Skip to content

Commit 6749f2b

Browse files
fhahnfhahn
committed
[LV] Add pointer induction test variant with inbounds, remove TODO.
The function doesn't crash any more with inbounds, add a variant with inbounds.
1 parent 76b54df commit 6749f2b

File tree

1 file changed

+94
-5
lines changed

1 file changed

+94
-5
lines changed

‎llvm/test/Transforms/LoopVectorize/opaque-ptr.ll

Lines changed: 94 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,8 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
22
; RUN: opt -S -passes=loop-vectorize -force-vector-width=2 < %s | FileCheck %s
33

4-
; TODO: This still crashes with inbounds on the GEPs.
5-
define void @test(ptr %p1.start, ptr %p2.start, ptr %p1.end) {
6-
; CHECK-LABEL: @test(
4+
define void @test_ptr_iv_no_inbounds(ptr %p1.start, ptr %p2.start, ptr %p1.end) {
5+
; CHECK-LABEL: @test_ptr_iv_no_inbounds(
76
; CHECK-NEXT: entry:
87
; CHECK-NEXT: br label [[LOOP:%.*]]
98
; CHECK: loop:
@@ -39,6 +38,96 @@ exit:
3938
ret void
4039
}
4140

41+
define void @test_ptr_iv_with_inbounds(ptr %p1.start, ptr %p2.start, ptr %p1.end) {
42+
; CHECK-LABEL: @test_ptr_iv_with_inbounds(
43+
; CHECK-NEXT: entry:
44+
; CHECK-NEXT: [[P1_START5:%.*]] = ptrtoint ptr [[P1_START:%.*]] to i64
45+
; CHECK-NEXT: [[P1_END4:%.*]] = ptrtoint ptr [[P1_END:%.*]] to i64
46+
; CHECK-NEXT: [[P1_START2:%.*]] = ptrtoint ptr [[P1_START]] to i64
47+
; CHECK-NEXT: [[P1_END1:%.*]] = ptrtoint ptr [[P1_END]] to i64
48+
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[P1_END4]], -4
49+
; CHECK-NEXT: [[TMP1:%.*]] = sub i64 [[TMP0]], [[P1_START5]]
50+
; CHECK-NEXT: [[TMP2:%.*]] = lshr i64 [[TMP1]], 2
51+
; CHECK-NEXT: [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
52+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 2
53+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
54+
; CHECK: vector.memcheck:
55+
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[P1_END1]], -4
56+
; CHECK-NEXT: [[TMP5:%.*]] = sub i64 [[TMP4]], [[P1_START2]]
57+
; CHECK-NEXT: [[TMP6:%.*]] = lshr i64 [[TMP5]], 2
58+
; CHECK-NEXT: [[TMP7:%.*]] = shl nuw i64 [[TMP6]], 2
59+
; CHECK-NEXT: [[TMP8:%.*]] = add i64 [[TMP7]], 4
60+
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP8]]
61+
; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[P2_START:%.*]], i64 [[TMP8]]
62+
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[P1_START]], [[SCEVGEP3]]
63+
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[P2_START]], [[SCEVGEP]]
64+
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
65+
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
66+
; CHECK: vector.ph:
67+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 2
68+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
69+
; CHECK-NEXT: [[TMP9:%.*]] = mul i64 [[N_VEC]], 4
70+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP9]]
71+
; CHECK-NEXT: [[TMP10:%.*]] = mul i64 [[N_VEC]], 4
72+
; CHECK-NEXT: [[IND_END6:%.*]] = getelementptr i8, ptr [[P2_START]], i64 [[TMP10]]
73+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
74+
; CHECK: vector.body:
75+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
76+
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
77+
; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[OFFSET_IDX]], 0
78+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[P1_START]], i64 [[TMP11]]
79+
; CHECK-NEXT: [[OFFSET_IDX8:%.*]] = mul i64 [[INDEX]], 4
80+
; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[OFFSET_IDX8]], 0
81+
; CHECK-NEXT: [[NEXT_GEP9:%.*]] = getelementptr i8, ptr [[P2_START]], i64 [[TMP12]]
82+
; CHECK-NEXT: [[TMP13:%.*]] = getelementptr float, ptr [[NEXT_GEP]], i32 0
83+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP13]], align 4, !alias.scope [[META0:![0-9]+]], !noalias [[META3:![0-9]+]]
