This is an archive of the discontinued LLVM Phabricator instance.

Differential D93692

[AMDGPU][GlobalISel] Fold flat vgpr + constant addresses
ClosedPublic

Authored by sebastian-ne on Dec 22 2020, 5:35 AM.

Details

Reviewers
arsenm
rampitec
Commits
rG221fdedc6926: [AMDGPU][GlobalISel] Fold flat vgpr + constant addresses
Summary

Use getPtrBaseWithConstantOffset in selectFlatOffsetImpl to fold more
vgpr+constant addresses.

Diff Detail

Event Timeline

sebastian-ne created this revision.Dec 22 2020, 5:35 AM
Herald added subscribers: kerbowa, jfb, hiraditya and 8 others. · View Herald TranscriptDec 22 2020, 5:35 AM
sebastian-ne requested review of this revision.Dec 22 2020, 5:35 AM
Herald added a project: Restricted Project. · View Herald TranscriptDec 22 2020, 5:35 AM
Herald added subscribers: llvm-commits, wdng. · View Herald Transcript
sebastian-ne added a parent revision: D93670: [AMDGPU][GlobalISel] GlobalISel for flat scratch.Dec 22 2020, 5:36 AM
arsenm accepted this revision.Dec 22 2020, 6:31 AM
This revision is now accepted and ready to land.Dec 22 2020, 6:31 AM
Harbormaster completed remote builds in B83250: Diff 313308.Dec 22 2020, 6:51 AM
rampitec accepted this revision.Dec 22 2020, 10:31 AM

LGTM

Closed by commit rG221fdedc6926: [AMDGPU][GlobalISel] Fold flat vgpr + constant addresses (authored by sebastian-ne). · Explain WhyDec 23 2020, 2:05 AM
This revision was automatically updated to reflect the committed changes.
sebastian-ne added a commit: rG221fdedc6926: [AMDGPU][GlobalISel] Fold flat vgpr + constant addresses.

Revision Contents

PathSize
llvm/
lib/
Target/
AMDGPU/
18 lines
test/
CodeGen/
AMDGPU/
GlobalISel/
1675 lines
16 lines
64 lines
260 lines
30 lines
89 lines

