::SelectINTRINSIC_WO_CHAIN(SDNode *N) {
unsigned IntrID = N->getConstantOperandVal(0);
- unsigned Opcode;
+ unsigned Opcode = AMDGPU::INSTRUCTION_LIST_END;
+ SDNode *ConvGlueNode = N->getGluedNode();
+ if (ConvGlueNode) {
+ // FIXME: Possibly iterate over multiple glue nodes?
+ assert(ConvGlueNode->getOpcode() == ISD::CONVERGENCECTRL_GLUE);
+ ConvGlueNode = ConvGlueNode->getOperand(0).getNode();
+ ConvGlueNode =
+ CurDAG->getMachineNode(TargetOpcode::CONVERGENCECTRL_GLUE, {},
+ MVT::Glue, SDValue(ConvGlueNode, 0));
+ } else {
+ ConvGlueNode = nullptr;
+ }
switch (IntrID) {
case Intrinsic::amdgcn_wqm:
Opcode = AMDGPU::WQM;
@@ -2720,11 +2731,19 @@ void AMDGPUDAGToDAGISel::SelectINTRINSIC_WO_CHAIN(SDNode *N) {
break;
default:
SelectCode(N);
- return;
+ break;
}
- SDValue Src = N->getOperand(1);
- CurDAG->SelectNodeTo(N, Opcode, N->getVTList(), {Src});
+ if (Opcode != AMDGPU::INSTRUCTION_LIST_END) {
+ SDValue Src = N->getOperand(1);
+ CurDAG->SelectNodeTo(N, Opcode, N->getVTList(), {Src});
+ }
+
+ if (ConvGlueNode) {
+ SmallVector<SDValue, 4> NewOps(N->op_begin(), N->op_end());
+ NewOps.push_back(SDValue(ConvGlueNode, 0));
+ CurDAG->MorphNodeTo(N, N->getOpcode(), N->getVTList(), NewOps);
+ }
}
void AMDGPUDAGToDAGISel::SelectINTRINSIC_VOID(SDNode *N) {
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 76e843455bab75c..7467c25a84d610f 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -245,6 +245,13 @@ static cl::opt<bool, true> LateCFGStructurize(
cl::location(AMDGPUTargetMachine::EnableLateStructurizeCFG),
cl::Hidden);
+// Disable structurizer-based control-flow lowering in order to test convergence
+// control tokens. This should eventually be replaced by the wave-transform.
+static cl::opt<bool, true> DisableStructurizer(
+ "amdgpu-disable-structurizer",
+ cl::desc("Disable structurizer for experiments; produces unusable code"),
+ cl::location(AMDGPUTargetMachine::DisableStructurizer), cl::ReallyHidden);
+
// Enable lib calls simplifications
static cl::opt<bool> EnableLibCallSimplify(
"amdgpu-simplify-libcall",
@@ -591,6 +598,7 @@ AMDGPUTargetMachine::AMDGPUTargetMachine(const Target &T, const Triple &TT,
bool AMDGPUTargetMachine::EnableLateStructurizeCFG = false;
bool AMDGPUTargetMachine::EnableFunctionCalls = false;
bool AMDGPUTargetMachine::EnableLowerModuleLDS = true;
+bool AMDGPUTargetMachine::DisableStructurizer = false;
AMDGPUTargetMachine::~AMDGPUTargetMachine() = default;
@@ -1186,7 +1194,7 @@ bool GCNPassConfig::addPreISel() {
// Merge divergent exit nodes. StructurizeCFG won't recognize the multi-exit
// regions formed by them.
addPass(&AMDGPUUnifyDivergentExitNodesID);
- if (!LateCFGStructurize) {
+ if (!LateCFGStructurize && !DisableStructurizer) {
if (EnableStructurizerWorkarounds) {
addPass(createFixIrreduciblePass());
addPass(createUnifyLoopExitsPass());
@@ -1194,7 +1202,7 @@ bool GCNPassConfig::addPreISel() {
addPass(createStructurizeCFGPass(false)); // true -> SkipUniformRegions
}
addPass(createAMDGPUAnnotateUniformValues());
- if (!LateCFGStructurize) {
+ if (!LateCFGStructurize && !DisableStructurizer) {
addPass(createSIAnnotateControlFlowPass());
// TODO: Move this right after structurizeCFG to avoid extra divergence
// analysis. This depends on stopping SIAnnotateControlFlow from making
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
index ce2dd2947daf650..30ab388c7d52e21 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.h
@@ -37,6 +37,7 @@ class AMDGPUTargetMachine : public LLVMTargetMachine {
static bool EnableLateStructurizeCFG;
static bool EnableFunctionCalls;
static bool EnableLowerModuleLDS;
+ static bool DisableStructurizer;
AMDGPUTargetMachine(const Target &T, const Triple &TT, StringRef CPU,
StringRef FS, const TargetOptions &Options,
diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index f30dc06ed408db6..23fe0a52a74bfc1 100644
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -98,6 +98,7 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM,
addRegisterClass(MVT::f64, V64RegClass);
addRegisterClass(MVT::v2f32, V64RegClass);
+ addRegisterClass(MVT::Untyped, V64RegClass);
addRegisterClass(MVT::v3i32, &AMDGPU::SGPR_96RegClass);
addRegisterClass(MVT::v3f32, TRI->getVGPRClassForBitWidth(96));
@@ -3828,6 +3829,9 @@ SDValue SITargetLowering::LowerCall(CallLoweringInfo &CLI,
Ops.push_back(DAG.getTargetConstant(0, DL, MVT::i64));
}
+ if (!IsTailCall)
+ Ops.push_back(CLI.ConvergenceControlToken);
+
if (IsTailCall) {
// Each tail call may have to adjust the stack by a
diff erent amount, so
// this information must travel along with the operation for eventual
@@ -5152,8 +5156,24 @@ MachineBasicBlock *SITargetLowering::EmitInstrWithCustomInserter(
MachineInstrBuilder MIB;
MIB = BuildMI(*BB, MI, DL, TII->get(AMDGPU::SI_CALL), ReturnAddrReg);
- for (const MachineOperand &MO : MI.operands())
- MIB.add(MO);
+ for (unsigned I = 0, E = MI.getNumOperands(); I != E; ++I) {
+ MachineOperand &MO = MI.getOperand(I);
+ if (I != 2) {
+ MIB.add(MO);
+ continue;
+ }
+ }
+
+ MachineOperand &MO = MI.getOperand(2);
+ MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+ // The token operand is always a register, whose definition is IMPLICIT_DEF
+ // iff there was no token on the call.
+ if (MachineInstr *Def = MRI.getVRegDef(MO.getReg())) {
+ if (Def->getOpcode() != TargetOpcode::IMPLICIT_DEF) {
+ MO.setImplicit();
+ MIB.add(MO);
+ }
+ }
MIB.cloneMemRefs(MI);
MI.eraseFromParent();
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 565af36bc523e4e..33c93cdf20c43b8 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -618,8 +618,8 @@ def SI_RETURN : SPseudoInstSI <
// This version is only needed so we can fill in the output register
// in the custom inserter.
def SI_CALL_ISEL : SPseudoInstSI <
- (outs), (ins SSrc_b64:$src0, unknown:$callee),
- [(AMDGPUcall i64:$src0, tglobaladdr:$callee)]> {
+ (outs), (ins SSrc_b64:$src0, unknown:$callee, unknown:$token),
+ [(AMDGPUcall i64:$src0, tglobaladdr:$callee, untyped:$token)]> {
let Size = 4;
let isCall = 1;
let SchedRW = [WriteBranch];
@@ -629,8 +629,8 @@ def SI_CALL_ISEL : SPseudoInstSI <
}
def : GCNPat<
- (AMDGPUcall i64:$src0, (i64 0)),
- (SI_CALL_ISEL $src0, (i64 0))
+ (AMDGPUcall i64:$src0, (i64 0), untyped:$token),
+ (SI_CALL_ISEL $src0, (i64 0), untyped:$token)
>;
// Wrapper around s_swappc_b64 with extra $callee parameter to track
diff --git a/llvm/test/CodeGen/AMDGPU/convergence-tokens.ll b/llvm/test/CodeGen/AMDGPU/convergence-tokens.ll
new file mode 100644
index 000000000000000..2ed6d7fd0f598d2
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/convergence-tokens.ll
@@ -0,0 +1,83 @@
+; RUN: llc --amdgpu-disable-structurizer -stop-after=amdgpu-isel -mtriple=amdgcn-- -mcpu=gfx900 -verify-machineinstrs -o - %s | FileCheck --check-prefixes=CHECK,ISEL %s
+; RUN: llc --amdgpu-disable-structurizer -stop-after=dead-mi-elimination -mtriple=amdgcn-- -mcpu=gfx900 -verify-machineinstrs -o - %s | FileCheck --check-prefixes=CHECK,DEADMI %s
+
+; CHECK-LABEL: name: basic_call
+; CHECK: [[TOKEN:%[0-9]+]]:sreg_64 = CONVERGENCECTRL_ENTRY
+; ISEL: {{.*}} SI_CALL_ISEL {{.*}}, @foo, [[TOKEN]], csr_amdgpu, {{.*}}
+; DEADMI: {{.*}} SI_CALL {{.*}}, @foo, csr_amdgpu, {{.*}}, implicit [[TOKEN]]
+define i32 @basic_call(i32 %src) #0 {
+ %t = call token @llvm.experimental.convergence.entry()
+ %r = call i32 @foo(i32 %src) [ "convergencectrl"(token %t) ]
+ ret i32 %r
+}
+
+; CHECK-LABEL: name: basic_intrinsic
+; CHECK: [[TOKEN:%[0-9]+]]:sreg_64 = CONVERGENCECTRL_ANCHOR
+; ISEL: CONVERGENCECTRL_GLUE [[TOKEN]]
+; DEADMI-NOT: CONVERGENCECTRL_GLUE
+; CHECK: {{.*}} = V_READFIRSTLANE_B32 {{.*}}, implicit [[TOKEN]]
+define i32 @basic_intrinsic(i32 %src) #0 {
+ %t = call token @llvm.experimental.convergence.anchor()
+ %r = call i32 @llvm.amdgcn.readfirstlane(i32 %src) [ "convergencectrl"(token %t) ]
+ ret i32 %r
+}
+
+; There's nothing to check here. The test is just meant to catch any crashes
+; when a convergent call has no token.
