[llvm] [SelectionDAG] Handle scalable INSERT_SUBVECTOR in ComputeNumSignBits (PR #221631)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 7 04:38:16 PDT 2026
https://github.com/pkuyo updated https://github.com/llvm/llvm-project/pull/221631
>From 004365541ea13bade0ba297d56f4276d4f3e66de Mon Sep 17 00:00:00 2001
From: Pkuyo <miku_mi at foxmail.com>
Date: Sun, 6 Sep 2026 20:52:18 -0700
Subject: [PATCH 1/3] [SelectionDAG] Handle scalable in ComputeNumSignBits
INSERT_SUBVECTOR
---
.../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 11 +++--
.../AArch64/AArch64SelectionDAGTest.cpp | 41 +++++++++++++++++++
2 files changed, 48 insertions(+), 4 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 1c5a162fdcc11..40a36bee3ad30 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5423,12 +5423,15 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, const APInt &DemandedElts,
return Tmp;
}
case ISD::INSERT_SUBVECTOR: {
- if (VT.isScalableVector())
- break;
- // Demand any elements from the subvector and the remainder from the src its
- // inserted into.
SDValue Src = Op.getOperand(0);
SDValue Sub = Op.getOperand(1);
+ if (VT.isScalableVector()) {
+ Tmp = ComputeNumSignBits(Sub, Depth + 1);
+ Tmp = std::min(Tmp, ComputeNumSignBits(Src, Depth + 1));
+ return Tmp;
+ }
+ // Demand any elements from the subvector and the remainder from the src its
+ // inserted into.
uint64_t Idx = Op.getConstantOperandVal(2);
unsigned NumSubElts = Sub.getValueType().getVectorNumElements();
APInt DemandedSubElts = DemandedElts.extractBits(NumSubElts, Idx);
diff --git a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
index be4e927eecb22..c13ff6c0d3231 100644
--- a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
+++ b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
@@ -211,6 +211,47 @@ TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_EXTRACT_SUBVECTOR) {
EXPECT_EQ(DAG->ComputeNumSignBits(Op, DemandedElts), 7u);
}
+TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_INSERT_SUBVECTOR) {
+ SDLoc Loc;
+ SDValue Src = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::nxv4i32,
+ DAG->getRegister(0, MVT::nxv4i8));
+ SDValue Sub = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::nxv2i32,
+ DAG->getRegister(1, MVT::nxv2i16));
+ SDValue SrcWithFewerSignBits = DAG->getNode(
+ ISD::SIGN_EXTEND, Loc, MVT::nxv4i32, DAG->getRegister(0, MVT::nxv4i16));
+ SDValue SubWithMoreSignBits = DAG->getNode(
+ ISD::SIGN_EXTEND, Loc, MVT::nxv2i32, DAG->getRegister(1, MVT::nxv2i8));
+
+ for (unsigned Idx : {0u, 2u}) {
+ SDValue Index = DAG->getConstant(Idx, Loc, MVT::i64);
+ SDValue Op =
+ DAG->getNode(ISD::INSERT_SUBVECTOR, Loc, MVT::nxv4i32, Src, Sub, Index);
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 17u);
+
+ Op = DAG->getNode(ISD::INSERT_SUBVECTOR, Loc, MVT::nxv4i32,
+ SrcWithFewerSignBits, SubWithMoreSignBits, Index);
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 17u);
+ }
+}
+
+TEST_F(AArch64SelectionDAGTest,
+ ComputeNumSignBitsSVE_INSERT_SUBVECTOR_FixedSubvector) {
+ SDLoc Loc;
+ SDValue Src = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::nxv4i32,
+ DAG->getRegister(0, MVT::nxv4i8));
+ SDValue Lane0 = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::i32,
+ DAG->getRegister(1, MVT::i8));
+ SDValue Lane1 = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::i32,
+ DAG->getRegister(2, MVT::i16));
+ SDValue Sub = DAG->getBuildVector(MVT::v2i32, Loc, {Lane0, Lane1});
+
+ for (unsigned Idx : {0u, 2u}) {
+ SDValue Op = DAG->getNode(ISD::INSERT_SUBVECTOR, Loc, MVT::nxv4i32, Src,
