xref: /llvm-project/llvm/test/CodeGen/RISCV/rvv/vp-combine-store-reverse.ll (revision 0c6e03eea04b93984d5fe562f64f1ce31f5cca09) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -mtriple=riscv64 -mattr=+f,+v -verify-machineinstrs < %s | FileCheck %s
3
4define void @test_store_reverse_combiner(<vscale x 2 x float> %val, <vscale x 2 x float>* %ptr, i32 zeroext %evl) {
5; CHECK-LABEL: test_store_reverse_combiner:
6; CHECK:       # %bb.0:
7; CHECK-NEXT:    slli a2, a1, 2
8; CHECK-NEXT:    add a0, a2, a0
9; CHECK-NEXT:    addi a0, a0, -4
10; CHECK-NEXT:    li a2, -4
11; CHECK-NEXT:    vsetvli zero, a1, e32, m1, ta, ma
12; CHECK-NEXT:    vsse32.v v8, (a0), a2
13; CHECK-NEXT:    ret
14  %rev = call <vscale x 2 x float> @llvm.experimental.vp.reverse.nxv2f32(<vscale x 2 x float> %val, <vscale x 2 x i1> splat (i1 true), i32 %evl)
15  call void @llvm.vp.store.nxv2f32.p0nxv2f32(<vscale x 2 x float> %rev, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> splat (i1 true), i32 %evl)
16  ret void
17}
18
19define void @test_store_mask_is_vp_reverse(<vscale x 2 x float> %val, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> %mask, i32 zeroext %evl) {
20; CHECK-LABEL: test_store_mask_is_vp_reverse:
21; CHECK:       # %bb.0:
22; CHECK-NEXT:    slli a2, a1, 2
23; CHECK-NEXT:    add a0, a2, a0
24; CHECK-NEXT:    addi a0, a0, -4
25; CHECK-NEXT:    li a2, -4
26; CHECK-NEXT:    vsetvli zero, a1, e32, m1, ta, ma
27; CHECK-NEXT:    vsse32.v v8, (a0), a2, v0.t
28; CHECK-NEXT:    ret
29  %storemask = call <vscale x 2 x i1> @llvm.experimental.vp.reverse.nxv2i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> splat (i1 true), i32 %evl)
30  %rev = call <vscale x 2 x float> @llvm.experimental.vp.reverse.nxv2f32(<vscale x 2 x float> %val, <vscale x 2 x i1> splat (i1 true), i32 %evl)
31  call void @llvm.vp.store.nxv2f32.p0nxv2f32(<vscale x 2 x float> %rev, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> %storemask, i32 %evl)
32  ret void
33}
34
35define void @test_store_mask_not_all_one(<vscale x 2 x float> %val, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> %notallones, i32 zeroext %evl) {
36; CHECK-LABEL: test_store_mask_not_all_one:
37; CHECK:       # %bb.0:
38; CHECK-NEXT:    vsetvli zero, a1, e32, m1, ta, ma
39; CHECK-NEXT:    vid.v v9, v0.t
40; CHECK-NEXT:    addi a1, a1, -1
41; CHECK-NEXT:    vrsub.vx v9, v9, a1, v0.t
42; CHECK-NEXT:    vrgather.vv v10, v8, v9, v0.t
43; CHECK-NEXT:    vse32.v v10, (a0), v0.t
44; CHECK-NEXT:    ret
45  %rev = call <vscale x 2 x float> @llvm.experimental.vp.reverse.nxv2f32(<vscale x 2 x float> %val, <vscale x 2 x i1> %notallones, i32 %evl)
46  call void @llvm.vp.store.nxv2f32.p0nxv2f32(<vscale x 2 x float> %rev, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> %notallones, i32 %evl)
47  ret void
48}
49
50define void @test_different_evl(<vscale x 2 x float> %val, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> %mask, i32 zeroext %evl1, i32 zeroext %evl2) {
51; CHECK-LABEL: test_different_evl:
52; CHECK:       # %bb.0:
53; CHECK-NEXT:    vsetvli zero, a1, e16, mf2, ta, ma
54; CHECK-NEXT:    vid.v v9
55; CHECK-NEXT:    addi a1, a1, -1
56; CHECK-NEXT:    vsetvli zero, zero, e8, mf4, ta, ma
57; CHECK-NEXT:    vmv.v.i v10, 0
58; CHECK-NEXT:    vmerge.vim v10, v10, 1, v0
59; CHECK-NEXT:    vsetvli zero, zero, e32, m1, ta, ma
60; CHECK-NEXT:    vid.v v11
61; CHECK-NEXT:    vsetvli zero, zero, e16, mf2, ta, ma
62; CHECK-NEXT:    vrsub.vx v9, v9, a1
63; CHECK-NEXT:    vsetvli zero, zero, e32, m1, ta, ma
64; CHECK-NEXT:    vrsub.vx v11, v11, a1
65; CHECK-NEXT:    vsetvli zero, zero, e8, mf4, ta, ma
66; CHECK-NEXT:    vrgatherei16.vv v12, v10, v9
67; CHECK-NEXT:    vmsne.vi v0, v12, 0
68; CHECK-NEXT:    vsetvli zero, zero, e32, m1, ta, ma
69; CHECK-NEXT:    vrgather.vv v9, v8, v11
70; CHECK-NEXT:    vsetvli zero, a2, e32, m1, ta, ma
71; CHECK-NEXT:    vse32.v v9, (a0), v0.t
72; CHECK-NEXT:    ret
73  %storemask = call <vscale x 2 x i1> @llvm.experimental.vp.reverse.nxv2i1(<vscale x 2 x i1> %mask, <vscale x 2 x i1> splat (i1 true), i32 %evl1)
74  %rev = call <vscale x 2 x float> @llvm.experimental.vp.reverse.nxv2f32(<vscale x 2 x float> %val, <vscale x 2 x i1> splat (i1 true), i32 %evl1)
75  call void @llvm.vp.store.nxv2f32.p0nxv2f32(<vscale x 2 x float> %rev, <vscale x 2 x float>* %ptr, <vscale x 2 x i1> %storemask, i32 %evl2)
76  ret void
77}
78
79declare <vscale x 2 x float> @llvm.experimental.vp.reverse.nxv2f32(<vscale x 2 x float>, <vscale x 2 x i1>, i32)
80declare <vscale x 2 x i1> @llvm.experimental.vp.reverse.nxv2i1(<vscale x 2 x i1>, <vscale x 2 x i1>, i32)
81declare void @llvm.vp.store.nxv2f32.p0nxv2f32(<vscale x 2 x float>, <vscale x 2 x float>* nocapture, <vscale x 2 x i1>, i32)
82