; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=nvptx64-- -mcpu=sm_90 < %s | FileCheck %s

; fold (srl (bitcast (build_vector e0, ..., eN)), (N-1) * eltsize) -> (zext eN)
;
; On NVPTX, i8 is promoted to i16 during type legalization. This means
; BUILD_VECTOR v4i8 has i16 operands (wider than the i8 element type).
; The fold must truncate the promoted operand to the element size before
; zero-extending to the result type.

define i32 @srl_bv_v4i8_extract_last(i8 %a, i8 %b, i8 %c, i8 %d) {
; CHECK-LABEL: srl_bv_v4i8_extract_last(
; CHECK:       {
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b8 %r1, [srl_bv_v4i8_extract_last_param_3];
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %bv0 = insertelement <4 x i8> poison, i8 %a, i32 0
  %bv1 = insertelement <4 x i8> %bv0, i8 %b, i32 1
  %bv2 = insertelement <4 x i8> %bv1, i8 %c, i32 2
  %bv3 = insertelement <4 x i8> %bv2, i8 %d, i32 3
  %bc = bitcast <4 x i8> %bv3 to i32
  %srl = lshr i32 %bc, 24
  ret i32 %srl
}

; v2i16 -> i32: i16 operands match element type (no promotion needed).
define i32 @srl_bv_v2i16_extract_last(i16 %a, i16 %b) {
; CHECK-LABEL: srl_bv_v2i16_extract_last(
; CHECK:       {
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0:
; CHECK-NEXT:    ld.param.b16 %r1, [srl_bv_v2i16_extract_last_param_1];
; CHECK-NEXT:    st.param.b32 [func_retval0], %r1;
; CHECK-NEXT:    ret;
  %bv0 = insertelement <2 x i16> poison, i16 %a, i32 0
  %bv1 = insertelement <2 x i16> %bv0, i16 %b, i32 1
  %bc = bitcast <2 x i16> %bv1 to i32
  %srl = lshr i32 %bc, 16
  ret i32 %srl
}