Skip to content

[AMDGPU] Allow potentially negative flat scratch offsets on GFX12 #78193

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jan 18, 2024
Merged

[AMDGPU] Allow potentially negative flat scratch offsets on GFX12 #78193

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
aefd530
[AMDGPU] Allow potentially negative flat scratch offsets on GFX12
rampitec Dec 6, 2023
f7f098e
Introduce GCNSubtarget::hasSignedScratchOffsets
jayfoad Jan 17, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
10 changes: 10 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1184,6 +1184,11 @@ bool AMDGPUDAGToDAGISel::isFlatScratchBaseLegalSV(SDValue Addr) const {
if (isNoUnsignedWrap(Addr))
return true;

// Starting with GFX12, VADDR and SADDR fields in VSCRATCH can use negative
// values.
if (AMDGPU::isGFX12Plus(*Subtarget))
return true;

auto LHS = Addr.getOperand(0);
auto RHS = Addr.getOperand(1);
return CurDAG->SignBitIsZero(RHS) && CurDAG->SignBitIsZero(LHS);
Expand All @@ -1192,6 +1197,11 @@ bool AMDGPUDAGToDAGISel::isFlatScratchBaseLegalSV(SDValue Addr) const {
// Check address value in SGPR/VGPR are legal for flat scratch in the form
// of: SGPR + VGPR + Imm.
bool AMDGPUDAGToDAGISel::isFlatScratchBaseLegalSVImm(SDValue Addr) const {
// Starting with GFX12, VADDR and SADDR fields in VSCRATCH can use negative
// values.
if (AMDGPU::isGFX12Plus(*Subtarget))
return true;

auto Base = Addr.getOperand(0);
auto *RHSImm = cast<ConstantSDNode>(Addr.getOperand(1));
// If the immediate offset is negative and within certain range, the base
Expand Down
10 changes: 10 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4586,6 +4586,11 @@ bool AMDGPUInstructionSelector::isFlatScratchBaseLegalSV(Register Addr) const {
if (isNoUnsignedWrap(AddrMI))
return true;

// Starting with GFX12, VADDR and SADDR fields in VSCRATCH can use negative
// values.
if (AMDGPU::isGFX12Plus(STI))
return true;

Register LHS = AddrMI->getOperand(1).getReg();
Register RHS = AddrMI->getOperand(2).getReg();
return KB->signBitIsZero(RHS) && KB->signBitIsZero(LHS);
Expand All @@ -4595,6 +4600,11 @@ bool AMDGPUInstructionSelector::isFlatScratchBaseLegalSV(Register Addr) const {
// of: SGPR + VGPR + Imm.
bool AMDGPUInstructionSelector::isFlatScratchBaseLegalSVImm(
Register Addr) const {
// Starting with GFX12, VADDR and SADDR fields in VSCRATCH can use negative
// values.
if (AMDGPU::isGFX12Plus(STI))
return true;