+define i32 @uncontrolled_call(i32 %src) #0 {
+ %r = call i32 @foo(i32 %src)
+ ret i32 %r
+}
+
+; CHECK-LABEL: name: basic_branch
+; CHECK: bb.0.entry:
+; CHECK: [[TOKEN:%[0-9]+]]:sreg_64 = CONVERGENCECTRL_ANCHOR
+; CHECK: bb.1.then:
+; ISEL: CONVERGENCECTRL_GLUE [[TOKEN]]
+; DEADMI-NOT: CONVERGENCECTRL_GLUE
+; CHECK: {{.*}} = V_READFIRSTLANE_B32 {{.*}}, implicit [[TOKEN]]
+define i32 @basic_branch(i32 %src, i1 %cond) #0 {
+entry:
+ %t = call token @llvm.experimental.convergence.anchor()
+ %x = add i32 %src, 1
+ br i1 %cond, label %then, label %else
+
+then:
+ %r = call i32 @llvm.amdgcn.readfirstlane(i32 %x) [ "convergencectrl"(token %t) ]
+ br label %else
+
+else:
+ %p = phi i32 [%r, %then], [%x, %entry]
+ ret i32 %p
+}
+
+; CHECK-LABEL: name: basic_loop
+; CHECK: [[TOKEN:%[0-9]+]]:sreg_64 = CONVERGENCECTRL_ANCHOR
+; CHECK: bb.1.loop:
+; CHECK: [[LOOP:%[0-9]+]]:sreg_64 = CONVERGENCECTRL_LOOP [[TOKEN]]
+; ISEL: CONVERGENCECTRL_GLUE [[LOOP]]
+; DEADMI-NOT: CONVERGENCECTRL_GLUE
+; CHECK: {{.*}} = V_READFIRSTLANE_B32 {{.*}}, implicit [[LOOP]]
+define i32 @basic_loop(i32 %src, i1 %cond) #0 {
+ %t1 = call token @llvm.experimental.convergence.anchor()
+ br label %loop
+
+loop:
+ %t2 = call token @llvm.experimental.convergence.loop() [ "convergencectrl"(token %t1) ]
+ %r = call i32 @llvm.amdgcn.readfirstlane(i32 %src) [ "convergencectrl"(token %t2) ]
+ br i1 %cond, label %loop, label %end
+
+end:
+ ret i32 %r
+}
+
+declare i32 @foo(i32 %x) #0
+
+declare i32 @llvm.amdgcn.readfirstlane(i32) #0
+
+declare token @llvm.experimental.convergence.entry()
+declare token @llvm.experimental.convergence.anchor()
+declare token @llvm.experimental.convergence.loop()
+
+attributes #0 = { nounwind readnone convergent }
+attributes #1 = { nounwind }
diff --git a/llvm/test/CodeGen/AMDGPU/isel-amdgpu-cs-chain-cc.ll b/llvm/test/CodeGen/AMDGPU/isel-amdgpu-cs-chain-cc.ll
index ab160ffc10ed073..e015095a4884a52 100644
--- a/llvm/test/CodeGen/AMDGPU/isel-amdgpu-cs-chain-cc.ll
+++ b/llvm/test/CodeGen/AMDGPU/isel-amdgpu-cs-chain-cc.ll
@@ -92,6 +92,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc(<4 x i32> inreg %a, <4 x i32> %b
; DAGISEL-GFX11-NEXT: $vgpr5 = COPY [[COPY2]]
; DAGISEL-GFX11-NEXT: $vgpr6 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr7 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -121,6 +122,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc(<4 x i32> inreg %a, <4 x i32> %b
; DAGISEL-GFX10-NEXT: $vgpr5 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr6 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr7 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -232,6 +234,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_ptr(ptr inreg %a, ptr %b, ptr ad
; DAGISEL-GFX11-NEXT: $vgpr9 = COPY [[COPY2]]
; DAGISEL-GFX11-NEXT: $vgpr10 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr11 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -269,6 +272,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_ptr(ptr inreg %a, ptr %b, ptr ad
; DAGISEL-GFX10-NEXT: $vgpr9 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr10 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr11 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -400,6 +404,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_struct( {ptr, i32, <4 x i32>} in
; DAGISEL-GFX11-NEXT: $vgpr11 = COPY [[COPY2]]
; DAGISEL-GFX11-NEXT: $vgpr12 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr13 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF2:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -449,6 +454,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_struct( {ptr, i32, <4 x i32>} in
; DAGISEL-GFX10-NEXT: $vgpr11 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr12 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr13 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF2:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -500,6 +506,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_float(float inreg %a, float %b)
; DAGISEL-GFX11-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM killed [[SI_PC_ADD_REL_OFFSET]], 0, 0 :: (dereferenceable invariant load (s64) from got, addrspace 4)
; DAGISEL-GFX11-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -517,6 +524,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_float(float inreg %a, float %b)
; DAGISEL-GFX10-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -568,6 +576,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_half(half inreg %a, half %b) {
; DAGISEL-GFX11-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM killed [[SI_PC_ADD_REL_OFFSET]], 0, 0 :: (dereferenceable invariant load (s64) from got, addrspace 4)
; DAGISEL-GFX11-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -585,6 +594,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_half(half inreg %a, half %b) {
; DAGISEL-GFX10-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -636,6 +646,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_bfloat(bfloat inreg %a, bfloat %
; DAGISEL-GFX11-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM killed [[SI_PC_ADD_REL_OFFSET]], 0, 0 :: (dereferenceable invariant load (s64) from got, addrspace 4)
; DAGISEL-GFX11-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -653,6 +664,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_bfloat(bfloat inreg %a, bfloat %
; DAGISEL-GFX10-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -704,6 +716,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_i16(i16 inreg %a, i16 %b) {
; DAGISEL-GFX11-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM killed [[SI_PC_ADD_REL_OFFSET]], 0, 0 :: (dereferenceable invariant load (s64) from got, addrspace 4)
; DAGISEL-GFX11-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -721,6 +734,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_i16(i16 inreg %a, i16 %b) {
; DAGISEL-GFX10-NEXT: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr0 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr1 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -856,6 +870,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_v16i16(<16 x i16> inreg %a, <16
; DAGISEL-GFX11-NEXT: $vgpr13 = COPY [[COPY2]]
; DAGISEL-GFX11-NEXT: $vgpr14 = COPY [[COPY1]]
; DAGISEL-GFX11-NEXT: $vgpr15 = COPY [[COPY]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13, implicit $vgpr14, implicit $vgpr15
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -901,6 +916,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_cc_v16i16(<16 x i16> inreg %a, <16
; DAGISEL-GFX10-NEXT: $vgpr13 = COPY [[COPY2]]
; DAGISEL-GFX10-NEXT: $vgpr14 = COPY [[COPY1]]
; DAGISEL-GFX10-NEXT: $vgpr15 = COPY [[COPY]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13, implicit $vgpr14, implicit $vgpr15
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
@@ -2464,6 +2480,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_many_regs(<36 x i32> inreg %a, <128
; DAGISEL-GFX11-NEXT: $vgpr29 = COPY [[COPY134]]
; DAGISEL-GFX11-NEXT: $vgpr30 = COPY [[COPY133]]
; DAGISEL-GFX11-NEXT: $vgpr31 = COPY [[COPY132]]
+ ; DAGISEL-GFX11-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX11-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13, implicit $vgpr14, implicit $vgpr15, implicit $vgpr16, implicit $vgpr17, implicit $vgpr18, implicit $vgpr19, implicit $vgpr20, implicit $vgpr21, implicit $vgpr22, implicit $vgpr23, implicit $vgpr24, implicit $vgpr25, implicit $vgpr26, implicit $vgpr27, implicit $vgpr28, implicit $vgpr29, implicit $vgpr30, implicit $vgpr31
; DAGISEL-GFX11-NEXT: ADJCALLSTACKDOWN 0, 528, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX11-NEXT: S_ENDPGM 0
@@ -2810,6 +2827,7 @@ define amdgpu_cs_chain void @amdgpu_cs_chain_many_regs(<36 x i32> inreg %a, <128
; DAGISEL-GFX10-NEXT: $vgpr29 = COPY [[COPY134]]
; DAGISEL-GFX10-NEXT: $vgpr30 = COPY [[COPY133]]
; DAGISEL-GFX10-NEXT: $vgpr31 = COPY [[COPY132]]
+ ; DAGISEL-GFX10-NEXT: [[DEF:%[0-9]+]]:sreg_64 = IMPLICIT_DEF
; DAGISEL-GFX10-NEXT: $sgpr30_sgpr31 = SI_CALL killed [[S_LOAD_DWORDX2_IMM]], @use, csr_amdgpu_si_gfx, implicit $sgpr0_sgpr1_sgpr2_sgpr3, implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3, implicit $vgpr4, implicit $vgpr5, implicit $vgpr6, implicit $vgpr7, implicit $vgpr8, implicit $vgpr9, implicit $vgpr10, implicit $vgpr11, implicit $vgpr12, implicit $vgpr13, implicit $vgpr14, implicit $vgpr15, implicit $vgpr16, implicit $vgpr17, implicit $vgpr18, implicit $vgpr19, implicit $vgpr20, implicit $vgpr21, implicit $vgpr22, implicit $vgpr23, implicit $vgpr24, implicit $vgpr25, implicit $vgpr26, implicit $vgpr27, implicit $vgpr28, implicit $vgpr29, implicit $vgpr30, implicit $vgpr31
