[AMDGPU] Enable serializing of argument info.

Summary:
- Support serialization of all arguments in machine function info. This
  enables fabricating MIR tests depending on argument info.

Reviewers: arsenm, rampitec

Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D64096

llvm-svn: 364995

3 files changed, 253 insertions, 1 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 8e4f35a..667c6f3 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -1059,5 +1059,84 @@ bool GCNTargetMachine::parseMachineFunctionInfo(
     return diagnoseRegisterClass(YamlMFI.StackPtrOffsetReg);
   }
 
+  auto parseAndCheckArgument = [&](const Optional<yaml::SIArgument> &A,
+                                   const TargetRegisterClass &RC,
+                                   ArgDescriptor &Arg) {
+    // Skip parsing if it's not present.
+    if (!A)
+      return false;
+
+    if (A->IsRegister) {
+      unsigned Reg;
+      if (parseNamedRegisterReference(PFS, Reg, A->RegisterName.Value,
+                                      Error)) {
+        SourceRange = A->RegisterName.SourceRange;
+        return true;
+      }
+      if (!RC.contains(Reg))
+        return diagnoseRegisterClass(A->RegisterName);
+      Arg = ArgDescriptor::createRegister(Reg);
+    } else
+      Arg = ArgDescriptor::createStack(A->StackOffset);
+    // Check and apply the optional mask.
+    if (A->Mask)
+      Arg = ArgDescriptor::createArg(Arg, A->Mask.getValue());
+
+    return false;
+  };
+
+  if (YamlMFI.ArgInfo &&
+      (parseAndCheckArgument(YamlMFI.ArgInfo->PrivateSegmentBuffer,
+                             AMDGPU::SReg_128RegClass,
+                             MFI->ArgInfo.PrivateSegmentBuffer) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->DispatchPtr,
+                             AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.DispatchPtr) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->QueuePtr, AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.QueuePtr) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->KernargSegmentPtr,
+                             AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.KernargSegmentPtr) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->DispatchID,
+                             AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.DispatchID) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->FlatScratchInit,
+                             AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.FlatScratchInit) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->PrivateSegmentSize,
+                             AMDGPU::SGPR_32RegClass,
+                             MFI->ArgInfo.PrivateSegmentSize) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkGroupIDX,
+                             AMDGPU::SGPR_32RegClass,
+                             MFI->ArgInfo.WorkGroupIDX) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkGroupIDY,
+                             AMDGPU::SGPR_32RegClass,
+                             MFI->ArgInfo.WorkGroupIDY) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkGroupIDZ,
+                             AMDGPU::SGPR_32RegClass,
+                             MFI->ArgInfo.WorkGroupIDZ) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkGroupInfo,
+                             AMDGPU::SGPR_32RegClass,
+                             MFI->ArgInfo.WorkGroupInfo) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->PrivateSegmentWaveByteOffset,
+                             AMDGPU::SGPR_32RegClass,
+                             MFI->ArgInfo.PrivateSegmentWaveByteOffset) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->ImplicitArgPtr,
+                             AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.ImplicitArgPtr) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->ImplicitBufferPtr,
+                             AMDGPU::SReg_64RegClass,
+                             MFI->ArgInfo.ImplicitBufferPtr) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkItemIDX,
+                             AMDGPU::VGPR_32RegClass,
+                             MFI->ArgInfo.WorkItemIDX) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkItemIDY,
+                             AMDGPU::VGPR_32RegClass,
+                             MFI->ArgInfo.WorkItemIDY) ||
+       parseAndCheckArgument(YamlMFI.ArgInfo->WorkItemIDZ,
+                             AMDGPU::VGPR_32RegClass,
+                             MFI->ArgInfo.WorkItemIDZ)))
+    return true;
+
   return false;
 }
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
index 8bb8ea7..f7d6172 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
@@ -324,6 +324,57 @@ static yaml::StringValue regToString(unsigned Reg,
   return Dest;
 }
 
