; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -global-isel=0 -mtriple=amdgcn-amd-amdpal -mcpu=hawaii %s -o - | FileCheck %s

; For all these tests we disable optimizations through function attributes
; because the code we are exercising here needs phis and we want to keep the
; IR small.

; This code used to crash in SDISel because f16 was promoted to f32 through
; a `f32 = vector_extract_elt <1 x f16>, i32 0`, which is illegal.
; The invalid SDNode and thus, the crash was only exposed by the constant
; folding.
define void @phi_vec1half_to_f32_with_const_folding(ptr addrspace(1) %dst) #0 {
; CHECK-LABEL: phi_vec1half_to_f32_with_const_folding:
; CHECK:       ; %bb.0: ; %entry
; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT:    s_mov_b32 s4, 0
; CHECK-NEXT:  ; %bb.1: ; %bb
; CHECK-NEXT:    v_cvt_f16_f32_e64 v2, s4
; CHECK-NEXT:    s_mov_b32 s7, 0xf000
; CHECK-NEXT:    s_mov_b32 s6, 0
; CHECK-NEXT:    s_mov_b32 s4, s6
; CHECK-NEXT:    s_mov_b32 s5, s6
; CHECK-NEXT:    buffer_store_short v2, v[0:1], s[4:7], 0 addr64 offset:2
; CHECK-NEXT:    v_cvt_f16_f32_e64 v2, s4
; CHECK-NEXT:    buffer_store_short v2, v[0:1], s[4:7], 0 addr64
; CHECK-NEXT:    s_waitcnt vmcnt(0)
; CHECK-NEXT:    s_setpc_b64 s[30:31]
entry:
  br label %bb

bb:
  %phi = phi <1 x half> [ zeroinitializer, %entry ]
  %res = shufflevector <1 x half> poison, <1 x half> %phi, <2 x i32> <i32 0, i32 1>
  store <2 x half> %res, ptr addrspace(1) %dst
  ret void
}

; Same as phi_vec1half_to_f32_with_const_folding but without the folding.
; This test exercises the same invalid SDNode, but it happened to work by
; accident before. Here we make sure the fix also work as expected in the
; non-constant folding case.
define void @phi_vec1half_to_f32(ptr addrspace(1) %src, ptr addrspace(1) %dst) #0 {
; CHECK-LABEL: phi_vec1half_to_f32:
; CHECK:       ; %bb.0: ; %entry
; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT:    s_mov_b32 s7, 0xf000
; CHECK-NEXT:    s_mov_b32 s6, 0
; CHECK-NEXT:    s_mov_b32 s4, s6
; CHECK-NEXT:    s_mov_b32 s5, s6
; CHECK-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
; CHECK-NEXT:    s_waitcnt vmcnt(0)
; CHECK-NEXT:    v_cvt_f32_f16_e64 v0, v0
; CHECK-NEXT:  ; %bb.1: ; %bb
; CHECK-NEXT:    v_cvt_f16_f32_e64 v0, v0
; CHECK-NEXT:    s_mov_b32 s7, 0xf000
; CHECK-NEXT:    s_mov_b32 s6, 0
; CHECK-NEXT:    s_mov_b32 s4, s6
; CHECK-NEXT:    s_mov_b32 s5, s6
; CHECK-NEXT:    buffer_store_short v0, v[2:3], s[4:7], 0 addr64 offset:2
; CHECK-NEXT:    v_cvt_f16_f32_e64 v0, s4
; CHECK-NEXT:    buffer_store_short v0, v[2:3], s[4:7], 0 addr64
; CHECK-NEXT:    s_waitcnt vmcnt(0)
; CHECK-NEXT:    s_setpc_b64 s[30:31]
entry:
  %input = load <1 x half>, ptr addrspace(1) %src
  br label %bb

bb:
  %phi = phi <1 x half> [ %input, %entry ]
  %res = shufflevector <1 x half> poison, <1 x half> %phi, <2 x i32> <i32 0, i32 1>
  store <2 x half> %res, ptr addrspace(1) %dst
  ret void
}