Diff 313308

llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp

Show First 20 Lines • Show All 3,419 Lines • ▼ Show 20 Lines
AMDGPUInstructionSelector::selectFlatOffsetImpl(MachineOperand &Root) const { AMDGPUInstructionSelector::selectFlatOffsetImpl(MachineOperand &Root) const {
MachineInstr *MI = Root.getParent(); MachineInstr *MI = Root.getParent();
auto Default = std::make_pair(Root.getReg(), 0); auto Default = std::make_pair(Root.getReg(), 0);
if (!STI.hasFlatInstOffsets()) if (!STI.hasFlatInstOffsets())
return Default; return Default;
const MachineInstr *OpDef = MRI->getVRegDef(Root.getReg()); Register PtrBase;
if (!OpDef || OpDef->getOpcode() != AMDGPU::G_PTR_ADD) int64_t ConstOffset;
return Default; std::tie(PtrBase, ConstOffset) =
getPtrBaseWithConstantOffset(Root.getReg(), *MRI);
Optional<int64_t> Offset = if (ConstOffset == 0)
getConstantVRegVal(OpDef->getOperand(2).getReg(), *MRI);
if (!Offset.hasValue())
return Default; return Default;
unsigned AddrSpace = (*MI->memoperands_begin())->getAddrSpace(); unsigned AddrSpace = (*MI->memoperands_begin())->getAddrSpace();
if (!TII.isLegalFLATOffset(Offset.getValue(), AddrSpace, Signed)) if (!TII.isLegalFLATOffset(ConstOffset, AddrSpace, Signed))
return Default; return Default;
Register BasePtr = OpDef->getOperand(1).getReg(); return std::make_pair(PtrBase, ConstOffset);
return std::make_pair(BasePtr, Offset.getValue());
} }
InstructionSelector::ComplexRendererFns InstructionSelector::ComplexRendererFns
AMDGPUInstructionSelector::selectFlatOffset(MachineOperand &Root) const { AMDGPUInstructionSelector::selectFlatOffset(MachineOperand &Root) const {
auto PtrWithOffset = selectFlatOffsetImpl<false>(Root); auto PtrWithOffset = selectFlatOffsetImpl<false>(Root);
return {{ return {{
[=](MachineInstrBuilder &MIB) { MIB.addReg(PtrWithOffset.first); }, [=](MachineInstrBuilder &MIB) { MIB.addReg(PtrWithOffset.first); },
▲ Show 20 Lines • Show All 929 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s ; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN %s
; Check lowering of some large extractelement that use the stack ; Check lowering of some large extractelement that use the stack
; instead of register indexing. ; instead of register indexing.
define i32 @v_extract_v64i32_varidx(<64 x i32> addrspace(1)* %ptr, i32 %idx) { define i32 @v_extract_v64i32_varidx(<64 x i32> addrspace(1)* %ptr, i32 %idx) {
; GCN-LABEL: v_extract_v64i32_varidx: ; GCN-LABEL: v_extract_v64i32_varidx:
; GCN: ; %bb.0: ; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v15, v0
; GCN-NEXT: s_add_u32 s4, s32, 0x3fc0 ; GCN-NEXT: s_add_u32 s4, s32, 0x3fc0
; GCN-NEXT: s_mov_b32 s5, 0 ; GCN-NEXT: v_add_co_u32_e32 v12, vcc, 64, v0
; GCN-NEXT: s_mov_b32 s6, s33 ; GCN-NEXT: s_mov_b32 s6, s33
; GCN-NEXT: s_and_b32 s33, s4, 0xffffc000 ; GCN-NEXT: s_and_b32 s33, s4, 0xffffc000
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 ; 4-byte Folded Spill
; GCN-NEXT: v_addc_co_u32_e32 v13, vcc, 0, v1, vcc
; GCN-NEXT: global_load_dwordx4 v[4:7], v[12:13], off offset:16
; GCN-NEXT: global_load_dwordx4 v[8:11], v[12:13], off offset:32
; GCN-NEXT: global_load_dwordx4 v[12:15], v[12:13], off offset:48
; GCN-NEXT: s_mov_b32 s5, 0
; GCN-NEXT: s_movk_i32 s4, 0x80 ; GCN-NEXT: s_movk_i32 s4, 0x80
; GCN-NEXT: v_mov_b32_e32 v12, s5 ; GCN-NEXT: v_mov_b32_e32 v17, s5
; GCN-NEXT: v_mov_b32_e32 v16, v1 ; GCN-NEXT: v_mov_b32_e32 v16, s4
; GCN-NEXT: v_add_co_u32_e32 v31, vcc, 64, v15
; GCN-NEXT: v_mov_b32_e32 v11, s4
; GCN-NEXT: v_addc_co_u32_e32 v32, vcc, 0, v16, vcc
; GCN-NEXT: v_add_co_u32_e32 v48, vcc, v15, v11
; GCN-NEXT: v_addc_co_u32_e32 v49, vcc, v16, v12, vcc
; GCN-NEXT: s_movk_i32 s4, 0xc0 ; GCN-NEXT: s_movk_i32 s4, 0xc0
; GCN-NEXT: v_mov_b32_e32 v12, s5
; GCN-NEXT: v_mov_b32_e32 v11, s4
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 ; 4-byte Folded Spill
; GCN-NEXT: v_add_co_u32_e32 v59, vcc, v15, v11
; GCN-NEXT: global_load_dwordx4 v[3:6], v[15:16], off
; GCN-NEXT: global_load_dwordx4 v[7:10], v[15:16], off offset:16
; GCN-NEXT: v_addc_co_u32_e32 v60, vcc, v16, v12, vcc
; GCN-NEXT: global_load_dwordx4 v[11:14], v[15:16], off offset:32
; GCN-NEXT: global_load_dwordx4 v[15:18], v[15:16], off offset:48
; GCN-NEXT: global_load_dwordx4 v[19:22], v[31:32], off
; GCN-NEXT: global_load_dwordx4 v[23:26], v[31:32], off offset:16
; GCN-NEXT: global_load_dwordx4 v[27:30], v[31:32], off offset:32
; GCN-NEXT: global_load_dwordx4 v[31:34], v[31:32], off offset:48
; GCN-NEXT: global_load_dwordx4 v[35:38], v[48:49], off
; GCN-NEXT: global_load_dwordx4 v[39:42], v[48:49], off offset:16
; GCN-NEXT: global_load_dwordx4 v[43:46], v[48:49], off offset:32
; GCN-NEXT: v_lshrrev_b32_e64 v0, 6, s33
; GCN-NEXT: v_add_u32_e32 v0, 0x100, v0
; GCN-NEXT: v_add_u32_e32 v1, 16, v0
; GCN-NEXT: s_add_u32 s32, s32, 0x10000 ; GCN-NEXT: s_add_u32 s32, s32, 0x10000
; GCN-NEXT: s_sub_u32 s32, s32, 0x10000 ; GCN-NEXT: s_sub_u32 s32, s32, 0x10000
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v36, off, s[0:3], s33 offset:580 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v37, off, s[0:3], s33 offset:584 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v38, off, s[0:3], s33 offset:588 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v39, off, s[0:3], s33 offset:592 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:596 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:600 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:604 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:608 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:612 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:616 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:620 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:624 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:628 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:632 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:636 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[47:50], v[48:49], off offset:48
; GCN-NEXT: global_load_dwordx4 v[43:46], v[59:60], off
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v51, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v52, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v53, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v54, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v55, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[51:54], v[59:60], off offset:16
; GCN-NEXT: global_load_dwordx4 v[55:58], v[59:60], off offset:32
; GCN-NEXT: global_load_dwordx4 v[59:62], v[59:60], off offset:48
; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 20, v0
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 24, v0
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 28, v0
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 32, v0
; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 36, v0
; GCN-NEXT: buffer_store_dword v12, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 40, v0
; GCN-NEXT: buffer_store_dword v13, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 44, v0
; GCN-NEXT: buffer_store_dword v14, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 48, v0
; GCN-NEXT: buffer_store_dword v15, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 52, v0
; GCN-NEXT: buffer_store_dword v16, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 56, v0
; GCN-NEXT: buffer_store_dword v17, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 60, v0
; GCN-NEXT: buffer_store_dword v18, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 64, v0
; GCN-NEXT: buffer_store_dword v19, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x44, v0
; GCN-NEXT: buffer_store_dword v20, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x48, v0
; GCN-NEXT: buffer_store_dword v21, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x4c, v0
; GCN-NEXT: buffer_store_dword v22, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x50, v0
; GCN-NEXT: buffer_store_dword v23, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x54, v0
; GCN-NEXT: buffer_store_dword v24, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x58, v0
; GCN-NEXT: buffer_store_dword v25, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x5c, v0
; GCN-NEXT: buffer_store_dword v26, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x60, v0
; GCN-NEXT: buffer_store_dword v27, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x64, v0
; GCN-NEXT: buffer_store_dword v28, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x68, v0
; GCN-NEXT: buffer_store_dword v29, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x6c, v0
; GCN-NEXT: buffer_store_dword v30, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x70, v0
; GCN-NEXT: buffer_store_dword v31, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x74, v0
; GCN-NEXT: buffer_store_dword v32, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x78, v0
; GCN-NEXT: buffer_store_dword v33, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x7c, v0
; GCN-NEXT: buffer_store_dword v34, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x80, v0
; GCN-NEXT: buffer_store_dword v35, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x84, v0
; GCN-NEXT: buffer_store_dword v36, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x88, v0
; GCN-NEXT: buffer_store_dword v37, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x8c, v0
; GCN-NEXT: buffer_store_dword v38, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x90, v0
; GCN-NEXT: buffer_store_dword v39, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x94, v0
; GCN-NEXT: buffer_store_dword v40, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x98, v0
; GCN-NEXT: buffer_store_dword v41, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x9c, v0
; GCN-NEXT: buffer_store_dword v42, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:576 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:580 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:584 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:588 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:592 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:596 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:600 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:604 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:608 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:612 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:616 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:620 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:624 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:628 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:632 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:636 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xa0, v0
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v8, v15
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v9, v16
; GCN-NEXT: v_add_u32_e32 v1, 0xa4, v0
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v10, v17
; GCN-NEXT: v_add_u32_e32 v1, 0xa8, v0
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v11, v18
; GCN-NEXT: v_add_u32_e32 v1, 0xac, v0
; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb0, v0
; GCN-NEXT: buffer_store_dword v47, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb4, v0
; GCN-NEXT: buffer_store_dword v48, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb8, v0
; GCN-NEXT: buffer_store_dword v49, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xbc, v0
; GCN-NEXT: buffer_store_dword v50, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xc0, v0
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:580 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0xc4, v0 ; GCN-NEXT: buffer_store_dword v2, off, s[0:3], s33 offset:584 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:588 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0xc8, v0 ; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:592 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:596 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0xcc, v0 ; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:600 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:604 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 4, v0 ; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:608 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v4, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:612 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 8, v0 ; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:616 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v5, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:620 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 12, v0 ; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:624 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v6, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:628 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:256 ; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:632 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0xd0, v0 ; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:636 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v51, v1, s[0:3], 0 offen ; GCN-NEXT: v_lshrrev_b32_e64 v3, 6, s33
; GCN-NEXT: v_add_u32_e32 v1, 0xd4, v0 ; GCN-NEXT: v_add_u32_e32 v3, 0x100, v3
; GCN-NEXT: buffer_store_dword v52, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v60, 16, v3
; GCN-NEXT: v_add_u32_e32 v1, 0xd8, v0 ; GCN-NEXT: v_add_co_u32_e32 v52, vcc, v0, v16
; GCN-NEXT: buffer_store_dword v53, v1, s[0:3], 0 offen ; GCN-NEXT: v_addc_co_u32_e32 v53, vcc, v1, v17, vcc
; GCN-NEXT: v_add_u32_e32 v1, 0xdc, v0 ; GCN-NEXT: v_mov_b32_e32 v17, s5
; GCN-NEXT: buffer_store_dword v54, v1, s[0:3], 0 offen ; GCN-NEXT: v_mov_b32_e32 v16, s4
; GCN-NEXT: v_add_u32_e32 v1, 0xe0, v0 ; GCN-NEXT: v_add_co_u32_e32 v56, vcc, v0, v16
; GCN-NEXT: buffer_store_dword v55, v1, s[0:3], 0 offen ; GCN-NEXT: v_addc_co_u32_e32 v57, vcc, v1, v17, vcc
; GCN-NEXT: v_add_u32_e32 v1, 0xe4, v0 ; GCN-NEXT: global_load_dwordx4 v[16:19], v[0:1], off
; GCN-NEXT: buffer_store_dword v56, v1, s[0:3], 0 offen ; GCN-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:16
; GCN-NEXT: v_add_u32_e32 v1, 0xe8, v0 ; GCN-NEXT: global_load_dwordx4 v[24:27], v[0:1], off offset:32
; GCN-NEXT: buffer_store_dword v57, v1, s[0:3], 0 offen ; GCN-NEXT: global_load_dwordx4 v[28:31], v[0:1], off offset:48
; GCN-NEXT: v_add_u32_e32 v1, 0xec, v0 ; GCN-NEXT: global_load_dwordx4 v[32:35], v[0:1], off offset:64
; GCN-NEXT: buffer_store_dword v58, v1, s[0:3], 0 offen ; GCN-NEXT: global_load_dwordx4 v[36:39], v[0:1], off offset:128
; GCN-NEXT: v_add_u32_e32 v1, 0xf0, v0 ; GCN-NEXT: global_load_dwordx4 v[40:43], v[0:1], off offset:192
; GCN-NEXT: buffer_store_dword v59, v1, s[0:3], 0 offen ; GCN-NEXT: global_load_dwordx4 v[44:47], v[52:53], off offset:16
; GCN-NEXT: v_add_u32_e32 v1, 0xf4, v0 ; GCN-NEXT: global_load_dwordx4 v[48:51], v[52:53], off offset:32
; GCN-NEXT: buffer_store_dword v60, v1, s[0:3], 0 offen ; GCN-NEXT: global_load_dwordx4 v[52:55], v[52:53], off offset:48
; GCN-NEXT: v_add_u32_e32 v1, 0xf8, v0 ; GCN-NEXT: global_load_dwordx4 v[12:15], v[56:57], off offset:16
; GCN-NEXT: buffer_store_dword v61, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 20, v3
; GCN-NEXT: v_add_u32_e32 v1, 0xfc, v0 ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v62, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:640 ; 4-byte Folded Spill
