shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Drop -CU suffix

Browse Source
This commit is contained in:
pvanhout 2025-08-21 12:44:08 +02:00
parent e83355bd69
commit 3b51b225ba
22 changed files with 14528 additions and 14528 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

240
llvm/test/CodeGen/AMDGPU/memory-legalizer-fence-mmra-global.ll
View File

@ -80,11 +80,11 @@ define amdgpu_kernel void @workgroup_acquire_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_acquire_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_acquire_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") acquire, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -153,11 +153,11 @@ define amdgpu_kernel void @workgroup_release_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") release, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -231,11 +231,11 @@ define amdgpu_kernel void @workgroup_acq_rel_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") acq_rel, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -309,11 +309,11 @@ define amdgpu_kernel void @workgroup_seq_cst_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") seq_cst, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -385,11 +385,11 @@ define amdgpu_kernel void @workgroup_one_as_acquire_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_acquire_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_acquire_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") acquire, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -458,11 +458,11 @@ define amdgpu_kernel void @workgroup_one_as_release_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") release, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -536,11 +536,11 @@ define amdgpu_kernel void @workgroup_one_as_acq_rel_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") acq_rel, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -614,11 +614,11 @@ define amdgpu_kernel void @workgroup_one_as_seq_cst_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") seq_cst, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -801,12 +801,12 @@ define amdgpu_kernel void @agent_release_fence() {
; GFX12-CU-NEXT: s_wait_storecnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent") release, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -906,13 +906,13 @@ define amdgpu_kernel void @agent_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent") acq_rel, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1012,13 +1012,13 @@ define amdgpu_kernel void @agent_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent") seq_cst, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1201,12 +1201,12 @@ define amdgpu_kernel void @agent_one_as_release_fence() {
; GFX12-CU-NEXT: s_wait_storecnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_one_as_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_one_as_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent-one-as") release, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1306,13 +1306,13 @@ define amdgpu_kernel void @agent_one_as_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_one_as_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_one_as_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent-one-as") acq_rel, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1412,13 +1412,13 @@ define amdgpu_kernel void @agent_one_as_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_one_as_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_one_as_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent-one-as") seq_cst, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1607,12 +1607,12 @@ define amdgpu_kernel void @system_release_fence() {
; GFX12-CU-NEXT: s_wait_storecnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence release, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1718,13 +1718,13 @@ define amdgpu_kernel void @system_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence acq_rel, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -1830,13 +1830,13 @@ define amdgpu_kernel void @system_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence seq_cst, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -2025,12 +2025,12 @@ define amdgpu_kernel void @system_one_as_release_fence() {
; GFX12-CU-NEXT: s_wait_storecnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_one_as_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_one_as_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("one-as") release, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -2136,13 +2136,13 @@ define amdgpu_kernel void @system_one_as_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_one_as_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_one_as_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("one-as") acq_rel, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void
@ -2248,13 +2248,13 @@ define amdgpu_kernel void @system_one_as_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_one_as_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_one_as_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("one-as") seq_cst, !mmra !{!"amdgpu-synchronize-as", !"global"}
ret void