MachineInstr *AddrMI = getDefIgnoringCopies(Addr, *MRI);
Register Base = AddrMI->getOperand(1).getReg();
std::optional<DefinitionAndSourceRegister> BaseDef =
Expand Down
105 changes: 51 additions & 54 deletions llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -79,16 +79,17 @@ define amdgpu_kernel void @store_load_sindex_kernel(i32 %idx) {
; GFX12-LABEL: store_load_sindex_kernel:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x24
; GFX12-NEXT: v_mov_b32_e32 v1, 15
; GFX12-NEXT: s_waitcnt lgkmcnt(0)
; GFX12-NEXT: s_and_b32 s1, s0, 15
; GFX12-NEXT: s_lshl_b32 s1, s0, 2
; GFX12-NEXT: s_and_b32 s0, s0, 15
; GFX12-NEXT: v_mov_b32_e32 v0, s1
; GFX12-NEXT: s_lshl_b32 s0, s0, 2
; GFX12-NEXT: s_lshl_b32 s1, s1, 2
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX12-NEXT: v_dual_mov_b32 v0, 15 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: s_add_co_i32 s0, s0, 4
; GFX12-NEXT: scratch_store_b32 off, v0, s0 th:TH_STORE_NT_RT
; GFX12-NEXT: v_mov_b32_e32 v2, s0
; GFX12-NEXT: scratch_store_b32 v0, v1, off offset:4 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v2, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down Expand Up @@ -170,8 +171,7 @@ define amdgpu_kernel void @store_load_vindex_kernel() {
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v1
; GFX12-NEXT: scratch_store_b32 v0, v2, off offset:4 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: v_add_nc_u32_e32 v1, 4, v1
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:124 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:128 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down Expand Up @@ -248,14 +248,13 @@ define void @store_load_vindex_foo(i32 %idx) {
; GFX12-LABEL: store_load_vindex_foo:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v0
; GFX12-NEXT: v_and_b32_e32 v0, 15, v0
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-NEXT: v_add_nc_u32_e32 v1, s32, v1
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_and_b32 v1, 15, v0
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_store_b32 v1, v2, off th:TH_STORE_NT_RT
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2)
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX12-NEXT: scratch_store_b32 v0, v2, s32 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v0, s32 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v1, s32 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_setpc_b64 s[30:31]
bb:
Expand Down Expand Up @@ -391,17 +390,19 @@ define amdgpu_kernel void @store_load_sindex_small_offset_kernel(i32 %idx) {
; GFX12-LABEL: store_load_sindex_small_offset_kernel:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x24
; GFX12-NEXT: scratch_load_b32 v2, off, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX12-NEXT: s_and_b32 s1, s0, 15
; GFX12-NEXT: scratch_load_b32 v3, off, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: v_mov_b32_e32 v1, 15
; GFX12-NEXT: s_waitcnt lgkmcnt(0)
; GFX12-NEXT: s_lshl_b32 s1, s0, 2
; GFX12-NEXT: s_and_b32 s0, s0, 15
; GFX12-NEXT: v_mov_b32_e32 v0, s1
; GFX12-NEXT: s_lshl_b32 s0, s0, 2
; GFX12-NEXT: s_lshl_b32 s1, s1, 2
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX12-NEXT: v_dual_mov_b32 v0, 15 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: s_addk_co_i32 s0, 0x104
; GFX12-NEXT: scratch_store_b32 off, v0, s0 th:TH_STORE_NT_RT
; GFX12-NEXT: v_mov_b32_e32 v2, s0
; GFX12-NEXT: scratch_store_b32 v0, v1, off offset:260 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:260 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v2, off offset:260 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down Expand Up @@ -490,13 +491,13 @@ define amdgpu_kernel void @store_load_vindex_small_offset_kernel() {
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_sub_nc_u32_e32 v1, 0, v0
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: v_mov_b32_e32 v2, 15
; GFX12-NEXT: scratch_load_b32 v3, off, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v1
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX12-NEXT: scratch_store_b32 v0, v2, off offset:260 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: v_add_nc_u32_e32 v1, 0x104, v1
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:124 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:384 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down Expand Up @@ -589,16 +590,14 @@ define void @store_load_vindex_small_offset_foo(i32 %idx) {
; GFX12-LABEL: store_load_vindex_small_offset_foo:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v0
; GFX12-NEXT: v_and_b32_e32 v0, 15, v0
; GFX12-NEXT: s_add_co_i32 s0, s32, 0x100
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_and_b32 v1, 15, v0
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_load_b32 v3, off, s32 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: v_add_nc_u32_e32 v1, s0, v1
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_store_b32 v1, v2, off th:TH_STORE_NT_RT
