[llvm] [SelectionDAG] Add GET_ACTIVE_LANE_MASK to ComputeNumSignBits (PR #208977)
Usman Nadeem via llvm-commits
llvm-commits at lists.llvm.org
Sat Jul 11 22:03:26 PDT 2026
https://github.com/UsmanNadeem created https://github.com/llvm/llvm-project/pull/208977
`GET_ACTIVE_LANE_MASK ` acts like `icmp ult` so code is similar to `SETCC`.
>From 3befef2349d804d48965d9c57c2807dc95f3e325 Mon Sep 17 00:00:00 2001
From: Usman Nadeem <mnadeem at qti.qualcomm.com>
Date: Sat, 11 Jul 2026 16:55:47 -0700
Subject: [PATCH 1/2] Precommit tests
Change-Id: Icd0172e3d9b147a8966d5d2c3aa53fcf6d136c77
---
...fold-sext-in-reg-predicate-fixed-length.ll | 41 +++++++++++++++++++
.../AArch64/AArch64SelectionDAGTest.cpp | 25 +++++++++++
2 files changed, 66 insertions(+)
diff --git a/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll b/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll
index 783cd6f693ed8..044e8ae82b694 100644
--- a/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll
+++ b/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll
@@ -17,3 +17,44 @@ entry:
%data = call <16 x i8> @llvm.masked.load.v16i8.p0(ptr %p, i32 1, <16 x i1> %mask, <16 x i8> zeroinitializer)
ret <16 x i8> %data
}
+
+define void @active_lane_mask_mstore_vscaleX2(ptr %p, i64 %n) #0 {
+; CHECK-LABEL: active_lane_mask_mstore_vscaleX2:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: whilelo p0.b, xzr, x1
+; CHECK-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT: ptrue p0.h
+; CHECK-NEXT: uunpklo z0.h, z0.b
+; CHECK-NEXT: lsl z0.h, z0.h, #15
+; CHECK-NEXT: asr z0.h, z0.h, #15
+; CHECK-NEXT: cmpne p1.h, p0/z, z0.h, #0
+; CHECK-NEXT: mov z0.h, #123 // =0x7b
+; CHECK-NEXT: st1h { z0.h }, p1, [x0]
+; CHECK-NEXT: ret
+entry:
+ %mask = call <16 x i1> @llvm.get.active.lane.mask.v16i1(i64 0, i64 %n)
+ call void @llvm.masked.store.v16i16.p0(<16 x i16> splat(i16 123), ptr %p, <16 x i1> %mask)
+ ret void
+}
+
+define void @active_lane_mask_mstore_vscaleX4(ptr %p, i64 %n) #1 {
+; CHECK-LABEL: active_lane_mask_mstore_vscaleX4:
+; CHECK: // %bb.0:
+; CHECK-NEXT: whilelo p0.b, xzr, x1
+; CHECK-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
+; CHECK-NEXT: ptrue p0.s
+; CHECK-NEXT: uunpklo z0.h, z0.b
+; CHECK-NEXT: uunpklo z0.s, z0.h
+; CHECK-NEXT: lsl z0.s, z0.s, #31
+; CHECK-NEXT: asr z0.s, z0.s, #31
+; CHECK-NEXT: cmpne p1.s, p0/z, z0.s, #0
+; CHECK-NEXT: mov z0.s, #123 // =0x7b
+; CHECK-NEXT: st1w { z0.s }, p1, [x0]
+; CHECK-NEXT: ret
+ %mask = call <16 x i1> @llvm.get.active.lane.mask.v16i1.i64(i64 0, i64 %n)
+ call void @llvm.masked.store.v16i32.p0(<16 x i32> splat(i32 123), ptr %p, <16 x i1> %mask)
+ ret void
+}
+
+attributes #0 = { vscale_range(2,2) }
+attributes #1 = { vscale_range(4,4) }
diff --git a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
index b144060afe925..6edbebb6e3a0b 100644
--- a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
+++ b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
@@ -128,6 +128,31 @@ TEST_F(AArch64SelectionDAGTest, computeKnownBits_EXTRACT_SUBVECTOR) {
EXPECT_TRUE(Known.isZero());
}
+TEST_F(AArch64SelectionDAGTest, ComputeNumSignBits_GET_ACTIVE_LANE_MASK) {
+ // GET_ACTIVE_LANE_MASK promoted/widened to a vector integer type wider
+ // than i1 (e.g. v8i8) should report that all bits of each lane are sign
+ // bits.
+ SDLoc Loc;
+ auto Int64VT = EVT::getIntegerVT(Context, 64);
+ auto MaskVT = MVT::v8i8;
+ auto Base = DAG->getConstant(0, Loc, Int64VT);
+ auto TripCount = DAG->getConstant(8, Loc, Int64VT);
+ auto Op =
+ DAG->getNode(ISD::GET_ACTIVE_LANE_MASK, Loc, MaskVT, Base, TripCount);
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 1u);
+}
+
+TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_GET_ACTIVE_LANE_MASK) {
+ SDLoc Loc;
+ auto Int64VT = EVT::getIntegerVT(Context, 64);
+ auto MaskVT = MVT::nxv8i16;
+ auto Base = DAG->getConstant(0, Loc, Int64VT);
+ auto TripCount = DAG->getConstant(8, Loc, Int64VT);
+ auto Op =
+ DAG->getNode(ISD::GET_ACTIVE_LANE_MASK, Loc, MaskVT, Base, TripCount);
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 1u);
+}
+
TEST_F(AArch64SelectionDAGTest, ComputeNumSignBits_SIGN_EXTEND_VECTOR_INREG) {
SDLoc Loc;
auto Int8VT = EVT::getIntegerVT(Context, 8);
>From 46df85d0557ffe2cdd63453ea435e6d86de62380 Mon Sep 17 00:00:00 2001
From: Usman Nadeem <mnadeem at qti.qualcomm.com>
Date: Sat, 11 Jul 2026 21:17:08 -0700
Subject: [PATCH 2/2] [SelectionDAG] Add GET_ACTIVE_LANE_MASK to
ComputeNumSignBits
This acts like `icmp ult` so code is similar to `SETCC`.
