test/CodeGen/AMDGPU/memory-legalizer-buffer-atomics.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1250 < %s | FileCheck --check-prefixes=GFX1250 %s

define void @buffer_fat_ptr_agent_atomic_add_noret_i32(ptr addrspace(7) inreg %ptr, i32 %val) {
; GFX1250-LABEL: buffer_fat_ptr_agent_atomic_add_noret_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s6, s2
; GFX1250-NEXT:    s_mov_b32 s4, s0
; GFX1250-NEXT:    ; kill: def $sgpr6 killed $sgpr6 def $sgpr6_sgpr7
; GFX1250-NEXT:    s_mov_b32 s7, s3
; GFX1250-NEXT:    s_mov_b32 s8, s7
; GFX1250-NEXT:    s_mov_b32 s9, s6
; GFX1250-NEXT:    ; kill: def $sgpr4 killed $sgpr4 def $sgpr4_sgpr5
; GFX1250-NEXT:    s_mov_b32 s5, s1
; GFX1250-NEXT:    s_mov_b32 s10, s5
; GFX1250-NEXT:    s_mov_b32 s0, s4
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s10
; GFX1250-NEXT:    s_mov_b32 s2, s9
; GFX1250-NEXT:    s_mov_b32 s3, s8
; GFX1250-NEXT:    ; kill: def $sgpr8 killed $sgpr16
; GFX1250-NEXT:    v_mov_b32_e32 v1, s16
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_atomic_add_u32 v0, v1, s[0:3], null offen
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %unused = atomicrmw add ptr addrspace(7) %ptr, i32 %val syncscope("agent") monotonic
  ret void
}

define i32 @buffer_fat_ptr_agent_atomic_add_ret_i32(ptr addrspace(7) inreg %ptr, i32 %val) {
; GFX1250-LABEL: buffer_fat_ptr_agent_atomic_add_ret_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s6, s2
; GFX1250-NEXT:    s_mov_b32 s4, s0
; GFX1250-NEXT:    ; kill: def $sgpr6 killed $sgpr6 def $sgpr6_sgpr7
; GFX1250-NEXT:    s_mov_b32 s7, s3
; GFX1250-NEXT:    s_mov_b32 s8, s7
; GFX1250-NEXT:    s_mov_b32 s9, s6
; GFX1250-NEXT:    ; kill: def $sgpr4 killed $sgpr4 def $sgpr4_sgpr5
; GFX1250-NEXT:    s_mov_b32 s5, s1
; GFX1250-NEXT:    s_mov_b32 s10, s5
; GFX1250-NEXT:    s_mov_b32 s0, s4
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s10
; GFX1250-NEXT:    s_mov_b32 s2, s9
; GFX1250-NEXT:    s_mov_b32 s3, s8
; GFX1250-NEXT:    ; kill: def $sgpr8 killed $sgpr16
; GFX1250-NEXT:    v_mov_b32_e32 v1, s16
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_atomic_add_u32 v0, v1, s[0:3], null offen th:TH_ATOMIC_RETURN
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %ret = atomicrmw add ptr addrspace(7) %ptr, i32 %val syncscope("agent") monotonic
  ret i32 %ret
}

define void @raw_buffer_atomic_add_v2f16_noret(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; GFX1250-LABEL: raw_buffer_atomic_add_v2f16_noret:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_atomic_pk_add_f16 v0, v1, s[0:3], s16 offen
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
  ret void
}

define <2 x half> @raw_buffer_atomic_add_v2f16_ret(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; GFX1250-LABEL: raw_buffer_atomic_add_v2f16_ret:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_atomic_pk_add_f16 v0, v1, s[0:3], s16 offen th:TH_ATOMIC_RETURN
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
  ret <2 x half> %ret
}

define float @struct_buffer_atomic_add_v2f16_ret(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX1250-LABEL: struct_buffer_atomic_add_v2f16_ret:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    scratch_store_b32 off, v2, s32 ; 4-byte Folded Spill
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    v_mov_b32_e32 v2, v1
; GFX1250-NEXT:    scratch_load_b32 v1, off, s32 ; 4-byte Folded Reload
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_mov_b32_e32 v3, v1
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_atomic_pk_add_f16 v0, v[2:3], s[0:3], s16 idxen offen th:TH_ATOMIC_RETURN
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %orig = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
  %r = bitcast <2 x half> %orig to float
  ret float %r
}

define void @struct_buffer_atomic_add_v2f16_noret(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX1250-LABEL: struct_buffer_atomic_add_v2f16_noret:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    scratch_store_b32 off, v2, s32 ; 4-byte Folded Spill
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    v_mov_b32_e32 v2, v1
; GFX1250-NEXT:    scratch_load_b32 v1, off, s32 ; 4-byte Folded Reload
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    ; kill: def $vgpr2 killed $vgpr2 def $vgpr2_vgpr3 killed $exec
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    v_mov_b32_e32 v3, v1
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_atomic_pk_add_f16 v0, v[2:3], s[0:3], s16 idxen offen
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %orig = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
  ret void
}

