[llvm-branch-commits] [llvm] [CodeGen] Add FREEZE target opcode (PR #240469)
Mitch Briles via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Mon Nov 9 13:47:01 PST 2026
https://github.com/MitchBriles updated https://github.com/llvm/llvm-project/pull/240469
>From cde7950690badf6c03ece4e8892a1238dcda9abf Mon Sep 17 00:00:00 2001
From: Mitch <mitchbriles at gmail.com>
Date: Fri, 23 Oct 2026 09:50:48 -0600
Subject: [PATCH] [CodeGen] Add FREEZE target opcode
Machine IR equivalent of `freeze` instruction. Unlike COPY,
ProcessImplicitDefs does not turn a `FREEZE` of an `IMPLICIT_DEF` into an
`IMPLICIT_DEF` and the register coalescer does not join it.
After register allocation, a `FREEZE` of a defined value is lowered to a
COPY. A `FREEZE` of an undefined value is kept until code emission, so
that no later pass sees the register as undefined. It emits a comment in
verbose asm (like `IMPLICIT_DEF`).
This PR doesn't select `freeze` -> `FREEZE` yet because there's some work to
do to avoid regressions.
Assisted-By: Claude Opus 5.5 <noreply at anthropic.com>
---
llvm/docs/CodeGenerator.md | 11 +-
llvm/include/llvm/Support/TargetOpcodes.def | 4 +
llvm/include/llvm/Target/Target.td | 8 ++
llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp | 16 +++
llvm/lib/CodeGen/ExpandPostRAPseudos.cpp | 6 +
llvm/lib/CodeGen/MachineVerifier.cpp | 4 +
llvm/lib/Target/Hexagon/HexagonAsmPrinter.cpp | 3 +-
.../Target/Hexagon/HexagonVLIWPacketizer.cpp | 10 +-
llvm/lib/Target/SystemZ/SystemZLongBranch.cpp | 3 +-
llvm/test/CodeGen/AMDGPU/hazard-freeze.mir | 22 +++
.../CodeGen/Mips/delay-slot-filler-freeze.mir | 21 +++
llvm/test/CodeGen/X86/freeze-pseudo-asm.mir | 20 +++
.../test/CodeGen/X86/freeze-pseudo-expand.mir | 29 ++++
llvm/test/CodeGen/X86/freeze-pseudo.mir | 134 ++++++++++++++++++
llvm/test/MachineVerifier/test_freeze.mir | 17 +++
.../match-table-imms.td | 4 +-
llvm/test/TableGen/get-named-operand-idx.td | 44 +++---
17 files changed, 323 insertions(+), 33 deletions(-)
create mode 100644 llvm/test/CodeGen/AMDGPU/hazard-freeze.mir
create mode 100644 llvm/test/CodeGen/Mips/delay-slot-filler-freeze.mir
create mode 100644 llvm/test/CodeGen/X86/freeze-pseudo-asm.mir
create mode 100644 llvm/test/CodeGen/X86/freeze-pseudo-expand.mir
create mode 100644 llvm/test/CodeGen/X86/freeze-pseudo.mir
create mode 100644 llvm/test/MachineVerifier/test_freeze.mir
diff --git a/llvm/docs/CodeGenerator.md b/llvm/docs/CodeGenerator.md
index cd58ae7b006fd4..227b16fe9b01db 100644
--- a/llvm/docs/CodeGenerator.md
+++ b/llvm/docs/CodeGenerator.md
@@ -464,7 +464,7 @@ are no virtual registers left in the code.
#### Undefined values, poison, and freeze
After instruction selection, machine code has no exact equivalent of LLVM IR's
-`undef`, `poison`, or `freeze`.
+`undef` or `poison`.
`IMPLICIT_DEF` defines a register without assigning a value. Currently, both
`undef` and `poison` select to it in SelectionDAG and GlobalISel. Before
@@ -474,7 +474,14 @@ marks a register use as having a value that doesn't matter. Different
instructions with `undef` uses of the same register may observe different
values.
-Selectors lower `freeze` to a `COPY`. If the copied value is undefined,
+`FREEZE` gives undefined bits an arbitrary, but fixed value, so every use of its
+result observes the same value. It isn't generally treated as copy-like, so
+`ProcessImplicitDefs` and the register coalescer don't look through it. After
+register allocation, a `FREEZE` of a defined value becomes a `COPY`, and a
+`FREEZE` of an undefined value stays until code emission, where it emits no
+code.