+static Optional<yaml::SIArgumentInfo>
+convertArgumentInfo(const AMDGPUFunctionArgInfo &ArgInfo,
+                    const TargetRegisterInfo &TRI) {
+  yaml::SIArgumentInfo AI;
+
+  auto convertArg = [&](Optional<yaml::SIArgument> &A,
+                        const ArgDescriptor &Arg) {
+    if (!Arg)
+      return false;
+
+    // Create a register or stack argument.
+    yaml::SIArgument SA = yaml::SIArgument::createArgument(Arg.isRegister());
+    if (Arg.isRegister()) {
+      raw_string_ostream OS(SA.RegisterName.Value);
+      OS << printReg(Arg.getRegister(), &TRI);
+    } else
+      SA.StackOffset = Arg.getStackOffset();
+    // Check and update the optional mask.
+    if (Arg.isMasked())
+      SA.Mask = Arg.getMask();
+
+    A = SA;
+    return true;
+  };
+
+  bool Any = false;
+  Any |= convertArg(AI.PrivateSegmentBuffer, ArgInfo.PrivateSegmentBuffer);
+  Any |= convertArg(AI.DispatchPtr, ArgInfo.DispatchPtr);
+  Any |= convertArg(AI.QueuePtr, ArgInfo.QueuePtr);
+  Any |= convertArg(AI.KernargSegmentPtr, ArgInfo.KernargSegmentPtr);
+  Any |= convertArg(AI.DispatchID, ArgInfo.DispatchID);
+  Any |= convertArg(AI.FlatScratchInit, ArgInfo.FlatScratchInit);
+  Any |= convertArg(AI.PrivateSegmentSize, ArgInfo.PrivateSegmentSize);
+  Any |= convertArg(AI.WorkGroupIDX, ArgInfo.WorkGroupIDX);
+  Any |= convertArg(AI.WorkGroupIDY, ArgInfo.WorkGroupIDY);
+  Any |= convertArg(AI.WorkGroupIDZ, ArgInfo.WorkGroupIDZ);
+  Any |= convertArg(AI.WorkGroupInfo, ArgInfo.WorkGroupInfo);
+  Any |= convertArg(AI.PrivateSegmentWaveByteOffset,
+                    ArgInfo.PrivateSegmentWaveByteOffset);
+  Any |= convertArg(AI.ImplicitArgPtr, ArgInfo.ImplicitArgPtr);
+  Any |= convertArg(AI.ImplicitBufferPtr, ArgInfo.ImplicitBufferPtr);
+  Any |= convertArg(AI.WorkItemIDX, ArgInfo.WorkItemIDX);
+  Any |= convertArg(AI.WorkItemIDY, ArgInfo.WorkItemIDY);
+  Any |= convertArg(AI.WorkItemIDZ, ArgInfo.WorkItemIDZ);
+
+  if (Any)
+    return AI;
+
+  return None;
+}
+
 yaml::SIMachineFunctionInfo::SIMachineFunctionInfo(
   const llvm::SIMachineFunctionInfo& MFI,
   const TargetRegisterInfo &TRI)
@@ -337,7 +388,8 @@ yaml::SIMachineFunctionInfo::SIMachineFunctionInfo(
     ScratchRSrcReg(regToString(MFI.getScratchRSrcReg(), TRI)),
     ScratchWaveOffsetReg(regToString(MFI.getScratchWaveOffsetReg(), TRI)),
     FrameOffsetReg(regToString(MFI.getFrameOffsetReg(), TRI)),
-    StackPtrOffsetReg(regToString(MFI.getStackPtrOffsetReg(), TRI)) {}
+    StackPtrOffsetReg(regToString(MFI.getStackPtrOffsetReg(), TRI)),
+    ArgInfo(convertArgumentInfo(MFI.getArgInfo(), TRI)) {}
 
 void yaml::SIMachineFunctionInfo::mappingImpl(yaml::IO &YamlIO) {
   MappingTraits<SIMachineFunctionInfo>::mapping(YamlIO, *this);
diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
index 329b38c..7339363 100644
--- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h
@@ -21,6 +21,7 @@
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/Optional.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/SparseBitVector.h"
 #include "llvm/CodeGen/MIRYamlMapping.h"
@@ -114,6 +115,123 @@ public:
 