; GCN-NEXT: v_and_b32_e32 v1, 63, v2 ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_lshlrev_b32_e32 v1, 2, v1 ; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:644 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v0, v0, v1 ; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:648 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:652 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:656 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:660 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:664 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:668 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:672 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:676 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:680 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:684 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:688 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:692 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:696 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v23, off, s[0:3], s33 offset:700 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[12:15], v[56:57], off offset:32
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[56:59], v[56:57], off offset:48
; GCN-NEXT: buffer_store_dword v21, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 24, v3
; GCN-NEXT: buffer_store_dword v22, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 28, v3
; GCN-NEXT: buffer_store_dword v23, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 32, v3
; GCN-NEXT: buffer_store_dword v24, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 36, v3
; GCN-NEXT: buffer_store_dword v25, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 40, v3
; GCN-NEXT: buffer_store_dword v26, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 44, v3
; GCN-NEXT: buffer_store_dword v27, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 48, v3
; GCN-NEXT: buffer_store_dword v28, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 52, v3
; GCN-NEXT: buffer_store_dword v29, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 56, v3
; GCN-NEXT: buffer_store_dword v30, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 60, v3
; GCN-NEXT: buffer_store_dword v31, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 64, v3
; GCN-NEXT: buffer_store_dword v32, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x44, v3
; GCN-NEXT: buffer_store_dword v33, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x48, v3
; GCN-NEXT: buffer_store_dword v34, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x4c, v3
; GCN-NEXT: buffer_store_dword v35, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x50, v3
; GCN-NEXT: buffer_store_dword v20, v60, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v4, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x54, v3
; GCN-NEXT: buffer_store_dword v5, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x58, v3
; GCN-NEXT: buffer_store_dword v6, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x5c, v3
; GCN-NEXT: buffer_store_dword v7, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x60, v3
; GCN-NEXT: buffer_store_dword v8, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x64, v3
; GCN-NEXT: buffer_store_dword v9, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x68, v3
; GCN-NEXT: buffer_store_dword v10, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x6c, v3
; GCN-NEXT: buffer_store_dword v11, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:576 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:580 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:584 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v23, off, s[0:3], s33 offset:588 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v24, off, s[0:3], s33 offset:592 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v25, off, s[0:3], s33 offset:596 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v26, off, s[0:3], s33 offset:600 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v27, off, s[0:3], s33 offset:604 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v28, off, s[0:3], s33 offset:608 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v29, off, s[0:3], s33 offset:612 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v30, off, s[0:3], s33 offset:616 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v31, off, s[0:3], s33 offset:620 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v32, off, s[0:3], s33 offset:624 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:628 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v34, off, s[0:3], s33 offset:632 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v35, off, s[0:3], s33 offset:636 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0x70, v3
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v12, v32
; GCN-NEXT: buffer_store_dword v12, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v13, v33
; GCN-NEXT: v_add_u32_e32 v0, 0x74, v3
; GCN-NEXT: buffer_store_dword v13, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v14, v34
; GCN-NEXT: v_add_u32_e32 v0, 0x78, v3
; GCN-NEXT: buffer_store_dword v14, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v15, v35
; GCN-NEXT: v_add_u32_e32 v0, 0x7c, v3
; GCN-NEXT: buffer_store_dword v15, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x80, v3
; GCN-NEXT: buffer_store_dword v36, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x84, v3
; GCN-NEXT: buffer_store_dword v37, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x88, v3
; GCN-NEXT: buffer_store_dword v38, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x8c, v3
; GCN-NEXT: buffer_store_dword v39, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x90, v3
; GCN-NEXT: buffer_store_dword v44, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x94, v3
; GCN-NEXT: buffer_store_dword v45, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x98, v3
; GCN-NEXT: buffer_store_dword v46, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x9c, v3
; GCN-NEXT: buffer_store_dword v47, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xa0, v3
; GCN-NEXT: buffer_store_dword v48, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xa4, v3
; GCN-NEXT: buffer_store_dword v49, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xa8, v3
; GCN-NEXT: buffer_store_dword v50, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xac, v3
; GCN-NEXT: buffer_store_dword v51, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xb0, v3
; GCN-NEXT: buffer_store_dword v52, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xb4, v3
; GCN-NEXT: buffer_store_dword v53, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xb8, v3
; GCN-NEXT: buffer_store_dword v54, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xbc, v3
; GCN-NEXT: buffer_store_dword v55, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xc0, v3
; GCN-NEXT: buffer_store_dword v40, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xc4, v3
; GCN-NEXT: buffer_store_dword v41, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xc8, v3
; GCN-NEXT: buffer_store_dword v42, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xcc, v3
; GCN-NEXT: buffer_store_dword v43, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 4, v3
; GCN-NEXT: buffer_store_dword v17, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 8, v3
; GCN-NEXT: buffer_store_dword v18, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 12, v3
; GCN-NEXT: buffer_store_dword v19, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:256
; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:640 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:644 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:648 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:652 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:656 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:660 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:664 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:668 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:672 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:676 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:680 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:684 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:688 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:692 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:696 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:700 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0xd0, v3
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v4, v8
; GCN-NEXT: buffer_store_dword v4, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v5, v9
; GCN-NEXT: v_add_u32_e32 v0, 0xd4, v3
; GCN-NEXT: buffer_store_dword v5, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v6, v10
; GCN-NEXT: v_add_u32_e32 v0, 0xd8, v3
; GCN-NEXT: buffer_store_dword v6, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v7, v11
; GCN-NEXT: v_add_u32_e32 v0, 0xdc, v3
; GCN-NEXT: buffer_store_dword v7, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0xe0, v3
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v8, v12
; GCN-NEXT: buffer_store_dword v8, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v9, v13
; GCN-NEXT: v_add_u32_e32 v0, 0xe4, v3
; GCN-NEXT: buffer_store_dword v9, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v10, v14
; GCN-NEXT: v_add_u32_e32 v0, 0xe8, v3
; GCN-NEXT: buffer_store_dword v10, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v11, v15
; GCN-NEXT: v_add_u32_e32 v0, 0xec, v3
; GCN-NEXT: buffer_store_dword v11, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xf0, v3
; GCN-NEXT: buffer_store_dword v56, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xf4, v3
; GCN-NEXT: buffer_store_dword v57, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xf8, v3
; GCN-NEXT: buffer_store_dword v58, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xfc, v3
; GCN-NEXT: buffer_store_dword v59, v0, s[0:3], 0 offen
; GCN-NEXT: v_and_b32_e32 v0, 63, v2
; GCN-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GCN-NEXT: v_add_u32_e32 v0, v3, v0
; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Reload
; GCN-NEXT: s_mov_b32 s33, s6 ; GCN-NEXT: s_mov_b32 s33, s6
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_setpc_b64 s[30:31] ; GCN-NEXT: s_setpc_b64 s[30:31]
%vec = load <64 x i32>, <64 x i32> addrspace(1)* %ptr %vec = load <64 x i32>, <64 x i32> addrspace(1)* %ptr
%elt = extractelement <64 x i32> %vec, i32 %idx %elt = extractelement <64 x i32> %vec, i32 %idx
ret i32 %elt ret i32 %elt
} }
define i16 @v_extract_v128i16_varidx(<128 x i16> addrspace(1)* %ptr, i32 %idx) { define i16 @v_extract_v128i16_varidx(<128 x i16> addrspace(1)* %ptr, i32 %idx) {
; GCN-LABEL: v_extract_v128i16_varidx: ; GCN-LABEL: v_extract_v128i16_varidx:
; GCN: ; %bb.0: ; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v15, v0
; GCN-NEXT: s_add_u32 s4, s32, 0x3fc0 ; GCN-NEXT: s_add_u32 s4, s32, 0x3fc0
; GCN-NEXT: s_mov_b32 s5, 0 ; GCN-NEXT: v_add_co_u32_e32 v12, vcc, 64, v0
; GCN-NEXT: s_mov_b32 s6, s33 ; GCN-NEXT: s_mov_b32 s6, s33
; GCN-NEXT: s_and_b32 s33, s4, 0xffffc000 ; GCN-NEXT: s_and_b32 s33, s4, 0xffffc000
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 ; 4-byte Folded Spill
; GCN-NEXT: v_addc_co_u32_e32 v13, vcc, 0, v1, vcc
; GCN-NEXT: global_load_dwordx4 v[4:7], v[12:13], off offset:16
; GCN-NEXT: global_load_dwordx4 v[8:11], v[12:13], off offset:32
; GCN-NEXT: global_load_dwordx4 v[12:15], v[12:13], off offset:48
; GCN-NEXT: s_mov_b32 s5, 0
; GCN-NEXT: s_movk_i32 s4, 0x80 ; GCN-NEXT: s_movk_i32 s4, 0x80
; GCN-NEXT: v_mov_b32_e32 v12, s5 ; GCN-NEXT: v_mov_b32_e32 v17, s5
; GCN-NEXT: v_mov_b32_e32 v16, v1 ; GCN-NEXT: v_mov_b32_e32 v16, s4
; GCN-NEXT: v_add_co_u32_e32 v31, vcc, 64, v15
; GCN-NEXT: v_mov_b32_e32 v11, s4
; GCN-NEXT: v_addc_co_u32_e32 v32, vcc, 0, v16, vcc
; GCN-NEXT: v_add_co_u32_e32 v48, vcc, v15, v11
; GCN-NEXT: v_addc_co_u32_e32 v49, vcc, v16, v12, vcc
; GCN-NEXT: s_movk_i32 s4, 0xc0 ; GCN-NEXT: s_movk_i32 s4, 0xc0
; GCN-NEXT: v_mov_b32_e32 v12, s5
; GCN-NEXT: v_mov_b32_e32 v11, s4
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 ; 4-byte Folded Spill
; GCN-NEXT: v_add_co_u32_e32 v59, vcc, v15, v11
; GCN-NEXT: global_load_dwordx4 v[3:6], v[15:16], off
; GCN-NEXT: global_load_dwordx4 v[7:10], v[15:16], off offset:16
; GCN-NEXT: v_addc_co_u32_e32 v60, vcc, v16, v12, vcc
; GCN-NEXT: global_load_dwordx4 v[11:14], v[15:16], off offset:32
; GCN-NEXT: global_load_dwordx4 v[15:18], v[15:16], off offset:48
; GCN-NEXT: global_load_dwordx4 v[19:22], v[31:32], off
; GCN-NEXT: global_load_dwordx4 v[23:26], v[31:32], off offset:16
; GCN-NEXT: global_load_dwordx4 v[27:30], v[31:32], off offset:32
; GCN-NEXT: global_load_dwordx4 v[31:34], v[31:32], off offset:48
; GCN-NEXT: global_load_dwordx4 v[35:38], v[48:49], off
; GCN-NEXT: global_load_dwordx4 v[39:42], v[48:49], off offset:16
; GCN-NEXT: global_load_dwordx4 v[43:46], v[48:49], off offset:32
; GCN-NEXT: v_lshrrev_b32_e64 v0, 6, s33
; GCN-NEXT: v_add_u32_e32 v0, 0x100, v0
; GCN-NEXT: v_add_u32_e32 v1, 16, v0
; GCN-NEXT: s_add_u32 s32, s32, 0x10000 ; GCN-NEXT: s_add_u32 s32, s32, 0x10000
; GCN-NEXT: s_sub_u32 s32, s32, 0x10000 ; GCN-NEXT: s_sub_u32 s32, s32, 0x10000
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:580 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v2, off, s[0:3], s33 offset:584 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:588 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:592 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:596 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:600 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:604 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:608 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:612 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:616 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:620 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:624 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:628 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:632 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:636 ; 4-byte Folded Spill
; GCN-NEXT: v_lshrrev_b32_e64 v3, 6, s33
; GCN-NEXT: v_add_u32_e32 v3, 0x100, v3
; GCN-NEXT: v_add_u32_e32 v60, 16, v3
; GCN-NEXT: v_add_co_u32_e32 v52, vcc, v0, v16
; GCN-NEXT: v_addc_co_u32_e32 v53, vcc, v1, v17, vcc
; GCN-NEXT: v_mov_b32_e32 v17, s5
; GCN-NEXT: v_mov_b32_e32 v16, s4
; GCN-NEXT: v_add_co_u32_e32 v56, vcc, v0, v16
; GCN-NEXT: v_addc_co_u32_e32 v57, vcc, v1, v17, vcc
; GCN-NEXT: global_load_dwordx4 v[16:19], v[0:1], off
; GCN-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:16
; GCN-NEXT: global_load_dwordx4 v[24:27], v[0:1], off offset:32
; GCN-NEXT: global_load_dwordx4 v[28:31], v[0:1], off offset:48
; GCN-NEXT: global_load_dwordx4 v[32:35], v[0:1], off offset:64
; GCN-NEXT: global_load_dwordx4 v[36:39], v[0:1], off offset:128
; GCN-NEXT: global_load_dwordx4 v[40:43], v[0:1], off offset:192
; GCN-NEXT: global_load_dwordx4 v[44:47], v[52:53], off offset:16
; GCN-NEXT: global_load_dwordx4 v[48:51], v[52:53], off offset:32
; GCN-NEXT: global_load_dwordx4 v[52:55], v[52:53], off offset:48
; GCN-NEXT: global_load_dwordx4 v[12:15], v[56:57], off offset:16
; GCN-NEXT: v_add_u32_e32 v0, 20, v3
; GCN-NEXT: v_add_u32_e32 v1, 0xd0, v3
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:640 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:644 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:648 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:652 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:656 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:660 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:664 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:668 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:672 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:676 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:680 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:684 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:688 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:692 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:696 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v23, off, s[0:3], s33 offset:700 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[12:15], v[56:57], off offset:32