; DAGISEL-GFX10-NEXT: ADJCALLSTACKDOWN 0, 528, implicit-def dead $scc, implicit-def $sgpr32, implicit $sgpr32
; DAGISEL-GFX10-NEXT: S_ENDPGM 0
diff --git a/llvm/test/CodeGen/AMDGPU/kernel-vgpr-spill-mubuf-with-voffset.ll b/llvm/test/CodeGen/AMDGPU/kernel-vgpr-spill-mubuf-with-voffset.ll
index 6e905542ce53c17..8b6b48bcdba0d1a 100644
--- a/llvm/test/CodeGen/AMDGPU/kernel-vgpr-spill-mubuf-with-voffset.ll
+++ b/llvm/test/CodeGen/AMDGPU/kernel-vgpr-spill-mubuf-with-voffset.ll
@@ -60,6 +60,7 @@ define amdgpu_kernel void @test_kernel(i32 %val) #0 {
; CHECK-NEXT: ; implicit-def: $sgpr15
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: s_or_saveexec_b64 s[34:35], -1
diff --git a/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll b/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll
index f70441e87a74b5d..5f507d482eeb6f2 100644
--- a/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll
+++ b/llvm/test/CodeGen/AMDGPU/need-fp-from-vgpr-spills.ll
@@ -27,7 +27,7 @@ define internal fastcc void @csr_vgpr_spill_fp_callee() #0 {
; CHECK-LABEL: csr_vgpr_spill_fp_callee:
; CHECK: ; %bb.0: ; %bb
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_mov_b32 s18, s33
+; CHECK-NEXT: s_mov_b32 s24, s33
; CHECK-NEXT: s_mov_b32 s33, s32
; CHECK-NEXT: s_xor_saveexec_b64 s[16:17], -1
; CHECK-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
@@ -43,6 +43,7 @@ define internal fastcc void @csr_vgpr_spill_fp_callee() #0 {
; CHECK-NEXT: s_mov_b64 s[20:21], s[0:1]
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: ;;#ASMSTART
; CHECK-NEXT: ; clobber csr v40
@@ -54,7 +55,7 @@ define internal fastcc void @csr_vgpr_spill_fp_callee() #0 {
; CHECK-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
; CHECK-NEXT: s_mov_b64 exec, s[4:5]
; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00
-; CHECK-NEXT: s_mov_b32 s33, s18
+; CHECK-NEXT: s_mov_b32 s33, s24
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: s_setpc_b64 s[30:31]
bb:
@@ -87,6 +88,7 @@ define amdgpu_kernel void @kernel_call() {
; CHECK-NEXT: ; implicit-def: $sgpr15
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: s_endpgm
bb:
@@ -146,6 +148,7 @@ define amdgpu_kernel void @kernel_tailcall() {
; CHECK-NEXT: ; implicit-def: $sgpr15
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: s_endpgm
bb:
@@ -170,7 +173,7 @@ define hidden i32 @caller_save_vgpr_spill_fp_tail_call() #0 {
; CHECK-LABEL: caller_save_vgpr_spill_fp_tail_call:
; CHECK: ; %bb.0: ; %entry
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_mov_b32 s18, s33
+; CHECK-NEXT: s_mov_b32 s24, s33
; CHECK-NEXT: s_mov_b32 s33, s32
; CHECK-NEXT: s_xor_saveexec_b64 s[16:17], -1
; CHECK-NEXT: buffer_store_dword v1, off, s[0:3], s33 ; 4-byte Folded Spill
@@ -185,6 +188,7 @@ define hidden i32 @caller_save_vgpr_spill_fp_tail_call() #0 {
; CHECK-NEXT: s_mov_b64 s[20:21], s[0:1]
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: v_readlane_b32 s31, v1, 1
; CHECK-NEXT: v_readlane_b32 s30, v1, 0
@@ -192,7 +196,7 @@ define hidden i32 @caller_save_vgpr_spill_fp_tail_call() #0 {
; CHECK-NEXT: buffer_load_dword v1, off, s[0:3], s33 ; 4-byte Folded Reload
; CHECK-NEXT: s_mov_b64 exec, s[4:5]
; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00
-; CHECK-NEXT: s_mov_b32 s33, s18
+; CHECK-NEXT: s_mov_b32 s33, s24
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: s_setpc_b64 s[30:31]
entry:
@@ -204,7 +208,7 @@ define hidden i32 @caller_save_vgpr_spill_fp() #0 {
; CHECK-LABEL: caller_save_vgpr_spill_fp:
; CHECK: ; %bb.0: ; %entry
; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; CHECK-NEXT: s_mov_b32 s19, s33
+; CHECK-NEXT: s_mov_b32 s25, s33
; CHECK-NEXT: s_mov_b32 s33, s32
; CHECK-NEXT: s_xor_saveexec_b64 s[16:17], -1
; CHECK-NEXT: buffer_store_dword v2, off, s[0:3], s33 ; 4-byte Folded Spill
@@ -219,6 +223,7 @@ define hidden i32 @caller_save_vgpr_spill_fp() #0 {
; CHECK-NEXT: s_mov_b64 s[20:21], s[0:1]
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: v_readlane_b32 s31, v2, 1
; CHECK-NEXT: v_readlane_b32 s30, v2, 0
@@ -226,7 +231,7 @@ define hidden i32 @caller_save_vgpr_spill_fp() #0 {
; CHECK-NEXT: buffer_load_dword v2, off, s[0:3], s33 ; 4-byte Folded Reload
; CHECK-NEXT: s_mov_b64 exec, s[4:5]
; CHECK-NEXT: s_add_i32 s32, s32, 0xfffffc00
-; CHECK-NEXT: s_mov_b32 s33, s19
+; CHECK-NEXT: s_mov_b32 s33, s25
; CHECK-NEXT: s_waitcnt vmcnt(0)
; CHECK-NEXT: s_setpc_b64 s[30:31]
entry:
@@ -258,6 +263,7 @@ define protected amdgpu_kernel void @kernel() {
; CHECK-NEXT: ; implicit-def: $sgpr15
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: s_endpgm
entry:
diff --git a/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll b/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll
index 9999cb9173b5db0..34e67d0993fb7a0 100644
--- a/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll
+++ b/llvm/test/CodeGen/AMDGPU/no-source-locations-in-prologue.ll
@@ -32,6 +32,7 @@ define hidden void @_ZL3barv() #0 !dbg !1644 {
; CHECK-NEXT: s_mov_b64 s[20:21], s[0:1]
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: .Ltmp1:
diff --git a/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll b/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll
index f523b4a2495f198..764f4942cbd03c2 100644
--- a/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll
+++ b/llvm/test/CodeGen/AMDGPU/sgpr-spills-split-regalloc.ll
@@ -16,7 +16,7 @@ define void @spill_sgpr_with_no_lower_vgpr_available() #0 {
; GCN-LABEL: spill_sgpr_with_no_lower_vgpr_available:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: s_mov_b32 s18, s33
+; GCN-NEXT: s_mov_b32 s24, s33
; GCN-NEXT: s_mov_b32 s33, s32
; GCN-NEXT: s_or_saveexec_b64 s[16:17], -1
; GCN-NEXT: buffer_store_dword v255, off, s[0:3], s33 offset:448 ; 4-byte Folded Spill
@@ -150,6 +150,7 @@ define void @spill_sgpr_with_no_lower_vgpr_available() #0 {
; GCN-NEXT: s_mov_b64 s[20:21], s[0:1]
; GCN-NEXT: s_mov_b64 s[0:1], s[20:21]
; GCN-NEXT: s_mov_b64 s[2:3], s[22:23]
+; GCN-NEXT: ; implicit-def: $sgpr18_sgpr19
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GCN-NEXT: v_readlane_b32 s31, v255, 1
@@ -269,7 +270,7 @@ define void @spill_sgpr_with_no_lower_vgpr_available() #0 {
; GCN-NEXT: buffer_load_dword v255, off, s[0:3], s33 offset:448 ; 4-byte Folded Reload
; GCN-NEXT: s_mov_b64 exec, s[4:5]
; GCN-NEXT: s_add_i32 s32, s32, 0xffff8c00
-; GCN-NEXT: s_mov_b32 s33, s18
+; GCN-NEXT: s_mov_b32 s33, s24
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_setpc_b64 s[30:31]
%alloca = alloca i32, align 4, addrspace(5)
@@ -310,7 +311,7 @@ define void @spill_to_lowest_available_vgpr() #0 {
; GCN-LABEL: spill_to_lowest_available_vgpr:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: s_mov_b32 s18, s33
+; GCN-NEXT: s_mov_b32 s24, s33
; GCN-NEXT: s_mov_b32 s33, s32
; GCN-NEXT: s_or_saveexec_b64 s[16:17], -1
; GCN-NEXT: buffer_store_dword v254, off, s[0:3], s33 offset:444 ; 4-byte Folded Spill
@@ -443,6 +444,7 @@ define void @spill_to_lowest_available_vgpr() #0 {
; GCN-NEXT: s_mov_b64 s[20:21], s[0:1]
; GCN-NEXT: s_mov_b64 s[0:1], s[20:21]
; GCN-NEXT: s_mov_b64 s[2:3], s[22:23]
+; GCN-NEXT: ; implicit-def: $sgpr18_sgpr19
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GCN-NEXT: v_readlane_b32 s31, v254, 1
@@ -561,7 +563,7 @@ define void @spill_to_lowest_available_vgpr() #0 {
; GCN-NEXT: buffer_load_dword v254, off, s[0:3], s33 offset:444 ; 4-byte Folded Reload
; GCN-NEXT: s_mov_b64 exec, s[4:5]
; GCN-NEXT: s_add_i32 s32, s32, 0xffff8c00
-; GCN-NEXT: s_mov_b32 s33, s18
+; GCN-NEXT: s_mov_b32 s33, s24
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_setpc_b64 s[30:31]
%alloca = alloca i32, align 4, addrspace(5)
@@ -1528,7 +1530,7 @@ define void @spill_sgpr_no_free_vgpr_ipra() #0 {
; GCN-LABEL: spill_sgpr_no_free_vgpr_ipra:
; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-NEXT: s_mov_b32 s18, s33
+; GCN-NEXT: s_mov_b32 s24, s33
; GCN-NEXT: s_mov_b32 s33, s32
; GCN-NEXT: s_add_i32 s32, s32, 0x7400
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:444 ; 4-byte Folded Spill
@@ -1666,6 +1668,7 @@ define void @spill_sgpr_no_free_vgpr_ipra() #0 {