 namespace yaml {
 
+struct SIArgument {
+  bool IsRegister;
+  union {
+    StringValue RegisterName;
+    unsigned StackOffset;
+  };
+  Optional<unsigned> Mask;
+
+  // Default constructor, which creates a stack argument.
+  SIArgument() : IsRegister(false), StackOffset(0) {}
+  SIArgument(const SIArgument &Other) {
+    IsRegister = Other.IsRegister;
+    if (IsRegister) {
+      ::new ((void *)std::addressof(RegisterName))
+          StringValue(Other.RegisterName);
+    } else
+      StackOffset = Other.StackOffset;
+    Mask = Other.Mask;
+  }
+  SIArgument &operator=(const SIArgument &Other) {
+    IsRegister = Other.IsRegister;
+    if (IsRegister) {
+      ::new ((void *)std::addressof(RegisterName))
+          StringValue(Other.RegisterName);
+    } else
+      StackOffset = Other.StackOffset;
+    Mask = Other.Mask;
+    return *this;
+  }
+  ~SIArgument() {
+    if (IsRegister)
+      RegisterName.~StringValue();
+  }
+
+  // Helper to create a register or stack argument.
+  static inline SIArgument createArgument(bool IsReg) {
+    if (IsReg)
+      return SIArgument(IsReg);
+    return SIArgument();
+  }
+
+private:
+  // Construct a register argument.
+  SIArgument(bool) : IsRegister(true), RegisterName() {}
+};
+
+template <> struct MappingTraits<SIArgument> {
+  static void mapping(IO &YamlIO, SIArgument &A) {
+    if (YamlIO.outputting()) {
+      if (A.IsRegister)
+        YamlIO.mapRequired("reg", A.RegisterName);
+      else
+        YamlIO.mapRequired("offset", A.StackOffset);
+    } else {
+      auto Keys = YamlIO.keys();
+      if (is_contained(Keys, "reg")) {
+        A = SIArgument::createArgument(true);
+        YamlIO.mapRequired("reg", A.RegisterName);
+      } else if (is_contained(Keys, "offset"))
+        YamlIO.mapRequired("offset", A.StackOffset);
+      else
+        YamlIO.setError("missing required key 'reg' or 'offset'");
+    }
+    YamlIO.mapOptional("mask", A.Mask);
+  }
+  static const bool flow = true;
+};
+
+struct SIArgumentInfo {
+  Optional<SIArgument> PrivateSegmentBuffer;
+  Optional<SIArgument> DispatchPtr;
+  Optional<SIArgument> QueuePtr;
+  Optional<SIArgument> KernargSegmentPtr;
+  Optional<SIArgument> DispatchID;
+  Optional<SIArgument> FlatScratchInit;
+  Optional<SIArgument> PrivateSegmentSize;
+
+  Optional<SIArgument> WorkGroupIDX;
+  Optional<SIArgument> WorkGroupIDY;
+  Optional<SIArgument> WorkGroupIDZ;
+  Optional<SIArgument> WorkGroupInfo;
+  Optional<SIArgument> PrivateSegmentWaveByteOffset;
+
+  Optional<SIArgument> ImplicitArgPtr;
+  Optional<SIArgument> ImplicitBufferPtr;
+
+  Optional<SIArgument> WorkItemIDX;
+  Optional<SIArgument> WorkItemIDY;
+  Optional<SIArgument> WorkItemIDZ;
+};
+
+template <> struct MappingTraits<SIArgumentInfo> {
+  static void mapping(IO &YamlIO, SIArgumentInfo &AI) {
+    YamlIO.mapOptional("privateSegmentBuffer", AI.PrivateSegmentBuffer);
+    YamlIO.mapOptional("dispatchPtr", AI.DispatchPtr);
+    YamlIO.mapOptional("queuePtr", AI.QueuePtr);
+    YamlIO.mapOptional("kernargSegmentPtr", AI.KernargSegmentPtr);
+    YamlIO.mapOptional("dispatchID", AI.DispatchID);
+    YamlIO.mapOptional("flatScratchInit", AI.FlatScratchInit);
+    YamlIO.mapOptional("privateSegmentSize", AI.PrivateSegmentSize);
+
+    YamlIO.mapOptional("workGroupIDX", AI.WorkGroupIDX);
+    YamlIO.mapOptional("workGroupIDY", AI.WorkGroupIDY);
+    YamlIO.mapOptional("workGroupIDZ", AI.WorkGroupIDZ);
+    YamlIO.mapOptional("workGroupInfo", AI.WorkGroupInfo);
+    YamlIO.mapOptional("privateSegmentWaveByteOffset",
+                       AI.PrivateSegmentWaveByteOffset);
+
+    YamlIO.mapOptional("implicitArgPtr", AI.ImplicitArgPtr);
+    YamlIO.mapOptional("implicitBufferPtr", AI.ImplicitBufferPtr);
+
+    YamlIO.mapOptional("workItemIDX", AI.WorkItemIDX);
+    YamlIO.mapOptional("workItemIDY", AI.WorkItemIDY);
+    YamlIO.mapOptional("workItemIDZ", AI.WorkItemIDZ);
+  }
+};
+
 struct SIMachineFunctionInfo final : public yaml::MachineFunctionInfo {
   uint64_t ExplicitKernArgSize = 0;
   unsigned MaxKernArgAlign = 0;
@@ -128,6 +246,8 @@ struct SIMachineFunctionInfo final : public yaml::MachineFunctionInfo {
   StringValue FrameOffsetReg = "$fp_reg";
   StringValue StackPtrOffsetReg = "$sp_reg";
 
+  Optional<SIArgumentInfo> ArgInfo;
+
   SIMachineFunctionInfo() = default;
   SIMachineFunctionInfo(const llvm::SIMachineFunctionInfo &,
                         const TargetRegisterInfo &TRI);
@@ -154,6 +274,7 @@ template <> struct MappingTraits<SIMachineFunctionInfo> {
                        StringValue("$fp_reg"));
     YamlIO.mapOptional("stackPtrOffsetReg", MFI.StackPtrOffsetReg,
                        StringValue("$sp_reg"));
+    YamlIO.mapOptional("argumentInfo", MFI.ArgInfo);
   }
 };