; Same as phi_vec1bf16_to_f32 but with bfloat instead of half.
define void @phi_vec1bf16_to_f32(ptr addrspace(1) %src, ptr addrspace(1) %dst) #0 {
; CHECK-LABEL: phi_vec1bf16_to_f32:
; CHECK:       ; %bb.0: ; %entry
; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT:    s_mov_b32 s7, 0xf000
; CHECK-NEXT:    s_mov_b32 s6, 0
; CHECK-NEXT:    s_mov_b32 s4, s6
; CHECK-NEXT:    s_mov_b32 s5, s6
; CHECK-NEXT:    buffer_load_ushort v0, v[0:1], s[4:7], 0 addr64
; CHECK-NEXT:    s_mov_b32 s4, 16
; CHECK-NEXT:    s_waitcnt vmcnt(0)
; CHECK-NEXT:    v_lshlrev_b32_e64 v0, s4, v0
; CHECK-NEXT:  ; %bb.1: ; %bb
; CHECK-NEXT:    v_mul_f32_e64 v0, 1.0, v0
; CHECK-NEXT:    s_mov_b32 s4, 16
; CHECK-NEXT:    v_lshrrev_b32_e64 v0, s4, v0
; CHECK-NEXT:    s_mov_b32 s7, 0xf000
; CHECK-NEXT:    s_mov_b32 s6, 0
; CHECK-NEXT:    s_mov_b32 s4, s6
; CHECK-NEXT:    s_mov_b32 s5, s6
; CHECK-NEXT:    buffer_store_short v0, v[2:3], s[4:7], 0 addr64 offset:2
; CHECK-NEXT:    s_waitcnt vmcnt(0)
; CHECK-NEXT:    s_setpc_b64 s[30:31]
entry:
  %input = load <1 x bfloat>, ptr addrspace(1) %src
  br label %bb

bb:
  %phi = phi <1 x bfloat> [ %input, %entry ]
  %res = shufflevector <1 x bfloat> poison, <1 x bfloat> %phi, <2 x i32> <i32 0, i32 1>
  store <2 x bfloat> %res, ptr addrspace(1) %dst
  ret void
}

; Same as phi_vec1half_to_f32_with_const_folding but with bfloat instead of half.
define void @phi_vec1bf16_to_f32_with_const_folding(ptr addrspace(1) %dst) #0 {
; CHECK-LABEL: phi_vec1bf16_to_f32_with_const_folding:
; CHECK:       ; %bb.0: ; %entry
; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; CHECK-NEXT:    s_mov_b32 s4, 0
; CHECK-NEXT:  ; %bb.1: ; %bb
; CHECK-NEXT:    v_mul_f32_e64 v2, 1.0, s4
; CHECK-NEXT:    s_mov_b32 s4, 16
; CHECK-NEXT:    v_lshrrev_b32_e32 v2, s4, v2
; CHECK-NEXT:    s_mov_b32 s7, 0xf000
; CHECK-NEXT:    s_mov_b32 s6, 0
; CHECK-NEXT:    s_mov_b32 s4, s6
; CHECK-NEXT:    s_mov_b32 s5, s6
; CHECK-NEXT:    buffer_store_short v2, v[0:1], s[4:7], 0 addr64 offset:2
; CHECK-NEXT:    s_waitcnt vmcnt(0)
; CHECK-NEXT:    s_setpc_b64 s[30:31]
entry:
  br label %bb

bb:
  %phi = phi <1 x bfloat> [ zeroinitializer, %entry ]
  %res = shufflevector <1 x bfloat> poison, <1 x bfloat> %phi, <2 x i32> <i32 0, i32 1>
  store <2 x bfloat> %res, ptr addrspace(1) %dst
  ret void
}

attributes #0 = { noinline optnone }