[llvm] [LV] Add support for partial alias masking with tail folding (PR #182457)

[Benjamin Maxwell via llvm-commits]

================
@@ -0,0 +1,435 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph:" --version 5
+; RUN: opt -S -mtriple=aarch64-unknown-linux-gnu -mattr=+sve2 -passes=loop-vectorize -force-partial-aliasing-vectorization -tail-folding-policy=must-fold-tail %s | FileCheck %s --check-prefix=CHECK-TF
+
+define void @alias_mask(ptr noalias %a, ptr %b, ptr %c, i64 %n) {
+; CHECK-TF-LABEL: define void @alias_mask(
+; CHECK-TF-SAME: ptr noalias [[A:%.*]], ptr [[B:%.*]], ptr [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-TF-NEXT:  [[ENTRY:.*:]]
+; CHECK-TF-NEXT:    [[B2:%.*]] = ptrtoaddr ptr [[B]] to i64
+; CHECK-TF-NEXT:    [[C1:%.*]] = ptrtoaddr ptr [[C]] to i64
+; CHECK-TF-NEXT:    [[CMP11:%.*]] = icmp sgt i64 [[N]], 0
+; CHECK-TF-NEXT:    br i1 [[CMP11]], label %[[FOR_BODY_PREHEADER:.*]], [[EXIT:label %.*]]
+; CHECK-TF:       [[FOR_BODY_PREHEADER]]:
+; CHECK-TF-NEXT:    br label %[[VECTOR_CLAMPED_VF_CHECK:.*]]
+; CHECK-TF:       [[VECTOR_CLAMPED_VF_CHECK]]:
+; CHECK-TF-NEXT:    [[ALIAS_MASK:%.*]] = call <vscale x 16 x i1> @llvm.loop.dependence.war.mask.nxv16i1.i64(i64 [[B2]], i64 [[C1]], i64 1)
+; CHECK-TF-NEXT:    [[TMP3:%.*]] = zext <vscale x 16 x i1> [[ALIAS_MASK]] to <vscale x 16 x i32>
+; CHECK-TF-NEXT:    [[TMP4:%.*]] = call i32 @llvm.vector.reduce.add.nxv16i32(<vscale x 16 x i32> [[TMP3]])
+; CHECK-TF-NEXT:    [[NUM_ACTIVE_LANES:%.*]] = zext i32 [[TMP4]] to i64
+; CHECK-TF-NEXT:    [[VF_IS_SCALAR:%.*]] = icmp ule i64 [[NUM_ACTIVE_LANES]], 1
+; CHECK-TF-NEXT:    [[TMP5:%.*]] = sub i64 -1, [[N]]
+; CHECK-TF-NEXT:    [[VF_STEP_OVERFLOW:%.*]] = icmp ult i64 [[TMP5]], [[NUM_ACTIVE_LANES]]
+; CHECK-TF-NEXT:    [[TMP6:%.*]] = or i1 [[VF_IS_SCALAR]], [[VF_STEP_OVERFLOW]]
+; CHECK-TF-NEXT:    br i1 [[TMP6]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK-TF:       [[VECTOR_PH]]:
+; CHECK-TF-NEXT:    [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i64(i64 0, i64 [[N]])
+; CHECK-TF-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK-TF:       [[VECTOR_BODY]]:
+; CHECK-TF-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-TF-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = phi <vscale x 16 x i1> [ [[ACTIVE_LANE_MASK_ENTRY]], %[[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-TF-NEXT:    [[TMP10:%.*]] = and <vscale x 16 x i1> [[ACTIVE_LANE_MASK]], [[ALIAS_MASK]]
+; CHECK-TF-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]]
+; CHECK-TF-NEXT:    [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], <vscale x 16 x i1> [[TMP10]], <vscale x 16 x i8> poison)
+; CHECK-TF-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]]
+; CHECK-TF-NEXT:    [[WIDE_MASKED_LOAD3:%.*]] = call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP12]], <vscale x 16 x i1> [[TMP10]], <vscale x 16 x i8> poison)
+; CHECK-TF-NEXT:    [[TMP13:%.*]] = select <vscale x 16 x i1> [[TMP10]], <vscale x 16 x i8> [[WIDE_MASKED_LOAD]], <vscale x 16 x i8> splat (i8 1)
+; CHECK-TF-NEXT:    [[TMP14:%.*]] = sdiv <vscale x 16 x i8> [[WIDE_MASKED_LOAD3]], [[TMP13]]
+; CHECK-TF-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[INDEX]]
+; CHECK-TF-NEXT:    call void @llvm.masked.store.nxv16i8.p0(<vscale x 16 x i8> [[TMP14]], ptr align 1 [[TMP15]], <vscale x 16 x i1> [[TMP10]])
+; CHECK-TF-NEXT:    [[INDEX_NEXT]] = add i64 [[INDEX]], [[NUM_ACTIVE_LANES]]
+; CHECK-TF-NEXT:    [[ACTIVE_LANE_MASK_NEXT]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i64(i64 [[INDEX_NEXT]], i64 [[N]])
+; CHECK-TF-NEXT:    [[TMP16:%.*]] = extractelement <vscale x 16 x i1> [[ACTIVE_LANE_MASK_NEXT]], i64 0
+; CHECK-TF-NEXT:    [[TMP17:%.*]] = xor i1 [[TMP16]], true
+; CHECK-TF-NEXT:    br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-TF:       [[MIDDLE_BLOCK]]:
+; CHECK-TF-NEXT:    br [[EXIT_LOOPEXIT:label %.*]]
+; CHECK-TF:       [[SCALAR_PH]]:
+;
+
+entry:
+  %cmp11 = icmp sgt i64 %n, 0
+  br i1 %cmp11, label %for.body, label %exit
----------------
MacDue wrote:

no (removed) 

https://github.com/llvm/llvm-project/pull/182457