[NVVM] Pretty Printing Immediate Arguments in LLVM Intrinsics

This patch adds LLVM infrastructure to support pretty printing of immediate arguments of the intrinsics. The motivation is to increase the readability of LLVM intrinsics and facilitate easy modifications and debugging of LLVM IR.
This adds a property PrettyPrintImmArg<ArgIndex, "functionName"> to the intrinsic ImmArgs that enables printing self-explanatory inline comment for the immediate arguments.
The addition of pretty print support can provide a simple, low-overhead feature that enhances usability of LLVM intrinsics without disrupting existing workflows.

Signed-off-by: Dharuni R Acharya <[email protected]>

Author: DharuniRAcharya

llvm/include/llvm/IR/Function.h

@@ -929,8 +929,8 @@ class LLVM_ABI Function : public GlobalObject, public ilist_node<Function> {
   /// Print the function to an output stream with an optional
   /// AssemblyAnnotationWriter.
   void print(raw_ostream &OS, AssemblyAnnotationWriter *AAW = nullptr,
-             bool ShouldPreserveUseListOrder = false,
-             bool IsForDebug = false) const;
+             bool ShouldPreserveUseListOrder = false, bool IsForDebug = false,
+             bool PrettyPrintIntrinsicArgs = false) const;
 
   /// viewCFG - This function is meant for use from the debugger.  You can just
   /// say 'call F->viewCFG()' and a ghostview window should pop up from the

llvm/include/llvm/IR/Intrinsics.h

@@ -30,6 +30,8 @@ class LLVMContext;
 class Module;
 class AttributeList;
 class AttributeSet;
+class raw_ostream;
+class Constant;
 
 /// This namespace contains an enum with a value for every intrinsic/builtin
 /// function known by LLVM. The enum values are returned by
@@ -81,6 +83,9 @@ namespace Intrinsic {
   /// Returns true if the intrinsic can be overloaded.
   LLVM_ABI bool isOverloaded(ID id);
 
+  /// Returns true if the intrinsic has pretty printed immediate arguments.
+  LLVM_ABI bool hasPrettyPrintedArgs(ID id);
+
   /// isTargetIntrinsic - Returns true if IID is an intrinsic specific to a
   /// certain target. If it is a generic intrinsic false is returned.
   LLVM_ABI bool isTargetIntrinsic(ID IID);
@@ -284,6 +289,9 @@ namespace Intrinsic {
   /// N.
   LLVM_ABI Intrinsic::ID getDeinterleaveIntrinsicID(unsigned Factor);
 
+  LLVM_ABI void printImmArg(ID IID, unsigned ArgIdx, raw_ostream &OS,
+                            const Constant *ImmArgVal);
+
   } // namespace Intrinsic
 
   } // namespace llvm

llvm/include/llvm/IR/Intrinsics.td

@@ -87,6 +87,13 @@ class NoUndef<AttrIndex idx> : IntrinsicProperty {
   int ArgNo = idx.Value;
 }
 
+// PrettyPrintImmArg - The specified immediate argument has a custom pretty-print
+// function for diagnostic output.
+class PrettyPrintImmArg<AttrIndex idx, string funcname> : IntrinsicProperty {
+  int ArgNo = idx.Value;
+  string FunctionName = funcname;
+}
+
 // NonNull - The return value or specified argument is not null.
 class NonNull<AttrIndex idx> : IntrinsicProperty {
   int ArgNo = idx.Value;

llvm/include/llvm/IR/IntrinsicsNVVM.td

@@ -2871,7 +2871,14 @@ foreach sp = [0, 1] in {
         defvar nargs = !size(args);
         defvar scale_d_imm = ArgIndex<!sub(nargs, 1)>;
         defvar scale_d_imm_range = [ImmArg<scale_d_imm>, Range<scale_d_imm, 0, 16>];
-        defvar intrinsic_properties = !listconcat(
+
+        // Check if this is the specific llvm.nvvm.tcgen05.mma.tensor intrinsic.
+        defvar is_target_intrinsic = !and(!eq(sp, 0), 
+                                          !eq(space, "tensor"), 
+                                          !eq(scale_d, 0), 
+                                          !eq(ashift, 0));
+
+        defvar base_properties = !listconcat(
           mma.common_intr_props,
           !if(!eq(scale_d, 1), scale_d_imm_range, []),
           [Range<ArgIndex<nargs>, 0, !if(!eq(scale_d, 1), 2, 4)>, // kind
@@ -2881,6 +2888,13 @@ foreach sp = [0, 1] in {
           ]
         );
 
