Skip to content

Commit 7ff8c04

Browse files
nikolaypanchenkonikolaypanchenko
committed
Refactored code and removed flag propagation as they're not yet
supported by vp intrinsics
1 parent 906c430 commit 7ff8c04

File tree

2 files changed

+20
-26
lines changed

2 files changed

+20
-26
lines changed

‎llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp

Lines changed: 15 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -1267,6 +1267,11 @@ InstructionCost VPWidenRecipe::computeCost(ElementCount VF,
12671267
}
12681268

12691269
void VPWidenEVLRecipe::execute(VPTransformState &State) {
1270+
unsigned Opcode = getOpcode();
1271+
// TODO: Support other opcodes
1272+
if (!Instruction::isBinaryOp(Opcode) && !Instruction::isUnaryOp(Opcode))
1273+
llvm_unreachable("Unsupported opcode in VPWidenEVLRecipe::execute");
1274+
12701275
State.setDebugLocFrom(getDebugLoc());
12711276
assert(State.UF == 1 && "Expected only UF == 1 when vectorizing with "
12721277
"explicit vector length.");
@@ -1277,34 +1282,23 @@ void VPWidenEVLRecipe::execute(VPTransformState &State) {
12771282

12781283
VPValue *EVL = getEVL();
12791284
Value *EVLArg = State.get(EVL, 0, /*NeedsScalar=*/true);
1280-
unsigned Opcode = getOpcode();
1281-
Instruction *I = getUnderlyingInstr();
12821285
IRBuilderBase &BuilderIR = State.Builder;
12831286
VectorBuilder Builder(BuilderIR);
12841287
Value *Mask = BuilderIR.CreateVectorSplat(State.VF, BuilderIR.getTrue());
1285-
Value *VPInst = nullptr;
1286-
1287-
//===------------------- Binary and Unary Ops ---------------------===//
1288-
if (Instruction::isBinaryOp(Opcode) || Instruction::isUnaryOp(Opcode)) {
1289-
// Just widen unops and binops.
12901288

1291-
SmallVector<Value *, 4> Ops;
1292-
for (unsigned I = 0, E = getNumOperands() - 1; I < E; ++I) {
1293-
VPValue *VPOp = getOperand(I);
1294-
Ops.push_back(State.get(VPOp, 0));
1295-
}
1289+
SmallVector<Value *, 4> Ops;
1290+
for (unsigned I = 0, E = getNumOperands() - 1; I < E; ++I) {
1291+
VPValue *VPOp = getOperand(I);
1292+
Ops.push_back(State.get(VPOp, 0));
1293+
}
12961294

1297-
Builder.setMask(Mask).setEVL(EVLArg);
1298-
VPInst = Builder.createVectorInstruction(Opcode, Ops[0]->getType(), Ops,
1299-
"vp.op");
1295+
Builder.setMask(Mask).setEVL(EVLArg);
1296+
Value *VPInst =
1297+
Builder.createVectorInstruction(Opcode, Ops[0]->getType(), Ops, "vp.op");
13001298

1301-
if (auto *VecOp = dyn_cast_or_null<Instruction>(VPInst))
1302-
VecOp->copyIRFlags(I);
1303-
} else {
1304-
llvm_unreachable("Unsupported opcode in VPWidenEVLRecipe::execute");
1305-
}
13061299
State.set(this, VPInst, 0);
1307-
State.addMetadata(VPInst, I);
1300+
State.addMetadata(VPInst,
1301+
dyn_cast_or_null<Instruction>(getUnderlyingValue()));
13081302
}
13091303

13101304
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)

‎llvm/test/Transforms/LoopVectorize/RISCV/vectorize-force-tail-with-evl-bin-unary-ops-args.ll

