[AMDGPU][GlobalISel] Add RegBankLegalize support for G_FENCE

llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp

@@ -913,6 +913,8 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST,
 
   addRulesForGOpcs({G_ABS}, Standard).Uni(S16, {{Sgpr32Trunc}, {Sgpr32SExt}});
 
+  addRulesForGOpcs({G_FENCE}).Any({{{}}, {{}, {}}});
+
   addRulesForGOpcs({G_READSTEADYCOUNTER, G_READCYCLECOUNTER}, Standard)
       .Uni(S64, {{Sgpr64}, {}});
 

llvm/test/CodeGen/AMDGPU/GlobalISel/fence.ll

@@ -0,0 +1,388 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GFX11 %s
+; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdhsa -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s
+
+@global_var = addrspace(1) global i32 0, align 4
+
+define amdgpu_kernel void @fence_release(ptr addrspace(1) %ptr, i32 %val1, i32 %val2) {
+; GFX9-LABEL: fence_release:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, global_var@gotpcrel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, global_var@gotpcrel32@hi+12
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    v_mov_b32_e32 v1, s3
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[4:5]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fence_release:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
+; GFX11-NEXT:    s_getpc_b64 s[4:5]
+; GFX11-NEXT:    s_add_u32 s4, s4, global_var@gotpcrel32@lo+4
+; GFX11-NEXT:    s_addc_u32 s5, s5, global_var@gotpcrel32@hi+12
+; GFX11-NEXT:    v_mov_b32_e32 v0, 0
+; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_mov_b32 v2, s3
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    global_store_b32 v0, v2, s[4:5] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: fence_release:
+; GFX12:       ; %bb.0:
+; GFX12-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
+; GFX12-NEXT:    s_getpc_b64 s[4:5]
+; GFX12-NEXT:    s_sext_i32_i16 s5, s5
+; GFX12-NEXT:    s_add_co_u32 s4, s4, global_var@gotpcrel32@lo+8
+; GFX12-NEXT:    s_add_co_ci_u32 s5, s5, global_var@gotpcrel32@hi+16
+; GFX12-NEXT:    v_mov_b32_e32 v0, 0
+; GFX12-NEXT:    s_load_b64 s[4:5], s[4:5], 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    v_dual_mov_b32 v1, s2 :: v_dual_mov_b32 v2, s3
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX12-NEXT:    global_wb scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    global_store_b32 v0, v2, s[4:5] scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+  store i32 %val1, ptr addrspace(1) %ptr
+  fence release
+  store volatile i32 %val2, ptr addrspace(1) @global_var
+  ret void
+}
+
+define amdgpu_kernel void @fence_acquire(ptr addrspace(1) %ptr) {
+; GFX9-LABEL: fence_acquire:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    buffer_wbinvl1_vol
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_load_dword s2, s[0:1], 0x0
+; GFX9-NEXT:    s_getpc_b64 s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    global_store_dword v1, v0, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fence_acquire:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    buffer_gl1_inv
+; GFX11-NEXT:    buffer_gl0_inv
+; GFX11-NEXT:    v_mov_b32_e32 v1, 0
+; GFX11-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX11-NEXT:    s_getpc_b64 s[0:1]
+; GFX11-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX11-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_mov_b32_e32 v0, s2
+; GFX11-NEXT:    global_store_b32 v1, v0, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: fence_acquire:
+; GFX12:       ; %bb.0:
+; GFX12-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX12-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-NEXT:    v_mov_b32_e32 v1, 0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    s_load_b32 s2, s[0:1], 0x0
+; GFX12-NEXT:    s_getpc_b64 s[0:1]
+; GFX12-NEXT:    s_sext_i32_i16 s1, s1
+; GFX12-NEXT:    s_add_co_u32 s0, s0, global_var@gotpcrel32@lo+8
+; GFX12-NEXT:    s_add_co_ci_u32 s1, s1, global_var@gotpcrel32@hi+16
+; GFX12-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    v_mov_b32_e32 v0, s2
+; GFX12-NEXT:    s_wait_loadcnt 0x0
