[InlineAsm][AArch64]Add backend support for flag output parameters

[mingmingl]

Summary

• The set of flag is from https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Flag-Output-Operands

Before:

• ARM64 GCC supports flag output constraints, while Clang doesn't parse condition code as shown in https://gcc.godbolt.org/z/7jzMEK796
• LLVM ISel won't lower them either (as shown in https://gcc.godbolt.org/z/Pv4PPf56c)

After:

• Given flag output constraints in LLVM IR, condition code is parsed and flag output is lowered to 'cset'.
• Clang parse is not added in this patch.

[nickdesaulniers]

very nice, thank you!

(looks like https://reviews.llvm.org/rGe5c37958f901cc9bec50624dbee85d40143e4bca for inspiration)

[mingmingl]

In D149032#4293831, @nickdesaulniers wrote:

very nice, thank you!

(looks like https://reviews.llvm.org/rGe5c37958f901cc9bec50624dbee85d40143e4bca for inspiration)

Thanks for taking a look! Yep https://github.com/llvm/llvm-project/commit/e5c37958f901c is where I learned this. Going to port the frontend change (D57394) to aarch64 in a follow-up patch.

[mingmingl]

While testing D149123 with ./bin/clang -cc1 -S -triple=aarch64 inline-asm-aarch64-flag-output.c (which invokes global-isel for instruction selection according to print-after-all output), turns out GlobalISel doesn't support flag output (for x86 and aarch64) yet.

/bin/clang -cc1 -O1 -S -triple=aarch64 file.c and /bin/clang -cc1 -O2 -S -triple=aarch64 file.c works (gives the assembly output)

I filed https://github.com/llvm/llvm-project/issues/62343 to track but think the issue is a non-blocker of this patch and D149123 for now.

[efriedma]

Adding a fallback from globalisel to selectiondag isel should be simple; I'd like to avoid advertising support for functionality that doesn't actually work at all optimization levels.

This is probably worth noting in release notes.

[mingmingl]

In D149032#4296049, @efriedma wrote:

Adding a fallback from globalisel to selectiondag isel should be simple;

Thanks for pointing out fallback mechanism (TargetLowering::fallBackToDAGISel)! Will do

I'd like to avoid advertising support for functionality that doesn't actually work at all optimization levels.

This is a valid point and sounds reasonable.

This is probably worth noting in release notes.

will do

In D149032#4296049, @efriedma wrote:

Adding a fallback from globalisel to selectiondag isel should be simple; I'd like to avoid advertising support for functionality that doesn't actually work at all optimization levels.

This is probably worth noting in release notes.

[mingmingl]

In llvm/lib/CodeGen/GlobalISel/InlineAsmLowering.cpp, allow Op ConstraintType to be Other. Add a test in 'llvm/test/CodeGen/AArch64/GlobalISel/arm64-fallback.ll' to show that fallback works with -global-isel-abort=2

Please take another look. thanks!

[mingmingl]

In D149032#4296325, @mingmingl wrote:

In D149032#4296049, @efriedma wrote:

Adding a fallback from globalisel to selectiondag isel should be simple;

Thanks for pointing out fallback mechanism (TargetLowering::fallBackToDAGISel)! Will do

I'd like to avoid advertising support for functionality that doesn't actually work at all optimization levels.

This is a valid point and sounds reasonable.

This is probably worth noting in release notes.

will do

In D149032#4296049, @efriedma wrote:

Adding a fallback from globalisel to selectiondag isel should be simple; I'd like to avoid advertising support for functionality that doesn't actually work at all optimization levels.

This is probably worth noting in release notes.

After spending some time on roughly figuring out how fallback works, relaxed the assertion as mentioned above and added a test to show that fallback works (with global-isel-abort=2 that enables fallback).

Currently global-isel is enabled for O0 with (the default) GlobalISelAbortMode GlobalISelAbort = GlobalISelAbortMode::Enable;, and explains the previous observation (O0 crashes and O1/O2 works).

So I'm thinking the change to 'clang/lib/Driver/ToolChains/Clang.cpp' (and release notes update) would goes to D149123

[efriedma]

clang shouldn't abort by default, as far as I know?

Anyway, this patch looks fine.

[mingmingl]

In D149032#4297534, @efriedma wrote:

clang shouldn't abort by default, as far as I know?

You are correct. After relaxing the assert condition in GlobalISel/InlineAsmLowering.cpp, the crashing O0 command generates valid output, indicating the abort option is not on -> after digging a little bit, the aarch64 target machine turns abort off. So no change needed in 'clang/lib/Driver/ToolChains/Clang.cpp'.

Anyway, this patch looks fine.