240
llvm/test/CodeGen/AMDGPU/memory-legalizer-fence.ll
View File

@ -1064,11 +1064,11 @@ define amdgpu_kernel void @workgroup_acquire_fence() {
; GFX12-CU-NEXT: s_wait_dscnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_acquire_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_acquire_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") acquire
ret void
@ -1145,11 +1145,11 @@ define amdgpu_kernel void @workgroup_release_fence() {
; GFX12-CU-NEXT: s_wait_dscnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") release
ret void
@ -1231,11 +1231,11 @@ define amdgpu_kernel void @workgroup_acq_rel_fence() {
; GFX12-CU-NEXT: s_wait_dscnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") acq_rel
ret void
@ -1317,11 +1317,11 @@ define amdgpu_kernel void @workgroup_seq_cst_fence() {
; GFX12-CU-NEXT: s_wait_dscnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup") seq_cst
ret void
@ -1393,11 +1393,11 @@ define amdgpu_kernel void @workgroup_one_as_acquire_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_acquire_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_acquire_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") acquire
ret void
@ -1466,11 +1466,11 @@ define amdgpu_kernel void @workgroup_one_as_release_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") release
ret void
@ -1544,11 +1544,11 @@ define amdgpu_kernel void @workgroup_one_as_acq_rel_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") acq_rel
ret void
@ -1622,11 +1622,11 @@ define amdgpu_kernel void @workgroup_one_as_seq_cst_fence() {
; GFX12-CU: ; %bb.0: ; %entry
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: workgroup_one_as_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: workgroup_one_as_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("workgroup-one-as") seq_cst
ret void
@ -1809,12 +1809,12 @@ define amdgpu_kernel void @agent_release_fence() {
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent") release
ret void
@ -1914,13 +1914,13 @@ define amdgpu_kernel void @agent_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent") acq_rel
ret void
@ -2020,13 +2020,13 @@ define amdgpu_kernel void @agent_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent") seq_cst
ret void
@ -2209,12 +2209,12 @@ define amdgpu_kernel void @agent_one_as_release_fence() {
; GFX12-CU-NEXT: s_wait_storecnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_one_as_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_one_as_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent-one-as") release
ret void
@ -2314,13 +2314,13 @@ define amdgpu_kernel void @agent_one_as_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_one_as_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_one_as_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent-one-as") acq_rel
ret void
@ -2420,13 +2420,13 @@ define amdgpu_kernel void @agent_one_as_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: agent_one_as_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: agent_one_as_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_DEV
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_DEV
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("agent-one-as") seq_cst
ret void
@ -2615,12 +2615,12 @@ define amdgpu_kernel void @system_release_fence() {
; GFX12-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence release
ret void
@ -2726,13 +2726,13 @@ define amdgpu_kernel void @system_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence acq_rel
ret void
@ -2838,13 +2838,13 @@ define amdgpu_kernel void @system_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence seq_cst
ret void
@ -3033,12 +3033,12 @@ define amdgpu_kernel void @system_one_as_release_fence() {
; GFX12-CU-NEXT: s_wait_storecnt 0x0
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_one_as_release_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_one_as_release_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("one-as") release
ret void
@ -3144,13 +3144,13 @@ define amdgpu_kernel void @system_one_as_acq_rel_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_one_as_acq_rel_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_one_as_acq_rel_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("one-as") acq_rel
ret void
@ -3256,13 +3256,13 @@ define amdgpu_kernel void @system_one_as_seq_cst_fence() {
; GFX12-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: system_one_as_seq_cst_fence:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: system_one_as_seq_cst_fence:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: global_wb scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: global_inv scope:SCOPE_SYS
; GFX1250-NEXT: s_endpgm
entry:
fence syncscope("one-as") seq_cst
ret void

2930
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-agent.ll
View File

File diff suppressed because it is too large Load Diff

20
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-lastuse.ll
View File

@ -108,16 +108,16 @@ define amdgpu_kernel void @flat_last_use_and_volatile_load(ptr %in, ptr %out) {
; GFX12-NEXT: flat_store_b32 v[0:1], v2
; GFX12-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: flat_last_use_and_volatile_load:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: flat_load_b32 v1, v0, s[2:3] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: flat_last_use_and_volatile_load:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: flat_load_b32 v1, v0, s[2:3] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-NEXT: s_endpgm
entry:
%val = load volatile i32, ptr %in, align 4, !amdgpu.last.use !{}
store i32 %val, ptr %out

20
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-nontemporal.ll
View File

@ -1346,16 +1346,16 @@ define amdgpu_kernel void @flat_nontemporal_volatile_load(
; GFX12-CU-NEXT: flat_store_b32 v[0:1], v2
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: flat_nontemporal_volatile_load:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: flat_load_b32 v1, v0, s[2:3] th:TH_LOAD_NT scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: flat_nontemporal_volatile_load:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: flat_load_b32 v1, v0, s[2:3] th:TH_LOAD_NT scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-NEXT: s_endpgm
ptr %in, ptr %out) {
entry:
%val = load volatile i32, ptr %in, align 4, !nontemporal !0

2304
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-singlethread.ll
View File

