1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
2 ; RUN: llc < %s -mtriple thumbv8.1m.main-arm-none-eabihf -mattr=+mve | FileCheck %s
4 declare void @use_int32x4_t(<4 x i32>)
6 ; A 256-bit addition, with the two halves of the result passed to function
7 ; calls to spill the carry bit out of FPSCR.
8 define void @add_256(<4 x i32> %a_low, <4 x i32> %a_high, <4 x i32> %b_low, <4 x i32> %b_high) {
9 ; CHECK-LABEL: add_256:
10 ; CHECK: @ %bb.0: @ %entry
11 ; CHECK-NEXT: .save {r7, lr}
12 ; CHECK-NEXT: push {r7, lr}
13 ; CHECK-NEXT: .vsave {d8, d9, d10, d11}
14 ; CHECK-NEXT: vpush {d8, d9, d10, d11}
16 ; CHECK-NEXT: sub sp, #8
17 ; CHECK-NEXT: vadci.i32 q0, q0, q2
18 ; CHECK-NEXT: vmov q4, q3
19 ; CHECK-NEXT: vmov q5, q1
20 ; CHECK-NEXT: vstr fpscr_nzcvqc, [sp, #4] @ 4-byte Spill
21 ; CHECK-NEXT: bl use_int32x4_t
22 ; CHECK-NEXT: vldr fpscr_nzcvqc, [sp, #4] @ 4-byte Reload
23 ; CHECK-NEXT: vadc.i32 q0, q5, q4
24 ; CHECK-NEXT: add sp, #8
25 ; CHECK-NEXT: vpop {d8, d9, d10, d11}
26 ; CHECK-NEXT: pop.w {r7, lr}
27 ; CHECK-NEXT: b use_int32x4_t
29 %adc_low = tail call { <4 x i32>, i32 } @llvm.arm.mve.vadc.v4i32(<4 x i32> %a_low, <4 x i32> %b_low, i32 0)
30 %carry = extractvalue { <4 x i32>, i32 } %adc_low, 1
31 %result_low = extractvalue { <4 x i32>, i32 } %adc_low, 0
32 tail call void @use_int32x4_t(<4 x i32> %result_low)
33 %adc_high = tail call { <4 x i32>, i32 } @llvm.arm.mve.vadc.v4i32(<4 x i32> %a_high, <4 x i32> %b_high, i32 %carry)
34 %result_high = extractvalue { <4 x i32>, i32 } %adc_high, 0
35 tail call void @use_int32x4_t(<4 x i32> %result_high)
39 ; A 256-bit subtraction, with the two halves of the result passed to function
40 ; calls to spill the carry bit out of FPSCR.
41 define void @sub_256(<4 x i32> %a_low, <4 x i32> %a_high, <4 x i32> %b_low, <4 x i32> %b_high) {
42 ; CHECK-LABEL: sub_256:
43 ; CHECK: @ %bb.0: @ %entry
44 ; CHECK-NEXT: .save {r7, lr}
45 ; CHECK-NEXT: push {r7, lr}
46 ; CHECK-NEXT: .vsave {d8, d9, d10, d11}
47 ; CHECK-NEXT: vpush {d8, d9, d10, d11}
49 ; CHECK-NEXT: sub sp, #8
50 ; CHECK-NEXT: vsbci.i32 q0, q0, q2
51 ; CHECK-NEXT: vmov q4, q3
52 ; CHECK-NEXT: vmov q5, q1
53 ; CHECK-NEXT: vstr fpscr_nzcvqc, [sp, #4] @ 4-byte Spill
54 ; CHECK-NEXT: bl use_int32x4_t
55 ; CHECK-NEXT: vldr fpscr_nzcvqc, [sp, #4] @ 4-byte Reload
56 ; CHECK-NEXT: vsbc.i32 q0, q5, q4
57 ; CHECK-NEXT: add sp, #8
58 ; CHECK-NEXT: vpop {d8, d9, d10, d11}
59 ; CHECK-NEXT: pop.w {r7, lr}
60 ; CHECK-NEXT: b use_int32x4_t
62 %adc_low = tail call { <4 x i32>, i32 } @llvm.arm.mve.vsbc.v4i32(<4 x i32> %a_low, <4 x i32> %b_low, i32 536870912)
63 %carry = extractvalue { <4 x i32>, i32 } %adc_low, 1
64 %result_low = extractvalue { <4 x i32>, i32 } %adc_low, 0
65 tail call void @use_int32x4_t(<4 x i32> %result_low)
66 %adc_high = tail call { <4 x i32>, i32 } @llvm.arm.mve.vsbc.v4i32(<4 x i32> %a_high, <4 x i32> %b_high, i32 %carry)
67 %result_high = extractvalue { <4 x i32>, i32 } %adc_high, 0
68 tail call void @use_int32x4_t(<4 x i32> %result_high)
72 ; The carry-out of the first VADC intrinsic call is used by two other VADCs,
73 ; both of which will modify FPSCR, so it must be spilled and reloaded.
74 ; Missed optimisation: the first VLDR isn't needed, because the carry bit is
76 define <4 x i32> @multiple_uses_of_carry_bit(<4 x i32> %a_low, <4 x i32> %a_high, <4 x i32> %b_low, <4 x i32> %b_high, <4 x i32> %a_high_2, <4 x i32> %b_high_2) {
77 ; CHECK-LABEL: multiple_uses_of_carry_bit:
78 ; CHECK: @ %bb.0: @ %entry
80 ; CHECK-NEXT: sub sp, #8
81 ; CHECK-NEXT: vadci.i32 q0, q0, q2
82 ; CHECK-NEXT: add r0, sp, #24
83 ; CHECK-NEXT: vstr fpscr_nzcvqc, [sp, #4] @ 4-byte Spill
84 ; CHECK-NEXT: vldr fpscr_nzcvqc, [sp, #4] @ 4-byte Reload
85 ; CHECK-NEXT: vadc.i32 q1, q1, q3
86 ; CHECK-NEXT: veor q0, q0, q1
87 ; CHECK-NEXT: vldrw.u32 q1, [r0]
88 ; CHECK-NEXT: add r0, sp, #8
89 ; CHECK-NEXT: vldr fpscr_nzcvqc, [sp, #4] @ 4-byte Reload
90 ; CHECK-NEXT: vldrw.u32 q2, [r0]
91 ; CHECK-NEXT: vadc.i32 q1, q2, q1
92 ; CHECK-NEXT: veor q0, q0, q1
93 ; CHECK-NEXT: add sp, #8
96 %adc_low = tail call { <4 x i32>, i32 } @llvm.arm.mve.vadc.v4i32(<4 x i32> %a_low, <4 x i32> %b_low, i32 0)
97 %carry = extractvalue { <4 x i32>, i32 } %adc_low, 1
98 %result_low = extractvalue { <4 x i32>, i32 } %adc_low, 0
99 %adc_high = tail call { <4 x i32>, i32 } @llvm.arm.mve.vadc.v4i32(<4 x i32> %a_high, <4 x i32> %b_high, i32 %carry)
100 %result_high = extractvalue { <4 x i32>, i32 } %adc_high, 0
101 %checksum_1 = xor <4 x i32> %result_low, %result_high
102 %adc_high_2 = tail call { <4 x i32>, i32 } @llvm.arm.mve.vadc.v4i32(<4 x i32> %a_high_2, <4 x i32> %b_high_2, i32 %carry)
103 %result_high_2 = extractvalue { <4 x i32>, i32 } %adc_high_2, 0
104 %checksum_2 = xor <4 x i32> %checksum_1, %result_high_2
105 ret <4 x i32> %checksum_2