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX12-NEXT: scratch_store_b32 v0, v2, s32 offset:256 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v0, s32 offset:256 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v1, s32 offset:256 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_setpc_b64 s[30:31]
bb:
Expand Down Expand Up @@ -697,17 +696,19 @@ define amdgpu_kernel void @store_load_sindex_large_offset_kernel(i32 %idx) {
; GFX12-LABEL: store_load_sindex_large_offset_kernel:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x24
; GFX12-NEXT: scratch_load_b32 v2, off, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX12-NEXT: s_and_b32 s1, s0, 15
; GFX12-NEXT: scratch_load_b32 v3, off, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: v_mov_b32_e32 v1, 15
; GFX12-NEXT: s_waitcnt lgkmcnt(0)
; GFX12-NEXT: s_lshl_b32 s1, s0, 2
; GFX12-NEXT: s_and_b32 s0, s0, 15
; GFX12-NEXT: v_mov_b32_e32 v0, s1
; GFX12-NEXT: s_lshl_b32 s0, s0, 2
; GFX12-NEXT: s_lshl_b32 s1, s1, 2
; GFX12-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX12-NEXT: v_dual_mov_b32 v0, 15 :: v_dual_mov_b32 v1, s1
; GFX12-NEXT: s_addk_co_i32 s0, 0x4004
; GFX12-NEXT: scratch_store_b32 off, v0, s0 th:TH_STORE_NT_RT
; GFX12-NEXT: v_mov_b32_e32 v2, s0
; GFX12-NEXT: scratch_store_b32 v0, v1, off offset:16388 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:16388 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v2, off offset:16388 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down Expand Up @@ -798,13 +799,13 @@ define amdgpu_kernel void @store_load_vindex_large_offset_kernel() {
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_sub_nc_u32_e32 v1, 0, v0
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: v_mov_b32_e32 v2, 15
; GFX12-NEXT: scratch_load_b32 v3, off, off offset:4 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v1
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX12-NEXT: scratch_store_b32 v0, v2, off offset:16388 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: v_add_nc_u32_e32 v1, 0x4004, v1
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:124 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v1, off offset:16512 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down Expand Up @@ -899,16 +900,14 @@ define void @store_load_vindex_large_offset_foo(i32 %idx) {
; GFX12-LABEL: store_load_vindex_large_offset_foo:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_lshlrev_b32 v1, 2, v0
; GFX12-NEXT: v_and_b32_e32 v0, 15, v0
; GFX12-NEXT: s_add_co_i32 s0, s32, 0x4000
; GFX12-NEXT: v_dual_mov_b32 v2, 15 :: v_dual_and_b32 v1, 15, v0
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_load_b32 v3, off, s32 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: v_add_nc_u32_e32 v1, s0, v1
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_store_b32 v1, v2, off th:TH_STORE_NT_RT
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX12-NEXT: scratch_store_b32 v0, v2, s32 offset:16384 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v0, s32 offset:16384 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v1, s32 offset:16384 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_setpc_b64 s[30:31]
bb:
Expand Down Expand Up @@ -1154,11 +1153,9 @@ define amdgpu_kernel void @store_load_vidx_sidx_offset(i32 %sidx) {
; GFX12-NEXT: v_mov_b32_e32 v1, 15
; GFX12-NEXT: s_waitcnt lgkmcnt(0)
; GFX12-NEXT: v_add_lshl_u32 v0, s0, v0, 2
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: v_add_nc_u32_e32 v0, 4, v0
; GFX12-NEXT: scratch_store_b32 v0, v1, off offset:1024 th:TH_STORE_NT_RT
; GFX12-NEXT: scratch_store_b32 v0, v1, off offset:1028 th:TH_STORE_NT_RT
; GFX12-NEXT: s_waitcnt_vscnt null, 0x0
; GFX12-NEXT: scratch_load_b32 v0, v0, off offset:1024 th:TH_LOAD_RT_NT
; GFX12-NEXT: scratch_load_b32 v0, v0, off offset:1028 th:TH_LOAD_RT_NT
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: s_endpgm
bb:
Expand Down
54 changes: 24 additions & 30 deletions llvm/test/CodeGen/AMDGPU/flat-scratch-i8-i16.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -894,8 +894,8 @@ define amdgpu_ps void @test_scratch_load_i8_zext_svs(ptr addrspace(5) inreg %in,
;
; GFX12-LABEL: test_scratch_load_i8_zext_svs:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: scratch_load_u8 v0, v0, off offset:1
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_load_u8 v0, v0, s0 offset:1
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v0
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -931,8 +931,8 @@ define amdgpu_ps void @test_scratch_load_i8_sext_svs(ptr addrspace(5) inreg %in,
;
; GFX12-LABEL: test_scratch_load_i8_sext_svs:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: scratch_load_i8 v0, v0, off offset:1
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_load_i8 v0, v0, s0 offset:1
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v0
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -968,8 +968,8 @@ define amdgpu_ps void @test_scratch_load_i16_zext_svs(ptr addrspace(5) inreg %in
;
; GFX12-LABEL: test_scratch_load_i16_zext_svs:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: scratch_load_u16 v0, v0, off offset:2