+        defvar intrinsic_properties = !if(is_target_intrinsic, 
+          !listconcat(base_properties,
+            [PrettyPrintImmArg<ArgIndex<nargs>, "printTcgen05MMAKind">,
+             PrettyPrintImmArg<ArgIndex<!add(nargs, 1)>, "printCTAGroupKind">,
+             PrettyPrintImmArg<ArgIndex<!add(nargs, 2)>, "printTcgen05CollectorUsageOp">]),
+          base_properties);
+
         def mma.record:
               DefaultAttrsIntrinsicFlags<[], args, flags, intrinsic_properties,
                 mma.intr>;

llvm/include/llvm/IR/Module.h

@@ -865,8 +865,8 @@ class LLVM_ABI Module {
   /// uselistorder directives so that use-lists can be recreated when reading
   /// the assembly.
   void print(raw_ostream &OS, AssemblyAnnotationWriter *AAW,
-             bool ShouldPreserveUseListOrder = false,
-             bool IsForDebug = false) const;
+             bool ShouldPreserveUseListOrder = false, bool IsForDebug = false,
+             bool PrettyPrintIntrinsicArgs = false) const;
 
   /// Dump the module to stderr (for debugging).
   void dump() const;

llvm/include/llvm/IR/NVVMIntrinsicUtils.h

@@ -18,8 +18,11 @@
 #include <stdint.h>
 
 #include "llvm/ADT/APFloat.h"
+#include "llvm/ADT/APInt.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/IntrinsicsNVPTX.h"
+#include "llvm/Support/raw_ostream.h"
 
 namespace llvm {
 namespace nvvm {
@@ -659,6 +662,70 @@ inline APFloat::roundingMode GetFMARoundingMode(Intrinsic::ID IntrinsicID) {
   llvm_unreachable("Invalid FP instrinsic rounding mode for NVVM fma");
 }
 
+inline void printTcgen05MMAKind(raw_ostream &OS, const Constant *ImmArgVal) {
+  if (const ConstantInt *CI = dyn_cast<ConstantInt>(ImmArgVal)) {
+    uint64_t Val = CI->getZExtValue();
+    switch (static_cast<Tcgen05MMAKind>(Val)) {
+    case Tcgen05MMAKind::F16:
+      OS << "kind::f16";
+      return;
+    case Tcgen05MMAKind::TF32:
+      OS << "kind::tf32";
+      return;
+    case Tcgen05MMAKind::F8F6F4:
+      OS << "kind::f8f6f4";
+      return;
+    case Tcgen05MMAKind::I8:
+      OS << "kind::i8";
+      return;
+    }
+  }
+  llvm_unreachable(
+      "printTcgen05MMAKind called with invalid value for immediate argument");
+}
+
+inline void printTcgen05CollectorUsageOp(raw_ostream &OS,
+                                         const Constant *ImmArgVal) {
+  if (const ConstantInt *CI = dyn_cast<ConstantInt>(ImmArgVal)) {
+    uint64_t Val = CI->getZExtValue();
+    switch (static_cast<Tcgen05CollectorUsageOp>(Val)) {
+    case Tcgen05CollectorUsageOp::DISCARD:
+      OS << "collector::a::discard";
+      return;
+    case Tcgen05CollectorUsageOp::LASTUSE:
+      OS << "collector::a::lastuse";
+      return;
+    case Tcgen05CollectorUsageOp::FILL:
+      OS << "collector::a::fill";
+      return;
+    case Tcgen05CollectorUsageOp::USE:
+      OS << "collector::a::use";
+      return;
+    }
+  }
+  llvm_unreachable("printTcgen05CollectorUsageOp called with invalid value for "
+                   "immediate argument");
+}
+
+inline void printCTAGroupKind(raw_ostream &OS, const Constant *ImmArgVal) {
+  if (const ConstantInt *CI = dyn_cast<ConstantInt>(ImmArgVal)) {
+    uint64_t Val = CI->getZExtValue();
+    switch (static_cast<CTAGroupKind>(Val)) {
+    case CTAGroupKind::CG_NONE:
+      OS << "cta_group::0";
+      return;
+    case CTAGroupKind::CG_1:
+      OS << "cta_group::1";
+      return;
+    case CTAGroupKind::CG_2:
+      OS << "cta_group::2";
+      return;
+    }
+  }
+  llvm_unreachable(
+      "printCTAGroupKind called with invalid value for immediate argument");
+}
+
 } // namespace nvvm
 } // namespace llvm
 #endif // LLVM_IR_NVVMINTRINSICUTILS_H