; GCN-NEXT: s_mov_b64 s[20:21], s[0:1]
; GCN-NEXT: s_mov_b64 s[0:1], s[20:21]
; GCN-NEXT: s_mov_b64 s[2:3], s[22:23]
+; GCN-NEXT: ; implicit-def: $sgpr18_sgpr19
; GCN-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GCN-NEXT: s_mov_b64 s[4:5], exec
; GCN-NEXT: s_mov_b64 exec, 1
@@ -1798,7 +1801,7 @@ define void @spill_sgpr_no_free_vgpr_ipra() #0 {
; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:440 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:444 ; 4-byte Folded Reload
; GCN-NEXT: s_add_i32 s32, s32, 0xffff8c00
-; GCN-NEXT: s_mov_b32 s33, s18
+; GCN-NEXT: s_mov_b32 s33, s24
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_setpc_b64 s[30:31]
call void @child_function_ipra()
diff --git a/llvm/test/CodeGen/AMDGPU/stacksave_stackrestore.ll b/llvm/test/CodeGen/AMDGPU/stacksave_stackrestore.ll
index 8c5b89429bcc1b4..33b5d6c6850bfe8 100644
--- a/llvm/test/CodeGen/AMDGPU/stacksave_stackrestore.ll
+++ b/llvm/test/CodeGen/AMDGPU/stacksave_stackrestore.ll
@@ -916,13 +916,13 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE32-O0-LABEL: kernel_stacksave_stackrestore_call_with_stack_objects:
; WAVE32-O0: ; %bb.0:
; WAVE32-O0-NEXT: s_mov_b32 s32, 0x1200
-; WAVE32-O0-NEXT: s_getpc_b64 s[20:21]
-; WAVE32-O0-NEXT: s_mov_b32 s20, s0
-; WAVE32-O0-NEXT: s_load_dwordx4 s[20:23], s[20:21], 0x0
+; WAVE32-O0-NEXT: s_getpc_b64 s[24:25]
+; WAVE32-O0-NEXT: s_mov_b32 s24, s0
+; WAVE32-O0-NEXT: s_load_dwordx4 s[24:27], s[24:25], 0x0
; WAVE32-O0-NEXT: s_waitcnt lgkmcnt(0)
-; WAVE32-O0-NEXT: s_bitset0_b32 s23, 21
-; WAVE32-O0-NEXT: s_add_u32 s20, s20, s9
-; WAVE32-O0-NEXT: s_addc_u32 s21, s21, 0
+; WAVE32-O0-NEXT: s_bitset0_b32 s27, 21
+; WAVE32-O0-NEXT: s_add_u32 s24, s24, s9
+; WAVE32-O0-NEXT: s_addc_u32 s25, s25, 0
; WAVE32-O0-NEXT: ; implicit-def: $vgpr3 : SGPR spill to VGPR lane
; WAVE32-O0-NEXT: s_mov_b32 s14, s8
; WAVE32-O0-NEXT: s_mov_b32 s13, s7
@@ -934,17 +934,17 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE32-O0-NEXT: v_writelane_b32 v3, s0, 0
; WAVE32-O0-NEXT: s_lshr_b32 s0, s0, 5
; WAVE32-O0-NEXT: v_writelane_b32 v3, s0, 1
-; WAVE32-O0-NEXT: s_or_saveexec_b32 s19, -1
-; WAVE32-O0-NEXT: buffer_store_dword v3, off, s[20:23], 0 offset:128 ; 4-byte Folded Spill
-; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s19
+; WAVE32-O0-NEXT: s_or_saveexec_b32 s20, -1
+; WAVE32-O0-NEXT: buffer_store_dword v3, off, s[24:27], 0 offset:128 ; 4-byte Folded Spill
+; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s20
; WAVE32-O0-NEXT: v_mov_b32_e32 v3, 42
-; WAVE32-O0-NEXT: buffer_store_dword v3, off, s[20:23], 0
+; WAVE32-O0-NEXT: buffer_store_dword v3, off, s[24:27], 0
; WAVE32-O0-NEXT: s_waitcnt_vscnt null, 0x0
-; WAVE32-O0-NEXT: s_mov_b64 s[0:1], s[20:21]
-; WAVE32-O0-NEXT: s_mov_b64 s[2:3], s[22:23]
+; WAVE32-O0-NEXT: s_mov_b64 s[0:1], s[24:25]
+; WAVE32-O0-NEXT: s_mov_b64 s[2:3], s[26:27]
; WAVE32-O0-NEXT: s_mov_b32 s6, s32
; WAVE32-O0-NEXT: v_mov_b32_e32 v3, 17
-; WAVE32-O0-NEXT: buffer_store_dword v3, off, s[20:23], s6 offset:4
+; WAVE32-O0-NEXT: buffer_store_dword v3, off, s[24:27], s6 offset:4
; WAVE32-O0-NEXT: s_mov_b32 s6, stack_passed_argument at abs32@hi
; WAVE32-O0-NEXT: s_mov_b32 s16, stack_passed_argument at abs32@lo
; WAVE32-O0-NEXT: ; kill: def $sgpr16 killed $sgpr16 def $sgpr16_sgpr17
@@ -1018,10 +1018,11 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE32-O0-NEXT: v_mov_b32_e32 v29, s18
; WAVE32-O0-NEXT: ; implicit-def: $sgpr18
; WAVE32-O0-NEXT: v_mov_b32_e32 v30, s18
+; WAVE32-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; WAVE32-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
-; WAVE32-O0-NEXT: s_or_saveexec_b32 s19, -1
-; WAVE32-O0-NEXT: buffer_load_dword v0, off, s[20:23], 0 offset:128 ; 4-byte Folded Reload
-; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s19
+; WAVE32-O0-NEXT: s_or_saveexec_b32 s20, -1
+; WAVE32-O0-NEXT: buffer_load_dword v0, off, s[24:27], 0 offset:128 ; 4-byte Folded Reload
+; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s20
; WAVE32-O0-NEXT: s_waitcnt vmcnt(0)
; WAVE32-O0-NEXT: v_readlane_b32 s1, v0, 1
; WAVE32-O0-NEXT: v_readlane_b32 s0, v0, 0
@@ -1136,6 +1137,7 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE64-O0-NEXT: v_mov_b32_e32 v29, s18
; WAVE64-O0-NEXT: ; implicit-def: $sgpr18
; WAVE64-O0-NEXT: v_mov_b32_e32 v30, s18
+; WAVE64-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; WAVE64-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; WAVE64-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
; WAVE64-O0-NEXT: buffer_load_dword v0, off, s[24:27], 0 offset:128 ; 4-byte Folded Reload
@@ -1153,13 +1155,13 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE32-WWM-PREALLOC-LABEL: kernel_stacksave_stackrestore_call_with_stack_objects:
; WAVE32-WWM-PREALLOC: ; %bb.0:
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s32, 0x1200
-; WAVE32-WWM-PREALLOC-NEXT: s_getpc_b64 s[20:21]
-; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s20, s0
-; WAVE32-WWM-PREALLOC-NEXT: s_load_dwordx4 s[20:23], s[20:21], 0x0
+; WAVE32-WWM-PREALLOC-NEXT: s_getpc_b64 s[24:25]
+; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s24, s0
+; WAVE32-WWM-PREALLOC-NEXT: s_load_dwordx4 s[24:27], s[24:25], 0x0
; WAVE32-WWM-PREALLOC-NEXT: s_waitcnt lgkmcnt(0)
-; WAVE32-WWM-PREALLOC-NEXT: s_bitset0_b32 s23, 21
-; WAVE32-WWM-PREALLOC-NEXT: s_add_u32 s20, s20, s9
-; WAVE32-WWM-PREALLOC-NEXT: s_addc_u32 s21, s21, 0
+; WAVE32-WWM-PREALLOC-NEXT: s_bitset0_b32 s27, 21
+; WAVE32-WWM-PREALLOC-NEXT: s_add_u32 s24, s24, s9
+; WAVE32-WWM-PREALLOC-NEXT: s_addc_u32 s25, s25, 0
; WAVE32-WWM-PREALLOC-NEXT: ; implicit-def: $vgpr32 : SGPR spill to VGPR lane
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s14, s8
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s13, s7
@@ -1172,13 +1174,13 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE32-WWM-PREALLOC-NEXT: s_lshr_b32 s0, s0, 5
; WAVE32-WWM-PREALLOC-NEXT: v_writelane_b32 v32, s0, 1
; WAVE32-WWM-PREALLOC-NEXT: v_mov_b32_e32 v3, 42
-; WAVE32-WWM-PREALLOC-NEXT: buffer_store_dword v3, off, s[20:23], 0
+; WAVE32-WWM-PREALLOC-NEXT: buffer_store_dword v3, off, s[24:27], 0
; WAVE32-WWM-PREALLOC-NEXT: s_waitcnt_vscnt null, 0x0
-; WAVE32-WWM-PREALLOC-NEXT: s_mov_b64 s[0:1], s[20:21]
-; WAVE32-WWM-PREALLOC-NEXT: s_mov_b64 s[2:3], s[22:23]
+; WAVE32-WWM-PREALLOC-NEXT: s_mov_b64 s[0:1], s[24:25]
+; WAVE32-WWM-PREALLOC-NEXT: s_mov_b64 s[2:3], s[26:27]
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s6, s32
; WAVE32-WWM-PREALLOC-NEXT: v_mov_b32_e32 v3, 17
-; WAVE32-WWM-PREALLOC-NEXT: buffer_store_dword v3, off, s[20:23], s6 offset:4
+; WAVE32-WWM-PREALLOC-NEXT: buffer_store_dword v3, off, s[24:27], s6 offset:4
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s6, stack_passed_argument at abs32@hi
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s16, stack_passed_argument at abs32@lo
; WAVE32-WWM-PREALLOC-NEXT: ; kill: def $sgpr16 killed $sgpr16 def $sgpr16_sgpr17
@@ -1252,6 +1254,7 @@ define amdgpu_kernel void @kernel_stacksave_stackrestore_call_with_stack_objects
; WAVE32-WWM-PREALLOC-NEXT: v_mov_b32_e32 v29, s18
; WAVE32-WWM-PREALLOC-NEXT: ; implicit-def: $sgpr18
; WAVE32-WWM-PREALLOC-NEXT: v_mov_b32_e32 v30, s18
+; WAVE32-WWM-PREALLOC-NEXT: ; implicit-def: $sgpr18_sgpr19
; WAVE32-WWM-PREALLOC-NEXT: s_swappc_b64 s[30:31], s[16:17]
; WAVE32-WWM-PREALLOC-NEXT: v_readlane_b32 s1, v32, 1
; WAVE32-WWM-PREALLOC-NEXT: v_readlane_b32 s0, v32, 0
@@ -1344,7 +1347,7 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE32-O0-LABEL: func_stacksave_stackrestore_call_with_stack_objects:
; WAVE32-O0: ; %bb.0:
; WAVE32-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WAVE32-O0-NEXT: s_mov_b32 s25, s33
+; WAVE32-O0-NEXT: s_mov_b32 s26, s33
; WAVE32-O0-NEXT: s_mov_b32 s33, s32
; WAVE32-O0-NEXT: s_xor_saveexec_b32 s16, -1
; WAVE32-O0-NEXT: buffer_store_dword v32, off, s[0:3], s33 offset:128 ; 4-byte Folded Spill
@@ -1358,9 +1361,9 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE32-O0-NEXT: v_writelane_b32 v0, s16, 0
; WAVE32-O0-NEXT: s_lshr_b32 s16, s16, 5
; WAVE32-O0-NEXT: v_writelane_b32 v0, s16, 1
-; WAVE32-O0-NEXT: s_or_saveexec_b32 s24, -1
+; WAVE32-O0-NEXT: s_or_saveexec_b32 s25, -1
; WAVE32-O0-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:132 ; 4-byte Folded Spill
-; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s24
+; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s25
; WAVE32-O0-NEXT: v_mov_b32_e32 v0, 42
; WAVE32-O0-NEXT: buffer_store_dword v0, off, s[0:3], s33
; WAVE32-O0-NEXT: s_waitcnt_vscnt null, 0x0
@@ -1437,10 +1440,11 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE32-O0-NEXT: v_mov_b32_e32 v29, s18
; WAVE32-O0-NEXT: ; implicit-def: $sgpr18
; WAVE32-O0-NEXT: v_mov_b32_e32 v30, s18
+; WAVE32-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; WAVE32-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
-; WAVE32-O0-NEXT: s_or_saveexec_b32 s24, -1
+; WAVE32-O0-NEXT: s_or_saveexec_b32 s25, -1
; WAVE32-O0-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:132 ; 4-byte Folded Reload
-; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s24
+; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s25
; WAVE32-O0-NEXT: s_waitcnt vmcnt(0)
; WAVE32-O0-NEXT: v_readlane_b32 s5, v0, 1
; WAVE32-O0-NEXT: v_readlane_b32 s4, v0, 0
@@ -1456,14 +1460,14 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE32-O0-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:136 ; 4-byte Folded Reload
; WAVE32-O0-NEXT: s_mov_b32 exec_lo, s4
; WAVE32-O0-NEXT: s_add_i32 s32, s32, 0xffffee00
-; WAVE32-O0-NEXT: s_mov_b32 s33, s25
+; WAVE32-O0-NEXT: s_mov_b32 s33, s26
; WAVE32-O0-NEXT: s_waitcnt vmcnt(0)
; WAVE32-O0-NEXT: s_setpc_b64 s[30:31]
;
; WAVE64-O0-LABEL: func_stacksave_stackrestore_call_with_stack_objects:
; WAVE64-O0: ; %bb.0:
; WAVE64-O0-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WAVE64-O0-NEXT: s_mov_b32 s19, s33
+; WAVE64-O0-NEXT: s_mov_b32 s28, s33
; WAVE64-O0-NEXT: s_mov_b32 s33, s32
; WAVE64-O0-NEXT: s_xor_saveexec_b64 s[16:17], -1
; WAVE64-O0-NEXT: buffer_store_dword v32, off, s[0:3], s33 offset:128 ; 4-byte Folded Spill
@@ -1556,6 +1560,7 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE64-O0-NEXT: v_mov_b32_e32 v29, s18
; WAVE64-O0-NEXT: ; implicit-def: $sgpr18
; WAVE64-O0-NEXT: v_mov_b32_e32 v30, s18
+; WAVE64-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; WAVE64-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; WAVE64-O0-NEXT: s_or_saveexec_b64 s[26:27], -1
; WAVE64-O0-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:132 ; 4-byte Folded Reload
@@ -1575,14 +1580,14 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE64-O0-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:136 ; 4-byte Folded Reload
; WAVE64-O0-NEXT: s_mov_b64 exec, s[4:5]
; WAVE64-O0-NEXT: s_add_i32 s32, s32, 0xffffdc00
-; WAVE64-O0-NEXT: s_mov_b32 s33, s19
+; WAVE64-O0-NEXT: s_mov_b32 s33, s28
; WAVE64-O0-NEXT: s_waitcnt vmcnt(0)
; WAVE64-O0-NEXT: s_setpc_b64 s[30:31]
;
; WAVE32-WWM-PREALLOC-LABEL: func_stacksave_stackrestore_call_with_stack_objects:
; WAVE32-WWM-PREALLOC: ; %bb.0:
; WAVE32-WWM-PREALLOC-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s24, s33
+; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s25, s33
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s33, s32
; WAVE32-WWM-PREALLOC-NEXT: s_xor_saveexec_b32 s16, -1
; WAVE32-WWM-PREALLOC-NEXT: buffer_store_dword v33, off, s[0:3], s33 offset:128 ; 4-byte Folded Spill
@@ -1672,6 +1677,7 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE32-WWM-PREALLOC-NEXT: v_mov_b32_e32 v29, s18
; WAVE32-WWM-PREALLOC-NEXT: ; implicit-def: $sgpr18
; WAVE32-WWM-PREALLOC-NEXT: v_mov_b32_e32 v30, s18
+; WAVE32-WWM-PREALLOC-NEXT: ; implicit-def: $sgpr18_sgpr19
; WAVE32-WWM-PREALLOC-NEXT: s_swappc_b64 s[30:31], s[16:17]
; WAVE32-WWM-PREALLOC-NEXT: v_readlane_b32 s5, v32, 1
; WAVE32-WWM-PREALLOC-NEXT: v_readlane_b32 s4, v32, 0
@@ -1687,7 +1693,7 @@ define void @func_stacksave_stackrestore_call_with_stack_objects() {
; WAVE32-WWM-PREALLOC-NEXT: buffer_load_dword v32, off, s[0:3], s33 offset:132 ; 4-byte Folded Reload
; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 exec_lo, s4
; WAVE32-WWM-PREALLOC-NEXT: s_add_i32 s32, s32, 0xffffee00
-; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s33, s24
+; WAVE32-WWM-PREALLOC-NEXT: s_mov_b32 s33, s25
; WAVE32-WWM-PREALLOC-NEXT: s_waitcnt vmcnt(0)
; WAVE32-WWM-PREALLOC-NEXT: s_setpc_b64 s[30:31]
%alloca = alloca [32 x i32], addrspace(5)
diff --git a/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll b/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
index bfc249e9081d226..d2364a61ed68622 100644
--- a/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
+++ b/llvm/test/CodeGen/AMDGPU/vgpr-liverange-ir.ll
@@ -233,10 +233,10 @@ define amdgpu_ps float @loop(i32 %z, float %v, i32 inreg %bound, ptr %extern_fun
; SI-NEXT: bb.1.Flow:
; SI-NEXT: successors: %bb.2(0x40000000), %bb.10(0x40000000)
; SI-NEXT: {{ $}}
- ; SI-NEXT: [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %47:vgpr_32, %bb.0, %4, %bb.9
- ; SI-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[COPY4]], %bb.0, undef %49:vgpr_32, %bb.9
- ; SI-NEXT: [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[COPY3]], %bb.0, undef %51:vgpr_32, %bb.9
- ; SI-NEXT: [[PHI3:%[0-9]+]]:vgpr_32 = PHI [[COPY2]], %bb.0, undef %53:vgpr_32, %bb.9
+ ; SI-NEXT: [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %49:vgpr_32, %bb.0, %4, %bb.9
+ ; SI-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[COPY4]], %bb.0, undef %51:vgpr_32, %bb.9
+ ; SI-NEXT: [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[COPY3]], %bb.0, undef %53:vgpr_32, %bb.9
+ ; SI-NEXT: [[PHI3:%[0-9]+]]:vgpr_32 = PHI [[COPY2]], %bb.0, undef %55:vgpr_32, %bb.9
; SI-NEXT: [[SI_ELSE:%[0-9]+]]:sreg_32 = SI_ELSE killed [[SI_IF]], %bb.10, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
; SI-NEXT: S_BRANCH %bb.2
; SI-NEXT: {{ $}}
@@ -249,8 +249,8 @@ define amdgpu_ps float @loop(i32 %z, float %v, i32 inreg %bound, ptr %extern_fun
; SI-NEXT: bb.3:
; SI-NEXT: successors: %bb.4(0x80000000)
; SI-NEXT: {{ $}}
- ; SI-NEXT: [[PHI4:%[0-9]+]]:vreg_64 = PHI undef %55:vreg_64, %bb.4, [[REG_SEQUENCE]], %bb.2
- ; SI-NEXT: [[PHI5:%[0-9]+]]:vgpr_32 = PHI undef %57:vgpr_32, %bb.4, [[PHI1]], %bb.2
+ ; SI-NEXT: [[PHI4:%[0-9]+]]:vreg_64 = PHI undef %57:vreg_64, %bb.4, [[REG_SEQUENCE]], %bb.2
+ ; SI-NEXT: [[PHI5:%[0-9]+]]:vgpr_32 = PHI undef %59:vgpr_32, %bb.4, [[PHI1]], %bb.2
; SI-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI4]].sub0, implicit $exec
; SI-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI4]].sub1, implicit $exec
; SI-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_1]], %subreg.sub1
@@ -286,8 +286,8 @@ define amdgpu_ps float @loop(i32 %z, float %v, i32 inreg %bound, ptr %extern_fun
; SI-NEXT: bb.7:
; SI-NEXT: successors: %bb.8(0x80000000)
; SI-NEXT: {{ $}}
- ; SI-NEXT: [[PHI6:%[0-9]+]]:vreg_64 = PHI undef %59:vreg_64, %bb.8, [[REG_SEQUENCE2]], %bb.6
- ; SI-NEXT: [[PHI7:%[0-9]+]]:vgpr_32 = PHI undef %61:vgpr_32, %bb.8, [[COPY4]], %bb.6
+ ; SI-NEXT: [[PHI6:%[0-9]+]]:vreg_64 = PHI undef %61:vreg_64, %bb.8, [[REG_SEQUENCE2]], %bb.6
+ ; SI-NEXT: [[PHI7:%[0-9]+]]:vgpr_32 = PHI undef %63:vgpr_32, %bb.8, [[COPY4]], %bb.6
; SI-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI6]].sub0, implicit $exec
; SI-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI6]].sub1, implicit $exec
; SI-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_2]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_3]], %subreg.sub1
@@ -356,9 +356,9 @@ define amdgpu_ps float @loop_with_use(i32 %z, float %v, i32 inreg %bound, ptr %e
; SI-NEXT: bb.1.Flow:
; SI-NEXT: successors: %bb.2(0x40000000), %bb.10(0x40000000)
; SI-NEXT: {{ $}}
- ; SI-NEXT: [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %48:vgpr_32, %bb.0, %4, %bb.9
- ; SI-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[COPY3]], %bb.0, undef %50:vgpr_32, %bb.9
- ; SI-NEXT: [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[COPY2]], %bb.0, undef %52:vgpr_32, %bb.9
+ ; SI-NEXT: [[PHI:%[0-9]+]]:vgpr_32 = PHI undef %50:vgpr_32, %bb.0, %4, %bb.9
+ ; SI-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[COPY3]], %bb.0, undef %52:vgpr_32, %bb.9
+ ; SI-NEXT: [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[COPY2]], %bb.0, undef %54:vgpr_32, %bb.9
; SI-NEXT: [[SI_ELSE:%[0-9]+]]:sreg_32 = SI_ELSE killed [[SI_IF]], %bb.10, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
; SI-NEXT: S_BRANCH %bb.2
; SI-NEXT: {{ $}}
@@ -371,7 +371,7 @@ define amdgpu_ps float @loop_with_use(i32 %z, float %v, i32 inreg %bound, ptr %e
; SI-NEXT: bb.3:
; SI-NEXT: successors: %bb.4(0x80000000)