+
+Selectors still lower `freeze` to a `COPY`. If the copied value is undefined,
different uses can observe different values, which violates the IR semantics of
`freeze`.
diff --git a/llvm/include/llvm/Support/TargetOpcodes.def b/llvm/include/llvm/Support/TargetOpcodes.def
index d9f18772e1a46c..852ab8f3c380e8 100644
--- a/llvm/include/llvm/Support/TargetOpcodes.def
+++ b/llvm/include/llvm/Support/TargetOpcodes.def
@@ -63,6 +63,10 @@ HANDLE_TARGET_OPCODE(IMPLICIT_DEF)
/// early-clobber result operand.
HANDLE_TARGET_OPCODE(INIT_UNDEF)
+/// FREEZE - Returns an arbitrary, but fixed value of the same type when the
+/// input is undefined/poison. Otherwise, copies the input.
+HANDLE_TARGET_OPCODE(FREEZE)
+
/// SUBREG_TO_REG - Expose the value of bits in a super register.
/// This is typically used after an instruction which writes its result to a
/// subregister but also, as a side effect, writes some value (often zero) into
diff --git a/llvm/include/llvm/Target/Target.td b/llvm/include/llvm/Target/Target.td
index d70e8fa9918797..b201ce2a379db3 100644
--- a/llvm/include/llvm/Target/Target.td
+++ b/llvm/include/llvm/Target/Target.td
@@ -1295,6 +1295,14 @@ def INIT_UNDEF : StandardPseudoInstruction {
let hasSideEffects = false;
let Size = 0;
}
+def FREEZE : StandardPseudoInstruction {
+ let OutOperandList = (outs unknown:$dst);
+ let InOperandList = (ins unknown:$src);
+ let AsmString = "";
+ let hasSideEffects = false;
+ let Size = 0;
+ let isMeta = true;
+}
def SUBREG_TO_REG : StandardPseudoInstruction {
let OutOperandList = (outs unknown:$dst);
let InOperandList = (ins unknown:$subsrc, i32imm:$subidx);
diff --git a/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp b/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
index 37822e36d6ed1f..b17125ef945b31 100644
--- a/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+++ b/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
@@ -1254,6 +1254,16 @@ void AsmPrinter::emitImplicitDef(const MachineInstr *MI) const {
OutStreamer->addBlankLine();
}
+static void emitFreeze(const MachineInstr *MI, AsmPrinter &AP) {
+ std::string Str;
+ raw_string_ostream OS(Str);
+ OS << "freeze: "
+ << printReg(MI->getOperand(0).getReg(),
+ AP.MF->getSubtarget().getRegisterInfo());
+ AP.OutStreamer->AddComment(Str);
+ AP.OutStreamer->addBlankLine();
+}
+
static void emitKill(const MachineInstr *MI, AsmPrinter &AP) {
std::string Str;
raw_string_ostream OS(Str);
@@ -2267,6 +2277,12 @@ void AsmPrinter::emitFunctionBody() {
// This is only used to influence register allocation behavior, no
// actual initialization is needed.
break;
+ case TargetOpcode::FREEZE:
+ // Only a FREEZE of an undefined value is left. Whatever the register
+ // holds is a valid frozen value.
+ if (isVerbose())
+ emitFreeze(&MI, *this);
+ break;
case TargetOpcode::RELOC_NONE: {
// Generate a temporary label for the current PC.
MCSymbol *Sym = OutContext.createTempSymbol("reloc_none");
diff --git a/llvm/lib/CodeGen/ExpandPostRAPseudos.cpp b/llvm/lib/CodeGen/ExpandPostRAPseudos.cpp
index fb65500e7dc7f8..4e7184a8f83aa1 100644
--- a/llvm/lib/CodeGen/ExpandPostRAPseudos.cpp
+++ b/llvm/lib/CodeGen/ExpandPostRAPseudos.cpp
@@ -145,6 +145,12 @@ bool ExpandPostRA::run(MachineFunction &MF) {
case TargetOpcode::SUBREG_TO_REG:
MadeChange |= LowerSubregToReg(&MI);
break;
+ case TargetOpcode::FREEZE:
+ // A FREEZE of an undefined value stays a definition until emission.