+ Sub, DAG->getConstant(Idx, Loc, MVT::i64));
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 17u);
+ }
+}
+
TEST_F(AArch64SelectionDAGTest, ComputeNumSignBits_VASHR) {
SDLoc Loc;
auto VecVT = MVT::v8i8;
>From 3b6522948084692909c5f1e9c87711050421bcbb Mon Sep 17 00:00:00 2001
From: Pkuyo <miku_mi at foxmail.com>
Date: Mon, 7 Sep 2026 02:17:12 -0700
Subject: [PATCH 2/3] [SelectionDAG] Add IR tests for scalable INSERT_SUBVECTOR
sign bits
---
.../test/CodeGen/AArch64/sve-insert-vector.ll | 75 +++++++++++++++++++
1 file changed, 75 insertions(+)
diff --git a/llvm/test/CodeGen/AArch64/sve-insert-vector.ll b/llvm/test/CodeGen/AArch64/sve-insert-vector.ll
index 9aebf291c0b3e..041a065687b05 100644
--- a/llvm/test/CodeGen/AArch64/sve-insert-vector.ll
+++ b/llvm/test/CodeGen/AArch64/sve-insert-vector.ll
@@ -1543,4 +1543,79 @@ define <vscale x 4 x float> @insert_nxv1f32_nxv4f32_3(<vscale x 4 x float> %vec,
ret <vscale x 4 x float> %retval
}
+define <vscale x 4 x i32> @insert_nxv4i32_nxv2i32_signbits(<vscale x 4 x i32> %vec, <vscale x 4 x i32> %subvec) {
+; CHECK-LABEL: insert_nxv4i32_nxv2i32_signbits:
+; CHECK: // %bb.0:
+; CHECK-NEXT: asr z0.s, z0.s, #24
+; CHECK-NEXT: asr z1.s, z1.s, #16
+; CHECK-NEXT: uunpklo z1.d, z1.s
+; CHECK-NEXT: uunpkhi z0.d, z0.s
+; CHECK-NEXT: uzp1 z0.s, z1.s, z0.s
+; CHECK-NEXT: ret
+ %base = ashr <vscale x 4 x i32> %vec, splat (i32 24)
+ %sub0 = ashr <vscale x 4 x i32> %subvec, splat (i32 16)
+ %sub = call <vscale x 2 x i32> @llvm.vector.extract.nxv2i32.nxv4i32(<vscale x 4 x i32> %sub0, i64 0)
+ %res = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.nxv2i32(<vscale x 4 x i32> %base, <vscale x 2 x i32> %sub, i64 0)
+ %shl = shl <vscale x 4 x i32> %res, splat (i32 16)
+ %sext = ashr <vscale x 4 x i32> %shl, splat (i32 16)
+ ret <vscale x 4 x i32> %sext
+}
+
+define <vscale x 4 x i32> @insert_nxv4i32_v2i32_signbits(<vscale x 4 x i32> %vec, <4 x i32> %subvec) {
+; CHECK-LABEL: insert_nxv4i32_v2i32_signbits:
+; CHECK: // %bb.0:
+; CHECK-NEXT: asr z0.s, z0.s, #24
+; CHECK-NEXT: sshr v1.2s, v1.2s, #16
+; CHECK-NEXT: ptrue p0.s, vl2
+; CHECK-NEXT: mov z0.s, p0/m, z1.s
+; CHECK-NEXT: ret
+ %base = ashr <vscale x 4 x i32> %vec, splat (i32 24)
+ %sub0 = ashr <4 x i32> %subvec, splat (i32 16)
+ %sub = call <2 x i32> @llvm.vector.extract.v2i32.v4i32(<4 x i32> %sub0, i64 0)
+ %res = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.v2i32(<vscale x 4 x i32> %base, <2 x i32> %sub, i64 0)
+ %shl = shl <vscale x 4 x i32> %res, splat (i32 16)
+ %sext = ashr <vscale x 4 x i32> %shl, splat (i32 16)
+ ret <vscale x 4 x i32> %sext
+}
+
+define <vscale x 4 x i32> @insert_nxv4i32_nxv2i32_signbits_insufficient_base(<vscale x 4 x i32> %vec, <vscale x 4 x i32> %subvec) {
+; CHECK-LABEL: insert_nxv4i32_nxv2i32_signbits_insufficient_base:
+; CHECK: // %bb.0:
+; CHECK-NEXT: asr z0.s, z0.s, #15
+; CHECK-NEXT: asr z1.s, z1.s, #16
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: uunpklo z1.d, z1.s
+; CHECK-NEXT: uunpkhi z0.d, z0.s
+; CHECK-NEXT: uzp1 z0.s, z1.s, z0.s
+; CHECK-NEXT: sxth z0.s, p0/m, z0.s
+; CHECK-NEXT: ret
+ %base = ashr <vscale x 4 x i32> %vec, splat (i32 15)
+ %sub0 = ashr <vscale x 4 x i32> %subvec, splat (i32 16)
+ %sub = call <vscale x 2 x i32> @llvm.vector.extract.nxv2i32.nxv4i32(<vscale x 4 x i32> %sub0, i64 0)
+ %res = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.nxv2i32(<vscale x 4 x i32> %base, <vscale x 2 x i32> %sub, i64 0)
+ %shl = shl <vscale x 4 x i32> %res, splat (i32 16)
+ %sext = ashr <vscale x 4 x i32> %shl, splat (i32 16)
+ ret <vscale x 4 x i32> %sext
