[AMDGPU] Add gfx12 run lines to fence MMRA tests (#94333)

2 files changed, 602 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-global.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-global.ll
index da9bc6b..7a68fb5 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-global.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-global.ll
@@ -10,6 +10,8 @@
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx940 -mattr=+tgsplit < %s | FileCheck -check-prefixes=GFX940-TGSPLIT %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 < %s | FileCheck --check-prefixes=GFX11-WGP %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX11-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 < %s | FileCheck --check-prefixes=GFX12-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX12-CU %s
 
 define amdgpu_kernel void @workgroup_acquire_fence() {
 ; GFX6-LABEL: workgroup_acquire_fence:
@@ -65,6 +67,19 @@ define amdgpu_kernel void @workgroup_acquire_fence() {
 ; GFX11-CU-LABEL: workgroup_acquire_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SE
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") acquire, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -120,6 +135,18 @@ define amdgpu_kernel void @workgroup_release_fence() {
 ; GFX11-CU-LABEL: workgroup_release_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") release, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -179,6 +206,19 @@ define amdgpu_kernel void @workgroup_acq_rel_fence() {
 ; GFX11-CU-LABEL: workgroup_acq_rel_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SE
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") acq_rel, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -238,6 +278,19 @@ define amdgpu_kernel void @workgroup_seq_cst_fence() {
 ; GFX11-CU-LABEL: workgroup_seq_cst_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SE
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") seq_cst, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -297,6 +350,19 @@ define amdgpu_kernel void @workgroup_one_as_acquire_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_acquire_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SE
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") acquire, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -352,6 +418,18 @@ define amdgpu_kernel void @workgroup_one_as_release_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_release_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") release, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -411,6 +489,19 @@ define amdgpu_kernel void @workgroup_one_as_acq_rel_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_acq_rel_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SE
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") acq_rel, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -470,6 +561,19 @@ define amdgpu_kernel void @workgroup_one_as_seq_cst_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_seq_cst_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SE
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") seq_cst, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -548,6 +652,24 @@ define amdgpu_kernel void @agent_acquire_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") acquire, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -614,6 +736,22 @@ define amdgpu_kernel void @agent_release_fence() {
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") release, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -694,6 +832,24 @@ define amdgpu_kernel void @agent_acq_rel_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") acq_rel, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -774,6 +930,24 @@ define amdgpu_kernel void @agent_seq_cst_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") seq_cst, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -852,6 +1026,24 @@ define amdgpu_kernel void @agent_one_as_acquire_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") acquire, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -918,6 +1110,22 @@ define amdgpu_kernel void @agent_one_as_release_fence() {
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") release, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -998,6 +1206,24 @@ define amdgpu_kernel void @agent_one_as_acq_rel_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") acq_rel, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1078,6 +1304,24 @@ define amdgpu_kernel void @agent_one_as_seq_cst_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_DEV
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") seq_cst, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1158,6 +1402,24 @@ define amdgpu_kernel void @system_acquire_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence acquire, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1226,6 +1488,22 @@ define amdgpu_kernel void @system_release_fence() {
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence release, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1310,6 +1588,24 @@ define amdgpu_kernel void @system_acq_rel_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence acq_rel, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1394,6 +1690,24 @@ define amdgpu_kernel void @system_seq_cst_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence seq_cst, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1474,6 +1788,24 @@ define amdgpu_kernel void @system_one_as_acquire_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") acquire, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1542,6 +1874,22 @@ define amdgpu_kernel void @system_one_as_release_fence() {
 ; GFX11-CU-NEXT:    s_waitcnt vmcnt(0)
 ; GFX11-CU-NEXT:    s_waitcnt_vscnt null, 0x0
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") release, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1626,6 +1974,24 @@ define amdgpu_kernel void @system_one_as_acq_rel_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") acq_rel, !mmra !{!"amdgpu-as", !"global"}
   ret void
@@ -1710,6 +2076,24 @@ define amdgpu_kernel void @system_one_as_seq_cst_fence() {
 ; GFX11-CU-NEXT:    buffer_gl1_inv
 ; GFX11-CU-NEXT:    buffer_gl0_inv
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_samplecnt 0x0
+; GFX12-WGP-NEXT:    s_wait_loadcnt 0x0
+; GFX12-WGP-NEXT:    s_wait_storecnt 0x0
+; GFX12-WGP-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_bvhcnt 0x0
+; GFX12-CU-NEXT:    s_wait_samplecnt 0x0
+; GFX12-CU-NEXT:    s_wait_loadcnt 0x0
+; GFX12-CU-NEXT:    s_wait_storecnt 0x0
+; GFX12-CU-NEXT:    global_inv scope:SCOPE_SYS
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") seq_cst, !mmra !{!"amdgpu-as", !"global"}
   ret void
diff --git a/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-local.ll b/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-local.ll
index 601a6a6..e1794db 100644
--- a/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-local.ll
+++ b/llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-local.ll
@@ -10,6 +10,8 @@
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx940 -mattr=+tgsplit < %s | FileCheck -check-prefixes=GFX940-TGSPLIT %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 < %s | FileCheck --check-prefixes=GFX11-WGP %s
 ; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1100 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX11-CU %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 < %s | FileCheck --check-prefixes=GFX12-WGP %s
+; RUN: llc -mtriple=amdgcn-amd-amdhsa -O0 -mcpu=gfx1200 -mattr=+cumode < %s | FileCheck --check-prefixes=GFX12-CU %s
 
