llvm · abhishek-kaushik22 · Aug 11, 2025 · Aug 3, 2025 · Aug 6, 2025
diff --git a/llvm/lib/CodeGen/PrologEpilogInserter.cpp b/llvm/lib/CodeGen/PrologEpilogInserter.cpp
@@ -1550,7 +1550,7 @@ void PEIImpl::replaceFrameIndices(MachineBasicBlock *BB, MachineFunction &MF,
       // If this instruction has a FrameIndex operand, we need to
       // use that target machine register info object to eliminate
       // it.
-      TRI.eliminateFrameIndex(MI, SPAdj, i);
+      TRI.eliminateFrameIndex(MI, SPAdj, i, RS);
 
       // Reset the iterator if we were at the beginning of the BB.
       if (AtBeginning) {

diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp
@@ -24,6 +24,7 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
 #include "llvm/CodeGen/WinEHFuncInfo.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/EHPersonalities.h"
@@ -2678,7 +2679,7 @@ StackOffset X86FrameLowering::getFrameIndexReference(const MachineFunction &MF,
   // object.
   // We need to factor in additional offsets applied during the prologue to the
   // frame, base, and stack pointer depending on which is used.
-  int Offset = MFI.getObjectOffset(FI) - getOffsetOfLocalArea();
+  int64_t Offset = MFI.getObjectOffset(FI) - getOffsetOfLocalArea();
   const X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
   unsigned CSSize = X86FI->getCalleeSavedFrameSize();
   uint64_t StackSize = MFI.getStackSize();
@@ -4212,6 +4213,14 @@ void X86FrameLowering::processFunctionBeforeFrameFinalized(
   // emitPrologue if it gets called and emits CFI.
   MF.setHasWinCFI(false);
 
+  MachineFrameInfo &MFI = MF.getFrameInfo();
+  // If the frame is big enough that we might need to scavenge a register to
+  // handle huge offsets, reserve a stack slot for that now.
+  if (!isInt<32>(MFI.estimateStackSize(MF))) {
+    int FI = MFI.CreateStackObject(SlotSize, Align(SlotSize), false);
+    RS->addScavengingFrameIndex(FI);
+  }
+
   // If we are using Windows x64 CFI, ensure that the stack is always 8 byte
   // aligned. The format doesn't support misaligned stack adjustments.
   if (MF.getTarget().getMCAsmInfo()->usesWindowsCFI())

diff --git a/llvm/lib/Target/X86/X86RegisterInfo.cpp b/llvm/lib/Target/X86/X86RegisterInfo.cpp
@@ -13,6 +13,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "X86RegisterInfo.h"
+#include "MCTargetDesc/X86BaseInfo.h"
 #include "X86FrameLowering.h"
 #include "X86MachineFunctionInfo.h"
 #include "X86Subtarget.h"
@@ -21,8 +22,8 @@
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/CodeGen/LiveRegMatrix.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
-#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
 #include "llvm/CodeGen/TargetFrameLowering.h"
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TileShapeInfo.h"
@@ -907,7 +908,7 @@ X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   int FrameIndex = MI.getOperand(FIOperandNum).getIndex();
 
   // Determine base register and offset.
-  int FIOffset;
+  int64_t FIOffset;
   Register BasePtr;
   if (MI.isReturn()) {
     assert((!hasStackRealignment(MF) ||
@@ -958,11 +959,41 @@ X86RegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
   }
 