+}
+
+define <vscale x 4 x i32> @insert_nxv4i32_nxv2i32_signbits_insufficient_sub(<vscale x 4 x i32> %vec, <vscale x 4 x i32> %subvec) {
+; CHECK-LABEL: insert_nxv4i32_nxv2i32_signbits_insufficient_sub:
+; CHECK: // %bb.0:
+; CHECK-NEXT: asr z0.s, z0.s, #24
+; CHECK-NEXT: asr z1.s, z1.s, #15
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: uunpklo z1.d, z1.s
+; CHECK-NEXT: uunpkhi z0.d, z0.s
+; CHECK-NEXT: uzp1 z0.s, z1.s, z0.s
+; CHECK-NEXT: sxth z0.s, p0/m, z0.s
+; CHECK-NEXT: ret
+ %base = ashr <vscale x 4 x i32> %vec, splat (i32 24)
+ %sub0 = ashr <vscale x 4 x i32> %subvec, splat (i32 15)
+ %sub = call <vscale x 2 x i32> @llvm.vector.extract.nxv2i32.nxv4i32(<vscale x 4 x i32> %sub0, i64 0)
+ %res = call <vscale x 4 x i32> @llvm.vector.insert.nxv4i32.nxv2i32(<vscale x 4 x i32> %base, <vscale x 2 x i32> %sub, i64 0)
+ %shl = shl <vscale x 4 x i32> %res, splat (i32 16)
+ %sext = ashr <vscale x 4 x i32> %shl, splat (i32 16)
+ ret <vscale x 4 x i32> %sext
+}
+
attributes #0 = { vscale_range(2,2) }
>From a3fae990b6653eb9783357c58278ce44ef2c96ab Mon Sep 17 00:00:00 2001
From: Pkuyo <miku_mi at foxmail.com>
Date: Mon, 7 Sep 2026 03:56:09 -0700
Subject: [PATCH 3/3] [SelectionDAG] Remove scalable INSERT_SUBVECTOR unit
tests
---
.../AArch64/AArch64SelectionDAGTest.cpp | 41 -------------------
1 file changed, 41 deletions(-)
diff --git a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
index c13ff6c0d3231..be4e927eecb22 100644
--- a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
+++ b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
@@ -211,47 +211,6 @@ TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_EXTRACT_SUBVECTOR) {
EXPECT_EQ(DAG->ComputeNumSignBits(Op, DemandedElts), 7u);
}
-TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_INSERT_SUBVECTOR) {
- SDLoc Loc;
- SDValue Src = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::nxv4i32,
- DAG->getRegister(0, MVT::nxv4i8));
- SDValue Sub = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::nxv2i32,
- DAG->getRegister(1, MVT::nxv2i16));
- SDValue SrcWithFewerSignBits = DAG->getNode(
- ISD::SIGN_EXTEND, Loc, MVT::nxv4i32, DAG->getRegister(0, MVT::nxv4i16));
- SDValue SubWithMoreSignBits = DAG->getNode(
- ISD::SIGN_EXTEND, Loc, MVT::nxv2i32, DAG->getRegister(1, MVT::nxv2i8));
-
- for (unsigned Idx : {0u, 2u}) {
- SDValue Index = DAG->getConstant(Idx, Loc, MVT::i64);
- SDValue Op =
- DAG->getNode(ISD::INSERT_SUBVECTOR, Loc, MVT::nxv4i32, Src, Sub, Index);
- EXPECT_EQ(DAG->ComputeNumSignBits(Op), 17u);
-
- Op = DAG->getNode(ISD::INSERT_SUBVECTOR, Loc, MVT::nxv4i32,
- SrcWithFewerSignBits, SubWithMoreSignBits, Index);
- EXPECT_EQ(DAG->ComputeNumSignBits(Op), 17u);
- }
-}
-
-TEST_F(AArch64SelectionDAGTest,
- ComputeNumSignBitsSVE_INSERT_SUBVECTOR_FixedSubvector) {
- SDLoc Loc;
- SDValue Src = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::nxv4i32,
- DAG->getRegister(0, MVT::nxv4i8));
- SDValue Lane0 = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::i32,
- DAG->getRegister(1, MVT::i8));
- SDValue Lane1 = DAG->getNode(ISD::SIGN_EXTEND, Loc, MVT::i32,
- DAG->getRegister(2, MVT::i16));
- SDValue Sub = DAG->getBuildVector(MVT::v2i32, Loc, {Lane0, Lane1});
-
- for (unsigned Idx : {0u, 2u}) {
- SDValue Op = DAG->getNode(ISD::INSERT_SUBVECTOR, Loc, MVT::nxv4i32, Src,
- Sub, DAG->getConstant(Idx, Loc, MVT::i64));
- EXPECT_EQ(DAG->ComputeNumSignBits(Op), 17u);
- }
-}
-
TEST_F(AArch64SelectionDAGTest, ComputeNumSignBits_VASHR) {
SDLoc Loc;
auto VecVT = MVT::v8i8;
More information about the llvm-commits
mailing list