[LV][EVL] Support icmp/fcmp instruction with EVL-vectorization #108533
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-backend-risc-v @llvm/pr-subscribers-llvm-transforms Author: LiqinWeng (LiqinWeng) ChangesPatch is 28.51 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/108533.diff 7 Files Affected:
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 00d9f2909d71e2..8bcf5f3553cd86 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -10546,13 +10546,13 @@ InstructionCost BoUpSLP::getSpillCost() const {
if (II->isAssumeLikeIntrinsic())
return true;
FastMathFlags FMF;
- SmallVector<Type *, 4> Tys;
+ SmallVector<Type *, 8> Tys;
for (auto &ArgOp : II->args())
Tys.push_back(ArgOp->getType());
if (auto *FPMO = dyn_cast<FPMathOperator>(II))
FMF = FPMO->getFastMathFlags();
IntrinsicCostAttributes ICA(II->getIntrinsicID(), II->getType(), Tys,
- FMF);
+ FMF, II);
InstructionCost IntrCost =
TTI->getIntrinsicInstrCost(ICA, TTI::TCK_RecipThroughput);
InstructionCost CallCost = TTI->getCallInstrCost(
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index 351f909ac0279d..6e0ea5673e5a4e 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -1269,6 +1269,33 @@ InstructionCost VPWidenRecipe::computeCost(ElementCount VF,
void VPWidenEVLRecipe::execute(VPTransformState &State) {
unsigned Opcode = getOpcode();
// TODO: Support other opcodes
+ if (Opcode == Instruction::ICmp || Opcode == Instruction::FCmp) {
+ Value *Op1 = State.get(getOperand(0), 0);
+ Value *Op2 = State.get(getOperand(1), 0);
+ auto &Ctx = State.Builder.getContext();
+ Value *Pred = MetadataAsValue::get(
+ Ctx, MDString::get(Ctx, CmpInst::getPredicateName(getPredicate())));
+
+ IRBuilderBase &BuilderIR = State.Builder;
+ VectorBuilder Builder(BuilderIR);
+ Value *Mask = BuilderIR.CreateVectorSplat(State.VF, BuilderIR.getTrue());
+ Builder.setMask(Mask).setEVL(State.get(getEVL(), 0, true));
+
+ VectorType *DataType = VectorType::get(Type::getInt1Ty(Ctx), State.VF);
+
+ Value *VPInst = Builder.createVectorInstruction(Opcode, DataType,
+ {Op1, Op2, Pred}, "vp.op");
+ // if (isa<FPMathOperator>(VPInst))
+ // setFlags(cast<Instruction>(VPInst));
+ if (VPInst) {
+ if (auto *VecOp = dyn_cast<CastInst>(VPInst))
+ VecOp->copyIRFlags(getUnderlyingInstr());
+ }
+ State.set(this, VPInst, 0);
+ State.addMetadata(VPInst,
+ dyn_cast_or_null<Instruction>(getUnderlyingValue()));
+ return;
+ }
if (!Instruction::isBinaryOp(Opcode) && !Instruction::isUnaryOp(Opcode))
llvm_unreachable("Unsupported opcode in VPWidenEVLRecipe::execute");
diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
index b722ec34ee6fb6..0b2c126210162f 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp
@@ -1345,7 +1345,8 @@ static void transformRecipestoEVLRecipes(VPlan &Plan, VPValue &EVL) {
.Case<VPWidenRecipe>([&](VPWidenRecipe *W) -> VPRecipeBase * {
unsigned Opcode = W->getOpcode();
if (!Instruction::isBinaryOp(Opcode) &&
- !Instruction::isUnaryOp(Opcode))
+ !Instruction::isUnaryOp(Opcode) &&
+ Opcode != Instruction::ICmp && Opcode != Instruction::FCmp)
return nullptr;
return new VPWidenEVLRecipe(*W, EVL);
})
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-cond-reduction.ll b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-cond-reduction.ll
index 41796e848632e4..a86f6b1d23ae7a 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-cond-reduction.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-cond-reduction.ll
@@ -69,7 +69,7 @@ define i32 @cond_add(ptr %a, i64 %n, i32 %start) {
; IF-EVL-INLOOP-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP13]]
; IF-EVL-INLOOP-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[TMP17]], i32 0