+; GFX12-NEXT:    global_store_b32 v1, v0, s[0:1] scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+  fence acquire
+  %val = load i32, ptr addrspace(1) %ptr
+  store volatile i32 %val, ptr addrspace(1) @global_var
+  ret void
+}
+
+define amdgpu_kernel void @fence_acq_rel(ptr addrspace(1) %ptr, i32 %val) {
+; GFX9-LABEL: fence_acq_rel:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s2, s[8:9], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    buffer_wbinvl1_vol
+; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
+; GFX9-NEXT:    s_getpc_b64 s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fence_acq_rel:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    buffer_gl1_inv
+; GFX11-NEXT:    buffer_gl0_inv
+; GFX11-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX11-NEXT:    s_getpc_b64 s[0:1]
+; GFX11-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX11-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: fence_acq_rel:
+; GFX12:       ; %bb.0:
+; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX12-NEXT:    global_wb scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX12-NEXT:    s_getpc_b64 s[0:1]
+; GFX12-NEXT:    s_sext_i32_i16 s1, s1
+; GFX12-NEXT:    s_add_co_u32 s0, s0, global_var@gotpcrel32@lo+8
+; GFX12-NEXT:    s_add_co_ci_u32 s1, s1, global_var@gotpcrel32@hi+16
+; GFX12-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX12-NEXT:    s_wait_loadcnt 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1] scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+  store i32 %val, ptr addrspace(1) %ptr
+  fence acq_rel
+  %load = load i32, ptr addrspace(1) %ptr
+  store volatile i32 %load, ptr addrspace(1) @global_var
+  ret void
+}
+
+define amdgpu_kernel void @fence_seq_cst(ptr addrspace(1) %ptr, i32 %val) {
+; GFX9-LABEL: fence_seq_cst:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s2, s[8:9], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    buffer_wbinvl1_vol
+; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
+; GFX9-NEXT:    s_getpc_b64 s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fence_seq_cst:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    buffer_gl1_inv
+; GFX11-NEXT:    buffer_gl0_inv
+; GFX11-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX11-NEXT:    s_getpc_b64 s[0:1]
+; GFX11-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX11-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: fence_seq_cst:
+; GFX12:       ; %bb.0:
+; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX12-NEXT:    global_wb scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX12-NEXT:    s_getpc_b64 s[0:1]
+; GFX12-NEXT:    s_sext_i32_i16 s1, s1
+; GFX12-NEXT:    s_add_co_u32 s0, s0, global_var@gotpcrel32@lo+8
+; GFX12-NEXT:    s_add_co_ci_u32 s1, s1, global_var@gotpcrel32@hi+16
+; GFX12-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX12-NEXT:    s_wait_loadcnt 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1] scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+  store i32 %val, ptr addrspace(1) %ptr
+  fence seq_cst
+  %load = load i32, ptr addrspace(1) %ptr
+  store volatile i32 %load, ptr addrspace(1) @global_var
+  ret void
+}
+
+define amdgpu_kernel void @fence_workgroup_release(ptr addrspace(3) %ptr, i32 %val1, i32 %val2) {
+; GFX9-LABEL: fence_workgroup_release:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dwordx4 s[0:3], s[8:9], 0x0
+; GFX9-NEXT:    s_getpc_b64 s[4:5]
+; GFX9-NEXT:    s_add_u32 s4, s4, global_var@gotpcrel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s5, s5, global_var@gotpcrel32@hi+12
+; GFX9-NEXT:    s_load_dwordx2 s[4:5], s[4:5], 0x0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v0, s1
+; GFX9-NEXT:    v_mov_b32_e32 v1, s0
+; GFX9-NEXT:    ds_write_b32 v1, v0
+; GFX9-NEXT:    v_mov_b32_e32 v0, s2
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v1, v0, s[4:5]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fence_workgroup_release:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_load_b128 s[0:3], s[4:5], 0x0