; SI-NEXT: {{ $}}
- ; SI-NEXT: [[PHI3:%[0-9]+]]:vreg_64 = PHI undef %54:vreg_64, %bb.4, [[REG_SEQUENCE]], %bb.2
+ ; SI-NEXT: [[PHI3:%[0-9]+]]:vreg_64 = PHI undef %56:vreg_64, %bb.4, [[REG_SEQUENCE]], %bb.2
; SI-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI3]].sub0, implicit $exec
; SI-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI3]].sub1, implicit $exec
; SI-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sgpr_64 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_1]], %subreg.sub1
@@ -407,7 +407,7 @@ define amdgpu_ps float @loop_with_use(i32 %z, float %v, i32 inreg %bound, ptr %e
; SI-NEXT: bb.7:
; SI-NEXT: successors: %bb.8(0x80000000)
; SI-NEXT: {{ $}}
- ; SI-NEXT: [[PHI4:%[0-9]+]]:vreg_64 = PHI undef %56:vreg_64, %bb.8, [[REG_SEQUENCE2]], %bb.6
+ ; SI-NEXT: [[PHI4:%[0-9]+]]:vreg_64 = PHI undef %58:vreg_64, %bb.8, [[REG_SEQUENCE2]], %bb.6
; SI-NEXT: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI4]].sub0, implicit $exec
; SI-NEXT: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sgpr_32 = V_READFIRSTLANE_B32 [[PHI4]].sub1, implicit $exec
; SI-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_64 = REG_SEQUENCE killed [[V_READFIRSTLANE_B32_2]], %subreg.sub0, killed [[V_READFIRSTLANE_B32_3]], %subreg.sub1
diff --git a/llvm/test/CodeGen/AMDGPU/vgpr_constant_to_sgpr.ll b/llvm/test/CodeGen/AMDGPU/vgpr_constant_to_sgpr.ll
index 7840559c78eb6b1..364ce82b2e997ee 100644
--- a/llvm/test/CodeGen/AMDGPU/vgpr_constant_to_sgpr.ll
+++ b/llvm/test/CodeGen/AMDGPU/vgpr_constant_to_sgpr.ll
@@ -47,6 +47,7 @@ define protected amdgpu_kernel void @kern(ptr %addr) !llvm.amdgcn.lds.kernel.id
; CHECK-NEXT: s_mov_b32 s15, 42
; CHECK-NEXT: s_mov_b64 s[0:1], s[20:21]
; CHECK-NEXT: s_mov_b64 s[2:3], s[22:23]
+; CHECK-NEXT: ; implicit-def: $sgpr18_sgpr19
; CHECK-NEXT: s_waitcnt lgkmcnt(0)
; CHECK-NEXT: s_swappc_b64 s[30:31], s[16:17]
; CHECK-NEXT: s_endpgm
diff --git a/llvm/test/CodeGen/AMDGPU/whole-wave-register-spill.ll b/llvm/test/CodeGen/AMDGPU/whole-wave-register-spill.ll
index 7eabe982ff2bcde..3a33194f17c875a 100644
--- a/llvm/test/CodeGen/AMDGPU/whole-wave-register-spill.ll
+++ b/llvm/test/CodeGen/AMDGPU/whole-wave-register-spill.ll
@@ -101,6 +101,7 @@ define void @test() #0 {
; GCN-O0-NEXT: s_mov_b64 s[20:21], s[0:1]
; GCN-O0-NEXT: s_mov_b64 s[0:1], s[20:21]
; GCN-O0-NEXT: s_mov_b64 s[2:3], s[22:23]
+; GCN-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; GCN-O0-NEXT: s_waitcnt lgkmcnt(0)
; GCN-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GCN-O0-NEXT: s_or_saveexec_b64 s[28:29], -1
diff --git a/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll b/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
index e79cb66dcd7760e..11f6a2960776b04 100644
--- a/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
+++ b/llvm/test/CodeGen/AMDGPU/wwm-reserved-spill.ll
@@ -406,6 +406,7 @@ define amdgpu_gfx void @strict_wwm_call(ptr addrspace(8) inreg %tmp14, i32 inreg
; GFX9-O0-NEXT: s_mov_b64 s[0:1], s[44:45]
; GFX9-O0-NEXT: s_mov_b64 s[2:3], s[46:47]
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v2
+; GFX9-O0-NEXT: ; implicit-def: $sgpr44_sgpr45
; GFX9-O0-NEXT: s_swappc_b64 s[30:31], s[42:43]
; GFX9-O0-NEXT: v_mov_b32_e32 v1, v0
; GFX9-O0-NEXT: v_add_u32_e64 v1, v1, v2
@@ -632,6 +633,7 @@ define amdgpu_gfx void @strict_wwm_call_i64(ptr addrspace(8) inreg %tmp14, i64 i
; GFX9-O0-NEXT: s_mov_b64 s[2:3], s[38:39]
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v2
; GFX9-O0-NEXT: v_mov_b32_e32 v1, v3
+; GFX9-O0-NEXT: ; implicit-def: $sgpr36_sgpr37
; GFX9-O0-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-O0-NEXT: s_swappc_b64 s[30:31], s[34:35]
; GFX9-O0-NEXT: s_or_saveexec_b64 s[46:47], -1
diff --git a/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll b/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
index 47c976d2a5c33d2..6ac61410a0e7d84 100644
--- a/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
+++ b/llvm/test/CodeGen/AMDGPU/wwm-reserved.ll
@@ -413,6 +413,7 @@ define amdgpu_kernel void @call(ptr addrspace(8) inreg %tmp14, i32 inreg %arg) {
; GFX9-O0-NEXT: ; implicit-def: $sgpr15
; GFX9-O0-NEXT: v_mov_b32_e32 v31, v3
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v6
+; GFX9-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; GFX9-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GFX9-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
; GFX9-O0-NEXT: buffer_load_dword v1, off, s[24:27], 0 ; 4-byte Folded Reload
@@ -656,6 +657,7 @@ define amdgpu_kernel void @call_i64(ptr addrspace(8) inreg %tmp14, i64 inreg %ar
; GFX9-O0-NEXT: v_mov_b32_e32 v31, v3
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v6
; GFX9-O0-NEXT: v_mov_b32_e32 v1, v7
+; GFX9-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; GFX9-O0-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GFX9-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
@@ -1283,6 +1285,7 @@ define amdgpu_kernel void @strict_wwm_call(ptr addrspace(8) inreg %tmp14, i32 in
; GFX9-O0-NEXT: ; implicit-def: $sgpr15
; GFX9-O0-NEXT: v_mov_b32_e32 v31, v3
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v6
+; GFX9-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; GFX9-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GFX9-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
; GFX9-O0-NEXT: buffer_load_dword v1, off, s[24:27], 0 ; 4-byte Folded Reload
@@ -1526,6 +1529,7 @@ define amdgpu_kernel void @strict_wwm_call_i64(ptr addrspace(8) inreg %tmp14, i6
; GFX9-O0-NEXT: v_mov_b32_e32 v31, v3
; GFX9-O0-NEXT: v_mov_b32_e32 v0, v6
; GFX9-O0-NEXT: v_mov_b32_e32 v1, v7
+; GFX9-O0-NEXT: ; implicit-def: $sgpr18_sgpr19
; GFX9-O0-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-O0-NEXT: s_swappc_b64 s[30:31], s[16:17]
; GFX9-O0-NEXT: s_or_saveexec_b64 s[20:21], -1
diff --git a/llvm/test/CodeGen/PowerPC/fmf-propagation.ll b/llvm/test/CodeGen/PowerPC/fmf-propagation.ll
index 58b3ee485ea4b1c..4e72a5ac5ede314 100644
--- a/llvm/test/CodeGen/PowerPC/fmf-propagation.ll
+++ b/llvm/test/CodeGen/PowerPC/fmf-propagation.ll
@@ -577,15 +577,15 @@ define double @fcmp_nnan(double %a, double %y, double %z) {
; FP library calls can have fast-math-flags.
; FMFDEBUG-LABEL: Optimized lowered selection DAG: %bb.0 'log2_approx:'
-; FMFDEBUG: ch,glue = PPCISD::CALL_NOP t11, TargetGlobalAddress:i64<ptr @log2>
-; FMFDEBUG: ch,glue = callseq_end t15, TargetConstant:i64<32>, TargetConstant:i64<0>, t15:1
-; FMFDEBUG: f64,ch,glue = CopyFromReg t16, Register:f64 $f1, t16:1
+; FMFDEBUG: ch,glue = PPCISD::CALL_NOP {{t[0-9]+}}, TargetGlobalAddress:i64<ptr @log2>
+; FMFDEBUG: ch,glue = callseq_end [[T15:t[0-9]+]], TargetConstant:i64<32>, TargetConstant:i64<0>, [[T15]]:1
+; FMFDEBUG: f64,ch,glue = CopyFromReg [[T16:t[0-9]+]], Register:f64 $f1, [[T16]]:1
; FMFDEBUG: Type-legalized selection DAG: %bb.0 'log2_approx:'
; GLOBALDEBUG-LABEL: Optimized lowered selection DAG: %bb.0 'log2_approx:'
-; GLOBALDEBUG: ch,glue = PPCISD::CALL_NOP t11, TargetGlobalAddress:i64<ptr @log2>
-; GLOBALDEBUG: ch,glue = callseq_end t15, TargetConstant:i64<32>, TargetConstant:i64<0>, t15:1
-; GLOBALDEBUG: f64,ch,glue = CopyFromReg t16, Register:f64 $f1, t16:1
+; GLOBALDEBUG: ch,glue = PPCISD::CALL_NOP {{t[0-9]+}}, TargetGlobalAddress:i64<ptr @log2>
+; GLOBALDEBUG: ch,glue = callseq_end [[T15:t[0-9]+]], TargetConstant:i64<32>, TargetConstant:i64<0>, [[T15]]:1
+; GLOBALDEBUG: f64,ch,glue = CopyFromReg [[T16:t[0-9]+]], Register:f64 $f1, [[T16]]:1
; GLOBALDEBUG: Type-legalized selection DAG: %bb.0 'log2_approx:'
declare double @log2(double)
diff --git a/llvm/test/MachineVerifier/convergencectrl/AMDGPU/basic.mir b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/basic.mir
new file mode 100644
index 000000000000000..94d0ddad25944d3
--- /dev/null
+++ b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/basic.mir
@@ -0,0 +1,37 @@
+# RUN: not --crash llc -march=amdgcn -run-pass=none -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck %s
+---
+name: basic
+tracksRegLiveness: true
+body: |
+ bb.0:
+ successors: %bb.1, %bb.2;
+ %0:sgpr_64 = CONVERGENCECTRL_ANCHOR
+ ; CHECK: Entry intrinsic cannot be preceded by a convergent operation in the same basic block.
+ ; CHECK: CONVERGENCECTRL_ENTRY
+ %1:sgpr_64 = CONVERGENCECTRL_ENTRY
+ ; CHECK: Loop intrinsic cannot be preceded by a convergent operation in the same basic block.
