test/CodeGen/AMDGPU/swdev380865.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -stress-regalloc=4 -o - %s | FileCheck %s

; Make sure we can rematerialize split 64-bit constants (which
; MachineLICM hoisted out of the loop) and avoid spilling inside the
; loop.
;
; MachineLICM originally believed the constant materializes to be
; rematerializable, but the lowered REG_SEQUENCE uses they coalesece
; into were not. The InlineSpiller also did not recognize redundant
; spills inside the loop, so we would repeatedly reload the same
; values.

define amdgpu_kernel void @_Z6kernelILi4000ELi1EEvPd(ptr addrspace(1) %x.coerce) {
; CHECK-LABEL: _Z6kernelILi4000ELi1EEvPd:
; CHECK:       ; %bb.0: ; %entry
; CHECK-NEXT:    s_mov_b64 s[0:1], 0
; CHECK-NEXT:    s_load_dword s2, s[0:1], 0x0
; CHECK-NEXT:    ; kill: killed $sgpr0_sgpr1
; CHECK-NEXT:    s_mov_b32 s7, 0x401c0000
; CHECK-NEXT:    s_mov_b32 s5, 0x40280000
; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
; CHECK-NEXT:    v_writelane_b32 v0, s2, 0
; CHECK-NEXT:    s_load_dwordx2 s[2:3], s[0:1], 0x0
; CHECK-NEXT:    s_mov_b32 s0, 0
; CHECK-NEXT:    s_mov_b32 s1, 0x40140000
; CHECK-NEXT:    s_mov_b32 s1, 0x40180000
; CHECK-NEXT:    v_writelane_b32 v0, s0, 1
; CHECK-NEXT:    v_writelane_b32 v0, s1, 2
; CHECK-NEXT:    s_mov_b32 s1, 0x40220000
; CHECK-NEXT:    v_writelane_b32 v0, s0, 3
; CHECK-NEXT:    v_writelane_b32 v0, s1, 4
; CHECK-NEXT:    s_mov_b32 s1, 0x40240000
; CHECK-NEXT:    v_writelane_b32 v0, s0, 5
; CHECK-NEXT:    v_writelane_b32 v0, s1, 6
; CHECK-NEXT:    s_mov_b32 s1, 0x40260000
; CHECK-NEXT:    v_writelane_b32 v0, s0, 7
; CHECK-NEXT:    s_waitcnt lgkmcnt(0)
; CHECK-NEXT:    v_mov_b32_e32 v1, s2
; CHECK-NEXT:    v_writelane_b32 v0, s1, 8
; CHECK-NEXT:    v_mov_b32_e32 v2, s3
; CHECK-NEXT:  .LBB0_1: ; %for.cond4.preheader
; CHECK-NEXT:    ; =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], 0
; CHECK-NEXT:    s_mov_b32 s2, 0
; CHECK-NEXT:    s_mov_b32 s3, 0x40140000
; CHECK-NEXT:    v_writelane_b32 v0, s0, 9
; CHECK-NEXT:    v_writelane_b32 v0, s6, 10
; CHECK-NEXT:    v_writelane_b32 v0, s7, 11
; CHECK-NEXT:    v_readlane_b32 s6, v0, 1
; CHECK-NEXT:    v_readlane_b32 s7, v0, 2
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[2:3]
; CHECK-NEXT:    s_mov_b32 s1, s7
; CHECK-NEXT:    s_mov_b32 s0, s2
; CHECK-NEXT:    v_writelane_b32 v0, s6, 1
; CHECK-NEXT:    v_writelane_b32 v0, s7, 2
; CHECK-NEXT:    v_readlane_b32 s6, v0, 10
; CHECK-NEXT:    v_readlane_b32 s7, v0, 11
; CHECK-NEXT:    s_mov_b32 s6, s2
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[0:1]
; CHECK-NEXT:    v_readlane_b32 s0, v0, 3
; CHECK-NEXT:    v_readlane_b32 s1, v0, 4
; CHECK-NEXT:    s_mov_b32 s3, s1
; CHECK-NEXT:    s_mov_b32 s0, 0
; CHECK-NEXT:    s_mov_b32 s1, 0x40140000
; CHECK-NEXT:    s_mov_b32 s2, s0
; CHECK-NEXT:    s_mov_b32 s1, s3
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[6:7]
; CHECK-NEXT:    v_writelane_b32 v0, s0, 3
; CHECK-NEXT:    v_writelane_b32 v0, s1, 4
; CHECK-NEXT:    v_readlane_b32 s0, v0, 5
; CHECK-NEXT:    v_readlane_b32 s1, v0, 6
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[2:3]
; CHECK-NEXT:    s_mov_b32 s3, s1
; CHECK-NEXT:    s_mov_b32 s0, 0
; CHECK-NEXT:    s_mov_b32 s1, 0x40140000
; CHECK-NEXT:    s_mov_b32 s2, s0
; CHECK-NEXT:    s_mov_b32 s1, s3
; CHECK-NEXT:    v_writelane_b32 v0, s0, 5
; CHECK-NEXT:    v_writelane_b32 v0, s1, 6
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[2:3]
; CHECK-NEXT:    v_readlane_b32 s0, v0, 7
; CHECK-NEXT:    v_readlane_b32 s1, v0, 8
; CHECK-NEXT:    s_mov_b32 s3, s1
; CHECK-NEXT:    s_mov_b32 s0, 0
; CHECK-NEXT:    s_mov_b32 s1, 0x40140000
; CHECK-NEXT:    s_mov_b32 s2, s0
; CHECK-NEXT:    s_mov_b32 s1, s3
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[2:3]
; CHECK-NEXT:    v_writelane_b32 v0, s0, 7
; CHECK-NEXT:    s_mov_b32 s4, s0
; CHECK-NEXT:    v_writelane_b32 v0, s1, 8
; CHECK-NEXT:    v_readlane_b32 s0, v0, 0
; CHECK-NEXT:    v_readlane_b32 s2, v0, 9
; CHECK-NEXT:    s_add_i32 s2, s2, s0
; CHECK-NEXT:    v_writelane_b32 v0, s2, 9
; CHECK-NEXT:    v_add_f64 v[1:2], v[1:2], s[4:5]
; CHECK-NEXT:    v_readlane_b32 s0, v0, 9
; CHECK-NEXT:    s_cmpk_lt_i32 s0, 0xa00
; CHECK-NEXT:    s_cbranch_scc1 .LBB0_1
; CHECK-NEXT:  ; %bb.2: ; %for.cond.cleanup.loopexit
; CHECK-NEXT:    v_mov_b32_e32 v3, 0
; CHECK-NEXT:    v_mov_b32_e32 v4, 0
; CHECK-NEXT:    global_store_dwordx2 v[3:4], v[1:2], off
; CHECK-NEXT:    s_endpgm
entry:
  %0 = load i32, ptr addrspace(4) null, align 4
  %cmp6 = icmp slt i32 0, 2560
  br i1 %cmp6, label %for.cond4.preheader, label %for.cond.cleanup

