xref: /llvm-project/llvm/test/CodeGen/RISCV/loop-strength-reduce-loop-invar.ll (revision 9122c5235ec85ce0c0ad337e862b006e7b349d84)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=riscv32 -verify-machineinstrs | FileCheck %s -check-prefixes=RV32
3; RUN: llc < %s -mtriple=riscv64 -verify-machineinstrs | FileCheck %s -check-prefixes=RV64
4
5; Test case:
6; - `A[row]` is loop invariant and should be hoisted up to preheader
7; FIXME: RV32 is working as expected, but RV64 doesn't
8
9; The following LLVM IR simulates:
10; int A[16][16];
11; void test(int row, int N) {
12; 	for (int i=0; i<N; ++I) {
13; 		A[row][i+1] = 4;
14; 		A[row][i+2] = 5;
15; 	}
16; }
17
18; After LSR:
19; int A[16][16];
20; void test(int row, int N) {
21; 	for (int *ptr = A[row][2]; N>0; N--) {
22; 		*(ptr-1) = 4;
23; 		*(ptr) = 5;
24; 		++ptr;
25; 	}
26; }
27
28@A = internal global [16 x [16 x i32]] zeroinitializer, align 32		; <ptr> [#uses=2]
29
30define void @test(i32 signext %row, i32 signext %N.in) nounwind {
31; RV32-LABEL: test:
32; RV32:       # %bb.0: # %entry
33; RV32-NEXT:    blez a1, .LBB0_3
34; RV32-NEXT:  # %bb.1: # %cond_true.preheader
35; RV32-NEXT:    slli a2, a0, 6
36; RV32-NEXT:    lui a3, %hi(A)
37; RV32-NEXT:    addi a3, a3, %lo(A)
38; RV32-NEXT:    li a0, 4
39; RV32-NEXT:    add a2, a2, a3
40; RV32-NEXT:    addi a2, a2, 8
41; RV32-NEXT:    li a3, 5
42; RV32-NEXT:  .LBB0_2: # %cond_true
43; RV32-NEXT:    # =>This Inner Loop Header: Depth=1
44; RV32-NEXT:    sw a0, -4(a2)
45; RV32-NEXT:    sw a3, 0(a2)
46; RV32-NEXT:    addi a1, a1, -1
47; RV32-NEXT:    addi a2, a2, 4
48; RV32-NEXT:    bnez a1, .LBB0_2
49; RV32-NEXT:  .LBB0_3: # %return
50; RV32-NEXT:    ret
51;
52; RV64-LABEL: test:
53; RV64:       # %bb.0: # %entry
54; RV64-NEXT:    blez a1, .LBB0_3
55; RV64-NEXT:  # %bb.1: # %cond_true.preheader
56; RV64-NEXT:    slli a3, a0, 6
57; RV64-NEXT:    lui a4, %hi(A)
58; RV64-NEXT:    addi a4, a4, %lo(A)
59; RV64-NEXT:    addiw a1, a1, 2
60; RV64-NEXT:    li a0, 2
61; RV64-NEXT:    li a2, 4
62; RV64-NEXT:    add a3, a3, a4
63; RV64-NEXT:    addi a4, a3, 4
64; RV64-NEXT:    li a5, 5
65; RV64-NEXT:  .LBB0_2: # %cond_true
66; RV64-NEXT:    # =>This Inner Loop Header: Depth=1
67; RV64-NEXT:    sw a2, 0(a4)
68; RV64-NEXT:    slli a6, a0, 2
69; RV64-NEXT:    addiw a0, a0, 1
70; RV64-NEXT:    add a6, a3, a6
71; RV64-NEXT:    sw a5, 0(a6)
72; RV64-NEXT:    addi a4, a4, 4
73; RV64-NEXT:    bne a0, a1, .LBB0_2
74; RV64-NEXT:  .LBB0_3: # %return
75; RV64-NEXT:    ret
76entry:
77	%N = bitcast i32 %N.in to i32
78	%tmp5 = icmp sgt i32 %N.in, 0
79	br i1 %tmp5, label %cond_true, label %return
80
81cond_true:
82	%indvar = phi i32 [ 0, %entry ], [ %indvar.next, %cond_true ]
83	%tmp2 = add i32 %indvar, 1
84	%tmp = getelementptr [16 x [16 x i32]], ptr @A, i32 0, i32 %row, i32 %tmp2
85	store i32 4, ptr %tmp
86	%tmp5.upgrd.1 = add i32 %indvar, 2
87	%tmp7 = getelementptr [16 x [16 x i32]], ptr @A, i32 0, i32 %row, i32 %tmp5.upgrd.1
88	store i32 5, ptr %tmp7
89	%indvar.next = add i32 %indvar, 1
90	%exitcond = icmp eq i32 %indvar.next, %N
91	br i1 %exitcond, label %return, label %cond_true
92
93return:
94	ret void
95}
96