1; RUN: not --crash llc -march=amdgcn -mcpu=tahiti -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s 2; RUN: not --crash llc -march=amdgcn -mcpu=hawaii -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s 3; RUN: not --crash llc -march=amdgcn -mcpu=fiji -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s 4; RUN: not --crash llc -march=amdgcn -mcpu=gfx900 -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s 5; RUN: not --crash llc -march=amdgcn -mcpu=gfx1010 -o /dev/null %s 2>&1 | FileCheck -check-prefix=FAIL %s 6 7; Make sure selection of these intrinsics fails on targets that do not 8; have the instruction available. 9; FIXME: Should also really make sure the v2f16 version fails. 10 11; FAIL: LLVM ERROR: Cannot select: {{.+}}: f32,ch = BUFFER_ATOMIC_FADD 12define amdgpu_cs void @atomic_fadd(<4 x i32> inreg %arg0) { 13 %ret = call float @llvm.amdgcn.buffer.atomic.fadd.f32(float 1.0, <4 x i32> %arg0, i32 0, i32 112, i1 false) 14 ret void 15} 16 17declare float @llvm.amdgcn.buffer.atomic.fadd.f32(float, <4 x i32>, i32, i32, i1 immarg) #0 18 19attributes #0 = { nounwind } 20