 define amdgpu_kernel void @workgroup_acquire_fence() {
 ; GFX6-LABEL: workgroup_acquire_fence:
@@ -64,6 +66,16 @@ define amdgpu_kernel void @workgroup_acquire_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") acquire, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -122,6 +134,16 @@ define amdgpu_kernel void @workgroup_release_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") release, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -180,6 +202,16 @@ define amdgpu_kernel void @workgroup_acq_rel_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") acq_rel, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -238,6 +270,16 @@ define amdgpu_kernel void @workgroup_seq_cst_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup") seq_cst, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -287,6 +329,14 @@ define amdgpu_kernel void @workgroup_one_as_acquire_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_acquire_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") acquire, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -336,6 +386,14 @@ define amdgpu_kernel void @workgroup_one_as_release_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_release_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") release, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -385,6 +443,14 @@ define amdgpu_kernel void @workgroup_one_as_acq_rel_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_acq_rel_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") acq_rel, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -434,6 +500,14 @@ define amdgpu_kernel void @workgroup_one_as_seq_cst_fence() {
 ; GFX11-CU-LABEL: workgroup_one_as_seq_cst_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: workgroup_one_as_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("workgroup-one-as") seq_cst, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -492,6 +566,16 @@ define amdgpu_kernel void @agent_acquire_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") acquire, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -550,6 +634,16 @@ define amdgpu_kernel void @agent_release_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") release, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -608,6 +702,16 @@ define amdgpu_kernel void @agent_acq_rel_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") acq_rel, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -666,6 +770,16 @@ define amdgpu_kernel void @agent_seq_cst_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent") seq_cst, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -715,6 +829,14 @@ define amdgpu_kernel void @agent_one_as_acquire_fence() {
 ; GFX11-CU-LABEL: agent_one_as_acquire_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") acquire, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -764,6 +886,14 @@ define amdgpu_kernel void @agent_one_as_release_fence() {
 ; GFX11-CU-LABEL: agent_one_as_release_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") release, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -813,6 +943,14 @@ define amdgpu_kernel void @agent_one_as_acq_rel_fence() {
 ; GFX11-CU-LABEL: agent_one_as_acq_rel_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") acq_rel, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -862,6 +1000,14 @@ define amdgpu_kernel void @agent_one_as_seq_cst_fence() {
 ; GFX11-CU-LABEL: agent_one_as_seq_cst_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: agent_one_as_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: agent_one_as_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("agent-one-as") seq_cst, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -920,6 +1066,16 @@ define amdgpu_kernel void @system_acquire_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence acquire, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -978,6 +1134,16 @@ define amdgpu_kernel void @system_release_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence release, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -1036,6 +1202,16 @@ define amdgpu_kernel void @system_acq_rel_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence acq_rel, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -1094,6 +1270,16 @@ define amdgpu_kernel void @system_seq_cst_fence() {
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_waitcnt lgkmcnt(0)
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_wait_dscnt 0x0
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_wait_dscnt 0x0
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence seq_cst, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -1143,6 +1329,14 @@ define amdgpu_kernel void @system_one_as_acquire_fence() {
 ; GFX11-CU-LABEL: system_one_as_acquire_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_acquire_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_acquire_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") acquire, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -1192,6 +1386,14 @@ define amdgpu_kernel void @system_one_as_release_fence() {
 ; GFX11-CU-LABEL: system_one_as_release_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_release_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_release_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") release, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -1241,6 +1443,14 @@ define amdgpu_kernel void @system_one_as_acq_rel_fence() {
 ; GFX11-CU-LABEL: system_one_as_acq_rel_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_acq_rel_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_acq_rel_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") acq_rel, !mmra !{!"amdgpu-as", !"local"}
   ret void
@@ -1290,6 +1500,14 @@ define amdgpu_kernel void @system_one_as_seq_cst_fence() {
 ; GFX11-CU-LABEL: system_one_as_seq_cst_fence:
 ; GFX11-CU:       ; %bb.0: ; %entry
 ; GFX11-CU-NEXT:    s_endpgm
+;
+; GFX12-WGP-LABEL: system_one_as_seq_cst_fence:
+; GFX12-WGP:       ; %bb.0: ; %entry
+; GFX12-WGP-NEXT:    s_endpgm
+;
+; GFX12-CU-LABEL: system_one_as_seq_cst_fence:
+; GFX12-CU:       ; %bb.0: ; %entry
+; GFX12-CU-NEXT:    s_endpgm
 entry:
   fence syncscope("one-as") seq_cst, !mmra !{!"amdgpu-as", !"local"}
   ret void