   if (MI.getOperand(FIOperandNum+3).isImm()) {
-    // Offset is a 32-bit integer.
-    int Imm = (int)(MI.getOperand(FIOperandNum + 3).getImm());
-    int Offset = FIOffset + Imm;
-    assert((!Is64Bit || isInt<32>((long long)FIOffset + Imm)) &&
-           "Requesting 64-bit offset in 32-bit immediate!");
+    const X86InstrInfo *TII = MF.getSubtarget<X86Subtarget>().getInstrInfo();
+    const DebugLoc &DL = MI.getDebugLoc();
+    int64_t Imm = MI.getOperand(FIOperandNum + 3).getImm();
+    int64_t Offset = FIOffset + Imm;
+    bool FitsIn32Bits = isInt<32>(Offset);
+    // If the offset will not fit in a 32-bit displacement, then for 64-bit
+    // targets, scavenge a register to hold it. Otherwise...
+    if (Is64Bit && !FitsIn32Bits) {
+      assert(RS && "RegisterScavenger was NULL");
+
+      RS->enterBasicBlockEnd(MBB);
+      RS->backward(std::next(II));
+
+      Register ScratchReg = RS->scavengeRegisterBackwards(
+          X86::GR64RegClass, II, /*RestoreAfter=*/false, /*SPAdj=*/0,
+          /*AllowSpill=*/true);
+      assert(ScratchReg != 0 && "scratch reg was 0");
+      RS->setRegUsed(ScratchReg);
+
+      BuildMI(MBB, II, DL, TII->get(X86::MOV64ri), ScratchReg).addImm(Offset);
+
+      MI.getOperand(FIOperandNum + 3).setImm(0);
+      MI.getOperand(FIOperandNum + 2).setReg(ScratchReg);
+
+      return false;
+    }
+
+    // ... for 32-bit targets, this is a bug!
+    if (!Is64Bit && !FitsIn32Bits) {
+      MI.emitGenericError("64-bit offset calculated but target is 32-bit");
+      // Trap so that the instruction verification pass does not fail if run.
+      BuildMI(MBB, MBBI, DL, TII->get(X86::TRAP));
+      return false;
+    }
+
     if (Offset != 0 || !tryOptimizeLEAtoMOV(II))
       MI.getOperand(FIOperandNum + 3).ChangeToImmediate(Offset);
   } else {

diff --git a/llvm/lib/Target/X86/X86RegisterInfo.h b/llvm/lib/Target/X86/X86RegisterInfo.h
@@ -13,6 +13,7 @@
 #ifndef LLVM_LIB_TARGET_X86_X86REGISTERINFO_H
 #define LLVM_LIB_TARGET_X86_X86REGISTERINFO_H
 
+#include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/TargetRegisterInfo.h"
 
 #define GET_REGINFO_HEADER
@@ -180,6 +181,10 @@ class X86RegisterInfo final : public X86GenRegisterInfo {
   constrainRegClassToNonRex2(const TargetRegisterClass *RC) const;
 
   bool isNonRex2RegClass(const TargetRegisterClass *RC) const;
+
+  bool requiresRegisterScavenging(const MachineFunction &MF) const override {
+    return true;
+  }
 };
 
