File: TailDuplication-convergent.ll

package info (click to toggle)
llvm-toolchain-15 1%3A15.0.6-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,554,644 kB
  • sloc: cpp: 5,922,452; ansic: 1,012,136; asm: 674,362; python: 191,568; objc: 73,855; f90: 42,327; lisp: 31,913; pascal: 11,973; javascript: 10,144; sh: 9,421; perl: 7,447; ml: 5,527; awk: 3,523; makefile: 2,520; xml: 885; cs: 573; fortran: 567
file content (46 lines) | stat: -rw-r--r-- 1,097 bytes parent folder | download
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
; RUN: llc -O2 -tail-dup-size=100 -enable-tail-merge=0 < %s | FileCheck %s
; RUN: %if ptxas %{ llc -O2 -tail-dup-size=100 -enable-tail-merge=0 < %s | %ptxas-verify %}
target triple = "nvptx64-nvidia-cuda"

declare void @foo()
declare void @llvm.nvvm.barrier0()

; syncthreads shouldn't be duplicated.
; CHECK: .func call_syncthreads
; CHECK: bar.sync
; CHECK-NOT: bar.sync
define void @call_syncthreads(i32* %a, i32* %b, i1 %cond, i1 %cond2) nounwind {
  br i1 %cond, label %L1, label %L2
  br i1 %cond2, label %Ret, label %L1
Ret:
  ret void
L1:
  store i32 0, i32* %a
  br label %L42
L2:
  store i32 1, i32* %a
  br label %L42
L42:
  call void @llvm.nvvm.barrier0()
  br label %Ret
}

; Check that call_syncthreads really does trigger tail duplication.
; CHECK: .func call_foo
; CHECK: call
; CHECK: call
define void @call_foo(i32* %a, i32* %b, i1 %cond, i1 %cond2) nounwind {
  br i1 %cond, label %L1, label %L2
  br i1 %cond2, label %Ret, label %L1
Ret:
  ret void
L1:
  store i32 0, i32* %a
  br label %L42
L2:
  store i32 1, i32* %a
  br label %L42
L42:
  call void @foo()
  br label %Ret
}