File: call-to-kernel-undefined.ll

package info (click to toggle)
llvm-toolchain-16 1%3A16.0.6-15~deb12u1
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,634,792 kB
  • sloc: cpp: 6,179,261; ansic: 1,216,205; asm: 741,319; python: 196,614; objc: 75,325; f90: 49,640; lisp: 32,396; pascal: 12,286; sh: 9,394; perl: 7,442; ml: 5,494; awk: 3,523; makefile: 2,723; javascript: 1,206; xml: 886; fortran: 581; cs: 573
file content (20 lines) | stat: -rw-r--r-- 695 bytes parent folder | download | duplicates (4) 
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(ptr addrspace(1) %out) #0 {
entry:
  store volatile i32 0, ptr 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(ptr addrspace(1) %out) #0 {
entry:
  call void @callee_kernel(ptr addrspace(1) %out)
  ret void
}

attributes #0 = { nounwind noinline }