+ ; CHECK: CONVERGENCECTRL_LOOP
+ %2:sgpr_64 = CONVERGENCECTRL_LOOP %0:sgpr_64
+ S_CBRANCH_EXECZ %bb.1, implicit $exec
+ S_BRANCH %bb.2
+
+ bb.1:
+ successors: %bb.2;
+ ; CHECK: Entry intrinsic can occur only in the entry block.
+ ; CHECK: CONVERGENCECTRL_ENTRY
+ %5:sgpr_64 = CONVERGENCECTRL_ENTRY
+
+ bb.2:
+ ; CHECK: Convergence control tokens can only be used by convergent operations.
+ ; CHECK: G_PHI
+ %6:sgpr_64 = G_PHI %0:sgpr_64, %bb.0, %0:sgpr_64, %bb.1
+ %7:sgpr_64 = CONVERGENCECTRL_ANCHOR
+ %8:sgpr_64 = IMPLICIT_DEF
+ %4:sgpr_64 = SI_CALL %8:sgpr_64, 1, implicit %7:sgpr_64
+ ; CHECK: An operation can use at most one convergence control token.
+ ; CHECK: SI_CALL %{{[0-9]}}:sgpr_64, 2
+ %9:sgpr_64 = SI_CALL %8:sgpr_64, 2, implicit %7:sgpr_64, implicit %7:sgpr_64
+ ; CHECK: Cannot mix controlled and uncontrolled convergence in the same function.
+ ; CHECK: SI_CALL %{{[0-9]}}:sgpr_64, 3
+ %10:sgpr_64 = SI_CALL %8:sgpr_64, 3
+...
diff --git a/llvm/test/MachineVerifier/convergencectrl/AMDGPU/cycles.mir b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/cycles.mir
new file mode 100644
index 000000000000000..87cf3e604929bb0
--- /dev/null
+++ b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/cycles.mir
@@ -0,0 +1,52 @@
+# RUN: not --crash llc -march=amdgcn -run-pass=none -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck %s
+---
+name: cycles
+body: |
+ bb.0:
+ %0:sgpr_64 = CONVERGENCECTRL_ANCHOR
+ %1:sgpr_64 = IMPLICIT_DEF
+ S_CBRANCH_EXECZ %bb.9, implicit $exec
+ S_BRANCH %bb.1
+
+ bb.1:
+ S_CBRANCH_EXECZ %bb.8, implicit $exec
+ S_BRANCH %bb.5
+
+ bb.2:
+ S_CBRANCH_EXECZ %bb.3, implicit $exec
+ S_BRANCH %bb.4
+
+ bb.3:
+ ; CHECK: Cycle heart must dominate all blocks in the cycle.
+ ; Irreducible cycle: entries(bb.4 bb.3)
+ %3:sgpr_64 = CONVERGENCECTRL_LOOP %0:sgpr_64
+ S_BRANCH %bb.4
+
+ bb.4:
+ S_BRANCH %bb.3
+
+ bb.5:
+ S_CBRANCH_EXECZ %bb.6, implicit $exec
+ S_BRANCH %bb.2
+
+ bb.6:
+ S_BRANCH %bb.7
+
+ bb.7:
+ ; CHECK: Cycle heart must dominate all blocks in the cycle.
+ ; Reducible cycle: entries(bb.6) bb.7
+ %4:sgpr_64 = CONVERGENCECTRL_LOOP %0:sgpr_64
+ S_BRANCH %bb.6
+
+ bb.8:
+ ; CHECK: Two static convergence token uses in a cycle that does not contain either token's definition.
+ %5:sgpr_64 = CONVERGENCECTRL_LOOP %0:sgpr_64
+ %6:sgpr_64 = CONVERGENCECTRL_LOOP %0:sgpr_64
+ S_BRANCH %bb.8
+
+ bb.9:
+ ; CHECK: Convergence token used by an instruction other than llvm.experimental.convergence.loop in a cycle that does not contain the token's definition.
+ %7:sgpr_64 = G_SI_CALL %1:sgpr_64, 3, implicit %0:sgpr_64
+ S_BRANCH %bb.9
+
+...
diff --git a/llvm/test/MachineVerifier/convergencectrl/AMDGPU/lit.local.cfg b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/lit.local.cfg
new file mode 100644
index 000000000000000..7c492428aec7614
--- /dev/null
+++ b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/lit.local.cfg
@@ -0,0 +1,2 @@
+if not "AMDGPU" in config.root.targets:
+ config.unsupported = True
diff --git a/llvm/test/MachineVerifier/convergencectrl/AMDGPU/mixed2.mir b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/mixed2.mir
new file mode 100644
index 000000000000000..c70a48bf21309e0
--- /dev/null
+++ b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/mixed2.mir
@@ -0,0 +1,15 @@
+# RUN: not --crash llc -march=amdgcn -run-pass=none -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck %s
+---
+name: mixed2
+body: |
+ bb.0:
+ %0:sgpr_64 = IMPLICIT_DEF
+ %1:sgpr_64 = SI_CALL %0, 1
+ ; CHECK: Cannot mix controlled and uncontrolled convergence in the same function.
+ ; CHECK: CONVERGENCECTRL_ANCHOR
+ %2:sgpr_64 = CONVERGENCECTRL_ANCHOR
+ ; CHECK: Cannot mix controlled and uncontrolled convergence in the same function.
+ ; CHECK: SI_CALL %{{[0-9]}}:sgpr_64, 2
+ %3:sgpr_64 = SI_CALL %0, 2, implicit %2:sgpr_64
+
+...
diff --git a/llvm/test/MachineVerifier/convergencectrl/AMDGPU/region-nesting.mir b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/region-nesting.mir
new file mode 100644
index 000000000000000..9e869acb3e9381e
--- /dev/null
+++ b/llvm/test/MachineVerifier/convergencectrl/AMDGPU/region-nesting.mir
@@ -0,0 +1,24 @@
+# RUN: not --crash llc -march=amdgcn -run-pass=none -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck %s
+---
+name: region_nesting
+body: |
+ bb.0:
+ %0:sgpr_64 = CONVERGENCECTRL_ANCHOR
+ %1:sgpr_64 = CONVERGENCECTRL_ANCHOR
+ %2:sgpr_64 = IMPLICIT_DEF
+ %3:sgpr_64 = SI_CALL %2, 1, implicit %0:sgpr_64
+ ; CHECK: Convergence region is not well-nested.
+ ; CHECK: SI_CALL %{{[0-9]}}:sgpr_64, 2
+ %4:sgpr_64 = SI_CALL %2, 2, implicit %1:sgpr_64
+ S_CBRANCH_EXECZ %bb.1, implicit $exec
+ S_BRANCH %bb.2
+
+ bb.1:
+ %5:sgpr_64 = SI_CALL %2, 3, implicit %0:sgpr_64
+
+ bb.2:
+ ; CHECK: Convergence region is not well-nested.
+ ; CHECK: SI_CALL %{{[0-9]}}:sgpr_64, 4
+ %6:sgpr_64 = SI_CALL %2, 4, implicit %1:sgpr_64
+
+...
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/builtins/match-table-replacerreg.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/builtins/match-table-replacerreg.td
index 622d1df7b381a8e..40a831d7e9e8f67 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/builtins/match-table-replacerreg.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/builtins/match-table-replacerreg.td
@@ -28,7 +28,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK: const uint8_t *GenMyCombiner::getMatchTable() const {
// CHECK-NEXT: constexpr static uint8_t MatchTable0[] = {
-// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(65), GIMT_Encode2(182), /*)*//*default:*//*Label 2*/ GIMT_Encode4(562),
+// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(69), GIMT_Encode2(186), /*)*//*default:*//*Label 2*/ GIMT_Encode4(562),
// CHECK-NEXT: /*TargetOpcode::G_UNMERGE_VALUES*//*Label 0*/ GIMT_Encode4(478), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
// CHECK-NEXT: /*TargetOpcode::G_FNEG*//*Label 1*/ GIMT_Encode4(530),
// CHECK-NEXT: // Label 0: @478
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
index f0ca65a87b76bc4..751b1318ecc01f0 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
@@ -34,12 +34,12 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK: const uint8_t *GenMyCombiner::getMatchTable() const {
// CHECK-NEXT: constexpr static uint8_t MatchTable0[] = {
-// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(19), GIMT_Encode2(128), /*)*//*default:*//*Label 3*/ GIMT_Encode4(563),
-// CHECK-NEXT: /*TargetOpcode::COPY*//*Label 0*/ GIMT_Encode4(446), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_CONSTANT*//*Label 1*/ GIMT_Encode4(477), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_ZEXT*//*Label 2*/ GIMT_Encode4(523),
-// CHECK-NEXT: // Label 0: @446
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 4*/ GIMT_Encode4(476), // Rule ID 0 //
+// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(19), GIMT_Encode2(132), /*)*//*default:*//*Label 3*/ GIMT_Encode4(579),
+// CHECK-NEXT: /*TargetOpcode::COPY*//*Label 0*/ GIMT_Encode4(462), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_CONSTANT*//*Label 1*/ GIMT_Encode4(493), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_ZEXT*//*Label 2*/ GIMT_Encode4(539),
+// CHECK-NEXT: // Label 0: @462
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 4*/ GIMT_Encode4(492), // Rule ID 0 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule0Enabled),
// CHECK-NEXT: GIM_CheckType, /*MI*/0, /*Op*/1, /*Type*/GILLT_s32,
// CHECK-NEXT: // MIs[0] a
@@ -51,10 +51,10 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIR_AddImm8, /*InsnID*/0, /*Imm*/0,
// CHECK-NEXT: GIR_EraseFromParent, /*InsnID*/0,
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 4: @476
+// CHECK-NEXT: // Label 4: @492
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 1: @477
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 5*/ GIMT_Encode4(522), // Rule ID 2 //
+// CHECK-NEXT: // Label 1: @493
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 5*/ GIMT_Encode4(538), // Rule ID 2 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule2Enabled),
// CHECK-NEXT: GIM_CheckType, /*MI*/0, /*Op*/1, /*Type*/GILLT_s32,
// CHECK-NEXT: // MIs[0] a
@@ -66,10 +66,10 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIR_AddCImm, /*InsnID*/0, /*Type*/GILLT_s32, /*Imm*/GIMT_Encode8(42),
// CHECK-NEXT: GIR_EraseFromParent, /*InsnID*/0,
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 5: @522
+// CHECK-NEXT: // Label 5: @538
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 2: @523
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 6*/ GIMT_Encode4(562), // Rule ID 1 //
+// CHECK-NEXT: // Label 2: @539
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 6*/ GIMT_Encode4(578), // Rule ID 1 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule1Enabled),
// CHECK-NEXT: // MIs[0] a
// CHECK-NEXT: // No operand predicates
@@ -83,10 +83,10 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIR_AddSimpleTempRegister, /*InsnID*/0, /*TempRegID*/0,
// CHECK-NEXT: GIR_EraseFromParent, /*InsnID*/0,
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 6: @562
+// CHECK-NEXT: // Label 6: @578
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 3: @563
+// CHECK-NEXT: // Label 3: @579
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: }; // Size: 564 bytes
+// CHECK-NEXT: }; // Size: 580 bytes
// CHECK-NEXT: return MatchTable0;
// CHECK-NEXT: }
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-intrinsics.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-intrinsics.td
index a446fb72298c25f..e8e6d3e74f40245 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-intrinsics.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-intrinsics.td
@@ -29,7 +29,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK: const uint8_t *GenMyCombiner::getMatchTable() const {
// CHECK-NEXT: constexpr static uint8_t MatchTable0[] = {
-// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(115), GIMT_Encode2(117), /*)*//*default:*//*Label 2*/ GIMT_Encode4(132),
+// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(119), GIMT_Encode2(121), /*)*//*default:*//*Label 2*/ GIMT_Encode4(132),
// CHECK-NEXT: /*TargetOpcode::G_INTRINSIC*//*Label 0*/ GIMT_Encode4(18),
// CHECK-NEXT: /*TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS*//*Label 1*/ GIMT_Encode4(73),
// CHECK-NEXT: // Label 0: @18
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-patfrag-root.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-patfrag-root.td
index d3c202c4cb01d44..26a0ec6235e309b 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-patfrag-root.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-patfrag-root.td
@@ -28,7 +28,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK: const uint8_t *GenMyCombiner::getMatchTable() const {
// CHECK-NEXT: constexpr static uint8_t MatchTable0[] = {
-// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(120), GIMT_Encode2(183), /*)*//*default:*//*Label 3*/ GIMT_Encode4(380),
+// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(124), GIMT_Encode2(187), /*)*//*default:*//*Label 3*/ GIMT_Encode4(380),
// CHECK-NEXT: /*TargetOpcode::G_TRUNC*//*Label 0*/ GIMT_Encode4(262), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
// CHECK-NEXT: /*TargetOpcode::G_ZEXT*//*Label 1*/ GIMT_Encode4(298), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
// CHECK-NEXT: /*TargetOpcode::G_FPEXT*//*Label 2*/ GIMT_Encode4(344),
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-variadics.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-variadics.td
index cc77bfdd29c38ec..83b77519bc73a81 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-variadics.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-variadics.td
@@ -37,7 +37,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK: const uint8_t *GenMyCombiner::getMatchTable() const {
// CHECK-NEXT: constexpr static uint8_t MatchTable0[] = {
-// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(65), GIMT_Encode2(69), /*)*//*default:*//*Label 2*/ GIMT_Encode4(88),
+// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(69), GIMT_Encode2(73), /*)*//*default:*//*Label 2*/ GIMT_Encode4(88),
// CHECK-NEXT: /*TargetOpcode::G_UNMERGE_VALUES*//*Label 0*/ GIMT_Encode4(26), GIMT_Encode4(0), GIMT_Encode4(0),
// CHECK-NEXT: /*TargetOpcode::G_BUILD_VECTOR*//*Label 1*/ GIMT_Encode4(57),
// CHECK-NEXT: // Label 0: @26
@@ -98,6 +98,6 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIM_Reject,
// CHECK-NEXT: // Label 2: @88
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: };
+// CHECK-NEXT: }; // Size: 89 bytes
// CHECK-NEXT: return MatchTable0;
// CHECK-NEXT: }
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
index 57ad0009b5bd6aa..5cf4e044a0fb8dd 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table.td
@@ -132,15 +132,15 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// Verify match table.
// CHECK: const uint8_t *GenMyCombiner::getMatchTable() const {
// CHECK-NEXT: constexpr static uint8_t MatchTable0[] = {
-// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(19), GIMT_Encode2(128), /*)*//*default:*//*Label 6*/ GIMT_Encode4(661),
-// CHECK-NEXT: /*TargetOpcode::COPY*//*Label 0*/ GIMT_Encode4(446), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_AND*//*Label 1*/ GIMT_Encode4(488), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_STORE*//*Label 2*/ GIMT_Encode4(541), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_TRUNC*//*Label 3*/ GIMT_Encode4(583), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_SEXT*//*Label 4*/ GIMT_Encode4(608), GIMT_Encode4(0),
-// CHECK-NEXT: /*TargetOpcode::G_ZEXT*//*Label 5*/ GIMT_Encode4(621),
-// CHECK-NEXT: // Label 0: @446
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 7*/ GIMT_Encode4(475), // Rule ID 4 //
+// CHECK-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(19), GIMT_Encode2(132), /*)*//*default:*//*Label 6*/ GIMT_Encode4(677),
+// CHECK-NEXT: /*TargetOpcode::COPY*//*Label 0*/ GIMT_Encode4(462), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_AND*//*Label 1*/ GIMT_Encode4(504), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_STORE*//*Label 2*/ GIMT_Encode4(557), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_TRUNC*//*Label 3*/ GIMT_Encode4(599), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_SEXT*//*Label 4*/ GIMT_Encode4(624), GIMT_Encode4(0),
+// CHECK-NEXT: /*TargetOpcode::G_ZEXT*//*Label 5*/ GIMT_Encode4(637),
+// CHECK-NEXT: // Label 0: @462
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 7*/ GIMT_Encode4(491), // Rule ID 4 //
// CHECK-NEXT: GIM_CheckFeatures, GIMT_Encode2(GIFBS_HasAnswerToEverything),
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule3Enabled),
// CHECK-NEXT: // MIs[0] a
@@ -155,8 +155,8 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: // Combiner Rule #3: InstTest1
// CHECK-NEXT: GIR_CustomAction, GIMT_Encode2(GICXXCustomAction_CombineApplyGICombiner0),
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 7: @475
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 8*/ GIMT_Encode4(487), // Rule ID 3 //
+// CHECK-NEXT: // Label 7: @491
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 8*/ GIMT_Encode4(503), // Rule ID 3 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule2Enabled),
// CHECK-NEXT: // MIs[0] a
// CHECK-NEXT: // No operand predicates
@@ -165,10 +165,10 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: // Combiner Rule #2: InstTest0
// CHECK-NEXT: GIR_CustomAction, GIMT_Encode2(GICXXCustomAction_CombineApplyGICombiner1),
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 8: @487
+// CHECK-NEXT: // Label 8: @503
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 1: @488
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 9*/ GIMT_Encode4(540), // Rule ID 6 //
+// CHECK-NEXT: // Label 1: @504
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 9*/ GIMT_Encode4(556), // Rule ID 6 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule5Enabled),
// CHECK-NEXT: GIM_CheckType, /*MI*/0, /*Op*/2, /*Type*/GILLT_s32,
// CHECK-NEXT: // MIs[0] dst
@@ -186,10 +186,10 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIR_Copy, /*NewInsnID*/0, /*OldInsnID*/1, /*OpIdx*/1, // z
// CHECK-NEXT: GIR_EraseFromParent, /*InsnID*/0,
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 9: @540
+// CHECK-NEXT: // Label 9: @556
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 2: @541
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 10*/ GIMT_Encode4(582), // Rule ID 5 //
+// CHECK-NEXT: // Label 2: @557
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 10*/ GIMT_Encode4(598), // Rule ID 5 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule4Enabled),
// CHECK-NEXT: // MIs[0] tmp
// CHECK-NEXT: GIM_RecordInsnIgnoreCopies, /*DefineMI*/1, /*MI*/0, /*OpIdx*/0, // MIs[1]
@@ -207,32 +207,32 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIR_CustomAction, GIMT_Encode2(GICXXCustomAction_CombineApplyGICombiner2),
// CHECK-NEXT: GIR_EraseFromParent, /*InsnID*/0,
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 10: @582
+// CHECK-NEXT: // Label 10: @598
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 3: @583
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 11*/ GIMT_Encode4(595), // Rule ID 0 //
+// CHECK-NEXT: // Label 3: @599
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 11*/ GIMT_Encode4(611), // Rule ID 0 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule0Enabled),
// CHECK-NEXT: // Combiner Rule #0: WipOpcodeTest0; wip_match_opcode 'G_TRUNC'
// CHECK-NEXT: GIR_CustomAction, GIMT_Encode2(GICXXCustomAction_CombineApplyGICombiner0),
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 11: @595
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 12*/ GIMT_Encode4(607), // Rule ID 1 //
+// CHECK-NEXT: // Label 11: @611
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 12*/ GIMT_Encode4(623), // Rule ID 1 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule1Enabled),
// CHECK-NEXT: // Combiner Rule #1: WipOpcodeTest1; wip_match_opcode 'G_TRUNC'
// CHECK-NEXT: GIR_CustomAction, GIMT_Encode2(GICXXCustomAction_CombineApplyGICombiner0),
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 12: @607
+// CHECK-NEXT: // Label 12: @623
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 4: @608
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 13*/ GIMT_Encode4(620), // Rule ID 2 //
+// CHECK-NEXT: // Label 4: @624
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 13*/ GIMT_Encode4(636), // Rule ID 2 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule1Enabled),
// CHECK-NEXT: // Combiner Rule #1: WipOpcodeTest1; wip_match_opcode 'G_SEXT'
// CHECK-NEXT: GIR_CustomAction, GIMT_Encode2(GICXXCustomAction_CombineApplyGICombiner0),
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 13: @620
+// CHECK-NEXT: // Label 13: @636
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 5: @621
-// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 14*/ GIMT_Encode4(660), // Rule ID 7 //
+// CHECK-NEXT: // Label 5: @637
+// CHECK-NEXT: GIM_Try, /*On fail goto*//*Label 14*/ GIMT_Encode4(676), // Rule ID 7 //
// CHECK-NEXT: GIM_CheckSimplePredicate, GIMT_Encode2(GICXXPred_Simple_IsRule6Enabled),
// CHECK-NEXT: // MIs[0] dst
// CHECK-NEXT: // No operand predicates
@@ -247,10 +247,10 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// CHECK-NEXT: GIR_AddSimpleTempRegister, /*InsnID*/0, /*TempRegID*/0,
// CHECK-NEXT: GIR_EraseFromParent, /*InsnID*/0,
// CHECK-NEXT: GIR_Done,
-// CHECK-NEXT: // Label 14: @660
+// CHECK-NEXT: // Label 14: @676
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: // Label 6: @661
+// CHECK-NEXT: // Label 6: @677
// CHECK-NEXT: GIM_Reject,
-// CHECK-NEXT: }; // Size: 662 bytes
+// CHECK-NEXT: }; // Size: 678 bytes
// CHECK-NEXT: return MatchTable0;
// CHECK-NEXT: }
More information about the llvm-commits
mailing list