llvm/test/CodeGen/AMDGPU/spill-offset-calculation.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 < %s | FileCheck -check-prefix=MUBUF %s
; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -enable-misched=0 -post-RA-scheduler=0 -stress-regalloc=8 -mattr=+enable-flat-scratch < %s | FileCheck -check-prefix=FLATSCR %s

; Test that the VGPR spiller correctly switches to SGPR offsets when the
; instruction offset field would overflow, and that it accounts for memory
; swizzling.

define amdgpu_kernel void @test_inst_offset_kernel() {
; MUBUF-LABEL: test_inst_offset_kernel:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_add_u32 s0, s0, s17
; MUBUF-NEXT:    s_addc_u32 s1, s1, 0
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:4 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:4088 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:4088 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:4
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_endpgm
;
; FLATSCR-LABEL: test_inst_offset_kernel:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s8, s13
; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s9, 0
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s0, 0xff8
; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 ; 4-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 ; 4-byte Folded Reload
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 offset:4
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_endpgm
entry:
  ; Occupy 4092 bytes of scratch, so the offset of the spill of %a just fits in
  ; the instruction offset field.
  %alloca = alloca i8, i32 4088, align 4, addrspace(5)

  %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1


  %a = load volatile i32, ptr addrspace(5) %aptr

  ; Force %a to spill.
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  %outptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  store volatile i32 %a, ptr addrspace(5) %outptr

  ret void
}

define amdgpu_kernel void @test_sgpr_offset_kernel() {
; MUBUF-LABEL: test_sgpr_offset_kernel:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_add_u32 s0, s0, s17
; MUBUF-NEXT:    s_addc_u32 s1, s1, 0
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_mov_b32 s4, 0x40000
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s4 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:8
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_endpgm
;
; FLATSCR-LABEL: test_sgpr_offset_kernel:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s8, s13
; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s9, 0
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s0, 0x1000
; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 ; 4-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 ; 4-byte Folded Reload
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 offset:8
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_endpgm
entry:
  ; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
  ; fit in the instruction, and has to live in the SGPR offset.
  %alloca = alloca i8, i32 4092, align 4, addrspace(5)

  %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  ; 0x40000 / 64 = 4096 (for wave64)
  %a = load volatile i32, ptr addrspace(5) %aptr
  ; Force %a to spill
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  %outptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  store volatile i32 %a, ptr addrspace(5) %outptr

  ret void
}

define void @test_sgpr_offset_function_scavenge_fail_func() #2 {
; MUBUF-LABEL: test_sgpr_offset_function_scavenge_fail_func:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_add_i32 s10, s32, 0x40100
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s10 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_add_i32 s10, s32, 0x40100
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s10 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_setpc_b64 s[30:31]
;
; FLATSCR-LABEL: test_sgpr_offset_function_scavenge_fail_func:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s32 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_add_i32 s8, s32, 0x1004
; FLATSCR-NEXT:    scratch_store_dword off, v0, s8 ; 4-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_add_i32 s8, s32, 0x1004
; FLATSCR-NEXT:    scratch_load_dword v0, off, s8 ; 4-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
entry:
  ; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
  ; fit in the instruction, and has to live in the SGPR offset.
  %alloca = alloca i8, i32 4096, align 4, addrspace(5)

  %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1

  %asm.0 = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
  %asm0.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 0
  %asm1.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 1
  %asm2.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 2
  %asm3.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 3
  %asm4.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 4
  %asm5.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 5
  %asm6.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 6
  %asm7.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 7

  ; 0x40000 / 64 = 4096 (for wave64)
  %a = load volatile i32, ptr addrspace(5) %aptr
  call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0.0, i32 %asm1.0, i32 %asm2.0, i32 %asm3.0, i32 %asm4.0, i32 %asm5.0, i32 %asm6.0, i32 %asm7.0, i32 %a)

  %asm = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
  %asm0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 0
  %asm1 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 1
  %asm2 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 2
  %asm3 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 3
  %asm4 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 4
  %asm5 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 5
  %asm6 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 6
  %asm7 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 7