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[56:59], v[56:57], off offset:48
; GCN-NEXT: buffer_store_dword v21, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 24, v3
; GCN-NEXT: buffer_store_dword v22, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 28, v3
; GCN-NEXT: buffer_store_dword v23, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 32, v3
; GCN-NEXT: buffer_store_dword v24, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 36, v3
; GCN-NEXT: buffer_store_dword v25, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 40, v3
; GCN-NEXT: buffer_store_dword v26, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 44, v3
; GCN-NEXT: buffer_store_dword v27, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 48, v3
; GCN-NEXT: buffer_store_dword v28, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 52, v3
; GCN-NEXT: buffer_store_dword v29, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 56, v3
; GCN-NEXT: buffer_store_dword v30, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 60, v3
; GCN-NEXT: buffer_store_dword v31, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 64, v3
; GCN-NEXT: buffer_store_dword v32, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x44, v3
; GCN-NEXT: buffer_store_dword v33, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x48, v3
; GCN-NEXT: buffer_store_dword v34, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x4c, v3
; GCN-NEXT: buffer_store_dword v35, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x50, v3
; GCN-NEXT: buffer_store_dword v20, v60, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v4, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x54, v3
; GCN-NEXT: buffer_store_dword v5, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x58, v3
; GCN-NEXT: buffer_store_dword v6, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x5c, v3
; GCN-NEXT: buffer_store_dword v7, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x60, v3
; GCN-NEXT: buffer_store_dword v8, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x64, v3
; GCN-NEXT: buffer_store_dword v9, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x68, v3
; GCN-NEXT: buffer_store_dword v10, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x6c, v3
; GCN-NEXT: buffer_store_dword v11, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:576 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:580 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:584 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v23, off, s[0:3], s33 offset:588 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v24, off, s[0:3], s33 offset:592 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v25, off, s[0:3], s33 offset:596 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v26, off, s[0:3], s33 offset:600 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v27, off, s[0:3], s33 offset:604 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v28, off, s[0:3], s33 offset:608 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v29, off, s[0:3], s33 offset:612 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v30, off, s[0:3], s33 offset:616 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v31, off, s[0:3], s33 offset:620 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v32, off, s[0:3], s33 offset:624 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v33, off, s[0:3], s33 offset:628 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v34, off, s[0:3], s33 offset:632 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v35, off, s[0:3], s33 offset:636 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0x70, v3
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v12, v32
; GCN-NEXT: buffer_store_dword v12, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v13, v33
; GCN-NEXT: v_add_u32_e32 v0, 0x74, v3
; GCN-NEXT: buffer_store_dword v13, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v14, v34
; GCN-NEXT: v_add_u32_e32 v0, 0x78, v3
; GCN-NEXT: buffer_store_dword v14, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v15, v35
; GCN-NEXT: v_add_u32_e32 v0, 0x7c, v3
; GCN-NEXT: buffer_store_dword v15, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x80, v3
; GCN-NEXT: buffer_store_dword v36, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x84, v3
; GCN-NEXT: buffer_store_dword v37, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x88, v3
; GCN-NEXT: buffer_store_dword v38, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x8c, v3
; GCN-NEXT: buffer_store_dword v39, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x90, v3
; GCN-NEXT: buffer_store_dword v44, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x94, v3
; GCN-NEXT: buffer_store_dword v45, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x98, v3
; GCN-NEXT: buffer_store_dword v46, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0x9c, v3
; GCN-NEXT: buffer_store_dword v47, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xa0, v3
; GCN-NEXT: buffer_store_dword v48, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xa4, v3
; GCN-NEXT: buffer_store_dword v49, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xa8, v3
; GCN-NEXT: buffer_store_dword v50, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xac, v3
; GCN-NEXT: buffer_store_dword v51, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xb0, v3
; GCN-NEXT: buffer_store_dword v52, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xb4, v3
; GCN-NEXT: buffer_store_dword v53, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xb8, v3
; GCN-NEXT: buffer_store_dword v54, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xbc, v3
; GCN-NEXT: buffer_store_dword v55, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xc0, v3
; GCN-NEXT: buffer_store_dword v40, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xc4, v3
; GCN-NEXT: buffer_store_dword v41, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xc8, v3
; GCN-NEXT: buffer_store_dword v42, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xcc, v3
; GCN-NEXT: buffer_store_dword v43, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 4, v3
; GCN-NEXT: buffer_store_dword v17, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 8, v3
; GCN-NEXT: buffer_store_dword v18, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 12, v3
; GCN-NEXT: buffer_store_dword v19, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:256
; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:640 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:644 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:648 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:652 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:656 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:660 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:664 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:668 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:672 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:676 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:680 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:684 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:688 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:692 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:696 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:700 ; 4-byte Folded Reload
; GCN-NEXT: v_lshrrev_b32_e32 v0, 1, v2
; GCN-NEXT: v_and_b32_e32 v0, 63, v0
; GCN-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GCN-NEXT: v_add_u32_e32 v0, v3, v0
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v36, off, s[0:3], s33 offset:580 ; 4-byte Folded Spill ; GCN-NEXT: v_mov_b32_e32 v4, v8
; GCN-NEXT: buffer_store_dword v37, off, s[0:3], s33 offset:584 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v4, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v38, off, s[0:3], s33 offset:588 ; 4-byte Folded Spill ; GCN-NEXT: v_mov_b32_e32 v5, v9
; GCN-NEXT: buffer_store_dword v39, off, s[0:3], s33 offset:592 ; 4-byte Folded Spill ; GCN-NEXT: v_add_u32_e32 v1, 0xd4, v3
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:596 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v5, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:600 ; 4-byte Folded Spill ; GCN-NEXT: v_mov_b32_e32 v6, v10
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:604 ; 4-byte Folded Spill ; GCN-NEXT: v_add_u32_e32 v1, 0xd8, v3
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:608 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v6, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:612 ; 4-byte Folded Spill ; GCN-NEXT: v_mov_b32_e32 v7, v11
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:616 ; 4-byte Folded Spill ; GCN-NEXT: v_add_u32_e32 v1, 0xdc, v3
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:620 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:624 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:628 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:632 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:636 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[47:50], v[48:49], off offset:48
; GCN-NEXT: global_load_dwordx4 v[43:46], v[59:60], off
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v51, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v52, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v53, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v54, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v55, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[51:54], v[59:60], off offset:16
; GCN-NEXT: global_load_dwordx4 v[55:58], v[59:60], off offset:32
; GCN-NEXT: global_load_dwordx4 v[59:62], v[59:60], off offset:48
; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 20, v0 ; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 24, v0 ; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 28, v0 ; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 32, v0 ; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 36, v0 ; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v12, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 40, v0 ; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v13, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 44, v0 ; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v14, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 48, v0 ; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v15, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 52, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xe0, v3
; GCN-NEXT: buffer_store_dword v16, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 56, v0
; GCN-NEXT: buffer_store_dword v17, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 60, v0
; GCN-NEXT: buffer_store_dword v18, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 64, v0
; GCN-NEXT: buffer_store_dword v19, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x44, v0
; GCN-NEXT: buffer_store_dword v20, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x48, v0
; GCN-NEXT: buffer_store_dword v21, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x4c, v0
; GCN-NEXT: buffer_store_dword v22, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x50, v0
; GCN-NEXT: buffer_store_dword v23, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x54, v0
; GCN-NEXT: buffer_store_dword v24, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x58, v0
; GCN-NEXT: buffer_store_dword v25, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x5c, v0
; GCN-NEXT: buffer_store_dword v26, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x60, v0
; GCN-NEXT: buffer_store_dword v27, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x64, v0
; GCN-NEXT: buffer_store_dword v28, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x68, v0
; GCN-NEXT: buffer_store_dword v29, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x6c, v0
; GCN-NEXT: buffer_store_dword v30, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x70, v0
; GCN-NEXT: buffer_store_dword v31, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x74, v0
; GCN-NEXT: buffer_store_dword v32, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x78, v0
; GCN-NEXT: buffer_store_dword v33, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x7c, v0
; GCN-NEXT: buffer_store_dword v34, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x80, v0
; GCN-NEXT: buffer_store_dword v35, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x84, v0
; GCN-NEXT: buffer_store_dword v36, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x88, v0
; GCN-NEXT: buffer_store_dword v37, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x8c, v0
; GCN-NEXT: buffer_store_dword v38, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x90, v0
; GCN-NEXT: buffer_store_dword v39, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x94, v0
; GCN-NEXT: buffer_store_dword v40, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x98, v0
; GCN-NEXT: buffer_store_dword v41, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x9c, v0
; GCN-NEXT: buffer_store_dword v42, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:576 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:580 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:584 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:588 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:592 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:596 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:600 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:604 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:608 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:612 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:616 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:620 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:624 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:628 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:632 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:636 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xa0, v0
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v8, v15 ; GCN-NEXT: v_mov_b32_e32 v8, v12
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v9, v16 ; GCN-NEXT: v_mov_b32_e32 v9, v13
; GCN-NEXT: v_add_u32_e32 v1, 0xa4, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xe4, v3
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v10, v17 ; GCN-NEXT: v_mov_b32_e32 v10, v14
; GCN-NEXT: v_add_u32_e32 v1, 0xa8, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xe8, v3
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v11, v18 ; GCN-NEXT: v_mov_b32_e32 v11, v15
; GCN-NEXT: v_add_u32_e32 v1, 0xac, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xec, v3
; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb0, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xf0, v3
; GCN-NEXT: buffer_store_dword v47, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v56, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb4, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xf4, v3
; GCN-NEXT: buffer_store_dword v48, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v57, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb8, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xf8, v3
; GCN-NEXT: buffer_store_dword v49, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v58, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xbc, v0 ; GCN-NEXT: v_add_u32_e32 v1, 0xfc, v3