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_load_u16 v0, v0, s0 offset:2
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v0
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1005,8 +1005,8 @@ define amdgpu_ps void @test_scratch_load_i16_sext_svs(ptr addrspace(5) inreg %in
;
; GFX12-LABEL: test_scratch_load_i16_sext_svs:
; GFX12: ; %bb.0:
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: scratch_load_i16 v0, v0, off offset:2
; GFX12-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX12-NEXT: scratch_load_i16 v0, v0, s0 offset:2
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v0
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1046,9 +1046,8 @@ define amdgpu_ps void @test_scratch_load_i8_zext_to_d16_lo_svs(ptr addrspace(5)
;
; GFX12-LABEL: test_scratch_load_i8_zext_to_d16_lo_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: v_mov_b32_e32 v3, 0xffff0000
; GFX12-NEXT: scratch_load_d16_u8 v3, v0, off offset:1
; GFX12-NEXT: v_dual_mov_b32 v3, 0xffff0000 :: v_dual_lshlrev_b32 v0, 2, v0
; GFX12-NEXT: scratch_load_d16_u8 v3, v0, s0 offset:1
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v3
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1090,9 +1089,8 @@ define amdgpu_ps void @test_scratch_load_i8_sext_to_d16_lo_svs(ptr addrspace(5)
;
; GFX12-LABEL: test_scratch_load_i8_sext_to_d16_lo_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: v_mov_b32_e32 v3, 0xffff0000
; GFX12-NEXT: scratch_load_d16_i8 v3, v0, off offset:1
; GFX12-NEXT: v_dual_mov_b32 v3, 0xffff0000 :: v_dual_lshlrev_b32 v0, 2, v0
; GFX12-NEXT: scratch_load_d16_i8 v3, v0, s0 offset:1
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v3
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1134,9 +1132,8 @@ define amdgpu_ps void @test_scratch_load_i16_to_d16_lo_svs(ptr addrspace(5) inre
;
; GFX12-LABEL: test_scratch_load_i16_to_d16_lo_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: v_mov_b32_e32 v3, 0xffff0000
; GFX12-NEXT: scratch_load_d16_b16 v3, v0, off offset:2
; GFX12-NEXT: v_dual_mov_b32 v3, 0xffff0000 :: v_dual_lshlrev_b32 v0, 2, v0
; GFX12-NEXT: scratch_load_d16_b16 v3, v0, s0 offset:2
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v3
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1178,9 +1175,8 @@ define amdgpu_ps void @test_scratch_load_i8_zext_to_d16_hi_svs(ptr addrspace(5)
;
; GFX12-LABEL: test_scratch_load_i8_zext_to_d16_hi_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: v_mov_b32_e32 v3, -1
; GFX12-NEXT: scratch_load_d16_hi_u8 v3, v0, off offset:1
; GFX12-NEXT: v_dual_mov_b32 v3, -1 :: v_dual_lshlrev_b32 v0, 2, v0
; GFX12-NEXT: scratch_load_d16_hi_u8 v3, v0, s0 offset:1
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v3
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1222,9 +1218,8 @@ define amdgpu_ps void @test_scratch_load_i8_sext_to_d16_hi_svs(ptr addrspace(5)
;
; GFX12-LABEL: test_scratch_load_i8_sext_to_d16_hi_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: v_mov_b32_e32 v3, -1
; GFX12-NEXT: scratch_load_d16_hi_i8 v3, v0, off offset:1
; GFX12-NEXT: v_dual_mov_b32 v3, -1 :: v_dual_lshlrev_b32 v0, 2, v0
; GFX12-NEXT: scratch_load_d16_hi_i8 v3, v0, s0 offset:1
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v3
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1266,9 +1261,8 @@ define amdgpu_ps void @test_scratch_load_i16_to_d16_hi_svs(ptr addrspace(5) inre
;
; GFX12-LABEL: test_scratch_load_i16_to_d16_hi_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: v_lshl_add_u32 v0, v0, 2, s0
; GFX12-NEXT: v_mov_b32_e32 v3, -1
; GFX12-NEXT: scratch_load_d16_hi_b16 v3, v0, off offset:2
; GFX12-NEXT: v_dual_mov_b32 v3, -1 :: v_dual_lshlrev_b32 v0, 2, v0
; GFX12-NEXT: scratch_load_d16_hi_b16 v3, v0, s0 offset:2
; GFX12-NEXT: s_waitcnt vmcnt(0)
; GFX12-NEXT: flat_store_b32 v[1:2], v3
; GFX12-NEXT: s_endpgm
Expand Down Expand Up @@ -1309,9 +1303,9 @@ define amdgpu_ps void @test_scratch_store_b8_from_d16_hi_svs(ptr %in, ptr addrsp
; GFX12-LABEL: test_scratch_store_b8_from_d16_hi_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: flat_load_b32 v0, v[0:1]
; GFX12-NEXT: v_lshl_add_u32 v1, v2, 2, s0
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v2
; GFX12-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX12-NEXT: scratch_store_d16_hi_b8 v1, v0, off offset:4
; GFX12-NEXT: scratch_store_d16_hi_b8 v1, v0, s0 offset:4
; GFX12-NEXT: s_endpgm
bb:
%load = load <4 x i8>, ptr %in
Expand Down Expand Up @@ -1350,9 +1344,9 @@ define amdgpu_ps void @test_scratch_store_b16_from_d16_hi_svs(ptr %in, ptr addrs
; GFX12-LABEL: test_scratch_store_b16_from_d16_hi_svs:
; GFX12: ; %bb.0: ; %bb
; GFX12-NEXT: flat_load_b32 v0, v[0:1]
; GFX12-NEXT: v_lshl_add_u32 v1, v2, 2, s0
; GFX12-NEXT: v_lshlrev_b32_e32 v1, 2, v2
; GFX12-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX12-NEXT: scratch_store_d16_hi_b16 v1, v0, off offset:2
; GFX12-NEXT: scratch_store_d16_hi_b16 v1, v0, s0 offset:2
; GFX12-NEXT: s_endpgm
bb:
%load = load <2 x i16>, ptr %in
Expand Down
Loading