Change-Id: I13a2248b1b50d38fc63c300bcdc86ea6c6f20f33
---
.../lib/CodeGen/SelectionDAG/SelectionDAG.cpp | 6 ++++++
...fold-sext-in-reg-predicate-fixed-length.ll | 20 +++++--------------
.../AArch64/AArch64SelectionDAGTest.cpp | 4 ++--
3 files changed, 13 insertions(+), 17 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 626803ed92a40..bf720addf5b8a 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -5225,6 +5225,12 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, const APInt &DemandedElts,
return VTBits;
break;
}
+ case ISD::GET_ACTIVE_LANE_MASK:
+ // Semantically similar to icmp ult.
+ if (TLI->getBooleanContents(VT.isVector(), /*isFloat=*/false) ==
+ TargetLowering::ZeroOrNegativeOneBooleanContent)
+ return VTBits;
+ break;
case ISD::ROTL:
case ISD::ROTR:
Tmp = ComputeNumSignBits(Op.getOperand(0), DemandedElts, Depth + 1);
diff --git a/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll b/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll
index 044e8ae82b694..747c1619187f4 100644
--- a/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll
+++ b/llvm/test/CodeGen/AArch64/fold-sext-in-reg-predicate-fixed-length.ll
@@ -22,14 +22,9 @@ define void @active_lane_mask_mstore_vscaleX2(ptr %p, i64 %n) #0 {
; CHECK-LABEL: active_lane_mask_mstore_vscaleX2:
; CHECK: // %bb.0: // %entry
; CHECK-NEXT: whilelo p0.b, xzr, x1
-; CHECK-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
-; CHECK-NEXT: ptrue p0.h
-; CHECK-NEXT: uunpklo z0.h, z0.b
-; CHECK-NEXT: lsl z0.h, z0.h, #15
-; CHECK-NEXT: asr z0.h, z0.h, #15
-; CHECK-NEXT: cmpne p1.h, p0/z, z0.h, #0
; CHECK-NEXT: mov z0.h, #123 // =0x7b
-; CHECK-NEXT: st1h { z0.h }, p1, [x0]
+; CHECK-NEXT: punpklo p0.h, p0.b
+; CHECK-NEXT: st1h { z0.h }, p0, [x0]
; CHECK-NEXT: ret
entry:
%mask = call <16 x i1> @llvm.get.active.lane.mask.v16i1(i64 0, i64 %n)
@@ -41,15 +36,10 @@ define void @active_lane_mask_mstore_vscaleX4(ptr %p, i64 %n) #1 {
; CHECK-LABEL: active_lane_mask_mstore_vscaleX4:
; CHECK: // %bb.0:
; CHECK-NEXT: whilelo p0.b, xzr, x1
-; CHECK-NEXT: mov z0.b, p0/z, #-1 // =0xffffffffffffffff
-; CHECK-NEXT: ptrue p0.s
-; CHECK-NEXT: uunpklo z0.h, z0.b
-; CHECK-NEXT: uunpklo z0.s, z0.h
-; CHECK-NEXT: lsl z0.s, z0.s, #31
-; CHECK-NEXT: asr z0.s, z0.s, #31
-; CHECK-NEXT: cmpne p1.s, p0/z, z0.s, #0
; CHECK-NEXT: mov z0.s, #123 // =0x7b
-; CHECK-NEXT: st1w { z0.s }, p1, [x0]
+; CHECK-NEXT: punpklo p0.h, p0.b
+; CHECK-NEXT: punpklo p0.h, p0.b
+; CHECK-NEXT: st1w { z0.s }, p0, [x0]
; CHECK-NEXT: ret
%mask = call <16 x i1> @llvm.get.active.lane.mask.v16i1.i64(i64 0, i64 %n)
call void @llvm.masked.store.v16i32.p0(<16 x i32> splat(i32 123), ptr %p, <16 x i1> %mask)
diff --git a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
index 6edbebb6e3a0b..2362500254dc5 100644
--- a/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
+++ b/llvm/unittests/Target/AArch64/AArch64SelectionDAGTest.cpp
@@ -139,7 +139,7 @@ TEST_F(AArch64SelectionDAGTest, ComputeNumSignBits_GET_ACTIVE_LANE_MASK) {
auto TripCount = DAG->getConstant(8, Loc, Int64VT);
auto Op =
DAG->getNode(ISD::GET_ACTIVE_LANE_MASK, Loc, MaskVT, Base, TripCount);
- EXPECT_EQ(DAG->ComputeNumSignBits(Op), 1u);
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 8u);
}
TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_GET_ACTIVE_LANE_MASK) {
@@ -150,7 +150,7 @@ TEST_F(AArch64SelectionDAGTest, ComputeNumSignBitsSVE_GET_ACTIVE_LANE_MASK) {
auto TripCount = DAG->getConstant(8, Loc, Int64VT);
auto Op =
DAG->getNode(ISD::GET_ACTIVE_LANE_MASK, Loc, MaskVT, Base, TripCount);
- EXPECT_EQ(DAG->ComputeNumSignBits(Op), 1u);
+ EXPECT_EQ(DAG->ComputeNumSignBits(Op), 16u);
}
TEST_F(AArch64SelectionDAGTest, ComputeNumSignBits_SIGN_EXTEND_VECTOR_INREG) {
More information about the llvm-commits
mailing list