Skip to content

Commit 98e747b

Browse files
pedroclobopedroclobo
authored
[clang] Use poison instead of undef as the placeholder when creating a new vector [NFC] (#117064)
Call `@llvm.vector.insert` with a `poison` vector when coercing a fixed vector to a scalable vector with the same element type.
1 parent 4851dbb commit 98e747b

15 files changed

+627
-627
lines changed

‎clang/lib/CodeGen/CGCall.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -1314,10 +1314,10 @@ static llvm::Value *CreateCoercedLoad(Address Src, llvm::Type *Ty,
13141314
}
13151315
if (ScalableDstTy->getElementType() == FixedSrcTy->getElementType()) {
13161316
auto *Load = CGF.Builder.CreateLoad(Src);
1317-
auto *UndefVec = llvm::UndefValue::get(ScalableDstTy);
1317+
auto *PoisonVec = llvm::PoisonValue::get(ScalableDstTy);
13181318
auto *Zero = llvm::Constant::getNullValue(CGF.CGM.Int64Ty);
13191319
llvm::Value *Result = CGF.Builder.CreateInsertVector(
1320-
ScalableDstTy, UndefVec, Load, Zero, "cast.scalable");
1320+
ScalableDstTy, PoisonVec, Load, Zero, "cast.scalable");
13211321
if (ScalableDstTy != Ty)
13221322
Result = CGF.Builder.CreateBitCast(Result, Ty);
13231323
return Result;

‎clang/test/CodeGen/AArch64/pure-scalable-args.c

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -70,23 +70,23 @@ void test_argpass_simple(PST *p) {
7070
// CHECK-AAPCS: define dso_local void @test_argpass_simple(ptr nocapture noundef readonly %p)
7171
// CHECK-AAPCS-NEXT: entry:
7272
// CHECK-AAPCS-NEXT: %0 = load <2 x i8>, ptr %p, align 16
73-
// CHECK-AAPCS-NEXT: %cast.scalable = tail call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> undef, <2 x i8> %0, i64 0)
73+
// CHECK-AAPCS-NEXT: %cast.scalable = tail call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> poison, <2 x i8> %0, i64 0)
7474
// CHECK-AAPCS-NEXT: %1 = bitcast <vscale x 2 x i8> %cast.scalable to <vscale x 16 x i1>
7575
// CHECK-AAPCS-NEXT: %2 = getelementptr inbounds nuw i8, ptr %p, i64 16
7676
// CHECK-AAPCS-NEXT: %3 = load <2 x double>, ptr %2, align 16
77-
// CHECK-AAPCS-NEXT: %cast.scalable1 = tail call <vscale x 2 x double> @llvm.vector.insert.nxv2f64.v2f64(<vscale x 2 x double> undef, <2 x double> %3, i64 0)
77+
// CHECK-AAPCS-NEXT: %cast.scalable1 = tail call <vscale x 2 x double> @llvm.vector.insert.nxv2f64.v2f64(<vscale x 2 x double> poison, <2 x double> %3, i64 0)
7878
// CHECK-AAPCS-NEXT: %4 = getelementptr inbounds nuw i8, ptr %p, i64 32
7979
// CHECK-AAPCS-NEXT: %5 = load <4 x float>, ptr %4, align 16
80-
// CHECK-AAPCS-NEXT: %cast.scalable2 = tail call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> undef, <4 x float> %5, i64 0)
80+
// CHECK-AAPCS-NEXT: %cast.scalable2 = tail call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> poison, <4 x float> %5, i64 0)
8181
// CHECK-AAPCS-NEXT: %6 = getelementptr inbounds nuw i8, ptr %p, i64 48
8282
// CHECK-AAPCS-NEXT: %7 = load <4 x float>, ptr %6, align 16
83-
// CHECK-AAPCS-NEXT: %cast.scalable3 = tail call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> undef, <4 x float> %7, i64 0)
83+
// CHECK-AAPCS-NEXT: %cast.scalable3 = tail call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> poison, <4 x float> %7, i64 0)
8484
// CHECK-AAPCS-NEXT: %8 = getelementptr inbounds nuw i8, ptr %p, i64 64
8585
// CHECK-AAPCS-NEXT: %9 = load <16 x i8>, ptr %8, align 16
86-
// CHECK-AAPCS-NEXT: %cast.scalable4 = tail call <vscale x 16 x i8> @llvm.vector.insert.nxv16i8.v16i8(<vscale x 16 x i8> undef, <16 x i8> %9, i64 0)
86+
// CHECK-AAPCS-NEXT: %cast.scalable4 = tail call <vscale x 16 x i8> @llvm.vector.insert.nxv16i8.v16i8(<vscale x 16 x i8> poison, <16 x i8> %9, i64 0)
8787
// CHECK-AAPCS-NEXT: %10 = getelementptr inbounds nuw i8, ptr %p, i64 80
8888
// CHECK-AAPCS-NEXT: %11 = load <2 x i8>, ptr %10, align 16
89-
// CHECK-AAPCS-NEXT: %cast.scalable5 = tail call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> undef, <2 x i8> %11, i64 0)
89+
// CHECK-AAPCS-NEXT: %cast.scalable5 = tail call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> poison, <2 x i8> %11, i64 0)
9090
// CHECK-AAPCS-NEXT: %12 = bitcast <vscale x 2 x i8> %cast.scalable5 to <vscale x 16 x i1>
9191
// CHECK-AAPCS-NEXT: tail call void @argpass_simple_callee(<vscale x 16 x i1> %1, <vscale x 2 x double> %cast.scalable1, <vscale x 4 x float> %cast.scalable2, <vscale x 4 x float> %cast.scalable3, <vscale x 16 x i8> %cast.scalable4, <vscale x 16 x i1> %12)
9292
// CHECK-AAPCS-NEXT: ret void
@@ -431,9 +431,9 @@ void test_va_arg(int n, ...) {
431431
// CHECK-AAPCS-NEXT: %v.sroa.43.0.vaarg.addr.sroa_idx = getelementptr inbounds i8, ptr %vaarg.addr, i64 48
432432
// CHECK-AAPCS-NEXT: %v.sroa.43.0.copyload = load <4 x float>, ptr %v.sroa.43.0.vaarg.addr.sroa_idx, align 16
433433
// CHECK-AAPCS-NEXT: call void @llvm.va_end.p0(ptr nonnull %ap)
434-
// CHECK-AAPCS-NEXT: %cast.scalable = call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> undef, <2 x i8> %v.sroa.0.0.copyload, i64 0)
434+
// CHECK-AAPCS-NEXT: %cast.scalable = call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> poison, <2 x i8> %v.sroa.0.0.copyload, i64 0)
435435
// CHECK-AAPCS-NEXT: %3 = bitcast <vscale x 2 x i8> %cast.scalable to <vscale x 16 x i1>
436-
// CHECK-AAPCS-NEXT: %cast.scalable2 = call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> undef, <4 x float> %v.sroa.43.0.copyload, i64 0)
436+
// CHECK-AAPCS-NEXT: %cast.scalable2 = call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> poison, <4 x float> %v.sroa.43.0.copyload, i64 0)
437437
// CHECK-AAPCS-NEXT: call void @use1(<vscale x 16 x i1> noundef %3, <vscale x 4 x float> noundef %cast.scalable2)
438438
// CHECK-AAPCS-NEXT: call void @llvm.lifetime.end.p0(i64 32, ptr nonnull %ap)
439439
// CHECK-AAPCS-NEXT: ret void
@@ -452,9 +452,9 @@ void test_va_arg(int n, ...) {
452452
// CHECK-DARWIN-NEXT: %v.sroa.43.0..sroa_idx = getelementptr inbounds i8, ptr %0, i64 48
453453
// CHECK-DARWIN-NEXT: %v.sroa.43.0.copyload = load <4 x float>, ptr %v.sroa.43.0..sroa_idx, align 16
454454
// CHECK-DARWIN-NEXT: call void @llvm.va_end.p0(ptr nonnull %ap)
455-
// CHECK-DARWIN-NEXT: %cast.scalable = call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> undef, <2 x i8> %v.sroa.0.0.copyload, i64 0)
455+
// CHECK-DARWIN-NEXT: %cast.scalable = call <vscale x 2 x i8> @llvm.vector.insert.nxv2i8.v2i8(<vscale x 2 x i8> poison, <2 x i8> %v.sroa.0.0.copyload, i64 0)
456456
// CHECK-DARWIN-NEXT: %1 = bitcast <vscale x 2 x i8> %cast.scalable to <vscale x 16 x i1>
457-
// CHECK-DARWIN-NEXT: %cast.scalable2 = call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> undef, <4 x float> %v.sroa.43.0.copyload, i64 0)
457+
// CHECK-DARWIN-NEXT: %cast.scalable2 = call <vscale x 4 x float> @llvm.vector.insert.nxv4f32.v4f32(<vscale x 4 x float> poison, <4 x float> %v.sroa.43.0.copyload, i64 0)
458458
// CHECK-DARWIN-NEXT: call void @use1(<vscale x 16 x i1> noundef %1, <vscale x 4 x float> noundef %cast.scalable2)
459459
// CHECK-DARWIN-NEXT: call void @llvm.lifetime.end.p0(i64 8, ptr nonnull %ap)
460460
// CHECK-DARWIN-NEXT: ret void