; GCN-NEXT: buffer_store_dword v50, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v59, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xc0, v0
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xc4, v0
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xc8, v0
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xcc, v0
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 4, v0
; GCN-NEXT: buffer_store_dword v4, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 8, v0
; GCN-NEXT: buffer_store_dword v5, v1, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 12, v0
; GCN-NEXT: buffer_store_dword v6, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:256
; GCN-NEXT: v_add_u32_e32 v3, 0xd0, v0
; GCN-NEXT: buffer_store_dword v51, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xd4, v0
; GCN-NEXT: buffer_store_dword v52, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xd8, v0
; GCN-NEXT: buffer_store_dword v53, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xdc, v0
; GCN-NEXT: buffer_store_dword v54, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xe0, v0
; GCN-NEXT: buffer_store_dword v55, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xe4, v0
; GCN-NEXT: buffer_store_dword v56, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xe8, v0
; GCN-NEXT: buffer_store_dword v57, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xec, v0
; GCN-NEXT: buffer_store_dword v58, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xf0, v0
; GCN-NEXT: v_lshrrev_b32_e32 v1, 1, v2
; GCN-NEXT: buffer_store_dword v59, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xf4, v0
; GCN-NEXT: v_and_b32_e32 v1, 63, v1
; GCN-NEXT: buffer_store_dword v60, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xf8, v0
; GCN-NEXT: buffer_store_dword v61, v3, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v3, 0xfc, v0
; GCN-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GCN-NEXT: v_add_u32_e32 v0, v0, v1
; GCN-NEXT: buffer_store_dword v62, v3, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Reload
; GCN-NEXT: v_and_b32_e32 v1, 1, v2 ; GCN-NEXT: v_and_b32_e32 v1, 1, v2
; GCN-NEXT: v_lshlrev_b32_e32 v1, 4, v1 ; GCN-NEXT: v_lshlrev_b32_e32 v1, 4, v1
; GCN-NEXT: s_mov_b32 s33, s6 ; GCN-NEXT: s_mov_b32 s33, s6
; GCN-NEXT: s_waitcnt vmcnt(15) ; GCN-NEXT: s_waitcnt vmcnt(13)
; GCN-NEXT: v_lshrrev_b32_e32 v0, v1, v0 ; GCN-NEXT: v_lshrrev_b32_e32 v0, v1, v0
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_setpc_b64 s[30:31] ; GCN-NEXT: s_setpc_b64 s[30:31]
%vec = load <128 x i16>, <128 x i16> addrspace(1)* %ptr %vec = load <128 x i16>, <128 x i16> addrspace(1)* %ptr
%elt = extractelement <128 x i16> %vec, i32 %idx %elt = extractelement <128 x i16> %vec, i32 %idx
ret i16 %elt ret i16 %elt
} }
define i64 @v_extract_v32i64_varidx(<32 x i64> addrspace(1)* %ptr, i32 %idx) { define i64 @v_extract_v32i64_varidx(<32 x i64> addrspace(1)* %ptr, i32 %idx) {
; GCN-LABEL: v_extract_v32i64_varidx: ; GCN-LABEL: v_extract_v32i64_varidx:
; GCN: ; %bb.0: ; GCN: ; %bb.0:
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v15, v0
; GCN-NEXT: s_add_u32 s4, s32, 0x3fc0 ; GCN-NEXT: s_add_u32 s4, s32, 0x3fc0
; GCN-NEXT: s_mov_b32 s5, 0 ; GCN-NEXT: v_add_co_u32_e32 v3, vcc, 64, v0
; GCN-NEXT: s_mov_b32 s6, s33 ; GCN-NEXT: s_mov_b32 s6, s33
; GCN-NEXT: s_and_b32 s33, s4, 0xffffc000 ; GCN-NEXT: s_and_b32 s33, s4, 0xffffc000
; GCN-NEXT: s_movk_i32 s4, 0x80
; GCN-NEXT: v_mov_b32_e32 v12, s5
; GCN-NEXT: v_mov_b32_e32 v16, v1
; GCN-NEXT: v_add_co_u32_e32 v31, vcc, 64, v15
; GCN-NEXT: v_mov_b32_e32 v11, s4
; GCN-NEXT: v_addc_co_u32_e32 v32, vcc, 0, v16, vcc
; GCN-NEXT: v_add_co_u32_e32 v48, vcc, v15, v11
; GCN-NEXT: v_addc_co_u32_e32 v49, vcc, v16, v12, vcc
; GCN-NEXT: s_movk_i32 s4, 0xc0
; GCN-NEXT: v_mov_b32_e32 v12, s5
; GCN-NEXT: v_mov_b32_e32 v11, s4
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 ; 4-byte Folded Spill
; GCN-NEXT: v_add_co_u32_e32 v59, vcc, v15, v11 ; GCN-NEXT: buffer_store_dword v2, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[3:6], v[15:16], off ; GCN-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v1, vcc
; GCN-NEXT: global_load_dwordx4 v[7:10], v[15:16], off offset:16 ; GCN-NEXT: global_load_dwordx4 v[7:10], v[3:4], off offset:16
; GCN-NEXT: v_addc_co_u32_e32 v60, vcc, v16, v12, vcc ; GCN-NEXT: global_load_dwordx4 v[11:14], v[3:4], off offset:32
; GCN-NEXT: global_load_dwordx4 v[11:14], v[15:16], off offset:32 ; GCN-NEXT: global_load_dwordx4 v[56:59], v[3:4], off offset:48
; GCN-NEXT: global_load_dwordx4 v[15:18], v[15:16], off offset:48 ; GCN-NEXT: global_load_dwordx4 v[15:18], v[0:1], off
; GCN-NEXT: global_load_dwordx4 v[19:22], v[31:32], off ; GCN-NEXT: s_movk_i32 s4, 0x80
; GCN-NEXT: global_load_dwordx4 v[23:26], v[31:32], off offset:16 ; GCN-NEXT: s_mov_b32 s5, 0
; GCN-NEXT: global_load_dwordx4 v[27:30], v[31:32], off offset:32 ; GCN-NEXT: v_mov_b32_e32 v3, s4
; GCN-NEXT: global_load_dwordx4 v[31:34], v[31:32], off offset:48 ; GCN-NEXT: v_mov_b32_e32 v4, s5
; GCN-NEXT: global_load_dwordx4 v[35:38], v[48:49], off ; GCN-NEXT: v_add_co_u32_e32 v3, vcc, v0, v3
; GCN-NEXT: global_load_dwordx4 v[39:42], v[48:49], off offset:16 ; GCN-NEXT: s_movk_i32 s4, 0xc0
; GCN-NEXT: global_load_dwordx4 v[43:46], v[48:49], off offset:32 ; GCN-NEXT: v_mov_b32_e32 v6, s5
; GCN-NEXT: v_lshrrev_b32_e64 v0, 6, s33 ; GCN-NEXT: v_addc_co_u32_e32 v4, vcc, v1, v4, vcc
; GCN-NEXT: v_add_u32_e32 v0, 0x100, v0 ; GCN-NEXT: v_mov_b32_e32 v5, s4
; GCN-NEXT: v_add_u32_e32 v1, 16, v0 ; GCN-NEXT: v_add_co_u32_e32 v60, vcc, v0, v5
; GCN-NEXT: s_add_u32 s32, s32, 0x10000 ; GCN-NEXT: v_addc_co_u32_e32 v61, vcc, v1, v6, vcc
; GCN-NEXT: s_sub_u32 s32, s32, 0x10000 ; GCN-NEXT: v_lshrrev_b32_e64 v62, 6, s33
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: v_add_u32_e32 v62, 0x100, v62
; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill ; GCN-NEXT: v_add_u32_e32 v2, 16, v62
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: s_add_u32 s32, s32, 0x14000
; GCN-NEXT: buffer_store_dword v36, off, s[0:3], s33 offset:580 ; 4-byte Folded Spill ; GCN-NEXT: s_sub_u32 s32, s32, 0x14000
; GCN-NEXT: buffer_store_dword v37, off, s[0:3], s33 offset:584 ; 4-byte Folded Spill ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v38, off, s[0:3], s33 offset:588 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:644 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v39, off, s[0:3], s33 offset:592 ; 4-byte Folded Spill ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:596 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:648 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:600 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:652 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:604 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:656 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:608 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:660 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:612 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:664 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:616 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:668 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:620 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:672 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:624 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v23, off, s[0:3], s33 offset:676 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:628 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v24, off, s[0:3], s33 offset:680 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:632 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v25, off, s[0:3], s33 offset:684 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:636 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v26, off, s[0:3], s33 offset:688 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[47:50], v[48:49], off offset:48 ; GCN-NEXT: buffer_store_dword v27, off, s[0:3], s33 offset:692 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[43:46], v[59:60], off ; GCN-NEXT: buffer_store_dword v28, off, s[0:3], s33 offset:696 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_store_dword v29, off, s[0:3], s33 offset:700 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v30, off, s[0:3], s33 offset:704 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:16
; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[24:27], v[0:1], off offset:32
; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[28:31], v[0:1], off offset:48
; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[32:35], v[0:1], off offset:64
; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[36:39], v[0:1], off offset:128
; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[40:43], v[0:1], off offset:192
; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[44:47], v[3:4], off offset:16
; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[48:51], v[3:4], off offset:32
; GCN-NEXT: buffer_store_dword v51, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[52:55], v[3:4], off offset:48
; GCN-NEXT: buffer_store_dword v52, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill ; GCN-NEXT: global_load_dwordx4 v[15:18], v[60:61], off offset:16
; GCN-NEXT: buffer_store_dword v53, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill ; GCN-NEXT: v_add_u32_e32 v0, 20, v62
; GCN-NEXT: buffer_store_dword v54, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v55, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:708 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:712 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill ; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:716 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[51:54], v[59:60], off offset:16 ; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:720 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[55:58], v[59:60], off offset:32 ; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:724 ; 4-byte Folded Spill
; GCN-NEXT: global_load_dwordx4 v[59:62], v[59:60], off offset:48 ; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:728 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:732 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 20, v0 ; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:736 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:740 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 24, v0 ; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:744 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:748 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 28, v0 ; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:752 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v23, off, s[0:3], s33 offset:756 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 32, v0 ; GCN-NEXT: buffer_store_dword v24, off, s[0:3], s33 offset:760 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v25, off, s[0:3], s33 offset:764 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 36, v0 ; GCN-NEXT: buffer_store_dword v26, off, s[0:3], s33 offset:768 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v12, v1, s[0:3], 0 offen ; GCN-NEXT: global_load_dwordx4 v[15:18], v[60:61], off offset:32
; GCN-NEXT: v_add_u32_e32 v1, 40, v0 ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v13, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:580 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 44, v0 ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: buffer_store_dword v14, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:584 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 48, v0 ; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:588 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v15, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:592 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 52, v0 ; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:596 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v16, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:600 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 56, v0 ; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:604 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v17, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:608 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 60, v0 ; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:612 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v18, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:616 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 64, v0 ; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:620 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v19, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:624 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x44, v0 ; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:628 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v20, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:632 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x48, v0 ; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:636 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v21, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:640 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x4c, v0 ; GCN-NEXT: global_load_dwordx4 v[15:18], v[60:61], off offset:48
; GCN-NEXT: buffer_store_dword v22, v1, s[0:3], 0 offen ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_add_u32_e32 v1, 0x50, v0 ; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v23, v1, s[0:3], 0 offen ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_add_u32_e32 v1, 0x54, v0 ; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v24, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x58, v0 ; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v25, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x5c, v0 ; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v26, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x60, v0 ; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v27, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x64, v0 ; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v28, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x68, v0 ; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v29, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x6c, v0 ; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v30, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
; GCN-NEXT: v_add_u32_e32 v1, 0x70, v0 ; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:576 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v31, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v20, v2, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0x74, v0 ; GCN-NEXT: buffer_store_dword v21, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v32, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 24, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x78, v0 ; GCN-NEXT: buffer_store_dword v22, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v33, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 28, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x7c, v0 ; GCN-NEXT: buffer_store_dword v23, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v34, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 32, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x80, v0 ; GCN-NEXT: buffer_store_dword v24, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v35, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 36, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x84, v0 ; GCN-NEXT: buffer_store_dword v25, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v36, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 40, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x88, v0 ; GCN-NEXT: buffer_store_dword v26, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v37, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 44, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x8c, v0 ; GCN-NEXT: buffer_store_dword v27, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v38, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 48, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x90, v0 ; GCN-NEXT: buffer_store_dword v28, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v39, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 52, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x94, v0 ; GCN-NEXT: buffer_store_dword v29, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v40, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 56, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x98, v0 ; GCN-NEXT: buffer_store_dword v30, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v41, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 60, v62
