[LV] Add test case for #121745.

Test for #121745.

Author: fhahn

llvm/test/Transforms/LoopVectorize/iv_outside_user.ll

@@ -979,3 +979,198 @@ loop:
 exit:
   ret float %add
 }
+
+; Test case for https://github.com/llvm/llvm-project/issues/121745.
+; FIXME: At the moment an incorrect exit value is used for %iv.next.
+define i32 @test_iv_uniform_with_outside_use_scev_simplification(ptr %dst) {
+; VEC-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification(
+; VEC-SAME: ptr [[DST:%.*]]) {
+; VEC-NEXT:  [[ENTRY:.*]]:
+; VEC-NEXT:    [[STEP_1:%.*]] = sext i8 0 to i32
+; VEC-NEXT:    [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1
+; VEC-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; VEC:       [[VECTOR_PH]]:
+; VEC-NEXT:    br label %[[VECTOR_BODY:.*]]
+; VEC:       [[VECTOR_BODY]]:
+; VEC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; VEC-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
+; VEC-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]]
+; VEC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[TMP1]], i32 0
+; VEC-NEXT:    store <2 x i16> zeroinitializer, ptr [[TMP2]], align 2
+; VEC-NEXT:    [[TMP4:%.*]] = add i32 [[STEP_2]], [[TMP0]]
+; VEC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; VEC-NEXT:    [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 8
+; VEC-NEXT:    br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
+; VEC:       [[MIDDLE_BLOCK]]:
+; VEC-NEXT:    br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]]
+; VEC:       [[SCALAR_PH]]:
+; VEC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; VEC-NEXT:    br label %[[LOOP:.*]]
+; VEC:       [[LOOP]]:
+; VEC-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; VEC-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]]
+; VEC-NEXT:    store i16 0, ptr [[GEP_DST]], align 2
+; VEC-NEXT:    [[IV_NEXT]] = add i32 [[STEP_2]], [[IV]]
+; VEC-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8
+; VEC-NEXT:    br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}}
+; VEC:       [[E_EXIT]]:
+; VEC-NEXT:    [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP4]], %[[MIDDLE_BLOCK]] ]
+; VEC-NEXT:    ret i32 [[RES]]
+;
+; INTERLEAVE-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification(
+; INTERLEAVE-SAME: ptr [[DST:%.*]]) {
+; INTERLEAVE-NEXT:  [[ENTRY:.*]]:
+; INTERLEAVE-NEXT:    [[STEP_1:%.*]] = sext i8 0 to i32
+; INTERLEAVE-NEXT:    [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1
+; INTERLEAVE-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; INTERLEAVE:       [[VECTOR_PH]]:
+; INTERLEAVE-NEXT:    br label %[[VECTOR_BODY:.*]]
+; INTERLEAVE:       [[VECTOR_BODY]]:
+; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
+; INTERLEAVE-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 1
+; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]]
+; INTERLEAVE-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP1]]
+; INTERLEAVE-NEXT:    store i16 0, ptr [[TMP2]], align 2
+; INTERLEAVE-NEXT:    store i16 0, ptr [[TMP3]], align 2
+; INTERLEAVE-NEXT:    [[TMP5:%.*]] = add i32 [[STEP_2]], [[TMP1]]
+; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; INTERLEAVE-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 8
+; INTERLEAVE-NEXT:    br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
+; INTERLEAVE:       [[MIDDLE_BLOCK]]:
+; INTERLEAVE-NEXT:    br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]]
+; INTERLEAVE:       [[SCALAR_PH]]:
+; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; INTERLEAVE-NEXT:    br label %[[LOOP:.*]]
+; INTERLEAVE:       [[LOOP]]:
+; INTERLEAVE-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; INTERLEAVE-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]]
+; INTERLEAVE-NEXT:    store i16 0, ptr [[GEP_DST]], align 2
+; INTERLEAVE-NEXT:    [[IV_NEXT]] = add i32 [[STEP_2]], [[IV]]
+; INTERLEAVE-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8
+; INTERLEAVE-NEXT:    br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}}
+; INTERLEAVE:       [[E_EXIT]]:
+; INTERLEAVE-NEXT:    [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP5]], %[[MIDDLE_BLOCK]] ]
+; INTERLEAVE-NEXT:    ret i32 [[RES]]
+;
+entry:
+  %step.1 = sext i8 0 to i32
+  %step.2 = add nsw i32 %step.1, 1
+  br label %loop
+
+loop:
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
+  %gep.dst = getelementptr inbounds i16, ptr %dst, i32 %iv
+  store i16 0, ptr %gep.dst, align 2
+  %iv.next = add i32 %step.2, %iv
+  %cmp.i = icmp slt i32 %iv.next, 8
+  br i1 %cmp.i, label %loop, label %e.exit
+
+e.exit:
+  %res = phi i32 [ %iv.next, %loop ]
+  ret i32 %res
+}
+
+; FIXME: At the moment an incorrect exit value is used for %iv.next.
+define i32 @test_iv_uniform_with_outside_use_scev_simplification_2(ptr %dst) {
+; VEC-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification_2(
+; VEC-SAME: ptr [[DST:%.*]]) {
+; VEC-NEXT:  [[ENTRY:.*]]:
+; VEC-NEXT:    [[STEP_1:%.*]] = sext i8 0 to i32
+; VEC-NEXT:    [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1
+; VEC-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; VEC:       [[VECTOR_PH]]:
+; VEC-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[STEP_2]], i64 0
+; VEC-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
+; VEC-NEXT:    br label %[[VECTOR_BODY:.*]]
+; VEC:       [[VECTOR_BODY]]:
+; VEC-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; VEC-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 2>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; VEC-NEXT:    [[OFFSET_IDX:%.*]] = mul i32 [[INDEX]], 2
+; VEC-NEXT:    [[TMP0:%.*]] = add i32 [[OFFSET_IDX]], 0
+; VEC-NEXT:    [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 2
+; VEC-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]]
+; VEC-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP1]]
+; VEC-NEXT:    store i16 0, ptr [[TMP2]], align 2
+; VEC-NEXT:    store i16 0, ptr [[TMP3]], align 2
+; VEC-NEXT:    [[TMP4:%.*]] = add <2 x i32> [[VEC_IND]], splat (i32 1)
+; VEC-NEXT:    [[TMP5:%.*]] = add <2 x i32> [[BROADCAST_SPLAT]], [[TMP4]]
+; VEC-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; VEC-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], splat (i32 4)
+; VEC-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4
+; VEC-NEXT:    br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
+; VEC:       [[MIDDLE_BLOCK]]:
+; VEC-NEXT:    [[TMP7:%.*]] = extractelement <2 x i32> [[TMP5]], i32 1
+; VEC-NEXT:    br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]]
+; VEC:       [[SCALAR_PH]]:
+; VEC-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; VEC-NEXT:    br label %[[LOOP:.*]]
+; VEC:       [[LOOP]]:
+; VEC-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; VEC-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]]
+; VEC-NEXT:    store i16 0, ptr [[GEP_DST]], align 2
+; VEC-NEXT:    [[INC:%.*]] = add i32 [[IV]], 1
+; VEC-NEXT:    [[IV_NEXT]] = add i32 [[STEP_2]], [[INC]]
+; VEC-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8
+; VEC-NEXT:    br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}}
+; VEC:       [[E_EXIT]]:
+; VEC-NEXT:    [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP7]], %[[MIDDLE_BLOCK]] ]
+; VEC-NEXT:    ret i32 [[RES]]
+;
+; INTERLEAVE-LABEL: define i32 @test_iv_uniform_with_outside_use_scev_simplification_2(
+; INTERLEAVE-SAME: ptr [[DST:%.*]]) {
+; INTERLEAVE-NEXT:  [[ENTRY:.*]]:
+; INTERLEAVE-NEXT:    [[STEP_1:%.*]] = sext i8 0 to i32
+; INTERLEAVE-NEXT:    [[STEP_2:%.*]] = add nsw i32 [[STEP_1]], 1
+; INTERLEAVE-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; INTERLEAVE:       [[VECTOR_PH]]:
+; INTERLEAVE-NEXT:    br label %[[VECTOR_BODY:.*]]
+; INTERLEAVE:       [[VECTOR_BODY]]:
+; INTERLEAVE-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; INTERLEAVE-NEXT:    [[OFFSET_IDX:%.*]] = mul i32 [[INDEX]], 2
+; INTERLEAVE-NEXT:    [[TMP0:%.*]] = add i32 [[OFFSET_IDX]], 0
+; INTERLEAVE-NEXT:    [[TMP1:%.*]] = add i32 [[OFFSET_IDX]], 2
+; INTERLEAVE-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP0]]
+; INTERLEAVE-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[TMP1]]
+; INTERLEAVE-NEXT:    store i16 0, ptr [[TMP2]], align 2
+; INTERLEAVE-NEXT:    store i16 0, ptr [[TMP3]], align 2
+; INTERLEAVE-NEXT:    [[TMP4:%.*]] = add i32 [[TMP1]], 1
+; INTERLEAVE-NEXT:    [[TMP5:%.*]] = add i32 [[STEP_2]], [[TMP4]]
+; INTERLEAVE-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
+; INTERLEAVE-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4
+; INTERLEAVE-NEXT:    br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], {{!llvm.loop ![0-9]+}}
+; INTERLEAVE:       [[MIDDLE_BLOCK]]:
+; INTERLEAVE-NEXT:    br i1 true, label %[[E_EXIT:.*]], label %[[SCALAR_PH]]
+; INTERLEAVE:       [[SCALAR_PH]]:
+; INTERLEAVE-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 8, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
+; INTERLEAVE-NEXT:    br label %[[LOOP:.*]]
+; INTERLEAVE:       [[LOOP]]:
+; INTERLEAVE-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
+; INTERLEAVE-NEXT:    [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i32 [[IV]]
+; INTERLEAVE-NEXT:    store i16 0, ptr [[GEP_DST]], align 2
+; INTERLEAVE-NEXT:    [[INC:%.*]] = add i32 [[IV]], 1
+; INTERLEAVE-NEXT:    [[IV_NEXT]] = add i32 [[STEP_2]], [[INC]]
+; INTERLEAVE-NEXT:    [[CMP_I:%.*]] = icmp slt i32 [[IV_NEXT]], 8
+; INTERLEAVE-NEXT:    br i1 [[CMP_I]], label %[[LOOP]], label %[[E_EXIT]], {{!llvm.loop ![0-9]+}}
+; INTERLEAVE:       [[E_EXIT]]:
+; INTERLEAVE-NEXT:    [[RES:%.*]] = phi i32 [ [[IV_NEXT]], %[[LOOP]] ], [ [[TMP5]], %[[MIDDLE_BLOCK]] ]
+; INTERLEAVE-NEXT:    ret i32 [[RES]]
+;
+entry:
+  %step.1 = sext i8 0 to i32
+  %step.2 = add nsw i32 %step.1, 1
+  br label %loop
+
+loop:
+  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
+  %gep.dst = getelementptr inbounds i16, ptr %dst, i32 %iv
+  store i16 0, ptr %gep.dst, align 2
+  %inc = add i32 %iv, 1
+  %iv.next = add i32 %step.2, %inc
+  %cmp.i = icmp slt i32 %iv.next, 8
+  br i1 %cmp.i, label %loop, label %e.exit
+
+e.exit:
+  %res = phi i32 [ %iv.next, %loop ]
+  ret i32 %res
+}