‎clang/test/CodeGen/AArch64/sve-acle-__ARM_FEATURE_SVE_VECTOR_OPERATORS.cpp

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -52,7 +52,7 @@ void test02() {
5252
// CHECK-NEXT: [[X:%.*]] = tail call <[[#div(VBITS, 32)]] x i32> @llvm.vector.extract.v[[#div(VBITS, 32)]]i32.nxv4i32(<vscale x 4 x i32> [[X_COERCE:%.*]], i64 0)
5353
// CHECK-NEXT: [[Y:%.*]] = tail call <[[#div(VBITS, 32)]] x i32> @llvm.vector.extract.v[[#div(VBITS, 32)]]i32.nxv4i32(<vscale x 4 x i32> [[X_COERCE1:%.*]], i64 0)
5454
// CHECK-NEXT: [[ADD:%.*]] = add <[[#div(VBITS, 32)]] x i32> [[Y]], [[X]]
55-
// CHECK-NEXT: [[CASTSCALABLESVE:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v[[#div(VBITS, 32)]]i32(<vscale x 4 x i32> undef, <[[#div(VBITS, 32)]] x i32> [[ADD]], i64 0)
55+
// CHECK-NEXT: [[CASTSCALABLESVE:%.*]] = tail call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v[[#div(VBITS, 32)]]i32(<vscale x 4 x i32> poison, <[[#div(VBITS, 32)]] x i32> [[ADD]], i64 0)
5656
// CHECK-NEXT: ret <vscale x 4 x i32> [[CASTSCALABLESVE]]
5757
typedef svint32_t vec __attribute__((arm_sve_vector_bits(N)));
5858
auto f(vec x, vec y) { return x + y; } // Returns a vec.

0 commit comments

Comments
 (0)