xref: /llvm-project/llvm/test/CodeGen/X86/fma-commute-loop.ll (revision e6bf48d11047e970cb24554a01b65b566d6b5d22) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-apple-darwin -mattr=avx512f | FileCheck %s
3
4define void @eggs(ptr %arg, ptr %arg1, ptr %arg2, ptr %arg3, ptr %arg4, ptr %arg5, i64 %arg6, i64 %arg7, i64 %arg8, i64 %arg9, i64 %arg10, i64 %arg11, i64 %arg12, ptr %arg13, ptr %arg14) nounwind {
5; CHECK-LABEL: eggs:
6; CHECK:       ## %bb.0: ## %bb
7; CHECK-NEXT:    pushq %r15
8; CHECK-NEXT:    pushq %r14
9; CHECK-NEXT:    pushq %r12
10; CHECK-NEXT:    pushq %rbx
11; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %r14
12; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rax
13; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %r15
14; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %r12
15; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %r11
16; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %r10
17; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %rbx
18; CHECK-NEXT:    leaq (%rbx,%r10,8), %r10
19; CHECK-NEXT:    leaq (%rbx,%r11,8), %r11
20; CHECK-NEXT:    vxorpd %xmm0, %xmm0, %xmm0
21; CHECK-NEXT:    xorl %ebx, %ebx
22; CHECK-NEXT:    vmovupd (%r14,%r15,8), %zmm1
23; CHECK-NEXT:    movq {{[0-9]+}}(%rsp), %r15
24; CHECK-NEXT:    addq %r12, %r15
25; CHECK-NEXT:    vmovupd (%r14,%r15,8), %zmm2
26; CHECK-NEXT:    addq {{[0-9]+}}(%rsp), %r12
27; CHECK-NEXT:    vmovupd (%r14,%r12,8), %zmm8
28; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
29; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
30; CHECK-NEXT:    vxorpd %xmm5, %xmm5, %xmm5
31; CHECK-NEXT:    vxorpd %xmm6, %xmm6, %xmm6
32; CHECK-NEXT:    vxorpd %xmm7, %xmm7, %xmm7
33; CHECK-NEXT:    .p2align 4
34; CHECK-NEXT:  LBB0_1: ## %bb15
35; CHECK-NEXT:    ## =>This Inner Loop Header: Depth=1
36; CHECK-NEXT:    vbroadcastsd (%r11,%rbx,8), %zmm9
37; CHECK-NEXT:    vfmadd231pd {{.*#+}} zmm0 = (zmm1 * zmm9) + zmm0
38; CHECK-NEXT:    vfmadd231pd {{.*#+}} zmm3 = (zmm2 * zmm9) + zmm3
39; CHECK-NEXT:    vfmadd231pd {{.*#+}} zmm4 = (zmm8 * zmm9) + zmm4
40; CHECK-NEXT:    vbroadcastsd (%r10,%rbx,8), %zmm9
41; CHECK-NEXT:    vfmadd231pd {{.*#+}} zmm5 = (zmm1 * zmm9) + zmm5
42; CHECK-NEXT:    vfmadd231pd {{.*#+}} zmm6 = (zmm2 * zmm9) + zmm6
43; CHECK-NEXT:    vfmadd231pd {{.*#+}} zmm7 = (zmm8 * zmm9) + zmm7
44; CHECK-NEXT:    incq %rbx
45; CHECK-NEXT:    cmpq %rbx, %rax
46; CHECK-NEXT:    jne LBB0_1
47; CHECK-NEXT:  ## %bb.2: ## %bb51
48; CHECK-NEXT:    vmovapd %zmm0, (%rdi)
49; CHECK-NEXT:    vmovapd %zmm3, (%rsi)
50; CHECK-NEXT:    vmovapd %zmm4, (%rdx)
51; CHECK-NEXT:    vmovapd %zmm5, (%rcx)
52; CHECK-NEXT:    vmovapd %zmm6, (%r8)
53; CHECK-NEXT:    vmovapd %zmm7, (%r9)
54; CHECK-NEXT:    popq %rbx
55; CHECK-NEXT:    popq %r12
56; CHECK-NEXT:    popq %r14
57; CHECK-NEXT:    popq %r15
58; CHECK-NEXT:    vzeroupper
59; CHECK-NEXT:    retq
60bb:
