File: loop_exit_with_xor.ll

package info (click to toggle)
llvm-toolchain-9 1%3A9.0.1-16
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 882,436 kB
  • sloc: cpp: 4,167,636; ansic: 714,256; asm: 457,610; python: 155,927; objc: 65,094; sh: 42,856; lisp: 26,908; perl: 7,786; pascal: 7,722; makefile: 6,881; ml: 5,581; awk: 3,648; cs: 2,027; xml: 888; javascript: 381; ruby: 156
file content (93 lines) | stat: -rw-r--r-- 2,960 bytes parent folder | download | duplicates (2)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s

; Where the mask of lanes wanting to exit the loop on this iteration is not
; obviously already masked by exec (in this case, the xor with -1 inserted by
; control flow annotation), then lower control flow must insert an S_AND_B64
; with exec.

; GCN-LABEL: {{^}}needs_and:
; GCN: s_xor_b64 [[REG1:[^ ,]*]], {{[^ ,]*, -1$}}
; GCN: s_and_b64 [[REG2:[^ ,]*]], exec, [[REG1]]
; GCN: s_or_b64 [[REG3:[^ ,]*]], [[REG2]],
; GCN: s_andn2_b64 exec, exec, [[REG3]]

define void @needs_and(i32 %arg) {
entry:
  br label %loop

loop:
  %tmp23phi = phi i32 [ %tmp23, %endif ], [ 0, %entry ]
  %tmp23 = add nuw i32 %tmp23phi, 1
  %tmp27 = icmp ult i32 %arg, %tmp23
  br i1 %tmp27, label %then, label %endif

then:                                             ; preds = %bb
  call void @llvm.amdgcn.buffer.store.f32(float undef, <4 x i32> undef, i32 0, i32 undef, i1 false, i1 false) #1
  br label %endif

endif:                                             ; preds = %bb28, %bb
  br i1 %tmp27, label %loop, label %loopexit

loopexit:
  ret void
}

; Where the mask of lanes wanting to exit the loop on this iteration is
; obviously already masked by exec (a V_CMP), then lower control flow can omit
; the S_AND_B64 to avoid an unnecessary instruction.

; GCN-LABEL: {{^}}doesnt_need_and:
; GCN: v_cmp{{[^ ]*}} [[REG1:[^ ,]*]]
; GCN: s_or_b64 [[REG2:[^ ,]*]], [[REG1]],
; GCN: s_andn2_b64 exec, exec, [[REG2]]

define void @doesnt_need_and(i32 %arg) {
entry:
  br label %loop

loop:
  %tmp23phi = phi i32 [ %tmp23, %loop ], [ 0, %entry ]
  %tmp23 = add nuw i32 %tmp23phi, 1
  %tmp27 = icmp ult i32 %arg, %tmp23
  call void @llvm.amdgcn.buffer.store.f32(float undef, <4 x i32> undef, i32 0, i32 undef, i1 false, i1 false) #1
  br i1 %tmp27, label %loop, label %loopexit

loopexit:
  ret void
}

; Another case where the mask of lanes wanting to exit the loop is not masked
; by exec, because it is a function parameter.

; GCN-LABEL: {{^}}break_cond_is_arg:
; GCN: s_xor_b64 [[REG1:[^ ,]*]], {{[^ ,]*, -1$}}
; GCN: s_andn2_b64 exec, exec, [[REG3:[^ ,]*]]
; GCN: s_and_b64 [[REG2:[^ ,]*]], exec, [[REG1]]
; GCN: s_or_b64 [[REG3]], [[REG2]],

define void @break_cond_is_arg(i32 %arg, i1 %breakcond) {
entry:
  br label %loop

loop:
  %tmp23phi = phi i32 [ %tmp23, %endif ], [ 0, %entry ]
  %tmp23 = add nuw i32 %tmp23phi, 1
  %tmp27 = icmp ult i32 %arg, %tmp23
  br i1 %tmp27, label %then, label %endif

then:                                             ; preds = %bb
  call void @llvm.amdgcn.buffer.store.f32(float undef, <4 x i32> undef, i32 0, i32 undef, i1 false, i1 false) #1
  br label %endif

endif:                                             ; preds = %bb28, %bb
  br i1 %breakcond, label %loop, label %loopexit

loopexit:
  ret void
}


declare void @llvm.amdgcn.buffer.store.f32(float, <4 x i32>, i32, i32, i1, i1) #3

attributes #3 = { nounwind writeonly }