File diff suppressed because it is too large Load Diff

2930
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-system.ll
View File

File diff suppressed because it is too large Load Diff

72
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-volatile.ll
View File

@ -145,16 +145,16 @@ define amdgpu_kernel void @flat_nontemporal_load_0(
; GFX12-CU-NEXT: flat_store_b32 v[0:1], v2
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: flat_nontemporal_load_0:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: flat_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: flat_nontemporal_load_0:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: flat_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-NEXT: s_endpgm
ptr %in, ptr %out) {
entry:
%val = load volatile i32, ptr %in, align 4
@ -428,20 +428,20 @@ define amdgpu_kernel void @flat_nontemporal_load_1(
; GFX12-CU-NEXT: flat_store_b32 v[0:1], v2
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: flat_nontemporal_load_1:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v1, v0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_wait_xcnt 0x0
; GFX1250-CU-NEXT: s_mov_b32 s4, 0x3ff
; GFX1250-CU-NEXT: v_and_b32_e64 v1, v1, s4
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: flat_load_b32 v1, v1, s[2:3] scale_offset scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: flat_nontemporal_load_1:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v1, v0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_wait_xcnt 0x0
; GFX1250-NEXT: s_mov_b32 s4, 0x3ff
; GFX1250-NEXT: v_and_b32_e64 v1, v1, s4
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: flat_load_b32 v1, v1, s[2:3] scale_offset scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-NEXT: s_endpgm
ptr %in, ptr %out) {
entry:
%tid = call i32 @llvm.amdgcn.workitem.id.x()
@ -1156,18 +1156,18 @@ define amdgpu_kernel void @flat_volatile_workgroup_release_store(
; GFX12-CU-NEXT: flat_store_b32 v[0:1], v2
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: flat_volatile_workgroup_release_store:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: v_mov_b32_e32 v1, s2
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_xcnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: flat_volatile_workgroup_release_store:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_xcnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: flat_store_b32 v0, v1, s[0:1] scope:SCOPE_SE
; GFX1250-NEXT: s_endpgm
i32 %in, ptr %out) {
entry:
store atomic volatile i32 %in, ptr %out syncscope("workgroup") release, align 4

2272
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-wavefront.ll
View File

File diff suppressed because it is too large Load Diff

2512
llvm/test/CodeGen/AMDGPU/memory-legalizer-flat-workgroup.ll
View File

File diff suppressed because it is too large Load Diff

2858
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-agent.ll
View File

File diff suppressed because it is too large Load Diff

20
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-lastuse.ll
View File

@ -87,16 +87,16 @@ define amdgpu_kernel void @global_last_use_and_volatile_load(ptr addrspace(1) %i
; GFX12-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX12-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: global_last_use_and_volatile_load:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: global_load_b32 v1, v0, s[2:3] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: global_last_use_and_volatile_load:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: global_load_b32 v1, v0, s[2:3] th:TH_LOAD_BYPASS scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT: s_endpgm
entry:
%val = load volatile i32, ptr addrspace(1) %in, align 4, !amdgpu.last.use !{}
store i32 %val, ptr addrspace(1) %out

20
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-nontemporal.ll
View File

@ -1111,16 +1111,16 @@ define amdgpu_kernel void @global_nontemporal_volatile_load(
; GFX12-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: global_nontemporal_volatile_load:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: global_load_b32 v1, v0, s[2:3] th:TH_LOAD_NT scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: global_nontemporal_volatile_load:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: global_load_b32 v1, v0, s[2:3] th:TH_LOAD_NT scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT: s_endpgm
ptr addrspace(1) %in, ptr addrspace(1) %out) {
entry:
%val = load volatile i32, ptr addrspace(1) %in, align 4, !nontemporal !0

2304
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-singlethread.ll
View File