  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}"() #0
   ; Force %a to spill with no free SGPRs
  call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0, i32 %asm1, i32 %asm2, i32 %asm3, i32 %asm4, i32 %asm5, i32 %asm6, i32 %asm7, i32 %a)
  ret void
}

define amdgpu_kernel void @test_sgpr_offset_function_scavenge_fail_kernel() #3 {
; MUBUF-LABEL: test_sgpr_offset_function_scavenge_fail_kernel:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_add_u32 s0, s0, s17
; MUBUF-NEXT:    s_addc_u32 s1, s1, 0
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_mov_b32 s10, 0x40100
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s10 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s10 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_endpgm
;
; FLATSCR-LABEL: test_sgpr_offset_function_scavenge_fail_kernel:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s8, s13
; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s9, 0
; FLATSCR-NEXT:    s_mov_b32 s8, 0
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s8 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s8, 0x1004
; FLATSCR-NEXT:    scratch_store_dword off, v0, s8 ; 4-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s8 ; 4-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_endpgm
entry:
  ; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
  ; fit in the instruction, and has to live in the SGPR offset.
  %alloca = alloca i8, i32 4096, align 4, addrspace(5)

  %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1

  %asm.0 = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
  %asm0.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 0
  %asm1.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 1
  %asm2.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 2
  %asm3.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 3
  %asm4.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 4
  %asm5.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 5
  %asm6.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 6
  %asm7.0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm.0, 7

  ; 0x40000 / 64 = 4096 (for wave64)
  %a = load volatile i32, ptr addrspace(5) %aptr
  call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0.0, i32 %asm1.0, i32 %asm2.0, i32 %asm3.0, i32 %asm4.0, i32 %asm5.0, i32 %asm6.0, i32 %asm7.0, i32 %a)

  %asm = call { i32, i32, i32, i32, i32, i32, i32, i32 } asm sideeffect "", "=s,=s,=s,=s,=s,=s,=s,=s"()
  %asm0 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 0
  %asm1 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 1
  %asm2 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 2
  %asm3 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 3
  %asm4 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 4
  %asm5 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 5
  %asm6 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 6
  %asm7 = extractvalue { i32, i32, i32, i32, i32, i32, i32, i32 } %asm, 7

  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}"() #0
   ; Force %a to spill with no free SGPRs
  call void asm sideeffect "", "s,s,s,s,s,s,s,s,v"(i32 %asm0, i32 %asm1, i32 %asm2, i32 %asm3, i32 %asm4, i32 %asm5, i32 %asm6, i32 %asm7, i32 %a)
  ret void
}

define amdgpu_kernel void @test_sgpr_offset_subregs_kernel() {
; MUBUF-LABEL: test_sgpr_offset_subregs_kernel:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_add_u32 s0, s0, s17
; MUBUF-NEXT:    s_addc_u32 s1, s1, 0
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], 0 offset:12 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], 0 offset:4084 ; 4-byte Folded Spill
; MUBUF-NEXT:    s_nop 0
; MUBUF-NEXT:    buffer_store_dword v1, off, s[0:3], 0 offset:4088 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:4 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:4084 ; 4-byte Folded Reload
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], 0 offset:4088 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ; v[0:1]
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_endpgm
;
; FLATSCR-LABEL: test_sgpr_offset_subregs_kernel:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s8, s13
; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s9, 0
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s0, 0xff4
; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 ; 8-byte Folded Spill
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:4 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s0, 0xff4
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 ; 8-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ; v[0:1]
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_endpgm
entry:
  ; Occupy 4088 bytes of scratch, so that the spill of the last subreg of %a
  ; still fits below offset 4096 (4088 + 8 - 4 = 4092), and can be placed in
  ; the instruction offset field.
  %alloca = alloca i8, i32 4084, align 4, addrspace(5)
  %aptr = getelementptr <2 x i32>, ptr addrspace(5) %alloca, i32 1
  %a = load volatile <2 x i32>, ptr addrspace(5) %aptr

  ; Force %a to spill.
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  ; Ensure the alloca sticks around.
  %bptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  %b = load volatile i32, ptr addrspace(5) %bptr