llvm/lib/IR/AsmWriter.cpp

@@ -53,6 +53,7 @@
 #include "llvm/IR/Instruction.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/IR/Module.h"
@@ -2831,6 +2832,7 @@ class AssemblyWriter {
   SetVector<const Comdat *> Comdats;
   bool IsForDebug;
   bool ShouldPreserveUseListOrder;
+  bool PrettyPrintIntrinsicArgs;
   UseListOrderMap UseListOrders;
   SmallVector<StringRef, 8> MDNames;
   /// Synchronization scope names registered with LLVMContext.
@@ -2841,7 +2843,8 @@ class AssemblyWriter {
   /// Construct an AssemblyWriter with an external SlotTracker
   AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac, const Module *M,
                  AssemblyAnnotationWriter *AAW, bool IsForDebug,
-                 bool ShouldPreserveUseListOrder = false);
+                 bool ShouldPreserveUseListOrder = false,
+                 bool PrettyPrintIntrinsicArgs = false);
 
   AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac,
                  const ModuleSummaryIndex *Index, bool IsForDebug);
@@ -2930,10 +2933,12 @@ class AssemblyWriter {
 
 AssemblyWriter::AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac,
                                const Module *M, AssemblyAnnotationWriter *AAW,
-                               bool IsForDebug, bool ShouldPreserveUseListOrder)
+                               bool IsForDebug, bool ShouldPreserveUseListOrder,
+                               bool PrettyPrintIntrinsicArgs)
     : Out(o), TheModule(M), Machine(Mac), TypePrinter(M), AnnotationWriter(AAW),
       IsForDebug(IsForDebug),
-      ShouldPreserveUseListOrder(ShouldPreserveUseListOrder) {
+      ShouldPreserveUseListOrder(ShouldPreserveUseListOrder),
+      PrettyPrintIntrinsicArgs(PrettyPrintIntrinsicArgs) {
   if (!TheModule)
     return;
   for (const GlobalObject &GO : TheModule->global_objects())
@@ -2944,7 +2949,8 @@ AssemblyWriter::AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac,
 AssemblyWriter::AssemblyWriter(formatted_raw_ostream &o, SlotTracker &Mac,
                                const ModuleSummaryIndex *Index, bool IsForDebug)
     : Out(o), TheIndex(Index), Machine(Mac), TypePrinter(/*Module=*/nullptr),
-      IsForDebug(IsForDebug), ShouldPreserveUseListOrder(false) {}
+      IsForDebug(IsForDebug), ShouldPreserveUseListOrder(false),
+      PrettyPrintIntrinsicArgs(false) {}
 
 void AssemblyWriter::writeOperand(const Value *Operand, bool PrintType) {
   if (!Operand) {
@@ -4561,12 +4567,40 @@ void AssemblyWriter::printInstruction(const Instruction &I) {
     Out << ' ';
     writeOperand(Operand, false);
     Out << '(';
+    bool HasPrettyPrintedArgs =
+        PrettyPrintIntrinsicArgs && isa<IntrinsicInst>(CI) &&
+        Intrinsic::hasPrettyPrintedArgs(CI->getIntrinsicID());
+
     ListSeparator LS;
-    for (unsigned op = 0, Eop = CI->arg_size(); op < Eop; ++op) {
-      Out << LS;
-      writeParamOperand(CI->getArgOperand(op), PAL.getParamAttrs(op));
-    }
+    if (HasPrettyPrintedArgs) {
+      Function *CalledFunc = CI->getCalledFunction();
+      auto PrintArgComment = [&](unsigned ArgNo) {
+        if (!CalledFunc->hasParamAttribute(ArgNo, Attribute::ImmArg))
+          return;
+        const Constant *ConstArg = dyn_cast<Constant>(CI->getArgOperand(ArgNo));
+        if (!ConstArg)
+          return;
+        std::string ArgComment;
+        raw_string_ostream ArgCommentStream(ArgComment);
+        Intrinsic::ID IID = CalledFunc->getIntrinsicID();
+        Intrinsic::printImmArg(IID, ArgNo, ArgCommentStream, ConstArg);
+        if (ArgComment.empty())
+          return;
+        Out << "/* " << ArgComment << " */ ";
+      };
 
+      for (unsigned ArgNo = 0, NumArgs = CI->arg_size(); ArgNo < NumArgs;
+           ++ArgNo) {
+        Out << LS;
+        PrintArgComment(ArgNo);
+        writeParamOperand(CI->getArgOperand(ArgNo), PAL.getParamAttrs(ArgNo));
+      }
+    } else {
+      for (unsigned op = 0, Eop = CI->arg_size(); op < Eop; ++op) {
+        Out << LS;
+        writeParamOperand(CI->getArgOperand(op), PAL.getParamAttrs(op));
+      }
+    }
     // Emit an ellipsis if this is a musttail call in a vararg function.  This
     // is only to aid readability, musttail calls forward varargs by default.
     if (CI->isMustTailCall() && CI->getParent() &&
@@ -4990,13 +5024,12 @@ void AssemblyWriter::printUseLists(const Function *F) {
 //===----------------------------------------------------------------------===//
 
 void Function::print(raw_ostream &ROS, AssemblyAnnotationWriter *AAW,
-                     bool ShouldPreserveUseListOrder,
-                     bool IsForDebug) const {
+                     bool ShouldPreserveUseListOrder, bool IsForDebug,
+                     bool PrettyPrintIntrinsicArgs) const {
   SlotTracker SlotTable(this->getParent());
   formatted_raw_ostream OS(ROS);
-  AssemblyWriter W(OS, SlotTable, this->getParent(), AAW,
-                   IsForDebug,
-                   ShouldPreserveUseListOrder);
+  AssemblyWriter W(OS, SlotTable, this->getParent(), AAW, IsForDebug,
+                   ShouldPreserveUseListOrder, PrettyPrintIntrinsicArgs);
   W.printFunction(this);
 }
 
@@ -5012,11 +5045,12 @@ void BasicBlock::print(raw_ostream &ROS, AssemblyAnnotationWriter *AAW,
 }
 
 void Module::print(raw_ostream &ROS, AssemblyAnnotationWriter *AAW,
-                   bool ShouldPreserveUseListOrder, bool IsForDebug) const {
+                   bool ShouldPreserveUseListOrder, bool IsForDebug,
+                   bool PrettyPrintIntrinsicArgs) const {
   SlotTracker SlotTable(this);
   formatted_raw_ostream OS(ROS);
   AssemblyWriter W(OS, SlotTable, this, AAW, IsForDebug,
-                   ShouldPreserveUseListOrder);
+                   ShouldPreserveUseListOrder, PrettyPrintIntrinsicArgs);
   W.printModule(this);
 }
 

llvm/lib/IR/Intrinsics.cpp

@@ -32,6 +32,7 @@
 #include "llvm/IR/IntrinsicsX86.h"
 #include "llvm/IR/IntrinsicsXCore.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/NVVMIntrinsicUtils.h"
 #include "llvm/IR/Type.h"
 
 using namespace llvm;
@@ -601,6 +602,12 @@ bool Intrinsic::isOverloaded(ID id) {
 #undef GET_INTRINSIC_OVERLOAD_TABLE
 }
 
+bool Intrinsic::hasPrettyPrintedArgs(ID id){
+#define GET_INTRINSIC_PRETTY_PRINT_TABLE
+#include "llvm/IR/IntrinsicImpl.inc"
+#undef GET_INTRINSIC_PRETTY_PRINT_TABLE
+}
+
 /// Table of per-target intrinsic name tables.
 #define GET_INTRINSIC_TARGET_DATA
 #include "llvm/IR/IntrinsicImpl.inc"
@@ -1129,3 +1136,7 @@ Intrinsic::ID Intrinsic::getDeinterleaveIntrinsicID(unsigned Factor) {
   assert(Factor >= 2 && Factor <= 8 && "Unexpected factor");
   return InterleaveIntrinsics[Factor - 2].Deinterleave;
 }
+
+#define GET_INTRINSIC_PRETTY_PRINT_ARGUMENTS
+#include "llvm/IR/IntrinsicImpl.inc"
+#undef GET_INTRINSIC_PRETTY_PRINT_ARGUMENTS