File diff suppressed because it is too large Load Diff

2706
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-system.ll
View File

File diff suppressed because it is too large Load Diff

72
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-volatile.ll
View File

@ -148,16 +148,16 @@ define amdgpu_kernel void @global_volatile_load_0(
; GFX12-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: global_volatile_load_0:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: global_volatile_load_0:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: global_load_b32 v1, v0, s[2:3] scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT: s_endpgm
ptr addrspace(1) %in, ptr addrspace(1) %out) {
entry:
%val = load volatile i32, ptr addrspace(1) %in, align 4
@ -358,20 +358,20 @@ define amdgpu_kernel void @global_volatile_load_1(
; GFX12-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: global_volatile_load_1:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: v_mov_b32_e32 v1, v0
; GFX1250-CU-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_wait_xcnt 0x0
; GFX1250-CU-NEXT: s_mov_b32 s4, 0x3ff
; GFX1250-CU-NEXT: v_and_b32_e64 v1, v1, s4
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: global_load_b32 v1, v1, s[2:3] scale_offset scope:SCOPE_SYS
; GFX1250-CU-NEXT: s_wait_loadcnt 0x0
; GFX1250-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: global_volatile_load_1:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: v_mov_b32_e32 v1, v0
; GFX1250-NEXT: s_load_b64 s[2:3], s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_wait_xcnt 0x0
; GFX1250-NEXT: s_mov_b32 s4, 0x3ff
; GFX1250-NEXT: v_and_b32_e64 v1, v1, s4
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: global_load_b32 v1, v1, s[2:3] scale_offset scope:SCOPE_SYS
; GFX1250-NEXT: s_wait_loadcnt 0x0
; GFX1250-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT: s_endpgm
ptr addrspace(1) %in, ptr addrspace(1) %out) {
entry:
%tid = call i32 @llvm.amdgcn.workitem.id.x()
@ -1036,18 +1036,18 @@ define amdgpu_kernel void @global_volatile_workgroup_release_store(
; GFX12-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: global_volatile_workgroup_release_store:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_load_b32 s2, s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: v_mov_b32_e32 v1, s2
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_xcnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: global_volatile_workgroup_release_store:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_load_b32 s2, s[4:5], 0x0
; GFX1250-NEXT: s_load_b64 s[0:1], s[4:5], 0x8
; GFX1250-NEXT: v_mov_b32_e32 v0, 0
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_mov_b32_e32 v1, s2
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_xcnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: global_store_b32 v0, v1, s[0:1]
; GFX1250-NEXT: s_endpgm
i32 %in, ptr addrspace(1) %out) {
entry:
store atomic volatile i32 %in, ptr addrspace(1) %out syncscope("workgroup") release, align 4

2304
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-wavefront.ll
View File

File diff suppressed because it is too large Load Diff

2648
llvm/test/CodeGen/AMDGPU/memory-legalizer-global-workgroup.ll
View File

File diff suppressed because it is too large Load Diff

854
llvm/test/CodeGen/AMDGPU/memory-legalizer-local-agent.ll
View File

File diff suppressed because it is too large Load Diff

854
llvm/test/CodeGen/AMDGPU/memory-legalizer-local-system.ll
View File

File diff suppressed because it is too large Load Diff

22
llvm/test/CodeGen/AMDGPU/memory-legalizer-local-volatile.ll
View File

@ -883,17 +883,17 @@ define amdgpu_kernel void @local_volatile_workgroup_release_store(
; GFX12-CU-NEXT: ds_store_b32 v0, v1
; GFX12-CU-NEXT: s_endpgm
;
; GFX1250-CU-LABEL: local_volatile_workgroup_release_store:
; GFX1250-CU: ; %bb.0: ; %entry
; GFX1250-CU-NEXT: s_load_b32 s0, s[4:5], 0x0
; GFX1250-CU-NEXT: s_load_b32 s1, s[4:5], 0x4
; GFX1250-CU-NEXT: s_wait_kmcnt 0x0
; GFX1250-CU-NEXT: v_mov_b32_e32 v0, s1
; GFX1250-CU-NEXT: v_mov_b32_e32 v1, s0
; GFX1250-CU-NEXT: s_wait_storecnt 0x0
; GFX1250-CU-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-CU-NEXT: ds_store_b32 v0, v1
; GFX1250-CU-NEXT: s_endpgm
; GFX1250-LABEL: local_volatile_workgroup_release_store:
; GFX1250: ; %bb.0: ; %entry
; GFX1250-NEXT: s_load_b32 s0, s[4:5], 0x0
; GFX1250-NEXT: s_load_b32 s1, s[4:5], 0x4
; GFX1250-NEXT: s_wait_kmcnt 0x0
; GFX1250-NEXT: v_mov_b32_e32 v0, s1
; GFX1250-NEXT: v_mov_b32_e32 v1, s0
; GFX1250-NEXT: s_wait_storecnt 0x0
; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX1250-NEXT: ds_store_b32 v0, v1
; GFX1250-NEXT: s_endpgm
i32 %in, ptr addrspace(3) %out) {
entry:
store atomic volatile i32 %in, ptr addrspace(3) %out syncscope("workgroup") release, align 4

854
llvm/test/CodeGen/AMDGPU/memory-legalizer-local-workgroup.ll
View File

File diff suppressed because it is too large Load Diff