Thanks for reviews!

llvm/lib/Target/AArch64/AArch64ISelLowering.h

unsigned getInlineAsmMemConstraint(StringRef ConstraintCode) const override {
     if (ConstraintCode == "Q")
       return InlineAsm::Constraint_Q;
     // FIXME: clang has code for 'Ump', 'Utf', 'Usa', and 'Ush' but these are
     //        followed by llvm_unreachable so we'll leave them unimplemented in
     //        the backend for now.
     return TargetLowering::getInlineAsmMemConstraint(ConstraintCode);
   }
 
+  /// Handle Lowering flag assembly outputs.
+  SDValue LowerAsmOutputForConstraint(SDValue &Chain, SDValue &Flag,
+                                      const SDLoc &DL,
+                                      const AsmOperandInfo &Constraint,
+                                      SelectionDAG &DAG) const override;
+
   bool shouldExtendGSIndex(EVT VT, EVT &EltTy) const override;
   bool shouldRemoveExtendFromGSIndex(EVT IndexVT, EVT DataVT) const override;
   bool isVectorLoadExtDesirable(SDValue ExtVal) const override;
   bool isUsedByReturnOnly(SDNode *N, SDValue &Chain) const override;
   bool mayBeEmittedAsTailCall(const CallInst *CI) const override;
   bool getIndexedAddressParts(SDNode *N, SDNode *Op, SDValue &Base,
                               SDValue &Offset, SelectionDAG &DAG) const;
   bool getPreIndexedAddressParts(SDNode *N, SDValue &Base, SDValue &Offset,

llvm/lib/Target/AArch64/AArch64ISelLowering.cpp

static PredicateConstraint parsePredicateConstraint(StringRef Constraint) {
   PredicateConstraint P = PredicateConstraint::Invalid;
   if (Constraint == "Upa")
     P = PredicateConstraint::Upa;
   if (Constraint == "Upl")
     P = PredicateConstraint::Upl;
   return P;
 }
 
+// The set of cc code supported is from
+// https://gcc.gnu.org/onlinedocs/gcc/Extended-Asm.html#Flag-Output-Operands
+static AArch64CC::CondCode parseConstraintCode(llvm::StringRef Constraint) {
+  AArch64CC::CondCode Cond = StringSwitch<AArch64CC::CondCode>(Constraint)
+                                 .Case("{@cchi}", AArch64CC::HI)
+                                 .Case("{@cccs}", AArch64CC::HS)
+                                 .Case("{@cclo}", AArch64CC::LO)
+                                 .Case("{@ccls}", AArch64CC::LS)
+                                 .Case("{@cccc}", AArch64CC::LO)
+                                 .Case("{@cceq}", AArch64CC::EQ)
+                                 .Case("{@ccgt}", AArch64CC::GT)
+                                 .Case("{@ccge}", AArch64CC::GE)
+                                 .Case("{@cclt}", AArch64CC::LT)
+                                 .Case("{@ccle}", AArch64CC::LE)
+                                 .Case("{@cchs}", AArch64CC::HS)
+                                 .Case("{@ccne}", AArch64CC::NE)
+                                 .Case("{@ccvc}", AArch64CC::VC)
+                                 .Case("{@ccpl}", AArch64CC::PL)
+                                 .Case("{@ccvs}", AArch64CC::VS)
+                                 .Case("{@ccmi}", AArch64CC::MI)
+                                 .Default(AArch64CC::Invalid);
+  return Cond;
+}
+
+/// Helper function to create 'CSET', which is equivalent to 'CSINC <Wd>, WZR,
+/// WZR, invert(<cond>)'.
+static SDValue getSETCC(AArch64CC::CondCode CC, SDValue NZCV, const SDLoc &DL,
+                        SelectionDAG &DAG) {
+  return DAG.getNode(
+      AArch64ISD::CSINC, DL, MVT::i32, DAG.getConstant(0, DL, MVT::i32),
+      DAG.getConstant(0, DL, MVT::i32),
+      DAG.getConstant(getInvertedCondCode(CC), DL, MVT::i32), NZCV);
+}
+
+// Lower @cc flag output via getSETCC.
+SDValue AArch64TargetLowering::LowerAsmOutputForConstraint(
+    SDValue &Chain, SDValue &Glue, const SDLoc &DL,
+    const AsmOperandInfo &OpInfo, SelectionDAG &DAG) const {
+  AArch64CC::CondCode Cond = parseConstraintCode(OpInfo.ConstraintCode);
+  if (Cond == AArch64CC::Invalid)
+    return SDValue();
+  // The output variable should be a scalar integer.
+  if (OpInfo.ConstraintVT.isVector() || !OpInfo.ConstraintVT.isInteger() ||
+      OpInfo.ConstraintVT.getSizeInBits() < 8)
+    report_fatal_error("Flag output operand is of invalid type");
+
+  // Get NZCV register. Only update chain when copyfrom is glued.
+  if (Glue.getNode()) {
+    Glue = DAG.getCopyFromReg(Chain, DL, AArch64::NZCV, MVT::i32, Glue);
+    Chain = Glue.getValue(1);
+  } else
+    Glue = DAG.getCopyFromReg(Chain, DL, AArch64::NZCV, MVT::i32);
+  // Extract CC code.
+  SDValue CC = getSETCC(Cond, Glue, DL, DAG);
+
+  SDValue Result;
+
+  // Truncate or ZERO_EXTEND based on value types.
+  if (OpInfo.ConstraintVT.getSizeInBits() <= 32)
+    Result = DAG.getNode(ISD::TRUNCATE, DL, OpInfo.ConstraintVT, CC);
+  else
+    Result = DAG.getNode(ISD::ZERO_EXTEND, DL, OpInfo.ConstraintVT, CC);
+
+  return Result;
+}
+
 /// getConstraintType - Given a constraint letter, return the type of
 /// constraint it is for this target.
 AArch64TargetLowering::ConstraintType
 AArch64TargetLowering::getConstraintType(StringRef Constraint) const {
   if (Constraint.size() == 1) {
     switch (Constraint[0]) {
     default:
       break;
case 'Z':
       return C_Immediate;
     case 'z':
     case 'S': // A symbolic address
       return C_Other;
     }
   } else if (parsePredicateConstraint(Constraint) !=
              PredicateConstraint::Invalid)
       return C_RegisterClass;
+  else if (parseConstraintCode(Constraint) !=
+           AArch64CC::Invalid)
+    return C_Other;
   return TargetLowering::getConstraintType(Constraint);
 }
 