; IF-EVL-INLOOP-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP18]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP12]])
-; IF-EVL-INLOOP-NEXT: [[TMP19:%.*]] = icmp sgt <vscale x 4 x i32> [[VP_OP_LOAD]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 3, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
+; IF-EVL-INLOOP-NEXT: [[TMP19:%.*]] = call <vscale x 4 x i1> @llvm.vp.icmp.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 3, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer), metadata !"sgt", <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP12]])
; IF-EVL-INLOOP-NEXT: [[TMP20:%.*]] = select <vscale x 4 x i1> [[TMP19]], <vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> zeroinitializer
; IF-EVL-INLOOP-NEXT: [[TMP21:%.*]] = call i32 @llvm.vp.reduce.add.nxv4i32(i32 0, <vscale x 4 x i32> [[TMP20]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP12]])
; IF-EVL-INLOOP-NEXT: [[TMP22]] = add i32 [[TMP21]], [[VEC_PHI]]
@@ -282,7 +282,7 @@ define i32 @cond_add_pred(ptr %a, i64 %n, i32 %start) {
; IF-EVL-INLOOP-NEXT: [[TMP17:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[TMP13]]
; IF-EVL-INLOOP-NEXT: [[TMP18:%.*]] = getelementptr inbounds i32, ptr [[TMP17]], i32 0
; IF-EVL-INLOOP-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP18]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP12]])
-; IF-EVL-INLOOP-NEXT: [[TMP19:%.*]] = icmp sgt <vscale x 4 x i32> [[VP_OP_LOAD]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 3, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
+; IF-EVL-INLOOP-NEXT: [[TMP19:%.*]] = call <vscale x 4 x i1> @llvm.vp.icmp.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 3, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer), metadata !"sgt", <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP12]])
; IF-EVL-INLOOP-NEXT: [[TMP20:%.*]] = select <vscale x 4 x i1> [[TMP16]], <vscale x 4 x i1> [[TMP19]], <vscale x 4 x i1> zeroinitializer
; IF-EVL-INLOOP-NEXT: [[TMP21:%.*]] = call i32 @llvm.vp.reduce.add.nxv4i32(i32 0, <vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> [[TMP20]], i32 [[TMP12]])
; IF-EVL-INLOOP-NEXT: [[TMP22]] = add i32 [[TMP21]], [[VEC_PHI]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-masked-loadstore.ll b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-masked-loadstore.ll
index 99da5058fbf922..c915ffd9a80d91 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-masked-loadstore.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-masked-loadstore.ll
@@ -45,18 +45,18 @@ define void @masked_loadstore(ptr noalias %a, ptr noalias %b, i64 %n) {
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[TMP11]]
; IF-EVL-NEXT: [[TMP16:%.*]] = getelementptr inbounds i32, ptr [[TMP15]], i32 0
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP16]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP10]])
-; IF-EVL-NEXT: [[TMP17:%.*]] = icmp ne <vscale x 4 x i32> [[VP_OP_LOAD]], zeroinitializer
-; IF-EVL-NEXT: [[TMP18:%.*]] = select <vscale x 4 x i1> [[TMP14]], <vscale x 4 x i1> [[TMP17]], <vscale x 4 x i1> zeroinitializer
-; IF-EVL-NEXT: [[TMP19:%.*]] = getelementptr i32, ptr [[A:%.*]], i64 [[TMP11]]
-; IF-EVL-NEXT: [[TMP20:%.*]] = getelementptr i32, ptr [[TMP19]], i32 0
-; IF-EVL-NEXT: [[VP_OP_LOAD3:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP20]], <vscale x 4 x i1> [[TMP18]], i32 [[TMP10]])
-; IF-EVL-NEXT: [[VP_OP:%.*]] = call <vscale x 4 x i32> @llvm.vp.add.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> [[VP_OP_LOAD3]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP10]])
-; IF-EVL-NEXT: call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_OP]], ptr align 4 [[TMP20]], <vscale x 4 x i1> [[TMP18]], i32 [[TMP10]])