  ; Ensure the spill is of the full super-reg.
  call void asm sideeffect "; $0", "r"(<2 x i32> %a)

  ret void
}

define amdgpu_kernel void @test_inst_offset_subregs_kernel() {
; MUBUF-LABEL: test_inst_offset_subregs_kernel:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_add_u32 s0, s0, s17
; MUBUF-NEXT:    s_addc_u32 s1, s1, 0
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:12 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], 0 offset:16 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_mov_b32 s4, 0x3ff00
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; MUBUF-NEXT:    s_nop 0
; MUBUF-NEXT:    buffer_store_dword v1, off, s[0:3], s4 offset:4 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s4 ; 4-byte Folded Reload
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], s4 offset:4 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ; v[0:1]
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_endpgm
;
; FLATSCR-LABEL: test_inst_offset_subregs_kernel:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_add_u32 flat_scratch_lo, s8, s13
; FLATSCR-NEXT:    s_addc_u32 flat_scratch_hi, s9, 0
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 offset:12 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s0, 0xffc
; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s0 ; 8-byte Folded Spill
; FLATSCR-NEXT:    s_mov_b32 s0, 0
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_movk_i32 s0, 0xffc
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s0 ; 8-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ; v[0:1]
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_endpgm
entry:
  ; Occupy 4092 bytes of scratch, so that the spill of the last subreg of %a
  ; does not fit below offset 4096 (4092 + 8 - 4 = 4096), and has to live
  ; in the SGPR offset.
  %alloca = alloca i8, i32 4088, align 4, addrspace(5)

  ; 0x3ff00 / 64 = 4092 (for wave64)
  %aptr = getelementptr <2 x i32>, ptr addrspace(5) %alloca, i32 1
  %a = load volatile <2 x i32>, ptr addrspace(5) %aptr

  ; Force %a to spill.
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  ; Ensure the alloca sticks around.
  %bptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  %b = load volatile i32, ptr addrspace(5) %bptr

  ; Ensure the spill is of the full super-reg.
  call void asm sideeffect "; $0", "r"(<2 x i32> %a)

  ret void
}

define void @test_inst_offset_function() {
; MUBUF-LABEL: test_inst_offset_function:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:4 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s32 offset:4088 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:4088 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s32 offset:4
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_setpc_b64 s[30:31]
;
; FLATSCR-LABEL: test_inst_offset_function:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; FLATSCR-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dword off, v0, s32 offset:4088 ; 4-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s32 offset:4088 ; 4-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dword off, v0, s32 offset:4
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
entry:
  ; Occupy enough bytes of scratch, so the offset of the spill of %a
  ; just fits in the instruction offset field when the emergency stack
  ; slot is added. It's hard to hit the actual limit since we're also
  ; going to insert the emergency stack slot for large frames.
  %alloca = alloca i8, i32 4088, align 4, addrspace(5)

  %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1


  %a = load volatile i32, ptr addrspace(5) %aptr

  ; Force %a to spill.
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  %outptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  store volatile i32 %a, ptr addrspace(5) %outptr

  ret void
}

define void @test_sgpr_offset_function() {
; MUBUF-LABEL: test_sgpr_offset_function:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_add_i32 s4, s32, 0x40100
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_add_i32 s4, s32, 0x40100
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s4 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s32 offset:8
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_setpc_b64 s[30:31]
;
; FLATSCR-LABEL: test_sgpr_offset_function:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; FLATSCR-NEXT:    scratch_load_dword v0, off, s32 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_add_i32 s0, s32, 0x1004
; FLATSCR-NEXT:    scratch_store_dword off, v0, s0 ; 4-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_add_i32 s0, s32, 0x1004
; FLATSCR-NEXT:    scratch_load_dword v0, off, s0 ; 4-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dword off, v0, s32 offset:8
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
entry:
  ; Occupy 4096 bytes of scratch, so the offset of the spill of %a does not
  ; fit in the instruction, and has to live in the SGPR offset.
  %alloca = alloca i8, i32 4096, align 4, addrspace(5)

  %aptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  ; 0x40000 / 64 = 4096 (for wave64)
  %a = load volatile i32, ptr addrspace(5) %aptr

  ; Force %a to spill
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  %outptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  store volatile i32 %a, ptr addrspace(5) %outptr