define void @raw_buffer_store_i32(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: raw_buffer_store_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    buffer_store_b32 v0, off, s[0:3], null
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  call void @llvm.amdgcn.raw.buffer.store.i32(i32 %v, <4 x i32> %rsrc, i32 0, i32 0, i32 0)
  ret void
}

define void @raw_buffer_store_i32_volatile(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: raw_buffer_store_i32_volatile:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_store_b32 v0, off, s[0:3], null scope:SCOPE_SYS
; GFX1250-NEXT:    s_wait_storecnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  call void @llvm.amdgcn.raw.buffer.store.i32(i32 %v, <4 x i32> %rsrc, i32 0, i32 0, i32 2147483648)
  ret void
}

define void @struct_buffer_store_i32(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: struct_buffer_store_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    v_mov_b32_e32 v1, 0
; GFX1250-NEXT:    buffer_store_b32 v0, v1, s[0:3], null idxen
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  call void @llvm.amdgcn.struct.buffer.store.i32(i32 %v, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
  ret void
}

define void @struct_buffer_store_i32_volatile(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: struct_buffer_store_i32_volatile:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    v_mov_b32_e32 v1, 0
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_store_b32 v0, v1, s[0:3], null idxen scope:SCOPE_SYS
; GFX1250-NEXT:    s_wait_storecnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  call void @llvm.amdgcn.struct.buffer.store.i32(i32 %v, <4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 2147483648)
  ret void
}

define i32 @raw_buffer_load_i32(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: raw_buffer_load_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    buffer_load_b32 v0, off, s[0:3], null
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %val = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %rsrc, i32 0, i32 0, i32 0)
  ret i32 %val
}

; FIXME?: Compiler strips volatile bit during lowering, we cannot emit volatile buffer loads this way.
define i32 @raw_buffer_load_i32_volatile(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: raw_buffer_load_i32_volatile:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    buffer_load_b32 v0, off, s[0:3], null
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %val = call i32 @llvm.amdgcn.raw.buffer.load.i32(<4 x i32> %rsrc, i32 0, i32 0, i32 21474836480)
  ret i32 %val
}

define i32 @struct_buffer_load_i32(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: struct_buffer_load_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    v_mov_b32_e32 v0, 0
; GFX1250-NEXT:    buffer_load_b32 v0, v0, s[0:3], null idxen
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %val = call i32 @llvm.amdgcn.struct.buffer.load.i32(<4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 0)
  ret i32 %val
}

; FIXME?: Compiler strips volatile bit during lowering, we cannot emit volatile buffer loads this way.
define i32 @struct_buffer_load_i32_volatile(<4 x i32> inreg %rsrc, i32 %v) {
; GFX1250-LABEL: struct_buffer_load_i32_volatile:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s4, s3
; GFX1250-NEXT:    s_mov_b32 s5, s2
; GFX1250-NEXT:    s_mov_b32 s6, s1
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s6
; GFX1250-NEXT:    s_mov_b32 s2, s5
; GFX1250-NEXT:    s_mov_b32 s3, s4
; GFX1250-NEXT:    ; kill: def $sgpr4_sgpr5_sgpr6_sgpr7 killed $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    v_mov_b32_e32 v0, 0
; GFX1250-NEXT:    buffer_load_b32 v0, v0, s[0:3], null idxen
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %val = call i32 @llvm.amdgcn.struct.buffer.load.i32(<4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 21474836480)
  ret i32 %val
}

define void @buffer_fat_ptr_store_i32(ptr addrspace(7) inreg %ptr, i32 %val) {
; GFX1250-LABEL: buffer_fat_ptr_store_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s6, s2
; GFX1250-NEXT:    s_mov_b32 s4, s0
; GFX1250-NEXT:    ; kill: def $sgpr6 killed $sgpr6 def $sgpr6_sgpr7
; GFX1250-NEXT:    s_mov_b32 s7, s3
; GFX1250-NEXT:    s_mov_b32 s8, s7
; GFX1250-NEXT:    s_mov_b32 s9, s6
; GFX1250-NEXT:    ; kill: def $sgpr4 killed $sgpr4 def $sgpr4_sgpr5
; GFX1250-NEXT:    s_mov_b32 s5, s1
; GFX1250-NEXT:    s_mov_b32 s10, s5
; GFX1250-NEXT:    s_mov_b32 s0, s4
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s10
; GFX1250-NEXT:    s_mov_b32 s2, s9
; GFX1250-NEXT:    s_mov_b32 s3, s8
; GFX1250-NEXT:    ; kill: def $sgpr8 killed $sgpr16
; GFX1250-NEXT:    v_mov_b32_e32 v1, s16
; GFX1250-NEXT:    buffer_store_b32 v0, v1, s[0:3], null offen
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  store i32 %val, ptr addrspace(7) %ptr
  ret void
}