+ if (MI.getOperand(1).isUndef())
+ continue;
+ MI.setDesc(TII->get(TargetOpcode::COPY));
+ [[fallthrough]];
case TargetOpcode::COPY:
TII->lowerCopy(&MI);
MadeChange = true;
diff --git a/llvm/lib/CodeGen/MachineVerifier.cpp b/llvm/lib/CodeGen/MachineVerifier.cpp
index 6e3504e2f460b7..e9dd71303f2528 100644
--- a/llvm/lib/CodeGen/MachineVerifier.cpp
+++ b/llvm/lib/CodeGen/MachineVerifier.cpp
@@ -2488,6 +2488,10 @@ void MachineVerifier::visitMachineInstrBefore(const MachineInstr *MI) {
}
break;
}
+ case TargetOpcode::FREEZE:
+ if (!MI->getOperand(0).isReg() || !MI->getOperand(1).isReg())
+ report("FREEZE operands must be registers", MI);
+ break;
case TargetOpcode::COPY_LANEMASK: {
const MachineOperand &DstOp = MI->getOperand(0);
const MachineOperand &SrcOp = MI->getOperand(1);
diff --git a/llvm/lib/Target/Hexagon/HexagonAsmPrinter.cpp b/llvm/lib/Target/Hexagon/HexagonAsmPrinter.cpp
index 143cee16477220..8ef0ae51e9ade5 100644
--- a/llvm/lib/Target/Hexagon/HexagonAsmPrinter.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonAsmPrinter.cpp
@@ -765,7 +765,8 @@ void HexagonAsmPrinter::emitInstruction(const MachineInstr *MI) {
MachineBasicBlock::const_instr_iterator MII = MI->getIterator();
for (++MII; MII != MBB->instr_end() && MII->isInsideBundle(); ++MII)
- if (!MII->isDebugInstr() && !MII->isImplicitDef())
+ if (!MII->isDebugInstr() && !MII->isImplicitDef() &&
+ MII->getOpcode() != TargetOpcode::FREEZE)
HexagonLowerToMC(MCII, &*MII, MCB, *this);
} else {
HexagonLowerToMC(MCII, MI, MCB, *this);
diff --git a/llvm/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp b/llvm/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
index ae50cd17ccb7af..3c4b320828f6cc 100644
--- a/llvm/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
+++ b/llvm/lib/Target/Hexagon/HexagonVLIWPacketizer.cpp
@@ -850,9 +850,9 @@ bool HexagonPacketizerList::canPromoteToDotNew(const MachineInstr &MI,
if (PI.isInlineAsm())
return false;
- // IMPLICIT_DEFs won't materialize as real instructions, so .new makes no
- // sense.
- if (PI.isImplicitDef())
+ // IMPLICIT_DEFs and FREEZEs won't materialize as real instructions, so .new
+ // makes no sense.
+ if (PI.isImplicitDef() || PI.getOpcode() == TargetOpcode::FREEZE)
return false;
// If dependency is through an implicitly defined register, we should not
@@ -1040,7 +1040,7 @@ bool HexagonPacketizerList::ignorePseudoInstruction(const MachineInstr &MI,
if (MI.isInlineAsm())
return false;
- if (MI.isImplicitDef())
+ if (MI.isImplicitDef() || MI.getOpcode() == TargetOpcode::FREEZE)
return false;
// We check if MI has any functional units mapped to it. If it doesn't,
@@ -1707,7 +1707,7 @@ HexagonPacketizerList::addToPacket(MachineInstr &MI) {
PacketStalls |= producesStall(MI);
PacketStallCycles = std::max(PacketStallCycles, calcStall(MI));
- if (MI.isImplicitDef()) {
+ if (MI.isImplicitDef() || MI.getOpcode() == TargetOpcode::FREEZE) {
// Add to the packet to allow subsequent instructions to be checked
// properly.