; GCN-NEXT: v_add_u32_e32 v1, 0x9c, v0 ; GCN-NEXT: buffer_store_dword v31, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v42, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 64, v62
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:576 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v32, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:580 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x44, v62
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:584 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v33, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:588 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x48, v62
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:592 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v34, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:596 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x4c, v62
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:600 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v35, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:604 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x50, v62
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:608 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v7, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:612 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x54, v62
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:616 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v8, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:620 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x58, v62
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:624 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v9, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:628 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x5c, v62
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:632 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v10, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:636 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x60, v62
; GCN-NEXT: v_add_u32_e32 v1, 0xa0, v0 ; GCN-NEXT: buffer_store_dword v11, v0, s[0:3], 0 offen
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: v_add_u32_e32 v0, 0x64, v62
; GCN-NEXT: v_mov_b32_e32 v8, v15 ; GCN-NEXT: buffer_store_dword v12, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0x68, v62
; GCN-NEXT: v_mov_b32_e32 v9, v16 ; GCN-NEXT: buffer_store_dword v13, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xa4, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0x6c, v62
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v14, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v10, v17 ; GCN-NEXT: v_add_u32_e32 v0, 0x70, v62
; GCN-NEXT: v_add_u32_e32 v1, 0xa8, v0 ; GCN-NEXT: buffer_store_dword v56, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0x74, v62
; GCN-NEXT: v_mov_b32_e32 v11, v18 ; GCN-NEXT: buffer_store_dword v57, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xac, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0x78, v62
; GCN-NEXT: buffer_store_dword v11, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v58, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb0, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0x7c, v62
; GCN-NEXT: buffer_store_dword v47, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v59, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb4, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0x80, v62
; GCN-NEXT: buffer_store_dword v48, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v36, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xb8, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0x84, v62
; GCN-NEXT: buffer_store_dword v49, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v37, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xbc, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0x88, v62
; GCN-NEXT: buffer_store_dword v50, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v38, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x8c, v62
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v39, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x90, v62
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v44, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x94, v62
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v45, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x98, v62
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v46, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0x9c, v62
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v47, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0xa0, v62
; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v48, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0xa4, v62
; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v49, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload ; GCN-NEXT: v_add_u32_e32 v0, 0xa8, v62
; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload ; GCN-NEXT: buffer_store_dword v50, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xc0, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0xac, v62
; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_store_dword v51, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v7, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xb0, v62
; GCN-NEXT: v_add_u32_e32 v1, 0xc4, v0 ; GCN-NEXT: buffer_store_dword v52, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v8, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xb4, v62
; GCN-NEXT: v_add_u32_e32 v1, 0xc8, v0 ; GCN-NEXT: buffer_store_dword v53, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v9, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xb8, v62
; GCN-NEXT: v_add_u32_e32 v1, 0xcc, v0 ; GCN-NEXT: buffer_store_dword v54, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v10, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xbc, v62
; GCN-NEXT: v_add_u32_e32 v1, 4, v0 ; GCN-NEXT: buffer_store_dword v55, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v4, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xc0, v62
; GCN-NEXT: v_add_u32_e32 v1, 8, v0 ; GCN-NEXT: buffer_store_dword v40, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v5, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xc4, v62
; GCN-NEXT: v_add_u32_e32 v1, 12, v0 ; GCN-NEXT: buffer_store_dword v41, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v6, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 0xc8, v62
; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:256 ; GCN-NEXT: buffer_store_dword v42, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xd0, v0 ; GCN-NEXT: v_add_u32_e32 v0, 0xcc, v62
; GCN-NEXT: buffer_store_dword v51, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_store_dword v43, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v1, 0xd4, v0 ; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:644 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v52, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v2, off, s[0:3], s33 offset:648 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xd8, v0 ; GCN-NEXT: buffer_load_dword v3, off, s[0:3], s33 offset:652 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v53, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:656 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xdc, v0 ; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:660 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v54, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:664 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xe0, v0 ; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:668 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v55, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:672 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xe4, v0 ; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:676 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v56, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:680 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xe8, v0 ; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:684 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v57, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:688 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xec, v0 ; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:692 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v58, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:696 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xf0, v0 ; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:700 ; 4-byte Folded Reload
; GCN-NEXT: buffer_store_dword v59, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:704 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v1, 0xf4, v0 ; GCN-NEXT: v_add_u32_e32 v0, 4, v62
; GCN-NEXT: buffer_store_dword v60, v1, s[0:3], 0 offen ; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_add_u32_e32 v1, 0xf8, v0 ; GCN-NEXT: buffer_store_dword v2, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v61, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 8, v62
; GCN-NEXT: v_add_u32_e32 v1, 0xfc, v0 ; GCN-NEXT: buffer_store_dword v3, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_store_dword v62, v1, s[0:3], 0 offen ; GCN-NEXT: v_add_u32_e32 v0, 12, v62
; GCN-NEXT: v_and_b32_e32 v1, 31, v2 ; GCN-NEXT: buffer_store_dword v4, v0, s[0:3], 0 offen
; GCN-NEXT: v_lshlrev_b32_e32 v1, 3, v1 ; GCN-NEXT: buffer_store_dword v1, off, s[0:3], s33 offset:256
; GCN-NEXT: v_add_u32_e32 v0, v0, v1 ; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:708 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v2, off, s[0:3], s33 offset:712 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v3, off, s[0:3], s33 offset:716 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:720 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:724 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:728 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:732 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:736 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:740 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:744 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:748 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:752 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:756 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:760 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:764 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:768 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0xd0, v62
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v4, v5
; GCN-NEXT: v_mov_b32_e32 v5, v6
; GCN-NEXT: buffer_store_dword v4, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xd4, v62
; GCN-NEXT: v_mov_b32_e32 v6, v7
; GCN-NEXT: buffer_store_dword v5, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xd8, v62
; GCN-NEXT: buffer_store_dword v6, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v7, v8
; GCN-NEXT: v_add_u32_e32 v0, 0xdc, v62
; GCN-NEXT: buffer_store_dword v7, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:580 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v2, off, s[0:3], s33 offset:584 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v3, off, s[0:3], s33 offset:588 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:592 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:596 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:600 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:604 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:608 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:612 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:616 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:620 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:624 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:628 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:632 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:636 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:640 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0xe0, v62
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v8, v9
; GCN-NEXT: v_mov_b32_e32 v9, v10
; GCN-NEXT: buffer_store_dword v8, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xe4, v62
; GCN-NEXT: v_mov_b32_e32 v10, v11
; GCN-NEXT: buffer_store_dword v9, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xe8, v62
; GCN-NEXT: buffer_store_dword v10, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v11, v12
; GCN-NEXT: v_add_u32_e32 v0, 0xec, v62
; GCN-NEXT: buffer_store_dword v11, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v1, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v2, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v3, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v7, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v8, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v9, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v10, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v11, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v12, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v13, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v14, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v15, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:576 ; 4-byte Folded Reload
; GCN-NEXT: v_add_u32_e32 v0, 0xf0, v62
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v12, v13
; GCN-NEXT: v_mov_b32_e32 v13, v14
; GCN-NEXT: buffer_store_dword v12, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xf4, v62
; GCN-NEXT: v_mov_b32_e32 v14, v15
; GCN-NEXT: buffer_store_dword v13, v0, s[0:3], 0 offen
; GCN-NEXT: v_add_u32_e32 v0, 0xf8, v62
; GCN-NEXT: buffer_store_dword v14, v0, s[0:3], 0 offen
; GCN-NEXT: v_mov_b32_e32 v15, v16
; GCN-NEXT: v_add_u32_e32 v0, 0xfc, v62
; GCN-NEXT: buffer_store_dword v15, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v0, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: v_and_b32_e32 v0, 31, v0
; GCN-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; GCN-NEXT: v_add_u32_e32 v0, v62, v0
; GCN-NEXT: v_add_u32_e32 v1, 4, v0 ; GCN-NEXT: v_add_u32_e32 v1, 4, v0
; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v1, v1, s[0:3], 0 offen ; GCN-NEXT: buffer_load_dword v1, v1, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v61, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v60, off, s[0:3], s33 offset:8 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v59, off, s[0:3], s33 offset:12 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload ; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:16 ; 4-byte Folded Reload
Show All 17 Lines