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1267,7 +1267,7 @@ define void @test_fadd(ptr nocapture %a, ptr nocapture readonly %b) {
12671267
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP13]]
12681268
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP14]], i32 0
12691269
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[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 [[TMP12]])
1270-
; IF-EVL-NEXT: [[TMP16:%.*]] = call fast <vscale x 4 x float> @llvm.vp.fadd.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
1270+
; IF-EVL-NEXT: [[TMP16:%.*]] = call <vscale x 4 x float> @llvm.vp.fadd.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
12711271
; IF-EVL-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP13]]
12721272
; IF-EVL-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP17]], i32 0
12731273
; IF-EVL-NEXT: call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[TMP16]], 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]])
@@ -1362,7 +1362,7 @@ define void @test_fsub(ptr nocapture %a, ptr nocapture readonly %b) {
13621362
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP13]]
13631363
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP14]], i32 0
13641364
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[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 [[TMP12]])
1365-
; IF-EVL-NEXT: [[TMP16:%.*]] = call fast <vscale x 4 x float> @llvm.vp.fsub.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
1365+
; IF-EVL-NEXT: [[TMP16:%.*]] = call <vscale x 4 x float> @llvm.vp.fsub.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
13661366
; IF-EVL-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP13]]
13671367
; IF-EVL-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP17]], i32 0
13681368
; IF-EVL-NEXT: call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[TMP16]], 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]])
@@ -1457,7 +1457,7 @@ define void @test_fmul(ptr nocapture %a, ptr nocapture readonly %b) {
14571457
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP13]]
14581458
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP14]], i32 0
14591459
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[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 [[TMP12]])
1460-
; IF-EVL-NEXT: [[TMP16:%.*]] = call fast <vscale x 4 x float> @llvm.vp.fmul.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
1460+
; IF-EVL-NEXT: [[TMP16:%.*]] = call <vscale x 4 x float> @llvm.vp.fmul.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
14611461
; IF-EVL-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP13]]
14621462
; IF-EVL-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP17]], i32 0
14631463
; IF-EVL-NEXT: call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[TMP16]], 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]])
@@ -1552,7 +1552,7 @@ define void @test_fdiv(ptr nocapture %a, ptr nocapture readonly %b) {
15521552
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP13]]
15531553
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP14]], i32 0
15541554
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[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 [[TMP12]])
1555-
; IF-EVL-NEXT: [[TMP16:%.*]] = call fast <vscale x 4 x float> @llvm.vp.fdiv.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
1555+
; IF-EVL-NEXT: [[TMP16:%.*]] = call <vscale x 4 x float> @llvm.vp.fdiv.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <vscale x 4 x float> shufflevector (<vscale x 4 x float> insertelement (<vscale x 4 x float> poison, float 3.000000e+00, i64 0), <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer), <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]])
15561556
; IF-EVL-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP13]]
15571557
; IF-EVL-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP17]], i32 0
15581558
; IF-EVL-NEXT: call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[TMP16]], 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]])
@@ -1700,7 +1700,7 @@ define void @test_fneg(ptr nocapture %a, ptr nocapture readonly %b) {
17001700
; IF-EVL-NEXT: [[TMP14:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP13]]
17011701
; IF-EVL-NEXT: [[TMP15:%.*]] = getelementptr inbounds float, ptr [[TMP14]], i32 0
17021702
; IF-EVL-NEXT: [[VP_OP_LOAD:%.*]] = call <vscale x 4 x float> @llvm.vp.load.nxv4f32.p0(ptr align 4 [[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 [[TMP12]])
1703-
; IF-EVL-NEXT: [[TMP16:%.*]] = call fast <vscale x 4 x float> @llvm.vp.fneg.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <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]])
1703+
; IF-EVL-NEXT: [[TMP16:%.*]] = call <vscale x 4 x float> @llvm.vp.fneg.nxv4f32(<vscale x 4 x float> [[VP_OP_LOAD]], <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]])
17041704
; IF-EVL-NEXT: [[TMP17:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[TMP13]]
17051705
; IF-EVL-NEXT: [[TMP18:%.*]] = getelementptr inbounds float, ptr [[TMP17]], i32 0
17061706
; IF-EVL-NEXT: call void @llvm.vp.store.nxv4f32.p0(<vscale x 4 x float> [[TMP16]], 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]])

0 commit comments

Comments
 (0)