 } // End llvm namespace

diff --git a/llvm/test/CodeGen/X86/avx512f-large-stack.ll b/llvm/test/CodeGen/X86/avx512f-large-stack.ll
@@ -0,0 +1,23 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --no_x86_scrub_sp --version 4
+; RUN: llc -O0 -mtriple=x86_64 -mattr=+avx512f -verify-machineinstrs < %s | FileCheck %s --check-prefix=CHECK
+define void @f(i16 %LGV2, i1 %LGV3) {
+; CHECK-LABEL: f:
+; CHECK:       # %bb.0: # %BB
+; CHECK-NEXT:    subq $2147483528, %rsp # imm = 0x7FFFFF88
+; CHECK-NEXT:    .cfi_def_cfa_offset 2147483536
+; CHECK-NEXT:    movb %sil, %cl
+; CHECK-NEXT:    movw %di, %ax
+; CHECK-NEXT:    movswq %ax, %rax
+; CHECK-NEXT:    andb $1, %cl
+; CHECK-NEXT:    movabsq $-2147483768, %rdx # imm = 0xFFFFFFFF7FFFFF88
+; CHECK-NEXT:    movb %cl, (%rsp,%rdx)
+; CHECK-NEXT:    addq $2147483528, %rsp # imm = 0x7FFFFF88
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    retq
+BB:
+  %A = alloca i1, i33 2147483648, align 1
+  %G = getelementptr i1, ptr %A, i16 %LGV2
+  %G4 = getelementptr i1, ptr %G, i32 -2147483648
+  store i1 %LGV3, ptr %G4, align 1
+  ret void
+}
diff --git a/llvm/test/CodeGen/X86/huge-stack.ll b/llvm/test/CodeGen/X86/huge-stack.ll
@@ -5,20 +5,70 @@
 define void @foo() unnamed_addr #0 {
 ; CHECK-LABEL: foo:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    movabsq $8589934462, %rax # imm = 0x1FFFFFF7E
+; CHECK-NEXT:    movabsq $8589934472, %rax # imm = 0x1FFFFFF88
 ; CHECK-NEXT:    subq %rax, %rsp
-; CHECK-NEXT:    .cfi_def_cfa_offset 8589934470
-; CHECK-NEXT:    movb $42, -129(%rsp)
-; CHECK-NEXT:    movb $43, -128(%rsp)
-; CHECK-NEXT:    movabsq $8589934462, %rax # imm = 0x1FFFFFF7E
+; CHECK-NEXT:    .cfi_def_cfa_offset 8589934480
+; CHECK-NEXT:    movabsq $4294967177, %rax # imm = 0xFFFFFF89
+; CHECK-NEXT:    movb $42, (%rsp,%rax)
+; CHECK-NEXT:    movb $43, -118(%rsp)
+; CHECK-NEXT:    movabsq $8589934472, %rax # imm = 0x1FFFFFF88
 ; CHECK-NEXT:    addq %rax, %rsp
 ; CHECK-NEXT:    .cfi_def_cfa_offset 8
 ; CHECK-NEXT:    retq
-  %1 = alloca %large, align 1
-  %2 = alloca %large, align 1
-  %3 = getelementptr inbounds %large, ptr %1, i64 0, i64 0
-  store i8 42, ptr %3, align 1
-  %4 = getelementptr inbounds %large, ptr %2, i64 0, i64 0
-  store i8 43, ptr %4, align 1
+  %large1 = alloca %large, align 1
+  %large2 = alloca %large, align 1
+  %ptrLarge1 = getelementptr inbounds %large, ptr %large1, i64 0, i64 0
+  store i8 42, ptr %ptrLarge1, align 1
+  %ptrLarge2 = getelementptr inbounds %large, ptr %large2, i64 0, i64 0
+  store i8 43, ptr %ptrLarge2, align 1
   ret void
 }
+
+declare ptr @baz(ptr, ptr, ptr, ptr)
+
+define ptr @scavenge_spill() unnamed_addr #0 {
+; CHECK-LABEL: scavenge_spill:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movabsq $25769803816, %rax # imm = 0x600000028
+; CHECK-NEXT:    subq %rax, %rsp
+; CHECK-NEXT:    .cfi_def_cfa_offset 25769803824
+; CHECK-NEXT:    movabsq $21474836521, %rax # imm = 0x500000029
+; CHECK-NEXT:    leaq (%rsp,%rax), %rdi
+; CHECK-NEXT:    movabsq $17179869226, %rax # imm = 0x40000002A
+; CHECK-NEXT:    leaq (%rsp,%rax), %rsi
+; CHECK-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; CHECK-NEXT:    movabsq $12884901931, %rax # imm = 0x30000002B