-; IF-EVL-NEXT: [[TMP21:%.*]] = zext i32 [[TMP10]] to i64
-; IF-EVL-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP21]], [[EVL_BASED_IV]]
+; IF-EVL-NEXT: [[VP_OP:%.*]] = call <vscale x 4 x i1> @llvm.vp.icmp.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> zeroinitializer, metadata !"ne", <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP10]])
+; IF-EVL-NEXT: [[TMP17:%.*]] = select <vscale x 4 x i1> [[TMP14]], <vscale x 4 x i1> [[VP_OP]], <vscale x 4 x i1> zeroinitializer
+; IF-EVL-NEXT: [[TMP18:%.*]] = getelementptr i32, ptr [[A:%.*]], i64 [[TMP11]]
+; IF-EVL-NEXT: [[TMP19:%.*]] = getelementptr i32, ptr [[TMP18]], i32 0
+; IF-EVL-NEXT: [[VP_OP_LOAD3:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP19]], <vscale x 4 x i1> [[TMP17]], i32 [[TMP10]])
+; IF-EVL-NEXT: [[VP_OP4:%.*]] = call <vscale x 4 x i32> @llvm.vp.add.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> [[VP_OP_LOAD3]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP10]])
+; IF-EVL-NEXT: call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_OP4]], ptr align 4 [[TMP19]], <vscale x 4 x i1> [[TMP17]], i32 [[TMP10]])
+; IF-EVL-NEXT: [[TMP20:%.*]] = zext i32 [[TMP10]] to i64
+; IF-EVL-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP20]], [[EVL_BASED_IV]]
; IF-EVL-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP8]]
-; IF-EVL-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; IF-EVL-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; IF-EVL-NEXT: [[TMP21:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; IF-EVL-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; IF-EVL: middle.block:
; IF-EVL-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
; IF-EVL: scalar.ph:
@@ -65,13 +65,13 @@ define void @masked_loadstore(ptr noalias %a, ptr noalias %b, i64 %n) {
; IF-EVL: for.body:
; IF-EVL-NEXT: [[I_011:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
; IF-EVL-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[I_011]]
-; IF-EVL-NEXT: [[TMP23:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
-; IF-EVL-NEXT: [[CMP1:%.*]] = icmp ne i32 [[TMP23]], 0
+; IF-EVL-NEXT: [[TMP22:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
+; IF-EVL-NEXT: [[CMP1:%.*]] = icmp ne i32 [[TMP22]], 0
; IF-EVL-NEXT: br i1 [[CMP1]], label [[IF_THEN:%.*]], label [[FOR_INC]]
; IF-EVL: if.then:
; IF-EVL-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[I_011]]
-; IF-EVL-NEXT: [[TMP24:%.*]] = load i32, ptr [[ARRAYIDX3]], align 4
-; IF-EVL-NEXT: [[ADD:%.*]] = add i32 [[TMP23]], [[TMP24]]
+; IF-EVL-NEXT: [[TMP23:%.*]] = load i32, ptr [[ARRAYIDX3]], align 4
+; IF-EVL-NEXT: [[ADD:%.*]] = add i32 [[TMP22]], [[TMP23]]
; IF-EVL-NEXT: store i32 [[ADD]], ptr [[ARRAYIDX3]], align 4
; IF-EVL-NEXT: br label [[FOR_INC]]
; IF-EVL: for.inc:
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-reverse-load-store.ll b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-reverse-load-store.ll
index c1cf8b0fc541e7..def9d8d4c19124 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-reverse-load-store.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-reverse-load-store.ll
@@ -147,33 +147,33 @@ define void @reverse_load_store_masked(i64 %startval, ptr noalias %ptr, ptr noal
; IF-EVL-NEXT: [[TMP13:%.*]] = getelementptr inbounds i32, ptr [[PTR:%.*]], i32 [[TMP8]]
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds i32, ptr [[TMP13]], i32 0
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP14]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
-; IF-EVL-NEXT: [[TMP15:%.*]] = icmp slt <vscale x 4 x i32> [[VP_OP_LOAD]], shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 100, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer)