 /// Examine constraint type and operand type and determine a weight value.
 /// This object must already have been set up with the operand type
 /// and the current alternative constraint selected.
 TargetLowering::ConstraintWeight
 AArch64TargetLowering::getSingleConstraintMatchWeight(
if (Constraint.size() == 1) {
     if (PC != PredicateConstraint::Invalid) {
       if (!VT.isScalableVector() || VT.getVectorElementType() != MVT::i1)
         return std::make_pair(0U, nullptr);
       bool restricted = (PC == PredicateConstraint::Upl);
       return restricted ? std::make_pair(0U, &AArch64::PPR_3bRegClass)
                         : std::make_pair(0U, &AArch64::PPRRegClass);
     }
   }
-  if (StringRef("{cc}").equals_insensitive(Constraint))
+  if (StringRef("{cc}").equals_insensitive(Constraint) ||
+      parseConstraintCode(Constraint) != AArch64CC::Invalid)
     return std::make_pair(unsigned(AArch64::NZCV), &AArch64::CCRRegClass);
 
   // Use the default implementation in TargetLowering to convert the register
   // constraint into a member of a register class.
   std::pair<unsigned, const TargetRegisterClass *> Res;
   Res = TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
 
   // Not found as a standard register?

llvm/test/CodeGen/AArch64/inline-asm-flag-output.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
+; RUN: llc -mtriple=aarch64 %s -o - | FileCheck %s
+define i32 @test_cchi(i64 %a) {
+; CHECK-LABEL: test_cchi:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, hi
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cchi},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_cccs(i64 %a) {
+; CHECK-LABEL: test_cccs:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, hs
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cccs},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_cclo(i64 %a) {
+; CHECK-LABEL: test_cclo:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, lo
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cclo},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccls(i64 %a) {
+; CHECK-LABEL: test_ccls:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, ls
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccls},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_cccc(i64 %a) {
+; CHECK-LABEL: test_cccc:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, lo
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cccc},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_cceq(i64 %a) {
+; CHECK-LABEL: test_cceq:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, eq
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cceq},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccgt(i64 %a) {
+; CHECK-LABEL: test_ccgt:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, gt
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccgt},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccge(i64 %a) {
+; CHECK-LABEL: test_ccge:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, ge
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccge},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_cclt(i64 %a) {
+; CHECK-LABEL: test_cclt:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, lt
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cclt},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccle(i64 %a) {
+; CHECK-LABEL: test_ccle:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, le
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccle},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_cchs(i64 %a) {
+; CHECK-LABEL: test_cchs:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, hs
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@cchs},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccne(i64 %a) {
+; CHECK-LABEL: test_ccne:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, ne
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccne},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccvc(i64 %a) {
+; CHECK-LABEL: test_ccvc:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, vc
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccvc},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccpl(i64 %a) {
+; CHECK-LABEL: test_ccpl:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, pl
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccpl},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccvs(i64 %a) {
+; CHECK-LABEL: test_ccvs:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, vs
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccvs},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+define i32 @test_ccmi(i64 %a) {
+; CHECK-LABEL: test_ccmi:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    //APP
+; CHECK-NEXT:    subs x0, x0, #3
+; CHECK-NEXT:    //NO_APP
+; CHECK-NEXT:    cset w0, mi
+; CHECK-NEXT:    ret
+entry:
+  %0 = tail call { i64, i32 } asm "subs $0, $0, #3", "=r,={@ccmi},0,~{dirflag},~{fpsr},~{flags}"(i64 %a)
+  %asmresult1 = extractvalue { i64, i32 } %0, 1
+  %1 = icmp ult i32 %asmresult1, 2
+  tail call void @llvm.assume(i1 %1)
+  ret i32 %asmresult1
+}
+
+declare void @llvm.assume(i1)