84+
; CHECK-NEXT: [[TMP14:%.*]] = getelementptr float, ptr [[NEXT_GEP9]], i32 0
85+
; CHECK-NEXT: [[WIDE_LOAD10:%.*]] = load <2 x float>, ptr [[TMP14]], align 4, !alias.scope [[META3]]
86+
; CHECK-NEXT: [[TMP15:%.*]] = fadd <2 x float> [[WIDE_LOAD]], [[WIDE_LOAD10]]
87+
; CHECK-NEXT: store <2 x float> [[TMP15]], ptr [[TMP13]], align 4, !alias.scope [[META0]], !noalias [[META3]]
88+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
89+
; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
90+
; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
91+
; CHECK: middle.block:
92+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
93+
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
94+
; CHECK: scalar.ph:
95+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P1_START]], [[ENTRY:%.*]] ], [ [[P1_START]], [[VECTOR_MEMCHECK]] ]
96+
; CHECK-NEXT: [[BC_RESUME_VAL7:%.*]] = phi ptr [ [[IND_END6]], [[MIDDLE_BLOCK]] ], [ [[P2_START]], [[ENTRY]] ], [ [[P2_START]], [[VECTOR_MEMCHECK]] ]
97+
; CHECK-NEXT: br label [[LOOP:%.*]]
98+
; CHECK: loop:
99+
; CHECK-NEXT: [[P1:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P1_NEXT:%.*]], [[LOOP]] ]
100+
; CHECK-NEXT: [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL7]], [[SCALAR_PH]] ], [ [[P2_NEXT:%.*]], [[LOOP]] ]
101+
; CHECK-NEXT: [[P1_VAL:%.*]] = load float, ptr [[P1]], align 4
102+
; CHECK-NEXT: [[P2_VAL:%.*]] = load float, ptr [[P2]], align 4
103+
; CHECK-NEXT: [[SUM:%.*]] = fadd float [[P1_VAL]], [[P2_VAL]]
104+
; CHECK-NEXT: store float [[SUM]], ptr [[P1]], align 4
105+
; CHECK-NEXT: [[P1_NEXT]] = getelementptr inbounds float, ptr [[P1]], i64 1
106+
; CHECK-NEXT: [[P2_NEXT]] = getelementptr inbounds float, ptr [[P2]], i64 1
107+
; CHECK-NEXT: [[C:%.*]] = icmp ne ptr [[P1_NEXT]], [[P1_END]]
108+
; CHECK-NEXT: br i1 [[C]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP8:![0-9]+]]
109+
; CHECK: exit:
110+
; CHECK-NEXT: ret void
111+
;
112+
entry:
113+
br label %loop
114+
115+
loop:
116+
%p1 = phi ptr [ %p1.start, %entry ], [ %p1.next, %loop ]
117+
%p2 = phi ptr [ %p2.start, %entry ], [ %p2.next, %loop ]
118+
%p1.val = load float, ptr %p1
119+
%p2.val = load float, ptr %p2
120+
%sum = fadd float %p1.val, %p2.val
121+
store float %sum, ptr %p1
122+
%p1.next = getelementptr inbounds float, ptr %p1, i64 1
123+
%p2.next = getelementptr inbounds float, ptr %p2, i64 1
124+
%c = icmp ne ptr %p1.next, %p1.end
125+
br i1 %c, label %loop, label %exit
126+
127+
exit:
128+
ret void
129+
}
130+
42131
define void @store_pointer_induction(ptr %start, ptr %end) {
43132
; CHECK-LABEL: @store_pointer_induction(
44133
; CHECK-NEXT: entry:
@@ -66,7 +155,7 @@ define void @store_pointer_induction(ptr %start, ptr %end) {
66155
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
67156
; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 16
68157
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
69-
; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
158+
; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]]
70159
; CHECK: middle.block:
71160
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
72161
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
@@ -78,7 +167,7 @@ define void @store_pointer_induction(ptr %start, ptr %end) {
78167
; CHECK-NEXT: store ptr [[IV]], ptr [[IV]], align 4
79168
; CHECK-NEXT: [[IV_NEXT]] = getelementptr inbounds ptr, ptr [[IV]], i32 1
80169
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq ptr [[IV_NEXT]], [[END]]
81-
; CHECK-NEXT: br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
170+
; CHECK-NEXT: br i1 [[EXITCOND]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP10:![0-9]+]]
82171
; CHECK: exit:
83172
; CHECK-NEXT: ret void
84173
;

0 commit comments

Comments
 (0)