CurrentPacketMIs.push_back(&MI);
diff --git a/llvm/lib/Target/SystemZ/SystemZLongBranch.cpp b/llvm/lib/Target/SystemZ/SystemZLongBranch.cpp
index ce05d7962b4929..7152df66c0ba46 100644
--- a/llvm/lib/Target/SystemZ/SystemZLongBranch.cpp
+++ b/llvm/lib/Target/SystemZ/SystemZLongBranch.cpp
@@ -215,7 +215,8 @@ static unsigned getInstSizeInBytes(const MachineInstr &MI,
// These do not have a size:
MI.isDebugOrPseudoInstr() || MI.isPosition() || MI.isKill() ||
MI.isImplicitDef() || MI.getOpcode() == TargetOpcode::MEMBARRIER ||
- MI.getOpcode() == TargetOpcode::INIT_UNDEF || MI.isFakeUse() ||
+ MI.getOpcode() == TargetOpcode::INIT_UNDEF ||
+ MI.getOpcode() == TargetOpcode::FREEZE || MI.isFakeUse() ||
MI.getOpcode() == TargetOpcode::RELOC_NONE ||
MI.getOpcode() == SystemZ::FENCE ||
// These have a size that may be zero:
diff --git a/llvm/test/CodeGen/AMDGPU/hazard-freeze.mir b/llvm/test/CodeGen/AMDGPU/hazard-freeze.mir
new file mode 100644
index 00000000000000..c174d2c77c8efb
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/hazard-freeze.mir
@@ -0,0 +1,22 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=amdgcn -mcpu=gfx906 -run-pass=post-RA-hazard-rec %s -o - | FileCheck %s
+
+---
+name: store_data_hazard_freeze
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+
+ ; CHECK-LABEL: name: store_data_hazard_freeze
+ ; CHECK: liveins: $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: GLOBAL_STORE_DWORDX4 $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit $exec
+ ; CHECK-NEXT: $vgpr6 = FREEZE undef $vgpr6
+ ; CHECK-NEXT: S_NOP 0
+ ; CHECK-NEXT: $vgpr2 = V_MOV_B32_e32 0, implicit $exec
+ ; CHECK-NEXT: S_ENDPGM 0, implicit $vgpr6
+ GLOBAL_STORE_DWORDX4 $vgpr0_vgpr1, $vgpr2_vgpr3_vgpr4_vgpr5, 0, 0, implicit $exec
+ $vgpr6 = FREEZE undef $vgpr6
+ $vgpr2 = V_MOV_B32_e32 0, implicit $exec
+ S_ENDPGM 0, implicit $vgpr6
+...
diff --git a/llvm/test/CodeGen/Mips/delay-slot-filler-freeze.mir b/llvm/test/CodeGen/Mips/delay-slot-filler-freeze.mir
new file mode 100644
index 00000000000000..ce5cdbb43bedbd
--- /dev/null
+++ b/llvm/test/CodeGen/Mips/delay-slot-filler-freeze.mir
@@ -0,0 +1,21 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=mipsel -run-pass=mips-delay-slot-filler %s -o - | FileCheck %s
+
+---
+name: freeze_not_in_delay_slot
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $r4, $r31
+
+ ; CHECK-LABEL: name: freeze_not_in_delay_slot
+ ; CHECK: liveins: $r4, $r31
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $r2 = FREEZE undef $r2
+ ; CHECK-NEXT: PseudoReturn undef $r31, implicit $r2, implicit $r3 {
+ ; CHECK-NEXT: $r3 = ADDiu $r4, 1
+ ; CHECK-NEXT: }
+ $r3 = ADDiu $r4, 1
+ $r2 = FREEZE undef $r2
+ PseudoReturn undef $r31, implicit $r2, implicit $r3
+...
diff --git a/llvm/test/CodeGen/X86/freeze-pseudo-asm.mir b/llvm/test/CodeGen/X86/freeze-pseudo-asm.mir
new file mode 100644
index 00000000000000..7d168a05a556cb
--- /dev/null
+++ b/llvm/test/CodeGen/X86/freeze-pseudo-asm.mir
@@ -0,0 +1,20 @@
+# NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=x86_64-- -start-before=post-ra-pseudos -o - %s | FileCheck %s
+
+--- |
+ define i32 @freeze_undef_reg() {
+ ; CHECK-LABEL: freeze_undef_reg:
+ ; CHECK: # %bb.0:
+ ; CHECK-NEXT: # freeze: $eax
+ ; CHECK-NEXT: retq
+ ret i32 0
+ }
+...