-; IF-EVL-NEXT: [[TMP16:%.*]] = select <vscale x 4 x i1> [[TMP11]], <vscale x 4 x i1> [[TMP15]], <vscale x 4 x i1> zeroinitializer
-; IF-EVL-NEXT: [[TMP17:%.*]] = getelementptr i32, ptr [[PTR1:%.*]], i64 [[TMP12]]
-; IF-EVL-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64()
-; IF-EVL-NEXT: [[TMP19:%.*]] = mul i64 [[TMP18]], 4
-; IF-EVL-NEXT: [[TMP20:%.*]] = mul i64 0, [[TMP19]]
-; IF-EVL-NEXT: [[TMP21:%.*]] = sub i64 1, [[TMP19]]
-; IF-EVL-NEXT: [[TMP22:%.*]] = getelementptr i32, ptr [[TMP17]], i64 [[TMP20]]
-; IF-EVL-NEXT: [[TMP23:%.*]] = getelementptr i32, ptr [[TMP22]], i64 [[TMP21]]
-; IF-EVL-NEXT: [[VP_REVERSE_MASK:%.*]] = call <vscale x 4 x i1> @llvm.experimental.vp.reverse.nxv4i1(<vscale x 4 x i1> [[TMP16]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
-; IF-EVL-NEXT: [[VP_OP_LOAD4:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP23]], <vscale x 4 x i1> [[VP_REVERSE_MASK]], i32 [[TMP6]])
+; IF-EVL-NEXT: [[VP_OP:%.*]] = call <vscale x 4 x i1> @llvm.vp.icmp.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i32> shufflevector (<vscale x 4 x i32> insertelement (<vscale x 4 x i32> poison, i32 100, i64 0), <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer), metadata !"slt", <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
+; IF-EVL-NEXT: [[TMP15:%.*]] = select <vscale x 4 x i1> [[TMP11]], <vscale x 4 x i1> [[VP_OP]], <vscale x 4 x i1> zeroinitializer
+; IF-EVL-NEXT: [[TMP16:%.*]] = getelementptr i32, ptr [[PTR1:%.*]], i64 [[TMP12]]
+; IF-EVL-NEXT: [[TMP17:%.*]] = call i64 @llvm.vscale.i64()
+; IF-EVL-NEXT: [[TMP18:%.*]] = mul i64 [[TMP17]], 4
+; IF-EVL-NEXT: [[TMP19:%.*]] = mul i64 0, [[TMP18]]
+; IF-EVL-NEXT: [[TMP20:%.*]] = sub i64 1, [[TMP18]]
+; IF-EVL-NEXT: [[TMP21:%.*]] = getelementptr i32, ptr [[TMP16]], i64 [[TMP19]]
+; IF-EVL-NEXT: [[TMP22:%.*]] = getelementptr i32, ptr [[TMP21]], i64 [[TMP20]]
+; IF-EVL-NEXT: [[VP_REVERSE_MASK:%.*]] = call <vscale x 4 x i1> @llvm.experimental.vp.reverse.nxv4i1(<vscale x 4 x i1> [[TMP15]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
+; IF-EVL-NEXT: [[VP_OP_LOAD4:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP22]], <vscale x 4 x i1> [[VP_REVERSE_MASK]], i32 [[TMP6]])
; IF-EVL-NEXT: [[VP_REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD4]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
-; IF-EVL-NEXT: [[TMP24:%.*]] = getelementptr i32, ptr [[PTR2:%.*]], i64 [[TMP12]]
-; IF-EVL-NEXT: [[TMP25:%.*]] = call i64 @llvm.vscale.i64()
-; IF-EVL-NEXT: [[TMP26:%.*]] = mul i64 [[TMP25]], 4
-; IF-EVL-NEXT: [[TMP27:%.*]] = mul i64 0, [[TMP26]]
-; IF-EVL-NEXT: [[TMP28:%.*]] = sub i64 1, [[TMP26]]
-; IF-EVL-NEXT: [[TMP29:%.*]] = getelementptr i32, ptr [[TMP24]], i64 [[TMP27]]
-; IF-EVL-NEXT: [[TMP30:%.*]] = getelementptr i32, ptr [[TMP29]], i64 [[TMP28]]
+; IF-EVL-NEXT: [[TMP23:%.*]] = getelementptr i32, ptr [[PTR2:%.*]], i64 [[TMP12]]
+; IF-EVL-NEXT: [[TMP24:%.*]] = call i64 @llvm.vscale.i64()
+; IF-EVL-NEXT: [[TMP25:%.*]] = mul i64 [[TMP24]], 4
+; IF-EVL-NEXT: [[TMP26:%.*]] = mul i64 0, [[TMP25]]
+; IF-EVL-NEXT: [[TMP27:%.*]] = sub i64 1, [[TMP25]]
+; IF-EVL-NEXT: [[TMP28:%.*]] = getelementptr i32, ptr [[TMP23]], i64 [[TMP26]]
+; IF-EVL-NEXT: [[TMP29:%.*]] = getelementptr i32, ptr [[TMP28]], i64 [[TMP27]]
; IF-EVL-NEXT: [[VP_REVERSE5:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_REVERSE]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
-; IF-EVL-NEXT: [[VP_REVERSE_MASK6:%.*]] = call <vscale x 4 x i1> @llvm.experimental.vp.reverse.nxv4i1(<vscale x 4 x i1> [[TMP16]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
-; IF-EVL-NEXT: call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE5]], ptr align 4 [[TMP30]], <vscale x 4 x i1> [[VP_REVERSE_MASK6]], i32 [[TMP6]])