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

Show First 20 Lines • Show All 2,514 Lines • ▼ Show 20 Lines; MOVREL-NEXT: s_setpc_b64 s[30:31]
%elt = extractelement <64 x i32> %vec, i32 7 %elt = extractelement <64 x i32> %vec, i32 7
ret i32 %elt ret i32 %elt
} }
define i32 @v_extract_v64i32_32(<64 x i32> addrspace(1)* %ptr) { define i32 @v_extract_v64i32_32(<64 x i32> addrspace(1)* %ptr) {
; GPRIDX-LABEL: v_extract_v64i32_32: ; GPRIDX-LABEL: v_extract_v64i32_32:
; GPRIDX: ; %bb.0: ; GPRIDX: ; %bb.0:
; GPRIDX-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GPRIDX-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GPRIDX-NEXT: s_movk_i32 s4, 0x80 ; GPRIDX-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:128
; GPRIDX-NEXT: s_mov_b32 s5, 0
; GPRIDX-NEXT: v_mov_b32_e32 v2, s4
; GPRIDX-NEXT: v_mov_b32_e32 v3, s5
; GPRIDX-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
; GPRIDX-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
; GPRIDX-NEXT: global_load_dwordx4 v[0:3], v[0:1], off
; GPRIDX-NEXT: s_waitcnt vmcnt(0) ; GPRIDX-NEXT: s_waitcnt vmcnt(0)
; GPRIDX-NEXT: s_setpc_b64 s[30:31] ; GPRIDX-NEXT: s_setpc_b64 s[30:31]
; ;
; MOVREL-LABEL: v_extract_v64i32_32: ; MOVREL-LABEL: v_extract_v64i32_32:
; MOVREL: ; %bb.0: ; MOVREL: ; %bb.0:
; MOVREL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; MOVREL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MOVREL-NEXT: s_movk_i32 s4, 0x80 ; MOVREL-NEXT: s_movk_i32 s4, 0x80
; MOVREL-NEXT: s_mov_b32 s5, 0 ; MOVREL-NEXT: s_mov_b32 s5, 0
; MOVREL-NEXT: v_mov_b32_e32 v2, s4 ; MOVREL-NEXT: v_mov_b32_e32 v2, s4
; MOVREL-NEXT: v_mov_b32_e32 v3, s5 ; MOVREL-NEXT: v_mov_b32_e32 v3, s5
; MOVREL-NEXT: v_add_u32_e32 v0, vcc, v0, v2 ; MOVREL-NEXT: v_add_u32_e32 v0, vcc, v0, v2
; MOVREL-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc ; MOVREL-NEXT: v_addc_u32_e32 v1, vcc, v1, v3, vcc
; MOVREL-NEXT: flat_load_dwordx4 v[0:3], v[0:1] ; MOVREL-NEXT: flat_load_dwordx4 v[0:3], v[0:1]
; MOVREL-NEXT: s_waitcnt vmcnt(0) ; MOVREL-NEXT: s_waitcnt vmcnt(0)
; MOVREL-NEXT: s_setpc_b64 s[30:31] ; MOVREL-NEXT: s_setpc_b64 s[30:31]
%vec = load <64 x i32>, <64 x i32> addrspace(1)* %ptr %vec = load <64 x i32>, <64 x i32> addrspace(1)* %ptr
%elt = extractelement <64 x i32> %vec, i32 32 %elt = extractelement <64 x i32> %vec, i32 32
ret i32 %elt ret i32 %elt
} }
define i32 @v_extract_v64i32_33(<64 x i32> addrspace(1)* %ptr) { define i32 @v_extract_v64i32_33(<64 x i32> addrspace(1)* %ptr) {
; GPRIDX-LABEL: v_extract_v64i32_33: ; GPRIDX-LABEL: v_extract_v64i32_33:
; GPRIDX: ; %bb.0: ; GPRIDX: ; %bb.0:
; GPRIDX-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GPRIDX-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GPRIDX-NEXT: s_movk_i32 s4, 0x80 ; GPRIDX-NEXT: global_load_dwordx4 v[0:3], v[0:1], off offset:128
; GPRIDX-NEXT: s_mov_b32 s5, 0
; GPRIDX-NEXT: v_mov_b32_e32 v2, s4
; GPRIDX-NEXT: v_mov_b32_e32 v3, s5
; GPRIDX-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
; GPRIDX-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v3, vcc
; GPRIDX-NEXT: global_load_dwordx4 v[0:3], v[0:1], off
; GPRIDX-NEXT: s_waitcnt vmcnt(0) ; GPRIDX-NEXT: s_waitcnt vmcnt(0)
; GPRIDX-NEXT: v_mov_b32_e32 v0, v1 ; GPRIDX-NEXT: v_mov_b32_e32 v0, v1
; GPRIDX-NEXT: s_setpc_b64 s[30:31] ; GPRIDX-NEXT: s_setpc_b64 s[30:31]
; ;
; MOVREL-LABEL: v_extract_v64i32_33: ; MOVREL-LABEL: v_extract_v64i32_33:
; MOVREL: ; %bb.0: ; MOVREL: ; %bb.0:
; MOVREL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; MOVREL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; MOVREL-NEXT: s_movk_i32 s4, 0x80 ; MOVREL-NEXT: s_movk_i32 s4, 0x80
▲ Show 20 Lines • Show All 48 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/flat-scratch.ll