+; CHECK-NEXT:    leaq (%rsp,%rax), %rdx
+; CHECK-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; CHECK-NEXT:    movabsq $8589934636, %rax # imm = 0x20000002C
+; CHECK-NEXT:    leaq (%rsp,%rax), %rcx
+; CHECK-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; CHECK-NEXT:    callq baz@PLT
+; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
+; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
+; CHECK-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; CHECK-NEXT:    leaq 46(%rsp), %rdi
+; CHECK-NEXT:    callq baz@PLT
+; CHECK-NEXT:    # kill: def $rcx killed $rax
+; CHECK-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; CHECK-NEXT:    movabsq $25769803816, %rcx # imm = 0x600000028
+; CHECK-NEXT:    addq %rcx, %rsp
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    retq
+  %large1 = alloca %large, align 1
+  %ptrLarge1 = getelementptr inbounds %large, ptr %large1, i64 0, i64 0
+  %large2 = alloca %large, align 1
+  %ptrLarge2 = getelementptr inbounds %large, ptr %large2, i64 0, i64 0
+  %large3 = alloca %large, align 1
+  %ptrLarge3 = getelementptr inbounds %large, ptr %large3, i64 0, i64 0
+  %large4 = alloca %large, align 1
+  %ptrLarge4 = getelementptr inbounds %large, ptr %large4, i64 0, i64 0
+  %large5 = alloca %large, align 1
+  %ptrLarge5 = getelementptr inbounds %large, ptr %large5, i64 0, i64 0
+  %ret1 = call ptr @baz(ptr %ptrLarge1, ptr %ptrLarge2, ptr %ptrLarge3, ptr %ptrLarge4)
+  %large6 = alloca %large, align 1
+  %ptrLarge6 = getelementptr inbounds %large, ptr %large6, i64 0, i64 0
+  %ret2 = call ptr @baz(ptr %ptrLarge6, ptr %ptrLarge2, ptr %ptrLarge3, ptr %ptrLarge4)
+  ret ptr %ret1
+}
diff --git a/llvm/test/CodeGen/X86/large-displacements-fastisel.ll b/llvm/test/CodeGen/X86/large-displacements-fastisel.ll
@@ -0,0 +1,18 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=x86_64 -O=0 -verify-machineinstrs | FileCheck %s
+@G = global i8 0
+
+; Regression test for PR113856 - incorrect FastISel assert
+
+define i32 @main() {
+; CHECK-LABEL: main:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    movabsq $-2147483652, %rax # imm = 0xFFFFFFFF7FFFFFFC
+; CHECK-NEXT:    movl $0, (%rsp,%rax)
+; CHECK-NEXT:    xorl %eax, %eax
+; CHECK-NEXT:    retq
+  %1 = alloca i32, align 4
+  %G = getelementptr i8, ptr %1, i32 -2147483648
+  store i32 0, ptr %G, align 4
+  ret i32 0
+}
diff --git a/llvm/test/CodeGen/X86/large-displacements.ll b/llvm/test/CodeGen/X86/large-displacements.ll
@@ -0,0 +1,82 @@
+; RUN: not llc < %s -mtriple=i686 -filetype=null -verify-machineinstrs 2>&1 | FileCheck %s -check-prefix=ERR-i686
+; RUN: llc < %s -mtriple=x86_64 -verify-machineinstrs | FileCheck %s -check-prefix=x86_64
+
+; Regression test for #121932, #113856, #106352, #69365, #25051 which are caused by
+; an incorrectly written assertion for 64-bit offsets when compiling for 32-bit X86.
+
+define i32 @main() #0 {
+; ERR-i686: error: <unknown>:0:0: 64-bit offset calculated but target is 32-bit
+;
+; x86_64-LABEL: main:
+; x86_64:       # %bb.0: # %entry
+; x86_64-NEXT:    movl $4294967192, %eax # imm = 0xFFFFFF98
+; x86_64-NEXT:    subq %rax, %rsp
