1 ; RUN: opt -S -instcombine < %s | FileCheck %s
3 target triple = "aarch64-unknown-linux-gnu"
5 ; Idempotent fmuls -- should compile to just a ret.
6 define <vscale x 8 x half> @idempotent_fmul_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) #0 {
7 ; CHECK-LABEL: @idempotent_fmul_f16(
8 ; CHECK-NEXT: ret <vscale x 8 x half> [[A:%.*]]
10 %1 = call <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half 1.0)
11 %2 = call <vscale x 8 x half> @llvm.aarch64.sve.fmul.nxv8f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %1)
12 ret <vscale x 8 x half> %2
15 define <vscale x 4 x float> @idempotent_fmul_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) #0 {
16 ; CHECK-LABEL: @idempotent_fmul_f32(
17 ; CHECK-NEXT: ret <vscale x 4 x float> [[A:%.*]]
19 %1 = call <vscale x 4 x float> @llvm.aarch64.sve.dup.x.nxv4f32(float 1.0)
20 %2 = call <vscale x 4 x float> @llvm.aarch64.sve.fmul.nxv4f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %1)
21 ret <vscale x 4 x float> %2
24 define <vscale x 2 x double> @idempotent_fmul_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) #0 {
25 ; CHECK-LABEL: @idempotent_fmul_f64(
26 ; CHECK-NEXT: ret <vscale x 2 x double> [[A:%.*]]
28 %1 = call <vscale x 2 x double> @llvm.aarch64.sve.dup.x.nxv2f64(double 1.0)
29 %2 = call <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %1)
30 ret <vscale x 2 x double> %2
33 define <vscale x 2 x double> @idempotent_fmul_different_argument_order(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) #0 {
34 ; CHECK-LABEL: @idempotent_fmul_different_argument_order(
35 ; CHECK-NEXT: ret <vscale x 2 x double> [[A:%.*]]
37 %1 = call <vscale x 2 x double> @llvm.aarch64.sve.dup.x.nxv2f64(double 1.0)
38 ; Different argument order to the above tests.
39 %2 = call <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %1, <vscale x 2 x double> %a)
40 ret <vscale x 2 x double> %2
43 define <vscale x 8 x half> @idempotent_fmul_with_predicated_dup(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) #0 {
44 ; CHECK-LABEL: @idempotent_fmul_with_predicated_dup(
45 ; CHECK-NEXT: ret <vscale x 8 x half> [[A:%.*]]
47 %1 = call <vscale x 8 x half> @llvm.aarch64.sve.dup.nxv8f16(<vscale x 8 x half> undef, <vscale x 8 x i1> %pg, half 1.0)
48 %2 = call <vscale x 8 x half> @llvm.aarch64.sve.fmul.nxv8f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %1)
49 ret <vscale x 8 x half> %2
52 define <vscale x 8 x half> @idempotent_fmul_two_dups(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) #0 {
53 ; Edge case -- make sure that the case where we're fmultiplying two dups
55 ; CHECK-LABEL: @idempotent_fmul_two_dups(
56 ; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half 0xH3C00)
57 ; CHECK-NEXT: ret <vscale x 8 x half> [[TMP1]]
59 %1 = call <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half 1.0)
60 %2 = call <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half 1.0)
61 %3 = call <vscale x 8 x half> @llvm.aarch64.sve.fmul.nxv8f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %1, <vscale x 8 x half> %2)
62 ret <vscale x 8 x half> %3
65 ; Non-idempotent fmuls -- we don't expect these to be optimised out.
