File: call-to-kernel.ll

package info (click to toggle)
llvm-toolchain-11 1%3A11.0.1-2
  • links: PTS, VCS
  • area: main
  • in suites: bullseye
  • size: 995,808 kB
  • sloc: cpp: 4,767,656; ansic: 760,916; asm: 477,436; python: 170,940; objc: 69,804; lisp: 29,914; sh: 23,855; f90: 18,173; pascal: 7,551; perl: 7,471; ml: 5,603; awk: 3,489; makefile: 2,573; xml: 915; cs: 573; fortran: 503; javascript: 452
file content (18 lines) | stat: -rw-r--r-- 639 bytes parent folder | download | duplicates (9) 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
 
; 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
}

; CHECK: LLVM ERROR: Unsupported calling convention for call
define amdgpu_kernel void @caller_kernel(i32 addrspace(1)* %out) #0 {
entry:
  call amdgpu_kernel void @callee_kernel(i32 addrspace(1)* %out)
  ret void
}

attributes #0 = { nounwind noinline }