[llvm] [AMDGPU][MIR] Serialize NumPhysicalVGPRSpillLanes (PR #115291)

[Akshat Oke via llvm-commits]

https://github.com/optimisan updated https://github.com/llvm/llvm-project/pull/115291

>From 88441d1c452908d3add3dd4f60577dcd43444975 Mon Sep 17 00:00:00 2001
From: Akshat Oke <Akshat.Oke at amd.com>
Date: Thu, 7 Nov 2024 10:37:54 +0000
Subject: [PATCH] [AMDGPU][MIR] Serialize NumPhysicalVGPRSpillLanes

---
 llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp  |  9 +++++----
 llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h    |  7 +++++++
 .../MIR/AMDGPU/num-phys-vgpr-spill-lanes.ll       | 15 +++++++++++++++
 3 files changed, 27 insertions(+), 4 deletions(-)
 create mode 100644 llvm/test/CodeGen/MIR/AMDGPU/num-phys-vgpr-spill-lanes.ll

diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 1e43d2727a00da..2ddcd5a799cb10 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -694,22 +694,22 @@ yaml::SIMachineFunctionInfo::SIMachineFunctionInfo(
     const llvm::MachineFunction &MF)
     : ExplicitKernArgSize(MFI.getExplicitKernArgSize()),
       MaxKernArgAlign(MFI.getMaxKernArgAlign()), LDSSize(MFI.getLDSSize()),
-      GDSSize(MFI.getGDSSize()),
-      DynLDSAlign(MFI.getDynLDSAlign()), IsEntryFunction(MFI.isEntryFunction()),
+      GDSSize(MFI.getGDSSize()), DynLDSAlign(MFI.getDynLDSAlign()),
+      IsEntryFunction(MFI.isEntryFunction()),
       NoSignedZerosFPMath(MFI.hasNoSignedZerosFPMath()),
       MemoryBound(MFI.isMemoryBound()), WaveLimiter(MFI.needsWaveLimiter()),
       HasSpilledSGPRs(MFI.hasSpilledSGPRs()),
       HasSpilledVGPRs(MFI.hasSpilledVGPRs()),
       HighBitsOf32BitAddress(MFI.get32BitAddressHighBits()),
       Occupancy(MFI.getOccupancy()),
+      NumPhysicalVGPRSpillLanes(MFI.getNumPhysicalVGPRSpillLanes()),
       ScratchRSrcReg(regToString(MFI.getScratchRSrcReg(), TRI)),
       FrameOffsetReg(regToString(MFI.getFrameOffsetReg(), TRI)),
       StackPtrOffsetReg(regToString(MFI.getStackPtrOffsetReg(), TRI)),
       BytesInStackArgArea(MFI.getBytesInStackArgArea()),
       ReturnsVoid(MFI.returnsVoid()),
       ArgInfo(convertArgumentInfo(MFI.getArgInfo(), TRI)),
-      PSInputAddr(MFI.getPSInputAddr()),
-      PSInputEnable(MFI.getPSInputEnable()),
+      PSInputAddr(MFI.getPSInputAddr()), PSInputEnable(MFI.getPSInputEnable()),
       Mode(MFI.getMode()) {
   for (Register Reg : MFI.getSGPRSpillPhysVGPRs())
     SpillPhysVGPRS.push_back(regToString(Reg, TRI));
@@ -754,6 +754,7 @@ bool SIMachineFunctionInfo::initializeBaseYamlFields(
   HasSpilledVGPRs = YamlMFI.HasSpilledVGPRs;
   BytesInStackArgArea = YamlMFI.BytesInStackArgArea;
   ReturnsVoid = YamlMFI.ReturnsVoid;
+  NumPhysicalVGPRSpillLanes = YamlMFI.NumPhysicalVGPRSpillLanes;
 
   if (YamlMFI.ScavengeFI) {
     auto FIOrErr = YamlMFI.ScavengeFI->getFI(MF.getFrameInfo());
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
index 018322eaa18665..db4aefadaee3e5 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
@@ -275,6 +275,7 @@ struct SIMachineFunctionInfo final : public yaml::MachineFunctionInfo {
   // TODO: 10 may be a better default since it's the maximum.
   unsigned Occupancy = 0;
 
+  unsigned NumPhysicalVGPRSpillLanes = 0;
   SmallVector<StringValue, 2> SpillPhysVGPRS;
   SmallVector<StringValue> WWMReservedRegs;
 
@@ -337,6 +338,8 @@ template <> struct MappingTraits<SIMachineFunctionInfo> {
     YamlIO.mapOptional("highBitsOf32BitAddress",
                        MFI.HighBitsOf32BitAddress, 0u);
     YamlIO.mapOptional("occupancy", MFI.Occupancy, 0);
+    YamlIO.mapOptional("numPhysicalVGPRSpillLanes",
+                       MFI.NumPhysicalVGPRSpillLanes);
     YamlIO.mapOptional("spillPhysVGPRs", MFI.SpillPhysVGPRS);
     YamlIO.mapOptional("wwmReservedRegs", MFI.WWMReservedRegs);
     YamlIO.mapOptional("scavengeFI", MFI.ScavengeFI);
@@ -614,6 +617,10 @@ class SIMachineFunctionInfo final : public AMDGPUMachineFunction,
   ArrayRef<Register> getSGPRSpillVGPRs() const { return SpillVGPRs; }
   ArrayRef<Register> getSGPRSpillPhysVGPRs() const { return SpillPhysVGPRs; }
 
+  unsigned getNumPhysicalVGPRSpillLanes() const {
+    return NumPhysicalVGPRSpillLanes;
+  }
+
   const WWMSpillsMap &getWWMSpills() const { return WWMSpills; }
   const ReservedRegSet &getWWMReservedRegs() const { return WWMReservedRegs; }
 
diff --git a/llvm/test/CodeGen/MIR/AMDGPU/num-phys-vgpr-spill-lanes.ll b/llvm/test/CodeGen/MIR/AMDGPU/num-phys-vgpr-spill-lanes.ll
new file mode 100644
index 00000000000000..a211427c5e0c17
--- /dev/null
+++ b/llvm/test/CodeGen/MIR/AMDGPU/num-phys-vgpr-spill-lanes.ll
@@ -0,0 +1,15 @@
+; RUN: llc -O0 -mtriple=amdgcn-amd-amdhsa --stop-after=prologepilog -o - %s | FileCheck %s
+
+; Spill the PC SGPR30_31 and FP to physical VGPR
+
+define void @test() #0 {
+; CHECK: machineFunctionInfo
+; CHECK: numPhysicalVGPRSpillLanes: 3
+entry:
+  %call = call i32 @ext_func()
+  ret void
+}
+
+declare i32 @ext_func();
+
+attributes #0 = { nounwind "amdgpu-num-vgpr"="41" "amdgpu-num-sgpr"="34" }