+; x86_64-NEXT:    .cfi_def_cfa_offset 4294967200
+; x86_64-NEXT:    movabsq $3221225318, %rax # imm = 0xBFFFFF66
+; x86_64-NEXT:    movb $32, (%rsp,%rax)
+; x86_64-NEXT:    movb $33, 2147483494(%rsp)
+; x86_64-NEXT:    movb $34, 1073741670(%rsp)
+; x86_64-NEXT:    movb $35, -154(%rsp)
+; x86_64-NEXT:    xorl %eax, %eax
+; x86_64-NEXT:    movl $4294967192, %ecx # imm = 0xFFFFFF98
+; x86_64-NEXT:    addq %rcx, %rsp
+; x86_64-NEXT:    .cfi_def_cfa_offset 8
+; x86_64-NEXT:    retq
+entry:
+  %a = alloca [1073741824 x i8], align 16
+  %b = alloca [1073741824 x i8], align 16
+  %c = alloca [1073741824 x i8], align 16
+  %d = alloca [1073741824 x i8], align 16
+
+  %arrayida = getelementptr inbounds [1073741824 x i8], ptr %a, i64 0, i64 -42
+  %arrayidb = getelementptr inbounds [1073741824 x i8], ptr %b, i64 0, i64 -42
+  %arrayidc = getelementptr inbounds [1073741824 x i8], ptr %c, i64 0, i64 -42
+  %arrayidd = getelementptr inbounds [1073741824 x i8], ptr %d, i64 0, i64 -42
+
+  store i8 32, ptr %arrayida, align 2
+  store i8 33, ptr %arrayidb, align 2
+  store i8 34, ptr %arrayidc, align 2
+  store i8 35, ptr %arrayidd, align 2
+
+  ret i32 0
+}
+
+; Same test as above but for an anonymous function.
+define i32 @0() #0 {
+; ERR-i686: error: <unknown>:0:0: 64-bit offset calculated but target is 32-bit
+;
+; x86_64-LABEL: __unnamed_1:
+; x86_64:       # %bb.0: # %entry
+; x86_64-NEXT:    movl $4294967192, %eax # imm = 0xFFFFFF98
+; x86_64-NEXT:    subq %rax, %rsp
+; x86_64-NEXT:    .cfi_def_cfa_offset 4294967200
+; x86_64-NEXT:    movabsq $3221225318, %rax # imm = 0xBFFFFF66
+; x86_64-NEXT:    movb $32, (%rsp,%rax)
+; x86_64-NEXT:    movb $33, 2147483494(%rsp)
+; x86_64-NEXT:    movb $34, 1073741670(%rsp)
+; x86_64-NEXT:    movb $35, -154(%rsp)
+; x86_64-NEXT:    xorl %eax, %eax
+; x86_64-NEXT:    movl $4294967192, %ecx # imm = 0xFFFFFF98
+; x86_64-NEXT:    addq %rcx, %rsp
+; x86_64-NEXT:    .cfi_def_cfa_offset 8
+; x86_64-NEXT:    retq
+entry:
+  %a = alloca [1073741824 x i8], align 16
+  %b = alloca [1073741824 x i8], align 16
+  %c = alloca [1073741824 x i8], align 16
+  %d = alloca [1073741824 x i8], align 16
+
+  %arrayida = getelementptr inbounds [1073741824 x i8], ptr %a, i64 0, i64 -42
+  %arrayidb = getelementptr inbounds [1073741824 x i8], ptr %b, i64 0, i64 -42
+  %arrayidc = getelementptr inbounds [1073741824 x i8], ptr %c, i64 0, i64 -42
+  %arrayidd = getelementptr inbounds [1073741824 x i8], ptr %d, i64 0, i64 -42
+
+  store i8 32, ptr %arrayida, align 2
+  store i8 33, ptr %arrayidb, align 2
+  store i8 34, ptr %arrayidc, align 2
+  store i8 35, ptr %arrayidd, align 2
+
+  ret i32 0
+}
+
+attributes #0 = { optnone noinline }
diff --git a/llvm/test/CodeGen/X86/merge-huge-sp-updates.ll b/llvm/test/CodeGen/X86/merge-huge-sp-updates.ll
@@ -22,8 +22,8 @@ entry:
   call void @bar(i64 0, i64 0, i64 0, i64 0, i64 0, ptr null, ptr %rhs, ptr null, ptr %rhs)
 ; CHECK: call{{.*}}bar
 ; CHECK: addq{{.*}}$2147483647, %rsp
-; CHECK: addq{{.*}}$372037585, %rsp
-; CHECK: .cfi_adjust_cfa_offset -2519521232
+; CHECK: addq{{.*}}$372037601, %rsp
+; CHECK: .cfi_adjust_cfa_offset -2519521248
   ret void
 }