File: call-to-kernel-undefined.ll

package info (click to toggle)
llvm-toolchain-13 1%3A13.0.1-6~deb11u1
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 1,418,812 kB
  • sloc: cpp: 5,290,827; ansic: 996,570; asm: 544,593; python: 188,212; objc: 72,027; lisp: 30,291; f90: 25,395; sh: 24,900; javascript: 9,780; pascal: 9,398; perl: 7,484; ml: 5,432; awk: 3,523; makefile: 2,892; xml: 953; cs: 573; fortran: 539
file content (20 lines) | stat: -rw-r--r-- 699 bytes parent folder | download | duplicates (9) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
 
; RUN: not --crash llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs -o /dev/null %s 2>&1 | FileCheck %s

; FIXME: It should be invalid IR to have a call to a kernel, but this
; is currently relied on, but should be eliminated before codegen.
define amdgpu_kernel void @callee_kernel(i32 addrspace(1)* %out) #0 {
entry:
  store volatile i32 0, i32 addrspace(1)* %out
  ret void
}

; Make sure there's no crash when the callsite calling convention
; doesn't match.
; CHECK: LLVM ERROR: invalid call to entry function
define amdgpu_kernel void @caller_kernel(i32 addrspace(1)* %out) #0 {
entry:
  call void @callee_kernel(i32 addrspace(1)* %out)
  ret void
}

attributes #0 = { nounwind noinline }