61  br label %bb15
62
63bb15:                                             ; preds = %bb15, %bb
64  %tmp = phi <8 x double> [ zeroinitializer, %bb ], [ %tmp38, %bb15 ]
65  %tmp16 = phi <8 x double> [ zeroinitializer, %bb ], [ %tmp39, %bb15 ]
66  %tmp17 = phi <8 x double> [ zeroinitializer, %bb ], [ %tmp40, %bb15 ]
67  %tmp18 = phi <8 x double> [ zeroinitializer, %bb ], [ %tmp46, %bb15 ]
68  %tmp19 = phi <8 x double> [ zeroinitializer, %bb ], [ %tmp47, %bb15 ]
69  %tmp20 = phi <8 x double> [ zeroinitializer, %bb ], [ %tmp48, %bb15 ]
70  %tmp21 = phi i64 [ 0, %bb ], [ %tmp49, %bb15 ]
71  %tmp22 = getelementptr inbounds double, ptr %arg14, i64 %arg11
72  %tmp24 = load <8 x double>, ptr %tmp22, align 8
73  %tmp25 = add i64 %arg10, %arg6
74  %tmp26 = getelementptr inbounds double, ptr %arg14, i64 %tmp25
75  %tmp28 = load <8 x double>, ptr %tmp26, align 8
76  %tmp29 = add i64 %arg10, %arg7
77  %tmp30 = getelementptr inbounds double, ptr %arg14, i64 %tmp29
78  %tmp32 = load <8 x double>, ptr %tmp30, align 8
79  %tmp33 = add i64 %tmp21, %arg8
80  %tmp34 = getelementptr inbounds double, ptr %arg13, i64 %tmp33
81  %tmp35 = load double, ptr %tmp34, align 8
82  %tmp36 = insertelement <8 x double> undef, double %tmp35, i32 0
83  %tmp37 = shufflevector <8 x double> %tmp36, <8 x double> undef, <8 x i32> zeroinitializer
84  %tmp38 = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> %tmp24, <8 x double> %tmp37, <8 x double> %tmp)
85  %tmp39 = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> %tmp28, <8 x double> %tmp37, <8 x double> %tmp16)
86  %tmp40 = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> %tmp32, <8 x double> %tmp37, <8 x double> %tmp17)
87  %tmp41 = add i64 %tmp21, %arg9
88  %tmp42 = getelementptr inbounds double, ptr %arg13, i64 %tmp41
89  %tmp43 = load double, ptr %tmp42, align 8
90  %tmp44 = insertelement <8 x double> undef, double %tmp43, i32 0
91  %tmp45 = shufflevector <8 x double> %tmp44, <8 x double> undef, <8 x i32> zeroinitializer
92  %tmp46 = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> %tmp24, <8 x double> %tmp45, <8 x double> %tmp18)
93  %tmp47 = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> %tmp28, <8 x double> %tmp45, <8 x double> %tmp19)
94  %tmp48 = call <8 x double> @llvm.fmuladd.v8f64(<8 x double> %tmp32, <8 x double> %tmp45, <8 x double> %tmp20)
95  %tmp49 = add nuw nsw i64 %tmp21, 1
96  %tmp50 = icmp eq i64 %tmp49, %arg12
97  br i1 %tmp50, label %bb51, label %bb15
98
99bb51:                                             ; preds = %bb15
100  store <8 x double> %tmp38, ptr %arg
101  store <8 x double> %tmp39, ptr %arg1
102  store <8 x double> %tmp40, ptr %arg2
103  store <8 x double> %tmp46, ptr %arg3
104  store <8 x double> %tmp47, ptr %arg4
105  store <8 x double> %tmp48, ptr %arg5
106  ret void
107}
108
109declare <8 x double> @llvm.fmuladd.v8f64(<8 x double>, <8 x double>, <8 x double>)
110