llvm/test/CodeGen/AMDGPU/llvm.amdgcn.buffer.load-last-use.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
;RUN: llc < %s -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -amdgpu-enable-delay-alu=0 | FileCheck %s --check-prefix=GFX12
;RUN: llc < %s -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -amdgpu-enable-delay-alu=0 | FileCheck %s --check-prefix=GFX12
;RUN: llc < %s -global-isel=0 -mtriple=amdgcn -mcpu=gfx1310 -amdgpu-enable-delay-alu=0 | FileCheck %s --check-prefix=GFX13
;RUN: llc < %s -global-isel=1 -mtriple=amdgcn -mcpu=gfx1310 -amdgpu-enable-delay-alu=0 | FileCheck %s --check-prefix=GFX13

define float @raw_buffer_load(<4 x i32> inreg) {
; GFX12-LABEL: raw_buffer_load:
; GFX12:       ; %bb.0: ; %main_body
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    buffer_load_b32 v0, off, s[0:3], null th:TH_LOAD_LU
; GFX12-NEXT:    s_wait_loadcnt 0x0
; GFX12-NEXT:    s_setpc_b64 s[30:31]
;
; GFX13-LABEL: raw_buffer_load:
; GFX13:       ; %bb.0: ; %main_body
; GFX13-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX13-NEXT:    s_wait_expcnt 0x0
; GFX13-NEXT:    s_wait_samplecnt 0x0
; GFX13-NEXT:    s_wait_bvhcnt 0x0
; GFX13-NEXT:    s_wait_kmcnt 0x0
; GFX13-NEXT:    buffer_load_b32 v0, off, s[0:3], null th:TH_LOAD_LU
; GFX13-NEXT:    s_wait_loadcnt 0x0
; GFX13-NEXT:    s_set_pc_i64 s[30:31]
main_body:
  %data = call float @llvm.amdgcn.raw.buffer.load.f32(<4 x i32> %0, i32 0, i32 0, i32 3)
  ret float %data
}

define float @struct_buffer_load(<4 x i32> inreg) {
; GFX12-LABEL: struct_buffer_load:
; GFX12:       ; %bb.0: ; %main_body
; GFX12-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT:    s_wait_expcnt 0x0
; GFX12-NEXT:    s_wait_samplecnt 0x0
; GFX12-NEXT:    s_wait_bvhcnt 0x0
; GFX12-NEXT:    s_wait_kmcnt 0x0
; GFX12-NEXT:    v_mov_b32_e32 v0, 0
; GFX12-NEXT:    buffer_load_b32 v0, v0, s[0:3], null idxen th:TH_LOAD_LU
; GFX12-NEXT:    s_wait_loadcnt 0x0
; GFX12-NEXT:    s_setpc_b64 s[30:31]
;
; GFX13-LABEL: struct_buffer_load:
; GFX13:       ; %bb.0: ; %main_body
; GFX13-NEXT:    s_wait_loadcnt_dscnt 0x0
; GFX13-NEXT:    s_wait_expcnt 0x0
; GFX13-NEXT:    s_wait_samplecnt 0x0
; GFX13-NEXT:    s_wait_bvhcnt 0x0
; GFX13-NEXT:    s_wait_kmcnt 0x0
; GFX13-NEXT:    v_mov_b32_e32 v0, 0
; GFX13-NEXT:    buffer_load_b32 v0, v0, s[0:3], null idxen th:TH_LOAD_LU
; GFX13-NEXT:    s_wait_loadcnt 0x0
; GFX13-NEXT:    s_set_pc_i64 s[30:31]
main_body:
  %data = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 3)
  ret float %data
}