Skip to content

Commit d92671c

Browse files
fhahnfhahn
committed
[PhaseOrdering] Add tests for optimizing std::find for AArch64.
1 parent 8de85e7 commit d92671c

File tree

1 file changed

+130
-0
lines changed

1 file changed

+130
-0
lines changed

‎llvm/test/Transforms/PhaseOrdering/AArch64/std-find.ll

Lines changed: 130 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,130 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2+
; RUN: opt -O3 -S %s | FileCheck %s
3+
4+
target triple = "arm64-apple-macosx15.0.0"
5+
6+
define i64 @std_find_i16_constant_offset_with_assumptions(ptr %first.coerce, i16 noundef signext %s) nofree nosync {
7+
; CHECK-LABEL: define i64 @std_find_i16_constant_offset_with_assumptions(
8+
; CHECK-SAME: ptr [[FIRST_COERCE:%.*]], i16 noundef signext [[S:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
9+
; CHECK-NEXT: [[ENTRY:.*]]:
10+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "align"(ptr [[FIRST_COERCE]], i64 2) ]
11+
; CHECK-NEXT: call void @llvm.assume(i1 true) [ "dereferenceable"(ptr [[FIRST_COERCE]], i64 256) ]
12+
; CHECK-NEXT: [[TMP0:%.*]] = ptrtoint ptr [[FIRST_COERCE]] to i64
13+
; CHECK-NEXT: [[COERCE_VAL_PI_I:%.*]] = add i64 [[TMP0]], 256
14+
; CHECK-NEXT: [[COERCE_VAL_IP:%.*]] = inttoptr i64 [[COERCE_VAL_PI_I]] to ptr
15+
; CHECK-NEXT: [[CMP_NOT6_I_I:%.*]] = icmp eq ptr [[FIRST_COERCE]], [[COERCE_VAL_IP]]
16+
; CHECK-NEXT: br i1 [[CMP_NOT6_I_I]], label %[[RETURN:.*]], label %[[LOOP_HEADER:.*]]
17+
; CHECK: [[LOOP_HEADER]]:
18+
; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ [[FIRST_COERCE]], %[[ENTRY]] ]
19+
; CHECK-NEXT: [[TMP1:%.*]] = load i16, ptr [[PTR_IV]], align 2
20+
; CHECK-NEXT: [[CMP2_I_I:%.*]] = icmp eq i16 [[TMP1]], [[S]]
21+
; CHECK-NEXT: br i1 [[CMP2_I_I]], label %[[RETURN_LOOPEXIT:.*]], label %[[LOOP_LATCH]]
22+
; CHECK: [[LOOP_LATCH]]:
23+
; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr inbounds nuw i8, ptr [[PTR_IV]], i64 2
24+
; CHECK-NEXT: [[CMP_NOT_I_I:%.*]] = icmp eq ptr [[PTR_IV_NEXT]], [[COERCE_VAL_IP]]
25+
; CHECK-NEXT: br i1 [[CMP_NOT_I_I]], label %[[RETURN_LOOPEXIT]], label %[[LOOP_HEADER]]
26+
; CHECK: [[RETURN_LOOPEXIT]]:
27+
; CHECK-NEXT: [[MERGE_PH:%.*]] = phi ptr [ [[COERCE_VAL_IP]], %[[LOOP_LATCH]] ], [ [[PTR_IV]], %[[LOOP_HEADER]] ]
28+
; CHECK-NEXT: [[DOTPRE:%.*]] = ptrtoint ptr [[MERGE_PH]] to i64
29+
; CHECK-NEXT: br label %[[RETURN]]
30+
; CHECK: [[RETURN]]:
31+
; CHECK-NEXT: [[RES_PRE_PHI:%.*]] = phi i64 [ [[DOTPRE]], %[[RETURN_LOOPEXIT]] ], [ [[TMP0]], %[[ENTRY]] ]
32+
; CHECK-NEXT: ret i64 [[RES_PRE_PHI]]
33+
;
34+
entry:
35+
%first = alloca { ptr }, align 8
36+
%s.addr = alloca i16, align 2
37+
store ptr %first.coerce, ptr %first, align 8
38+
store i16 %s, ptr %s.addr, align 2
39+
%0 = load ptr, ptr %first, align 8
40+
call void @llvm.assume(i1 true) [ "align"(ptr %0, i64 2) ]
41+
call void @llvm.assume(i1 true) [ "dereferenceable"(ptr %0, i64 256) ]
42+
%start.ptr = load ptr, ptr %first, align 8
43+
%1 = load i64, ptr %first, align 8
44+
%coerce.val.pi.i = add i64 %1, 256
45+
%coerce.val.ip = inttoptr i64 %coerce.val.pi.i to ptr
46+
%cmp.not6.i.i = icmp eq ptr %start.ptr, %coerce.val.ip
47+
br i1 %cmp.not6.i.i, label %return, label %loop.ph
48+
49+
loop.ph:
50+
%2 = load i16, ptr %s.addr, align 2
51+
br label %loop.header
52+
53+
loop.header:
54+
%ptr.iv = phi ptr [ %start.ptr, %loop.ph ], [ %ptr.iv.next, %loop.latch ]
55+
%3 = load i16, ptr %ptr.iv, align 2
56+
%cmp2.i.i = icmp eq i16 %3, %2
57+
br i1 %cmp2.i.i, label %return, label %loop.latch
58+
59+
loop.latch:
60+
%ptr.iv.next = getelementptr inbounds nuw i8, ptr %ptr.iv, i64 2
61+
%cmp.not.i.i = icmp eq ptr %ptr.iv.next, %coerce.val.ip
62+
br i1 %cmp.not.i.i, label %return, label %loop.header
63+
64+
return:
65+
%merge = phi ptr [ %start.ptr, %entry ], [ %coerce.val.ip, %loop.latch ], [ %ptr.iv, %loop.header ]
66+
%res = ptrtoint ptr %merge to i64
67+
ret i64 %res
68+
}
69+
70+
define i64 @std_find_i16_constant_offset_no_assumptions(ptr %first.coerce, i16 noundef signext %s) nofree nosync {
71+
; CHECK-LABEL: define i64 @std_find_i16_constant_offset_no_assumptions(
72+
; CHECK-SAME: ptr [[FIRST_COERCE:%.*]], i16 noundef signext [[S:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] {
73+
; CHECK-NEXT: [[ENTRY:.*]]:
74+
; CHECK-NEXT: [[TMP0:%.*]] = ptrtoint ptr [[FIRST_COERCE]] to i64
75+
; CHECK-NEXT: [[COERCE_VAL_PI_I:%.*]] = add i64 [[TMP0]], 256
76+
; CHECK-NEXT: [[COERCE_VAL_IP:%.*]] = inttoptr i64 [[COERCE_VAL_PI_I]] to ptr
77+
; CHECK-NEXT: [[CMP_NOT6_I_I:%.*]] = icmp eq ptr [[FIRST_COERCE]], [[COERCE_VAL_IP]]
78+
; CHECK-NEXT: br i1 [[CMP_NOT6_I_I]], label %[[RETURN:.*]], label %[[LOOP_HEADER:.*]]
79+
; CHECK: [[LOOP_HEADER]]:
80+
; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[PTR_IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ], [ [[FIRST_COERCE]], %[[ENTRY]] ]
81+
; CHECK-NEXT: [[TMP1:%.*]] = load i16, ptr [[PTR_IV]], align 2
82+
; CHECK-NEXT: [[CMP2_I_I:%.*]] = icmp eq i16 [[TMP1]], [[S]]
83+
; CHECK-NEXT: br i1 [[CMP2_I_I]], label %[[RETURN_LOOPEXIT:.*]], label %[[LOOP_LATCH]]
84+
; CHECK: [[LOOP_LATCH]]:
85+
; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr inbounds nuw i8, ptr [[PTR_IV]], i64 2
86+
; CHECK-NEXT: [[CMP_NOT_I_I:%.*]] = icmp eq ptr [[PTR_IV_NEXT]], [[COERCE_VAL_IP]]
87+
; CHECK-NEXT: br i1 [[CMP_NOT_I_I]], label %[[RETURN_LOOPEXIT]], label %[[LOOP_HEADER]]
88+
; CHECK: [[RETURN_LOOPEXIT]]:
89+
; CHECK-NEXT: [[MERGE_PH:%.*]] = phi ptr [ [[COERCE_VAL_IP]], %[[LOOP_LATCH]] ], [ [[PTR_IV]], %[[LOOP_HEADER]] ]
90+
; CHECK-NEXT: [[DOTPRE:%.*]] = ptrtoint ptr [[MERGE_PH]] to i64
91+
; CHECK-NEXT: br label %[[RETURN]]
92+
; CHECK: [[RETURN]]:
93+
; CHECK-NEXT: [[RES_PRE_PHI:%.*]] = phi i64 [ [[DOTPRE]], %[[RETURN_LOOPEXIT]] ], [ [[TMP0]], %[[ENTRY]] ]
94+
; CHECK-NEXT: ret i64 [[RES_PRE_PHI]]
95+
;
96+
entry:
97+
%first = alloca { ptr }, align 8
98+
%s.addr = alloca i16, align 2
99+
store ptr %first.coerce, ptr %first, align 8
100+
store i16 %s, ptr %s.addr, align 2
101+
%0 = load ptr, ptr %first, align 8
102+
%start.ptr = load ptr, ptr %first, align 8
103+
%1 = load i64, ptr %first, align 8
104+
%coerce.val.pi.i = add i64 %1, 256
105+
%coerce.val.ip = inttoptr i64 %coerce.val.pi.i to ptr
106+
%cmp.not6.i.i = icmp eq ptr %start.ptr, %coerce.val.ip
107+
br i1 %cmp.not6.i.i, label %return, label %loop.ph
108+
109+
loop.ph:
110+
%2 = load i16, ptr %s.addr, align 2
111+
br label %loop.header
112+
113+
loop.header:
114+
%ptr.iv = phi ptr [ %start.ptr, %loop.ph ], [ %ptr.iv.next, %loop.latch ]
115+
%3 = load i16, ptr %ptr.iv, align 2
116+
%cmp2.i.i = icmp eq i16 %3, %2
117+
br i1 %cmp2.i.i, label %return, label %loop.latch
118+
119+
loop.latch:
120+
%ptr.iv.next = getelementptr inbounds nuw i8, ptr %ptr.iv, i64 2
121+
%cmp.not.i.i = icmp eq ptr %ptr.iv.next, %coerce.val.ip
122+
br i1 %cmp.not.i.i, label %return, label %loop.header
123+
124+
return:
125+
%merge = phi ptr [ %start.ptr, %entry ], [ %coerce.val.ip, %loop.latch ], [ %ptr.iv, %loop.header ]
126+
%res = ptrtoint ptr %merge to i64
127+
ret i64 %res
128+
}
129+
130+
declare void @llvm.assume(i1 noundef)

0 commit comments

Comments
 (0)