llvm/test/CodeGen/AMDGPU/srl-bitcast-bv.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=amdgpu9.50-- < %s | FileCheck -check-prefix=GFX950 %s
; RUN: llc -mtriple=amdgpu12.50-- < %s | FileCheck -check-prefix=GFX1250 %s

define i64 @srl_bv_v4i16_extract_last(i16 %a, i16 %b, i16 %c, i16 %d) {
; GFX950-LABEL: srl_bv_v4i16_extract_last:
; GFX950:       ; %bb.0:
; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX950-NEXT:    v_and_b32_e32 v0, 0xffff, v3
; GFX950-NEXT:    v_mov_b32_e32 v1, 0
; GFX950-NEXT:    s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: srl_bv_v4i16_extract_last:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    v_cvt_u32_u16_e32 v0, v3.l
; GFX1250-NEXT:    v_mov_b32_e32 v1, 0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %bv0 = insertelement <4 x i16> poison, i16 %a, i32 0
  %bv1 = insertelement <4 x i16> %bv0, i16 %b, i32 1
  %bv2 = insertelement <4 x i16> %bv1, i16 %c, i32 2
  %bv3 = insertelement <4 x i16> %bv2, i16 %d, i32 3
  %bc = bitcast <4 x i16> %bv3 to i64
  %srl = lshr i64 %bc, 48
  ret i64 %srl
}

define i64 @srl_bv_v2f32_extract_last(float %a, float %b) {
; GFX950-LABEL: srl_bv_v2f32_extract_last:
; GFX950:       ; %bb.0:
; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX950-NEXT:    v_mov_b32_e32 v0, v1
; GFX950-NEXT:    v_mov_b32_e32 v1, 0
; GFX950-NEXT:    s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: srl_bv_v2f32_extract_last:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    v_dual_mov_b32 v0, v1 :: v_dual_mov_b32 v1, 0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %bv0 = insertelement <2 x float> poison, float %a, i32 0
  %bv1 = insertelement <2 x float> %bv0, float %b, i32 1
  %bc = bitcast <2 x float> %bv1 to i64
  %srl = lshr i64 %bc, 32
  ret i64 %srl
}

; Negative test: shift amount is not (NumElts-1) * EltSize.
define i64 @srl_bv_not_last(i16 %a, i16 %b, i16 %c, i16 %d) {
; GFX950-LABEL: srl_bv_not_last:
; GFX950:       ; %bb.0:
; GFX950-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX950-NEXT:    s_mov_b32 s0, 0x5040100
; GFX950-NEXT:    v_perm_b32 v2, v3, v2, s0
; GFX950-NEXT:    v_perm_b32 v0, v1, v0, s0
; GFX950-NEXT:    v_alignbit_b32 v0, v2, v0, 16
; GFX950-NEXT:    v_and_b32_e32 v1, 0xffff, v3
; GFX950-NEXT:    s_setpc_b64 s[30:31]
;
; GFX1250-LABEL: srl_bv_not_last:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    v_mov_b16_e32 v2.h, v3.l
; GFX1250-NEXT:    v_mov_b16_e32 v0.h, v1.l
; GFX1250-NEXT:    v_cvt_u32_u16_e32 v1, v3.l
; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2)
; GFX1250-NEXT:    v_alignbit_b32 v0, v2, v0, 16
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %bv0 = insertelement <4 x i16> poison, i16 %a, i32 0
  %bv1 = insertelement <4 x i16> %bv0, i16 %b, i32 1
  %bv2 = insertelement <4 x i16> %bv1, i16 %c, i32 2
  %bv3 = insertelement <4 x i16> %bv2, i16 %d, i32 3
  %bc = bitcast <4 x i16> %bv3 to i64
  %srl = lshr i64 %bc, 16
  ret i64 %srl
}