+---
+name: freeze_undef_reg
+tracksRegLiveness: true
+body: |
+ bb.0:
+ $eax = FREEZE undef $ecx
+ RET 0, implicit $eax
+...
diff --git a/llvm/test/CodeGen/X86/freeze-pseudo-expand.mir b/llvm/test/CodeGen/X86/freeze-pseudo-expand.mir
new file mode 100644
index 00000000000000..cb7d415823baa9
--- /dev/null
+++ b/llvm/test/CodeGen/X86/freeze-pseudo-expand.mir
@@ -0,0 +1,29 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=x86_64-- -run-pass=post-ra-pseudos -o - %s | FileCheck %s
+
+---
+name: freeze_reg
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $edi
+
+ ; CHECK-LABEL: name: freeze_reg
+ ; CHECK: liveins: $edi
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: $eax = MOV32rr $edi
+ ; CHECK-NEXT: RET 0, implicit $eax
+ $eax = FREEZE $edi
+ RET 0, implicit $eax
+...
+---
+name: freeze_undef_reg
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; CHECK-LABEL: name: freeze_undef_reg
+ ; CHECK: $eax = FREEZE undef $ecx
+ ; CHECK-NEXT: RET 0, implicit $eax
+ $eax = FREEZE undef $ecx
+ RET 0, implicit $eax
+...
diff --git a/llvm/test/CodeGen/X86/freeze-pseudo.mir b/llvm/test/CodeGen/X86/freeze-pseudo.mir
new file mode 100644
index 00000000000000..56f37c9d97f413
--- /dev/null
+++ b/llvm/test/CodeGen/X86/freeze-pseudo.mir
@@ -0,0 +1,134 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
+# RUN: llc -mtriple=x86_64-- -run-pass=process-implicit-defs -o - %s | FileCheck %s --check-prefix=IMPDEF
+# RUN: llc -mtriple=x86_64-- -run-pass=process-implicit-defs,two-address-instruction,register-coalescer,greedy,virt-reg-rewriter -o - %s | FileCheck %s --check-prefix=RA
+
+# A COPY of an IMPLICIT_DEF is undefined, so each use may read a different
+# register. Every use of a FREEZE of an IMPLICIT_DEF reads the same register.
+
+---
+name: copy_undef
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; IMPDEF-LABEL: name: copy_undef
+ ; IMPDEF: RET 0, implicit undef $eax, implicit undef $ecx
+ ;
+ ; RA-LABEL: name: copy_undef
+ ; RA: RET 0, implicit undef $eax, implicit undef $ecx
+ %0:gr32 = IMPLICIT_DEF
+ %1:gr32 = COPY %0
+ $eax = COPY %1
+ $ecx = COPY %1
+ RET 0, implicit $eax, implicit $ecx
+...
+---
+name: freeze_undef
+tracksRegLiveness: true
+body: |
+ bb.0:
+ ; IMPDEF-LABEL: name: freeze_undef
+ ; IMPDEF: [[FREEZE:%[0-9]+]]:gr32 = FREEZE undef %0:gr32
+ ; IMPDEF-NEXT: $eax = COPY [[FREEZE]]
+ ; IMPDEF-NEXT: $ecx = COPY [[FREEZE]]
+ ; IMPDEF-NEXT: RET 0, implicit $eax, implicit $ecx
+ ;
+ ; RA-LABEL: name: freeze_undef
+ ; RA: renamable $eax = FREEZE undef renamable $eax
+ ; RA-NEXT: $ecx = COPY renamable $eax
+ ; RA-NEXT: RET 0, implicit $eax, implicit $ecx
+ %0:gr32 = IMPLICIT_DEF
+ %1:gr32 = FREEZE %0
+ $eax = COPY %1
+ $ecx = COPY %1
+ RET 0, implicit $eax, implicit $ecx
+...
