1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -march=ve -mattr=+vpu | FileCheck %s
4 declare <256 x float> @llvm.vp.merge.v256f32(<256 x i1>, <256 x float>, <256 x float>, i32)
5 declare <256 x float> @llvm.vp.fsub.v256f32(<256 x float>, <256 x float>, <256 x i1>, i32)
7 define fastcc <256 x float> @test_vp_fsub_v256f32_vv_merge(<256 x float> %passthru, <256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n) {
8 ; CHECK-LABEL: test_vp_fsub_v256f32_vv_merge:
10 ; CHECK-NEXT: and %s0, %s0, (32)0
12 ; CHECK-NEXT: pvfsub.up %v0, %v1, %v2, %vm1
13 ; CHECK-NEXT: b.l.t (, %s10)
14 %vr = call <256 x float> @llvm.vp.fsub.v256f32(<256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n)
15 %r0 = call <256 x float> @llvm.vp.merge.v256f32(<256 x i1> %m, <256 x float> %vr, <256 x float> %passthru, i32 %n)
19 define fastcc <256 x float> @test_vp_fsub_v256f32_rv_merge(<256 x float> %passthru, float %s0, <256 x float> %i1, <256 x i1> %m, i32 %n) {
20 ; CHECK-LABEL: test_vp_fsub_v256f32_rv_merge:
22 ; CHECK-NEXT: and %s1, %s1, (32)0
24 ; CHECK-NEXT: pvfsub.up %v0, %s0, %v1, %vm1
25 ; CHECK-NEXT: b.l.t (, %s10)
26 %xins = insertelement <256 x float> undef, float %s0, i32 0
27 %i0 = shufflevector <256 x float> %xins, <256 x float> undef, <256 x i32> zeroinitializer
28 %vr = call <256 x float> @llvm.vp.fsub.v256f32(<256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n)
29 %r0 = call <256 x float> @llvm.vp.merge.v256f32(<256 x i1> %m, <256 x float> %vr, <256 x float> %passthru, i32 %n)
33 define fastcc <256 x float> @test_vp_fsub_v256f32_vr_merge(<256 x float> %passthru, <256 x float> %i0, float %s1, <256 x i1> %m, i32 %n) {
34 ; CHECK-LABEL: test_vp_fsub_v256f32_vr_merge:
36 ; CHECK-NEXT: and %s1, %s1, (32)0
37 ; CHECK-NEXT: lea %s2, 256
39 ; CHECK-NEXT: vbrd %v2, %s0
41 ; CHECK-NEXT: pvfsub.up %v0, %v1, %v2, %vm1
42 ; CHECK-NEXT: b.l.t (, %s10)
43 %yins = insertelement <256 x float> undef, float %s1, i32 0
44 %i1 = shufflevector <256 x float> %yins, <256 x float> undef, <256 x i32> zeroinitializer
45 %vr = call <256 x float> @llvm.vp.fsub.v256f32(<256 x float> %i0, <256 x float> %i1, <256 x i1> %m, i32 %n)
46 %r0 = call <256 x float> @llvm.vp.merge.v256f32(<256 x i1> %m, <256 x float> %vr, <256 x float> %passthru, i32 %n)
51 declare <256 x double> @llvm.vp.merge.v256f64(<256 x i1>, <256 x double>, <256 x double>, i32)
52 declare <256 x double> @llvm.vp.fsub.v256f64(<256 x double>, <256 x double>, <256 x i1>, i32)
54 define fastcc <256 x double> @test_vp_fsub_v256f64_vv_merge(<256 x double> %passthru, <256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n) {
55 ; CHECK-LABEL: test_vp_fsub_v256f64_vv_merge:
57 ; CHECK-NEXT: and %s0, %s0, (32)0
59 ; CHECK-NEXT: vfsub.d %v0, %v1, %v2, %vm1
60 ; CHECK-NEXT: b.l.t (, %s10)
61 %vr = call <256 x double> @llvm.vp.fsub.v256f64(<256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n)
62 %r0 = call <256 x double> @llvm.vp.merge.v256f64(<256 x i1> %m, <256 x double> %vr, <256 x double> %passthru, i32 %n)
63 ret <256 x double> %r0
66 define fastcc <256 x double> @test_vp_fsub_v256f64_rv_merge(<256 x double> %passthru, double %s0, <256 x double> %i1, <256 x i1> %m, i32 %n) {
67 ; CHECK-LABEL: test_vp_fsub_v256f64_rv_merge:
69 ; CHECK-NEXT: and %s1, %s1, (32)0
71 ; CHECK-NEXT: vfsub.d %v0, %s0, %v1, %vm1
72 ; CHECK-NEXT: b.l.t (, %s10)
73 %xins = insertelement <256 x double> undef, double %s0, i32 0
74 %i0 = shufflevector <256 x double> %xins, <256 x double> undef, <256 x i32> zeroinitializer
75 %vr = call <256 x double> @llvm.vp.fsub.v256f64(<256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n)
76 %r0 = call <256 x double> @llvm.vp.merge.v256f64(<256 x i1> %m, <256 x double> %vr, <256 x double> %passthru, i32 %n)
77 ret <256 x double> %r0
80 define fastcc <256 x double> @test_vp_fsub_v256f64_vr_merge(<256 x double> %passthru, <256 x double> %i0, double %s1, <256 x i1> %m, i32 %n) {
81 ; CHECK-LABEL: test_vp_fsub_v256f64_vr_merge:
83 ; CHECK-NEXT: and %s1, %s1, (32)0
84 ; CHECK-NEXT: lea %s2, 256
86 ; CHECK-NEXT: vbrd %v2, %s0
88 ; CHECK-NEXT: vfsub.d %v0, %v1, %v2, %vm1
89 ; CHECK-NEXT: b.l.t (, %s10)
90 %yins = insertelement <256 x double> undef, double %s1, i32 0
91 %i1 = shufflevector <256 x double> %yins, <256 x double> undef, <256 x i32> zeroinitializer
92 %vr = call <256 x double> @llvm.vp.fsub.v256f64(<256 x double> %i0, <256 x double> %i1, <256 x i1> %m, i32 %n)
93 %r0 = call <256 x double> @llvm.vp.merge.v256f64(<256 x i1> %m, <256 x double> %vr, <256 x double> %passthru, i32 %n)
94 ret <256 x double> %r0