+; GFX11-NEXT:    s_getpc_b64 s[4:5]
+; GFX11-NEXT:    s_add_u32 s4, s4, global_var@gotpcrel32@lo+4
+; GFX11-NEXT:    s_addc_u32 s5, s5, global_var@gotpcrel32@hi+12
+; GFX11-NEXT:    v_mov_b32_e32 v2, 0
+; GFX11-NEXT:    s_load_b64 s[4:5], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_dual_mov_b32 v0, s1 :: v_dual_mov_b32 v1, s0
+; GFX11-NEXT:    v_mov_b32_e32 v3, s2
+; GFX11-NEXT:    ds_store_b32 v1, v0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v2, v3, s[4:5] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: fence_workgroup_release:
+; GFX12:       ; %bb.0:
+; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    s_getpc_b64 s[4:5]
+; GFX12-NEXT:    s_sext_i32_i16 s5, s5
+; GFX12-NEXT:    s_add_co_u32 s4, s4, global_var@gotpcrel32@lo+8
+; GFX12-NEXT:    s_add_co_ci_u32 s5, s5, global_var@gotpcrel32@hi+16
+; GFX12-NEXT:    v_mov_b32_e32 v2, 0
+; GFX12-NEXT:    s_load_b64 s[4:5], s[4:5], 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    v_dual_mov_b32 v0, s1 :: v_dual_mov_b32 v1, s0
+; GFX12-NEXT:    v_mov_b32_e32 v3, s2
+; GFX12-NEXT:    ds_store_b32 v1, v0
+; GFX12-NEXT:    s_wait_dscnt 0x0
+; GFX12-NEXT:    global_store_b32 v2, v3, s[4:5] scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+  store i32 %val1, ptr addrspace(3) %ptr
+  fence syncscope("workgroup") release
+  store volatile i32 %val2, ptr addrspace(1) @global_var
+  ret void
+}
+
+define amdgpu_kernel void @fence_agent_seq_cst(ptr addrspace(1) %ptr, i32 %val) {
+; GFX9-LABEL: fence_agent_seq_cst:
+; GFX9:       ; %bb.0:
+; GFX9-NEXT:    s_load_dword s2, s[8:9], 0x8
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[8:9], 0x0
+; GFX9-NEXT:    v_mov_b32_e32 v0, 0
+; GFX9-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX9-NEXT:    v_mov_b32_e32 v1, s2
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    buffer_wbinvl1_vol
+; GFX9-NEXT:    global_load_dword v1, v0, s[0:1]
+; GFX9-NEXT:    s_getpc_b64 s[0:1]
+; GFX9-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX9-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX9-NEXT:    s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX9-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX9-NEXT:    global_store_dword v0, v1, s[0:1]
+; GFX9-NEXT:    s_waitcnt vmcnt(0)
+; GFX9-NEXT:    s_endpgm
+;
+; GFX11-LABEL: fence_agent_seq_cst:
+; GFX11:       ; %bb.0:
+; GFX11-NEXT:    s_clause 0x1
+; GFX11-NEXT:    s_load_b32 s2, s[4:5], 0x8
+; GFX11-NEXT:    s_load_b64 s[0:1], s[4:5], 0x0
+; GFX11-NEXT:    s_waitcnt lgkmcnt(0)
+; GFX11-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    buffer_gl1_inv
+; GFX11-NEXT:    buffer_gl0_inv
+; GFX11-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX11-NEXT:    s_getpc_b64 s[0:1]
+; GFX11-NEXT:    s_add_u32 s0, s0, global_var@gotpcrel32@lo+4
+; GFX11-NEXT:    s_addc_u32 s1, s1, global_var@gotpcrel32@hi+12
+; GFX11-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX11-NEXT:    global_store_b32 v0, v1, s[0:1] dlc
+; GFX11-NEXT:    s_waitcnt_vscnt null, 0x0
+; GFX11-NEXT:    s_endpgm
+;
+; GFX12-LABEL: fence_agent_seq_cst:
+; GFX12:       ; %bb.0:
+; GFX12-NEXT:    s_load_b96 s[0:2], s[4:5], 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    v_dual_mov_b32 v0, 0 :: v_dual_mov_b32 v1, s2
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1]
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-NEXT:    global_load_b32 v1, v0, s[0:1]
+; GFX12-NEXT:    s_getpc_b64 s[0:1]
+; GFX12-NEXT:    s_sext_i32_i16 s1, s1
+; GFX12-NEXT:    s_add_co_u32 s0, s0, global_var@gotpcrel32@lo+8
+; GFX12-NEXT:    s_add_co_ci_u32 s1, s1, global_var@gotpcrel32@hi+16
+; GFX12-NEXT:    s_load_b64 s[0:1], s[0:1], 0x0
+; GFX12-NEXT:    s_wait_loadcnt 0x0
+; GFX12-NEXT:    s_wait_kmcnt 0x0
+; GFX12-NEXT:    global_store_b32 v0, v1, s[0:1] scope:SCOPE_SYS
+; GFX12-NEXT:    s_wait_storecnt 0x0
+; GFX12-NEXT:    s_endpgm
+  store i32 %val, ptr addrspace(1) %ptr
+  fence syncscope("agent") seq_cst
+  %load = load i32, ptr addrspace(1) %ptr
+  store volatile i32 %load, ptr addrspace(1) @global_var
+  ret void
+}