  ret void
}

define void @test_sgpr_offset_subregs_function() {
; MUBUF-LABEL: test_sgpr_offset_subregs_function:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:12 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s32 offset:4084 ; 4-byte Folded Spill
; MUBUF-NEXT:    s_nop 0
; MUBUF-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:4088 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:4 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:4084 ; 4-byte Folded Reload
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:4088 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ; v[0:1]
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_setpc_b64 s[30:31]
;
; FLATSCR-LABEL: test_sgpr_offset_subregs_function:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s32 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s32 offset:4084 ; 8-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s32 offset:4 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s32 offset:4084 ; 8-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ; v[0:1]
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
entry:
  ; We want to test the spill of the last subreg of %a is the highest
  ; valid value for the immediate offset. We enable the emergency
  ; stack slot for large frames, so it's hard to get the frame layout
  ; exactly as we want to test it.
  ; Occupy 4084 bytes of scratch, so that the spill of the last subreg of %a
  ; still fits below offset 4096 (4084 + 8 - 4 = 4092), and can be placed in
  ; the instruction offset field.
  %alloca = alloca i8, i32 4084, align 4, addrspace(5)
  %aptr = getelementptr <2 x i32>, ptr addrspace(5) %alloca, i32 1
  %a = load volatile <2 x i32>, ptr addrspace(5) %aptr

  ; Force %a to spill.
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  ; Ensure the alloca sticks around.
  %bptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  %b = load volatile i32, ptr addrspace(5) %bptr

  ; Ensure the spill is of the full super-reg.
  call void asm sideeffect "; $0", "r"(<2 x i32> %a)

  ret void
}

define void @test_inst_offset_subregs_function() {
; MUBUF-LABEL: test_inst_offset_subregs_function:
; MUBUF:       ; %bb.0: ; %entry
; MUBUF-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:12 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:16 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_add_i32 s4, s32, 0x3ff00
; MUBUF-NEXT:    buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; MUBUF-NEXT:    s_nop 0
; MUBUF-NEXT:    buffer_store_dword v1, off, s[0:3], s4 offset:4 ; 4-byte Folded Spill
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s32 offset:8 glc
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    s_add_i32 s4, s32, 0x3ff00
; MUBUF-NEXT:    buffer_load_dword v0, off, s[0:3], s4 ; 4-byte Folded Reload
; MUBUF-NEXT:    buffer_load_dword v1, off, s[0:3], s4 offset:4 ; 4-byte Folded Reload
; MUBUF-NEXT:    s_waitcnt vmcnt(0)
; MUBUF-NEXT:    ;;#ASMSTART
; MUBUF-NEXT:    ; v[0:1]
; MUBUF-NEXT:    ;;#ASMEND
; MUBUF-NEXT:    s_setpc_b64 s[30:31]
;
; FLATSCR-LABEL: test_inst_offset_subregs_function:
; FLATSCR:       ; %bb.0: ; %entry
; FLATSCR-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s32 offset:12 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_store_dwordx2 off, v[0:1], s32 offset:4092 ; 8-byte Folded Spill
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    scratch_load_dword v0, off, s32 offset:8 glc
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    scratch_load_dwordx2 v[0:1], off, s32 offset:4092 ; 8-byte Folded Reload
; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
; FLATSCR-NEXT:    ;;#ASMSTART
; FLATSCR-NEXT:    ; v[0:1]
; FLATSCR-NEXT:    ;;#ASMEND
; FLATSCR-NEXT:    s_setpc_b64 s[30:31]
entry:
  ; Occupy 4088 bytes of scratch, so that the spill of the last subreg of %a
  ; does not fit below offset 4096 (408 + 4 + 8 - 4 = 4096), and has to live
  ; in the SGPR offset.
  %alloca = alloca i8, i32 4088, align 4, addrspace(5)

  ; 0x3ff0000 / 64 = 4092 (for wave64)
  %aptr = getelementptr <2 x i32>, ptr addrspace(5) %alloca, i32 1
  %a = load volatile <2 x i32>, ptr addrspace(5) %aptr

  ; Force %a to spill.
  call void asm sideeffect "", "~{v0},~{v1},~{v2},~{v3},~{v4},~{v5},~{v6},~{v7}" ()

  ; Ensure the alloca sticks around.
  %bptr = getelementptr i32, ptr addrspace(5) %alloca, i32 1
  %b = load volatile i32, ptr addrspace(5) %bptr

  ; Ensure the spill is of the full super-reg.
  call void asm sideeffect "; $0", "r"(<2 x i32> %a)

  ret void
}

attributes #0 = { nounwind }
attributes #1 = { nounwind "amdgpu-num-sgpr"="17" "amdgpu-num-vgpr"="8" }
attributes #2 = { nounwind "amdgpu-num-sgpr"="16" "amdgpu-num-vgpr"="8" }
attributes #3 = { nounwind "amdgpu-num-sgpr"="18" "amdgpu-num-vgpr"="8" }