[llvm-mca] Add -skip-unsupported-instructions option

Prior to this patch, if llvm-mca encountered an instruction which parses
but has no scheduler info, the instruction is always reported as
unsupported, and llvm-mca halts with an error.

However, it would still be useful to allow MCA to continue even in the
case of instructions lacking scheduling information. Obviously if
scheduling information is lacking, it's not possible to give an accurate
analysis for those instructions, and therefore a warning is emitted.

A user could previously have worked around such unsupported instructions
manually by deleting such instructions from the input, but this provides
them a way of doing this for bulk inputs where they may not have a list
of such unsupported instructions to drop up front.

Note that this behaviour of instructions with no scheduling information
under -skip-unsupported-instructions is analagous to current
instructions which fail to parse: those are currently dropped from the
input with a message printed, after which the analysis continues.

Author: peterwaller-arm

llvm/lib/MCA/InstrBuilder.cpp

@@ -542,8 +542,7 @@ InstrBuilder::createInstrDescImpl(const MCInst &MCI,
   const MCSchedClassDesc &SCDesc = *SM.getSchedClassDesc(SchedClassID);
   if (SCDesc.NumMicroOps == MCSchedClassDesc::InvalidNumMicroOps) {
     return make_error<InstructionError<MCInst>>(
-        "found an unsupported instruction in the input assembly sequence.",
-        MCI);
+        "found an unsupported instruction in the input assembly sequence", MCI);
   }
 
   LLVM_DEBUG(dbgs() << "\n\t\tOpcode Name= " << MCII.getName(Opcode) << '\n');

llvm/test/tools/llvm-mca/X86/BtVer2/unsupported-instruction.s

@@ -1,6 +1,22 @@
-# RUN: not llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 %s 2>&1 | FileCheck %s
+# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -skip-unsupported-instructions %s |& FileCheck --check-prefix=CHECK-SKIP %s
+# RUN: not llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 %s |& FileCheck --check-prefix=CHECK-ERROR %s
 
+# CHECK-SKIP: warning: found an unsupported instruction in the input assembly sequence, skipping with -skip-unsupported-instructions, note accuracy will be impacted:
+# CHECK-ERROR: error: found an unsupported instruction in the input assembly sequence, use -skip-unsupported-instructions to ignore.
+
+# Currently lacks scheduling information and is therefore reported as unsupported by llvm-mca.
+# This may change some day in which case an alternative unsupported input would need to be found or the test removed.
 bzhi %eax, %ebx, %ecx
 
-# CHECK: error: found an unsupported instruction in the input assembly sequence.
-# CHECK-NEXT: note: instruction: 	bzhil	%eax, %ebx, %ecx
+# Supported instruction that may be analysed.
+add %eax, %eax
+
+# CHECK-SKIP: Iterations:        100
+# CHECK-SKIP: Instructions:      100
+# CHECK-SKIP: Total Cycles:      103
+# CHECK-SKIP: Total uOps:        100
+
+# CHECK-SKIP: Dispatch Width:    2
+# CHECK-SKIP: uOps Per Cycle:    0.97
+# CHECK-SKIP: IPC:               0.97
+# CHECK-SKIP: Block RThroughput: 0.5

llvm/tools/llvm-mca/CodeRegion.h

@@ -59,6 +59,7 @@
 #define LLVM_TOOLS_LLVM_MCA_CODEREGION_H
 
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/ADT/StringRef.h"
@@ -97,6 +98,20 @@ class CodeRegion {
     Instructions.emplace_back(Instruction);
   }
 
+  // Remove the given instructions from the set, for unsupported instructions
+  // being skipped. Returns an ArrayRef for the updated vector of Instructions.
+  [[nodiscard]] llvm::ArrayRef<llvm::MCInst>
+  dropInstructions(const llvm::SmallPtrSetImpl<const llvm::MCInst *> &Insts) {
+    if (Insts.empty())
+      return Instructions;
+    Instructions.erase(std::remove_if(Instructions.begin(), Instructions.end(),
+                                      [&Insts](const llvm::MCInst &Inst) {
+                                        return Insts.contains(&Inst);
+                                      }),
+                       Instructions.end());
+    return Instructions;
+  }
+
   llvm::SMLoc startLoc() const { return RangeStart; }
   llvm::SMLoc endLoc() const { return RangeEnd; }
 

llvm/tools/llvm-mca/llvm-mca.cpp

@@ -237,6 +237,11 @@ static cl::opt<bool> DisableInstrumentManager(
              "ignores instruments.)."),
     cl::cat(ViewOptions), cl::init(false));
 
+static cl::opt<bool> SkipUnsupportedInstructions(
+    "skip-unsupported-instructions",
+    cl::desc("Make unsupported instruction errors into warnings."),
+    cl::cat(ViewOptions), cl::init(false));
+
 namespace {
 
 const Target *getTarget(const char *ProgName) {
@@ -558,6 +563,7 @@ int main(int argc, char **argv) {
   assert(MAB && "Unable to create asm backend!");
 
   json::Object JSONOutput;
+  int NonEmptyRegions = 0;
   for (const std::unique_ptr<mca::AnalysisRegion> &Region : Regions) {
     // Skip empty code regions.
     if (Region->empty())
@@ -571,14 +577,13 @@ int main(int argc, char **argv) {
 
     IPP->resetState();
 
-    DenseMap<const MCInst *, SmallVector<mca::Instrument *>>
-        InstToInstruments;
+    DenseMap<const MCInst *, SmallVector<mca::Instrument *>> InstToInstruments;
     SmallVector<std::unique_ptr<mca::Instruction>> LoweredSequence;
+    SmallPtrSet<const MCInst *, 16> DroppedInsts;
     for (const MCInst &MCI : Insts) {
       SMLoc Loc = MCI.getLoc();
       const SmallVector<mca::Instrument *> Instruments =
           InstrumentRegions.getActiveInstruments(Loc);
-      InstToInstruments.insert({&MCI, Instruments});
 
       Expected<std::unique_ptr<mca::Instruction>> Inst =
           IB.createInstruction(MCI, Instruments);
@@ -588,7 +593,15 @@ int main(int argc, char **argv) {
                 [&IP, &STI](const mca::InstructionError<MCInst> &IE) {
                   std::string InstructionStr;
                   raw_string_ostream SS(InstructionStr);
-                  WithColor::error() << IE.Message << '\n';
+                  if (SkipUnsupportedInstructions)
+                    WithColor::warning()
+                        << IE.Message
+                        << ", skipping with -skip-unsupported-instructions, "
+                           "note accuracy will be impacted:\n";
+                  else
+                    WithColor::error()
+                        << IE.Message
+                        << ", use -skip-unsupported-instructions to ignore.\n";
                   IP->printInst(&IE.Inst, 0, "", *STI, SS);
                   SS.flush();
                   WithColor::note()
@@ -597,14 +610,25 @@ int main(int argc, char **argv) {
           // Default case.
           WithColor::error() << toString(std::move(NewE));
         }
+        if (SkipUnsupportedInstructions) {
+          DroppedInsts.insert(&MCI);
+          continue;
+        }
         return 1;
       }
 
       IPP->postProcessInstruction(Inst.get(), MCI);
-
+      InstToInstruments.insert({&MCI, Instruments});
       LoweredSequence.emplace_back(std::move(Inst.get()));
     }
 
+    Insts = Region->dropInstructions(DroppedInsts);
+
+    // Skip empty regions.
+    if (Insts.empty())
+      continue;
+    NonEmptyRegions++;
+
     mca::CircularSourceMgr S(LoweredSequence,
                              PrintInstructionTables ? 1 : Iterations);
 
@@ -759,6 +783,11 @@ int main(int argc, char **argv) {
     ++RegionIdx;
   }
 
+  if (NonEmptyRegions == 0) {
+    WithColor::error() << "no assembly instructions found.\n";
+    return 1;
+  }
+
   if (PrintJson)
     TOF->os() << formatv("{0:2}", json::Value(std::move(JSONOutput))) << "\n";