xref: /llvm-project/llvm/test/CodeGen/RISCV/loop-strength-reduce-add-cheaper-than-mul.ll (revision 1456b68686808fa7c6ed7327aba65b639f81d5b8)
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=riscv32 -verify-machineinstrs -mattr=+m | FileCheck %s -check-prefixes=RV32
3; RUN: llc < %s -mtriple=riscv64 -verify-machineinstrs -mattr=+m | FileCheck %s -check-prefixes=RV64
4
5; Test case:
6; - Addition should be cheaper than multiplication
7
8; The following LLVM IR simulates:
9; int8_t flag2[8193];
10; void test(int i) {
11;   int tmp = i * 2;
12; 	if (i * 2 > 8192) return ;
13;   for (int j = 0; ; ++j) {
14; 		int offset = j * i + tmp;
15; 		flag2[offset] = 0;
16; 		if (offset + i > 8192) break;
17;   }
18; }
19
20; After LSR:
21; int8_t flag2[8193];
22; void test(int i) {
23; 	int j = i * 2;
24; 	if (j > 8193) return ;
25;   do {
26; 		flag2[j] = 0;
27; 		j += i;
28;   } while (j < 8193);
29; }
30
31@flags2 = internal global [8193 x i8] zeroinitializer, align 32		; <ptr> [#uses=1]
32
33define void @test(i32 signext %i) nounwind {
34; RV32-LABEL: test:
35; RV32:       # %bb.0: # %entry
36; RV32-NEXT:    slli a1, a0, 1
37; RV32-NEXT:    lui a3, 2
38; RV32-NEXT:    blt a3, a1, .LBB0_3
39; RV32-NEXT:  # %bb.1: # %bb.preheader
40; RV32-NEXT:    lui a2, %hi(flags2)
41; RV32-NEXT:    addi a2, a2, %lo(flags2)
42; RV32-NEXT:    addi a3, a3, 1
43; RV32-NEXT:  .LBB0_2: # %bb
44; RV32-NEXT:    # =>This Inner Loop Header: Depth=1
45; RV32-NEXT:    add a4, a2, a1
46; RV32-NEXT:    add a1, a1, a0
47; RV32-NEXT:    sb zero, 0(a4)
48; RV32-NEXT:    blt a1, a3, .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:    slliw a1, a0, 1
55; RV64-NEXT:    lui a3, 2
56; RV64-NEXT:    blt a3, a1, .LBB0_3
57; RV64-NEXT:  # %bb.1: # %bb.preheader
58; RV64-NEXT:    lui a2, %hi(flags2)
59; RV64-NEXT:    addi a2, a2, %lo(flags2)
60; RV64-NEXT:    addiw a3, a3, 1
61; RV64-NEXT:  .LBB0_2: # %bb
62; RV64-NEXT:    # =>This Inner Loop Header: Depth=1
63; RV64-NEXT:    slli a4, a1, 32
64; RV64-NEXT:    srli a4, a4, 32
65; RV64-NEXT:    add a4, a2, a4
66; RV64-NEXT:    addw a1, a1, a0
67; RV64-NEXT:    sb zero, 0(a4)
68; RV64-NEXT:    blt a1, a3, .LBB0_2
69; RV64-NEXT:  .LBB0_3: # %return
70; RV64-NEXT:    ret
71entry:
72	%k_addr.012 = shl i32 %i, 1
73	%tmp14 = icmp sgt i32 %k_addr.012, 8192
74	br i1 %tmp14, label %return, label %bb
75
76bb:
77	%indvar = phi i32 [ 0, %entry ], [ %indvar.next, %bb ]
78	%tmp.15 = mul i32 %indvar, %i
79	%tmp.16 = add i32 %tmp.15, %k_addr.012
80	%gep.upgrd.1 = zext i32 %tmp.16 to i64
81	%tmp = getelementptr [8193 x i8], ptr @flags2, i32 0, i64 %gep.upgrd.1
82	store i8 0, ptr %tmp
83	%tmp.17 = add i32 %tmp.16, %i
84	%tmp.upgrd.2 = icmp sgt i32 %tmp.17, 8192
85	%indvar.next = add i32 %indvar, 1
86	br i1 %tmp.upgrd.2, label %return, label %bb
87
88return:
89	ret void
90}
91