for.cond4.preheader:                              ; preds = %for.cond4.preheader, %entry
  %idx.07 = phi i32 [ %add13, %for.cond4.preheader ], [ 0, %entry ]
  %arrayidx.promoted = load double, ptr addrspace(1) null, align 8
  %add9 = fadd contract double %arrayidx.promoted, 0.000000e+00
  %add9.1 = fadd contract double %add9, 5.000000e+00
  %add9.2 = fadd contract double %add9.1, 6.000000e+00
  %add9.3 = fadd contract double %add9.2, 7.000000e+00
  %add9.4 = fadd contract double %add9.3, 9.000000e+00
  %add9.5 = fadd contract double %add9.4, 1.000000e+01
  %add9.6 = fadd contract double %add9.5, 1.100000e+01
  %add9.7 = fadd contract double %add9.6, 1.200000e+01
  store double %add9.7, ptr addrspace(1) null, align 8
  %add13 = add i32 %idx.07, %0
  %cmp = icmp slt i32 %add13, 2560
  br i1 %cmp, label %for.cond4.preheader, label %for.cond.cleanup

for.cond.cleanup:                                 ; preds = %for.cond4.preheader, %entry
  ret void
}

declare i32 @llvm.amdgcn.workitem.id.x() #0
declare i32 @llvm.amdgcn.workgroup.id.x() #0
declare align 4 ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() #0

attributes #0 = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }