; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_35 -verify-machineinstrs | FileCheck %s
; RUN: %if ptxas %{ llc < %s -mtriple=nvptx64 -mcpu=sm_35 | %ptxas-verify %}

; Verify that we correctly emit code for extending ldg/ldu. We do not expose
; extending variants in the backend, but the ldg/ldu selection code may pick
; extending loads as candidates. We do want to support this, so make sure we
; emit the necessary cvt.* instructions to implement the extension and let ptxas
; emit the real extending loads.

target datalayout = "e-i64:64-v16:16-v32:32-n16:32:64"
target triple = "nvptx64-nvidia-cuda"

define ptx_kernel void @spam(ptr addrspace(1) noalias nocapture readonly %arg, ptr addrspace(1) noalias nocapture %arg1, i64 %arg2, i64 %arg3) #0 {
; CHECK-LABEL: spam(
; CHECK:       .maxntid 1, 1, 1
; CHECK-NEXT:  {
; CHECK-NEXT:    .reg .b32 %r<2>;
; CHECK-NEXT:    .reg .b64 %rd<9>;
; CHECK-EMPTY:
; CHECK-NEXT:  // %bb.0: // %bb
; CHECK-NEXT:    ld.param.b64 %rd1, [spam_param_0];
; CHECK-NEXT:    ld.param.b64 %rd2, [spam_param_3];
; CHECK-NEXT:    shl.b64 %rd3, %rd2, 1;
; CHECK-NEXT:    add.s64 %rd4, %rd1, %rd3;
; CHECK-NEXT:    ld.param.b64 %rd5, [spam_param_1];
; CHECK-NEXT:    ld.global.nc.s16 %r1, [%rd4+16];
; CHECK-NEXT:    mul.wide.s32 %rd6, %r1, %r1;
; CHECK-NEXT:    ld.global.b64 %rd7, [%rd5];
; CHECK-NEXT:    add.s64 %rd8, %rd6, %rd7;
; CHECK-NEXT:    st.global.b64 [%rd5], %rd8;
; CHECK-NEXT:    ret;
bb:
  %tmp5 = add nsw i64 %arg3, 8
  %tmp6 = getelementptr i16, ptr addrspace(1) %arg, i64 %tmp5
  %tmp7 = load i16, ptr addrspace(1) %tmp6, align 2
  %tmp8 = sext i16 %tmp7 to i64
  %tmp9 = mul nsw i64 %tmp8, %tmp8
  %tmp10 = load i64, ptr addrspace(1) %arg1, align 8
  %tmp11 = add nsw i64 %tmp9, %tmp10
  store i64 %tmp11, ptr addrspace(1) %arg1, align 8
  ret void
}

attributes #0 = { norecurse nounwind "polly.skip.fn" "nvvm.maxntid"="1,1,1" }