-; IF-EVL-NEXT: [[TMP31:%.*]] = zext i32 [[TMP6]] to i64
-; IF-EVL-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP31]], [[EVL_BASED_IV]]
+; IF-EVL-NEXT: [[VP_REVERSE_MASK6:%.*]] = call <vscale x 4 x i1> @llvm.experimental.vp.reverse.nxv4i1(<vscale x 4 x i1> [[TMP15]], <vscale x 4 x i1> shufflevector (<vscale x 4 x i1> insertelement (<vscale x 4 x i1> poison, i1 true, i64 0), <vscale x 4 x i1> poison, <vscale x 4 x i32> zeroinitializer), i32 [[TMP6]])
+; IF-EVL-NEXT: call void @llvm.vp.store.nxv4i32.p0(<vscale x 4 x i32> [[VP_REVERSE5]], ptr align 4 [[TMP29]], <vscale x 4 x i1> [[VP_REVERSE_MASK6]], i32 [[TMP6]])
+; IF-EVL-NEXT: [[TMP30:%.*]] = zext i32 [[TMP6]] to i64
+; IF-EVL-NEXT: [[INDEX_EVL_NEXT]] = add i64 [[TMP30]], [[EVL_BASED_IV]]
; IF-EVL-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP4]]
-; IF-EVL-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
-; IF-EVL-NEXT: br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+; IF-EVL-NEXT: [[TMP31:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; IF-EVL-NEXT: br i1 [[TMP31]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; IF-EVL: middle.block:
; IF-EVL-NEXT: br i1 true, label [[LOOPEND:%.*]], label [[SCALAR_PH]]
; IF-EVL: scalar.ph:
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/vplan-vp-cmp-intrinsics.ll b/llvm/test/Transforms/LoopVe...
[truncated]
|
LiqinWeng
force-pushed
the
widen-vp-cmp-with-evl
branch
from
2d4782f to
2e5a44d
Compare
LiqinWeng
force-pushed
the
widen-vp-cmp-with-evl
branch
from
2e5a44d to
4b50426
Compare
artagnon
reviewed
Oct 15, 2024
llvm/test/Transforms/LoopVectorize/RISCV/vplan-vp-cmp-intrinsics.ll
Outdated
Show resolved
Hide resolved
llvm/test/Transforms/LoopVectorize/RISCV/vplan-vp-cmp-intrinsics.ll
Outdated
Show resolved
Hide resolved
LiqinWeng
force-pushed
the
widen-vp-cmp-with-evl
branch
from
4b50426 to
f788c9c
Compare
|
ping |
Mel-Chen
reviewed
Oct 23, 2024
| VectorType *RetType = VectorType::get(Type::getInt1Ty(Ctx), State.VF); | ||
| Value *VPInst = Builder.createVectorInstruction(Opcode, RetType, | ||
| {Op1, Op2, Pred}, "vp.op"); | ||
| if (auto *VecOp = dyn_cast<CastInst>(VPInst)) |
There was a problem hiding this comment.
Why createVectorInstruction may return CastInst if Opcode is ICmp/FCmp?
Comment on lines
1433
to
1484
| if (!Instruction::isBinaryOp(Opcode) && !Instruction::isUnaryOp(Opcode)) | ||
| llvm_unreachable("Unsupported opcode in VPWidenEVLRecipe::execute"); |
LiqinWeng
force-pushed
the
widen-vp-cmp-with-evl
branch
3 times, most recently
from
63f1ab9 to
d50dd93
Compare
| // TODO: Support other opcodes | ||
| if (!Instruction::isBinaryOp(Opcode) && !Instruction::isUnaryOp(Opcode)) | ||
| llvm_unreachable("Unsupported opcode in VPWidenEVLRecipe::execute"); | ||
| if (Opcode == Instruction::ICmp || Opcode == Instruction::FCmp) { |
There was a problem hiding this comment.
Shall we instead represent it as WPWidentIntrisic?
There was a problem hiding this comment.
If possible that would be great, given that there are quite a few changes to the existing code here
There was a problem hiding this comment.
…ions with EVL-vectorization
LiqinWeng
force-pushed
the
widen-vp-cmp-with-evl
branch
from
d50dd93 to
74a4321
Compare
LiqinWeng
force-pushed
the
widen-vp-cmp-with-evl
branch
from
6c89f9d to
f219d2c
Compare
|
JFYI, we've now landed a change which moves away from representing most non-trapping instructions using VP intrinsics, and instead relying on the VLOptimizer to reduce the VL for normal IR nodes instead. Skimming this review title, I suspect this patch is no longer needed and can be closed. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
No description provided.