llvm/test/CodeGen/AMDGPU/fcanonicalize-elimination.bf16.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1250 %s -o - | FileCheck -check-prefix=GCN %s

define float @test_canonicalize_amdgcn_tanh_f32(float %a) {
; GCN-LABEL: test_canonicalize_amdgcn_tanh_f32:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_wait_loadcnt_dscnt 0x0
; GCN-NEXT:    s_wait_kmcnt 0x0
; GCN-NEXT:    v_tanh_f32_e32 v0, v0
; GCN-NEXT:    s_set_pc_i64 s[30:31]
  %tanh = call float @llvm.amdgcn.tanh.f32(float %a)
  %canonicalized = call float @llvm.canonicalize.f32(float %tanh)
  ret float %canonicalized
}

define bfloat @test_canonicalize_amdgcn_tanh_bf16(bfloat %a) {
; GCN-LABEL: test_canonicalize_amdgcn_tanh_bf16:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_wait_loadcnt_dscnt 0x0
; GCN-NEXT:    s_wait_kmcnt 0x0
; GCN-NEXT:    v_tanh_bf16_e32 v0, v0
; GCN-NEXT:    v_nop
; GCN-NEXT:    s_delay_alu instid0(TRANS32_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
; GCN-NEXT:    v_lshlrev_b32_e32 v0, 16, v0
; GCN-NEXT:    v_max_num_f32_e32 v0, v0, v0
; GCN-NEXT:    s_delay_alu instid0(VALU_DEP_1)
; GCN-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
; GCN-NEXT:    s_set_pc_i64 s[30:31]
  %tanh = call bfloat @llvm.amdgcn.tanh.bf16(bfloat %a)
  %canonicalized = call bfloat @llvm.canonicalize.bf16(bfloat %tanh)
  ret bfloat %canonicalized
}

define half @test_canonicalize_amdgcn_tanh_f16(half %a) {
; GCN-LABEL: test_canonicalize_amdgcn_tanh_f16:
; GCN:       ; %bb.0:
; GCN-NEXT:    s_wait_loadcnt_dscnt 0x0
; GCN-NEXT:    s_wait_kmcnt 0x0
; GCN-NEXT:    v_tanh_f16_e32 v0, v0
; GCN-NEXT:    s_set_pc_i64 s[30:31]
  %tanh = call half @llvm.amdgcn.tanh.f16(half %a)
  %canonicalized = call half @llvm.canonicalize.f16(half %tanh)
  ret half %canonicalized
}