+---
+name: freeze_defined
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $edi
+
+ ; IMPDEF-LABEL: name: freeze_defined
+ ; IMPDEF: liveins: $edi
+ ; IMPDEF-NEXT: {{ $}}
+ ; IMPDEF-NEXT: [[COPY:%[0-9]+]]:gr32 = COPY $edi
+ ; IMPDEF-NEXT: [[FREEZE:%[0-9]+]]:gr32 = FREEZE [[COPY]]
+ ; IMPDEF-NEXT: $eax = COPY [[FREEZE]]
+ ; IMPDEF-NEXT: $ecx = COPY [[FREEZE]]
+ ; IMPDEF-NEXT: RET 0, implicit $eax, implicit $ecx
+ ;
+ ; RA-LABEL: name: freeze_defined
+ ; RA: liveins: $edi
+ ; RA-NEXT: {{ $}}
+ ; RA-NEXT: renamable $eax = FREEZE killed renamable $edi
+ ; RA-NEXT: $ecx = COPY renamable $eax
+ ; RA-NEXT: RET 0, implicit $eax, implicit $ecx
+ %0:gr32 = COPY $edi
+ %1:gr32 = FREEZE %0
+ $eax = COPY %1
+ $ecx = COPY %1
+ RET 0, implicit $eax, implicit $ecx
+...
+---
+name: freeze_undef_cross_block
+tracksRegLiveness: true
+body: |
+ ; IMPDEF-LABEL: name: freeze_undef_cross_block
+ ; IMPDEF: bb.0:
+ ; IMPDEF-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
+ ; IMPDEF-NEXT: liveins: $edi
+ ; IMPDEF-NEXT: {{ $}}
+ ; IMPDEF-NEXT: [[COPY:%[0-9]+]]:gr32 = COPY $edi
+ ; IMPDEF-NEXT: [[FREEZE:%[0-9]+]]:gr32 = FREEZE undef %1:gr32
+ ; IMPDEF-NEXT: CMP32rr [[FREEZE]], [[COPY]], implicit-def $eflags
+ ; IMPDEF-NEXT: JCC_1 %bb.2, 4, implicit $eflags
+ ; IMPDEF-NEXT: JMP_1 %bb.1
+ ; IMPDEF-NEXT: {{ $}}
+ ; IMPDEF-NEXT: bb.1:
+ ; IMPDEF-NEXT: $eax = COPY [[FREEZE]]
+ ; IMPDEF-NEXT: RET 0, implicit $eax
+ ; IMPDEF-NEXT: {{ $}}
+ ; IMPDEF-NEXT: bb.2:
+ ; IMPDEF-NEXT: $eax = COPY [[COPY]]
+ ; IMPDEF-NEXT: RET 0, implicit $eax
+ ;
+ ; RA-LABEL: name: freeze_undef_cross_block
+ ; RA: bb.0:
+ ; RA-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
+ ; RA-NEXT: liveins: $edi
+ ; RA-NEXT: {{ $}}
+ ; RA-NEXT: renamable $eax = FREEZE undef renamable $eax
+ ; RA-NEXT: CMP32rr renamable $eax, renamable $edi, implicit-def $eflags
+ ; RA-NEXT: JCC_1 %bb.2, 4, implicit $eflags
+ ; RA-NEXT: JMP_1 %bb.1
+ ; RA-NEXT: {{ $}}
+ ; RA-NEXT: bb.1:
+ ; RA-NEXT: liveins: $eax
+ ; RA-NEXT: {{ $}}
+ ; RA-NEXT: RET 0, implicit $eax
+ ; RA-NEXT: {{ $}}
+ ; RA-NEXT: bb.2:
+ ; RA-NEXT: liveins: $edi
+ ; RA-NEXT: {{ $}}
+ ; RA-NEXT: $eax = COPY killed renamable $edi
+ ; RA-NEXT: RET 0, implicit $eax
+ bb.0:
+ successors: %bb.1, %bb.2
+ liveins: $edi
+
+ %0:gr32 = COPY $edi
+ %1:gr32 = IMPLICIT_DEF
+ %2:gr32 = FREEZE %1
+ CMP32rr %2, %0, implicit-def $eflags
+ JCC_1 %bb.2, 4, implicit $eflags
+ JMP_1 %bb.1
+
+ bb.1:
+ $eax = COPY %2
+ RET 0, implicit $eax
+
+ bb.2:
+ $eax = COPY %0
+ RET 0, implicit $eax
+...
