[SPIRV] Added support for the constrained comparison intrinsics (#157439)

Added SPIR-V support for constrained floating-point comparison
intrinsics (fcmp, fcmps) with lowering and tests.

Author: SubashBoopathi

llvm/lib/Target/SPIRV/SPIRVPrepareFunctions.cpp

@@ -335,6 +335,21 @@ static void lowerFunnelShifts(IntrinsicInst *FSHIntrinsic) {
   FSHIntrinsic->setCalledFunction(FSHFunc);
 }
 
+static void lowerConstrainedFPCmpIntrinsic(
+    ConstrainedFPCmpIntrinsic *ConstrainedCmpIntrinsic,
+    SmallVector<Instruction *> &EraseFromParent) {
+  if (!ConstrainedCmpIntrinsic)
+    return;
+  // Extract the floating-point values being compared
+  Value *LHS = ConstrainedCmpIntrinsic->getArgOperand(0);
+  Value *RHS = ConstrainedCmpIntrinsic->getArgOperand(1);
+  FCmpInst::Predicate Pred = ConstrainedCmpIntrinsic->getPredicate();
+  IRBuilder<> Builder(ConstrainedCmpIntrinsic);
+  Value *FCmp = Builder.CreateFCmp(Pred, LHS, RHS);
+  ConstrainedCmpIntrinsic->replaceAllUsesWith(FCmp);
+  EraseFromParent.push_back(dyn_cast<Instruction>(ConstrainedCmpIntrinsic));
+}
+
 static void lowerExpectAssume(IntrinsicInst *II) {
   // If we cannot use the SPV_KHR_expect_assume extension, then we need to
   // ignore the intrinsic and move on. It should be removed later on by LLVM.
@@ -376,6 +391,7 @@ static bool toSpvLifetimeIntrinsic(IntrinsicInst *II, Intrinsic::ID NewID) {
 bool SPIRVPrepareFunctions::substituteIntrinsicCalls(Function *F) {
   bool Changed = false;
   const SPIRVSubtarget &STI = TM.getSubtarget<SPIRVSubtarget>(*F);
+  SmallVector<Instruction *> EraseFromParent;
   for (BasicBlock &BB : *F) {
     for (Instruction &I : make_early_inc_range(BB)) {
       auto Call = dyn_cast<CallInst>(&I);
@@ -423,9 +439,17 @@ bool SPIRVPrepareFunctions::substituteIntrinsicCalls(Function *F) {
         lowerPtrAnnotation(II);
         Changed = true;
         break;
+      case Intrinsic::experimental_constrained_fcmp:
+      case Intrinsic::experimental_constrained_fcmps:
+        lowerConstrainedFPCmpIntrinsic(dyn_cast<ConstrainedFPCmpIntrinsic>(II),
+                                       EraseFromParent);
+        Changed = true;
+        break;
       }
     }
   }
+  for (auto *I : EraseFromParent)
+    I->eraseFromParent();
   return Changed;
 }
 

llvm/test/CodeGen/SPIRV/llvm-intrinsics/constrained-comparison.ll

@@ -0,0 +1,56 @@
+; RUN: llc -verify-machineinstrs -O0 -mtriple=spirv64-unknown-unknown %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv64-unknown-unknown %s -o - -filetype=obj | spirv-val %}
+
+; CHECK-DAG: OpFOrdEqual
+; CHECK-DAG: OpFOrdGreaterThan
+; CHECK-DAG: OpFOrdGreaterThanEqual
+; CHECK-DAG: OpFOrdLessThan
+; CHECK-DAG: OpFOrdLessThanEqual
+; CHECK-DAG: OpFOrdNotEqual
+; CHECK-DAG: OpOrdered
+; CHECK-DAG: OpFUnordEqual
+; CHECK-DAG: OpFUnordGreaterThan
+; CHECK-DAG: OpFUnordGreaterThanEqual
+; CHECK-DAG: OpFUnordLessThan
+; CHECK-DAG: OpFUnordLessThanEqual
+; CHECK-DAG: OpFUnordNotEqual
+; CHECK-DAG: OpUnordered
+
+define dso_local spir_kernel void @test(float %a){
+entry:
+  %cmp = tail call i1 @llvm.experimental.constrained.fcmps.f32(float %a, float %a, metadata !"oeq", metadata !"fpexcept.strict") 
+  %cmp1 = tail call i1 @llvm.experimental.constrained.fcmps.f32(float %a, float %a, metadata !"ogt", metadata !"fpexcept.strict") 
+  %cmp2 = tail call i1 @llvm.experimental.constrained.fcmps.f32(float %a, float %a, metadata !"oge", metadata !"fpexcept.strict") 
+  %cmp3 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"olt", metadata !"fpexcept.strict") 
+  %cmp4 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"ole", metadata !"fpexcept.strict") 
+  %cmp5 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"one", metadata !"fpexcept.strict") 
+  %cmp6 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"ord", metadata !"fpexcept.strict") 
+  %cmp7 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"ueq", metadata !"fpexcept.strict") 
+  %cmp8 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"ugt", metadata !"fpexcept.strict") 
+  %cmp9 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"uge", metadata !"fpexcept.strict") 
+  %cmp10 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"ult", metadata !"fpexcept.strict") 
+  %cmp11 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"ule", metadata !"fpexcept.strict") 
+  %cmp12 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"une", metadata !"fpexcept.strict") 
+  %cmp13 = tail call i1 @llvm.experimental.constrained.fcmp.f32(float %a, float %a, metadata !"uno", metadata !"fpexcept.strict") 
+
+  %or1 = or i1 %cmp, %cmp1
+  %or2 = or i1 %or1, %cmp2
+  %or3 = or i1 %or2, %cmp3
+  %or4 = or i1 %or3, %cmp4
+  %or5 = or i1 %or4, %cmp5
+  %or6 = or i1 %or5, %cmp6
+  %or7 = or i1 %or6, %cmp7
+  %or8 = or i1 %or7, %cmp8
+  %or9 = or i1 %or8, %cmp9
+  %or10 = or i1 %or9, %cmp10
+  %or11 = or i1 %or10, %cmp11
+  %or12 = or i1 %or11, %cmp12
+  %or13 = or i1 %or12, %cmp13
+  br i1 %or13, label %true_block, label %false_block
+true_block:
+  ret void
+false_block:
+  ret void
+}
+declare i1 @llvm.experimental.constrained.fcmps.f32(float, float, metadata, metadata) 
+declare i1 @llvm.experimental.constrained.fcmp.f32(float, float, metadata, metadata)