define void @buffer_fat_ptr_store_i32_volatile(ptr addrspace(7) inreg %ptr, i32 %val) {
; GFX1250-LABEL: buffer_fat_ptr_store_i32_volatile:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s6, s2
; GFX1250-NEXT:    s_mov_b32 s4, s0
; GFX1250-NEXT:    ; kill: def $sgpr6 killed $sgpr6 def $sgpr6_sgpr7
; GFX1250-NEXT:    s_mov_b32 s7, s3
; GFX1250-NEXT:    s_mov_b32 s8, s7
; GFX1250-NEXT:    s_mov_b32 s9, s6
; GFX1250-NEXT:    ; kill: def $sgpr4 killed $sgpr4 def $sgpr4_sgpr5
; GFX1250-NEXT:    s_mov_b32 s5, s1
; GFX1250-NEXT:    s_mov_b32 s10, s5
; GFX1250-NEXT:    s_mov_b32 s0, s4
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s10
; GFX1250-NEXT:    s_mov_b32 s2, s9
; GFX1250-NEXT:    s_mov_b32 s3, s8
; GFX1250-NEXT:    ; kill: def $sgpr8 killed $sgpr16
; GFX1250-NEXT:    v_mov_b32_e32 v1, s16
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_store_b32 v0, v1, s[0:3], null offen scope:SCOPE_SYS
; GFX1250-NEXT:    s_wait_storecnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  store volatile i32 %val, ptr addrspace(7) %ptr
  ret void
}

define i32 @buffer_fat_ptr_load_i32(ptr addrspace(7) inreg %ptr) {
; GFX1250-LABEL: buffer_fat_ptr_load_i32:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s6, s2
; GFX1250-NEXT:    s_mov_b32 s4, s0
; GFX1250-NEXT:    ; kill: def $sgpr6 killed $sgpr6 def $sgpr6_sgpr7
; GFX1250-NEXT:    s_mov_b32 s7, s3
; GFX1250-NEXT:    s_mov_b32 s8, s7
; GFX1250-NEXT:    s_mov_b32 s9, s6
; GFX1250-NEXT:    ; kill: def $sgpr4 killed $sgpr4 def $sgpr4_sgpr5
; GFX1250-NEXT:    s_mov_b32 s5, s1
; GFX1250-NEXT:    s_mov_b32 s10, s5
; GFX1250-NEXT:    s_mov_b32 s0, s4
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s10
; GFX1250-NEXT:    s_mov_b32 s2, s9
; GFX1250-NEXT:    s_mov_b32 s3, s8
; GFX1250-NEXT:    ; kill: def $sgpr8 killed $sgpr16
; GFX1250-NEXT:    v_mov_b32_e32 v0, s16
; GFX1250-NEXT:    buffer_load_b32 v0, v0, s[0:3], null offen
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %val = load i32, ptr addrspace(7) %ptr
  ret i32 %val
}

define i32 @buffer_fat_ptr_load_i32_volatile(ptr addrspace(7) inreg %ptr) {
; GFX1250-LABEL: buffer_fat_ptr_load_i32_volatile:
; GFX1250:       ; %bb.0:
; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT:    s_wait_kmcnt 0x0
; GFX1250-NEXT:    s_mov_b32 s6, s2
; GFX1250-NEXT:    s_mov_b32 s4, s0
; GFX1250-NEXT:    ; kill: def $sgpr6 killed $sgpr6 def $sgpr6_sgpr7
; GFX1250-NEXT:    s_mov_b32 s7, s3
; GFX1250-NEXT:    s_mov_b32 s8, s7
; GFX1250-NEXT:    s_mov_b32 s9, s6
; GFX1250-NEXT:    ; kill: def $sgpr4 killed $sgpr4 def $sgpr4_sgpr5
; GFX1250-NEXT:    s_mov_b32 s5, s1
; GFX1250-NEXT:    s_mov_b32 s10, s5
; GFX1250-NEXT:    s_mov_b32 s0, s4
; GFX1250-NEXT:    ; kill: def $sgpr0 killed $sgpr0 def $sgpr0_sgpr1_sgpr2_sgpr3
; GFX1250-NEXT:    s_mov_b32 s1, s10
; GFX1250-NEXT:    s_mov_b32 s2, s9
; GFX1250-NEXT:    s_mov_b32 s3, s8
; GFX1250-NEXT:    ; kill: def $sgpr8 killed $sgpr16
; GFX1250-NEXT:    v_mov_b32_e32 v0, s16
; GFX1250-NEXT:    s_wait_xcnt 0x0
; GFX1250-NEXT:    buffer_load_b32 v0, v0, s[0:3], null offen scope:SCOPE_SYS
; GFX1250-NEXT:    s_wait_loadcnt 0x0
; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
  %val = load volatile i32, ptr addrspace(7) %ptr
  ret i32 %val
}