diff --git a/llvm/test/MachineVerifier/test_freeze.mir b/llvm/test/MachineVerifier/test_freeze.mir
new file mode 100644
index 00000000000000..aa736438b4e288
--- /dev/null
+++ b/llvm/test/MachineVerifier/test_freeze.mir
@@ -0,0 +1,17 @@
+# RUN: not --crash llc -mtriple=x86_64 -run-pass=none -filetype=null %s 2>&1 | FileCheck %s -implicit-check-not="Bad machine code"
+# REQUIRES: x86-registered-target
+
+---
+name: test_freeze
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $edi
+
+ %0:gr32 = COPY $edi
+ %1:gr32 = FREEZE %0
+
+ ; CHECK: Bad machine code: FREEZE operands must be registers
+ ; CHECK: - instruction: %2:gr32 = FREEZE 0
+ %2:gr32 = FREEZE 0
+...
diff --git a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
index e20c0440726dd0..c4e3fffc3b96cf 100644
--- a/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
+++ b/llvm/test/TableGen/GlobalISelCombinerEmitter/match-table-imms.td
@@ -87,7 +87,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// tell them apart and assumes IEEEhalf.
// DEFAULT: const uint8_t *GenMyCombiner::getMatchTable() const {
// DEFAULT-NEXT: constexpr static uint8_t MatchTable0[] = {
-// DEFAULT-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(20), GIMT_Encode2({{[0-9]+}}), /*)*//*default:*//*Label 6*/ GIMT_Encode4([[LBL6:[0-9]+]]),
+// DEFAULT-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(21), GIMT_Encode2({{[0-9]+}}), /*)*//*default:*//*Label 6*/ GIMT_Encode4([[LBL6:[0-9]+]]),
// DEFAULT-NEXT: /*TargetOpcode::COPY*//*Label 0*/ GIMT_Encode4([[LBL0:[0-9]+]]), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
// DEFAULT-NEXT: /*TargetOpcode::G_CONSTANT*//*Label 1*/ GIMT_Encode4([[LBL1:[0-9]+]]),
// DEFAULT-NEXT: /*TargetOpcode::G_FCONSTANT*//*Label 2*/ GIMT_Encode4([[LBL2:[0-9]+]]), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
@@ -247,7 +247,7 @@ def MyCombiner: GICombiner<"GenMyCombiner", [
// EXTLLT-NEXT: };
// EXTLLT: const uint8_t *GenMyCombiner::getMatchTable() const {
// EXTLLT-NEXT: constexpr static uint8_t MatchTable0[] = {
-// EXTLLT-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(20), GIMT_Encode2({{[0-9]+}}), /*)*//*default:*//*Label 6*/ GIMT_Encode4([[LBL6:[0-9]+]]),
+// EXTLLT-NEXT: GIM_SwitchOpcode, /*MI*/0, /*[*/GIMT_Encode2(21), GIMT_Encode2({{[0-9]+}}), /*)*//*default:*//*Label 6*/ GIMT_Encode4([[LBL6:[0-9]+]]),
// EXTLLT-NEXT: /*TargetOpcode::COPY*//*Label 0*/ GIMT_Encode4([[LBL0:[0-9]+]]), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
// EXTLLT-NEXT: /*TargetOpcode::G_CONSTANT*//*Label 1*/ GIMT_Encode4([[LBL1:[0-9]+]]),
// EXTLLT-NEXT: /*TargetOpcode::G_FCONSTANT*//*Label 2*/ GIMT_Encode4([[LBL2:[0-9]+]]), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0), GIMT_Encode4(0),
diff --git a/llvm/test/TableGen/get-named-operand-idx.td b/llvm/test/TableGen/get-named-operand-idx.td
index 165d3f705c04c0..80f0f8b66b8ab0 100644
--- a/llvm/test/TableGen/get-named-operand-idx.td
+++ b/llvm/test/TableGen/get-named-operand-idx.td
@@ -78,28 +78,28 @@ defm : RemapAllTargetPseudoPointerOperands<RegClass>;
// CHECK-EMPTY:
// CHECK-NEXT: LLVM_READONLY static uint8_t getInstructionIndexForOpLookup(uint32_t Opcode) {
// CHECK-NEXT: static constexpr uint8_t InstructionIndex[] = {
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
-// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
-// CHECK-NEXT: 2, 2, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+// CHECK-NEXT: 1, 2, 2, 0,
// CHECK-NEXT: };
// CHECK-NEXT: return InstructionIndex[Opcode];
// CHECK-NEXT: }
More information about the llvm-branch-commits
mailing list