66 define <vscale x 8 x half> @non_idempotent_fmul_f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a) #0 {
67 ; CHECK-LABEL: @non_idempotent_fmul_f16(
68 ; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half 0xH4000)
69 ; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.fmul.nxv8f16(<vscale x 8 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[TMP1]])
70 ; CHECK-NEXT: ret <vscale x 8 x half> [[TMP2]]
72 %1 = call <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half 2.0)
73 %2 = call <vscale x 8 x half> @llvm.aarch64.sve.fmul.nxv8f16(<vscale x 8 x i1> %pg, <vscale x 8 x half> %a, <vscale x 8 x half> %1)
74 ret <vscale x 8 x half> %2
77 define <vscale x 4 x float> @non_idempotent_fmul_f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a) #0 {
78 ; CHECK-LABEL: @non_idempotent_fmul_f32(
79 ; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.dup.x.nxv4f32(float 2.000000e+00)
80 ; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.fmul.nxv4f32(<vscale x 4 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[TMP1]])
81 ; CHECK-NEXT: ret <vscale x 4 x float> [[TMP2]]
83 %1 = call <vscale x 4 x float> @llvm.aarch64.sve.dup.x.nxv4f32(float 2.0)
84 %2 = call <vscale x 4 x float> @llvm.aarch64.sve.fmul.nxv4f32(<vscale x 4 x i1> %pg, <vscale x 4 x float> %a, <vscale x 4 x float> %1)
85 ret <vscale x 4 x float> %2
88 define <vscale x 2 x double> @non_idempotent_fmul_f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a) #0 {
89 ; CHECK-LABEL: @non_idempotent_fmul_f64(
90 ; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.dup.x.nxv2f64(double 2.000000e+00)
91 ; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1> [[PG:%.*]], <vscale x 2 x double> [[A:%.*]], <vscale x 2 x double> [[TMP1]])
92 ; CHECK-NEXT: ret <vscale x 2 x double> [[TMP2]]
94 %1 = call <vscale x 2 x double> @llvm.aarch64.sve.dup.x.nxv2f64(double 2.0)
95 %2 = call <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1> %pg, <vscale x 2 x double> %a, <vscale x 2 x double> %1)
96 ret <vscale x 2 x double> %2
99 define <vscale x 2 x double> @non_idempotent_fmul_with_predicated_dup(<vscale x 2 x i1> %pg1, <vscale x 2 x i1> %pg2, <vscale x 2 x double> %a) #0 {
100 ; Different predicates
101 ; CHECK-LABEL: @non_idempotent_fmul_with_predicated_dup(
102 ; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.dup.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> [[PG1:%.*]], double 1.000000e+00)
103 ; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1> [[PG2:%.*]], <vscale x 2 x double> [[A:%.*]], <vscale x 2 x double> [[TMP1]])
104 ; CHECK-NEXT: ret <vscale x 2 x double> [[TMP2]]
106 %1 = call <vscale x 2 x double> @llvm.aarch64.sve.dup.nxv2f64(<vscale x 2 x double> undef, <vscale x 2 x i1> %pg1, double 1.0)
107 %2 = call <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1> %pg2, <vscale x 2 x double> %a, <vscale x 2 x double> %1)
108 ret <vscale x 2 x double> %2
111 declare <vscale x 8 x half> @llvm.aarch64.sve.dup.x.nxv8f16(half)
112 declare <vscale x 4 x float> @llvm.aarch64.sve.dup.x.nxv4f32(float)
113 declare <vscale x 2 x double> @llvm.aarch64.sve.dup.x.nxv2f64(double)
115 declare <vscale x 2 x double> @llvm.aarch64.sve.dup.nxv2f64(<vscale x 2 x double>, <vscale x 2 x i1>, double)
116 declare <vscale x 8 x half> @llvm.aarch64.sve.dup.nxv8f16(<vscale x 8 x half>, <vscale x 8 x i1>, half)
118 declare <vscale x 8 x half> @llvm.aarch64.sve.fmul.nxv8f16(<vscale x 8 x i1>, <vscale x 8 x half>, <vscale x 8 x half>)
119 declare <vscale x 4 x float> @llvm.aarch64.sve.fmul.nxv4f32(<vscale x 4 x i1>, <vscale x 4 x float>, <vscale x 4 x float>)
120 declare <vscale x 2 x double> @llvm.aarch64.sve.fmul.nxv2f64(<vscale x 2 x i1>, <vscale x 2 x double>, <vscale x 2 x double>)
122 attributes #0 = { "target-features"="+sve" }