Show First 20 Lines • Show All 46 Lines • ▼ Show 20 Linesbb:
%i11 = bitcast float addrspace(5)* %i10 to i32 addrspace(5)* %i11 = bitcast float addrspace(5)* %i10 to i32 addrspace(5)*
%i12 = load volatile i32, i32 addrspace(5)* %i11, align 4 %i12 = load volatile i32, i32 addrspace(5)* %i11, align 4
ret void ret void
} }
define amdgpu_kernel void @store_load_vindex_kernel() { define amdgpu_kernel void @store_load_vindex_kernel() {
; GFX9-LABEL: store_load_vindex_kernel: ; GFX9-LABEL: store_load_vindex_kernel:
; GFX9: ; %bb.0: ; %bb ; GFX9: ; %bb.0: ; %bb
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s0, s3
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; GFX9-NEXT: v_sub_u32_e32 v0, 0, v0 ; GFX9-NEXT: v_sub_u32_e32 v0, 0, v0
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s0, s3
; GFX9-NEXT: v_mov_b32_e32 v2, 4 ; GFX9-NEXT: v_mov_b32_e32 v2, 4
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_add_u32_e32 v0, v2, v0
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s1, 0 ; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s1, 0
; GFX9-NEXT: v_add_u32_e32 v1, v2, v1 ; GFX9-NEXT: v_add_u32_e32 v1, v2, v1
; GFX9-NEXT: v_mov_b32_e32 v3, 15 ; GFX9-NEXT: v_mov_b32_e32 v3, 15
; GFX9-NEXT: scratch_store_dword v1, v3, off ; GFX9-NEXT: scratch_store_dword v1, v3, off
; GFX9-NEXT: v_add_u32_e32 v0, 0x7c, v0 ; GFX9-NEXT: v_add_u32_e32 v0, v2, v0
; GFX9-NEXT: scratch_load_dword v0, v0, off ; GFX9-NEXT: scratch_load_dword v0, v0, off offset:124
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
; ;
; GFX10-LABEL: store_load_vindex_kernel: ; GFX10-LABEL: store_load_vindex_kernel:
; GFX10: ; %bb.0: ; %bb ; GFX10: ; %bb.0: ; %bb
; GFX10-NEXT: s_add_u32 s0, s0, s3 ; GFX10-NEXT: s_add_u32 s0, s0, s3
; GFX10-NEXT: s_addc_u32 s1, s1, 0 ; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
; GFX10-NEXT: v_sub_nc_u32_e32 v1, 0, v0 ; GFX10-NEXT: v_sub_nc_u32_e32 v1, 0, v0
; GFX10-NEXT: v_mov_b32_e32 v2, 4
; GFX10-NEXT: v_lshlrev_b32_e32 v0, 2, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX10-NEXT: v_mov_b32_e32 v2, 4
; GFX10-NEXT: v_mov_b32_e32 v3, 15
; GFX10-NEXT: v_lshlrev_b32_e32 v1, 2, v1 ; GFX10-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX10-NEXT: v_add_nc_u32_e32 v0, v2, v0 ; GFX10-NEXT: v_add_nc_u32_e32 v0, v2, v0
; GFX10-NEXT: v_add_nc_u32_e32 v1, v2, v1 ; GFX10-NEXT: v_add_nc_u32_e32 v1, v2, v1
; GFX10-NEXT: v_mov_b32_e32 v2, 15 ; GFX10-NEXT: scratch_store_dword v0, v3, off
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0x7c, v1 ; GFX10-NEXT: scratch_load_dword v0, v1, off offset:124
; GFX10-NEXT: scratch_store_dword v0, v2, off
; GFX10-NEXT: scratch_load_dword v0, v1, off
; GFX10-NEXT: s_endpgm ; GFX10-NEXT: s_endpgm
bb: bb:
%i = alloca [32 x float], align 4, addrspace(5) %i = alloca [32 x float], align 4, addrspace(5)
%i1 = bitcast [32 x float] addrspace(5)* %i to i8 addrspace(5)* %i1 = bitcast [32 x float] addrspace(5)* %i to i8 addrspace(5)*
%i2 = tail call i32 @llvm.amdgcn.workitem.id.x() %i2 = tail call i32 @llvm.amdgcn.workitem.id.x()
%i3 = zext i32 %i2 to i64 %i3 = zext i32 %i2 to i64
%i7 = getelementptr inbounds [32 x float], [32 x float] addrspace(5)* %i, i32 0, i32 %i2 %i7 = getelementptr inbounds [32 x float], [32 x float] addrspace(5)* %i, i32 0, i32 %i2
%i8 = bitcast float addrspace(5)* %i7 to i32 addrspace(5)* %i8 = bitcast float addrspace(5)* %i7 to i32 addrspace(5)*
▲ Show 20 Lines • Show All 49 Lines • ▼ Show 20 Linesbb:
%i12 = load volatile i32, i32 addrspace(5)* %i11, align 4 %i12 = load volatile i32, i32 addrspace(5)* %i11, align 4
ret void ret void
} }
define void @private_ptr_foo(float addrspace(5)* nocapture %arg) { define void @private_ptr_foo(float addrspace(5)* nocapture %arg) {
; GFX9-LABEL: private_ptr_foo: ; GFX9-LABEL: private_ptr_foo:
; GFX9: ; %bb.0: ; GFX9: ; %bb.0:
; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NEXT: v_add_u32_e32 v0, 4, v0
; GFX9-NEXT: v_mov_b32_e32 v1, 0x41200000 ; GFX9-NEXT: v_mov_b32_e32 v1, 0x41200000
; GFX9-NEXT: scratch_store_dword v0, v1, off ; GFX9-NEXT: scratch_store_dword v0, v1, off offset:4
; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: s_setpc_b64 s[30:31] ; GFX9-NEXT: s_setpc_b64 s[30:31]
; ;
; GFX10-LABEL: private_ptr_foo: ; GFX10-LABEL: private_ptr_foo:
; GFX10: ; %bb.0: ; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-NEXT: v_add_nc_u32_e32 v0, 4, v0
; GFX10-NEXT: v_mov_b32_e32 v1, 0x41200000 ; GFX10-NEXT: v_mov_b32_e32 v1, 0x41200000
; GFX10-NEXT: scratch_store_dword v0, v1, off ; GFX10-NEXT: scratch_store_dword v0, v1, off offset:4
; GFX10-NEXT: s_waitcnt_vscnt null, 0x0 ; GFX10-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-NEXT: s_setpc_b64 s[30:31] ; GFX10-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr inbounds float, float addrspace(5)* %arg, i32 1 %gep = getelementptr inbounds float, float addrspace(5)* %arg, i32 1
store float 1.000000e+01, float addrspace(5)* %gep, align 4 store float 1.000000e+01, float addrspace(5)* %gep, align 4
ret void ret void
} }
define amdgpu_kernel void @store_load_sindex_small_offset_kernel(i32 %idx) { define amdgpu_kernel void @store_load_sindex_small_offset_kernel(i32 %idx) {
▲ Show 20 Lines • Show All 58 Lines • ▼ Show 20 Lines
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s1, 0 ; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s1, 0
; GFX9-NEXT: s_add_u32 s0, 4, 0 ; GFX9-NEXT: s_add_u32 s0, 4, 0
; GFX9-NEXT: scratch_load_dword v1, off, s0 ; GFX9-NEXT: scratch_load_dword v1, off, s0
; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; GFX9-NEXT: v_sub_u32_e32 v0, 0, v0 ; GFX9-NEXT: v_sub_u32_e32 v0, 0, v0
; GFX9-NEXT: v_mov_b32_e32 v2, 0x104 ; GFX9-NEXT: v_mov_b32_e32 v2, 0x104
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_add_u32_e32 v0, v2, v0
; GFX9-NEXT: v_add_u32_e32 v1, v2, v1 ; GFX9-NEXT: v_add_u32_e32 v1, v2, v1
; GFX9-NEXT: v_mov_b32_e32 v3, 15 ; GFX9-NEXT: v_mov_b32_e32 v3, 15
; GFX9-NEXT: scratch_store_dword v1, v3, off ; GFX9-NEXT: scratch_store_dword v1, v3, off
; GFX9-NEXT: v_add_u32_e32 v0, 0x7c, v0 ; GFX9-NEXT: v_add_u32_e32 v0, v2, v0
; GFX9-NEXT: scratch_load_dword v0, v0, off ; GFX9-NEXT: scratch_load_dword v0, v0, off offset:124
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
; ;
; GFX10-LABEL: store_load_vindex_small_offset_kernel: ; GFX10-LABEL: store_load_vindex_small_offset_kernel:
; GFX10: ; %bb.0: ; %bb ; GFX10: ; %bb.0: ; %bb
; GFX10-NEXT: s_add_u32 s0, s0, s3 ; GFX10-NEXT: s_add_u32 s0, s0, s3
; GFX10-NEXT: s_addc_u32 s1, s1, 0 ; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
; GFX10-NEXT: v_sub_nc_u32_e32 v1, 0, v0 ; GFX10-NEXT: v_sub_nc_u32_e32 v1, 0, v0
; GFX10-NEXT: v_mov_b32_e32 v2, 0x104
; GFX10-NEXT: v_lshlrev_b32_e32 v0, 2, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX10-NEXT: v_mov_b32_e32 v2, 0x104
; GFX10-NEXT: v_mov_b32_e32 v3, 15
; GFX10-NEXT: s_add_u32 s0, 4, 0 ; GFX10-NEXT: s_add_u32 s0, 4, 0
; GFX10-NEXT: v_lshlrev_b32_e32 v1, 2, v1 ; GFX10-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX10-NEXT: scratch_load_dword v3, off, s0
; GFX10-NEXT: v_add_nc_u32_e32 v0, v2, v0 ; GFX10-NEXT: v_add_nc_u32_e32 v0, v2, v0
; GFX10-NEXT: v_add_nc_u32_e32 v1, v2, v1 ; GFX10-NEXT: v_add_nc_u32_e32 v1, v2, v1
; GFX10-NEXT: v_mov_b32_e32 v2, 15 ; GFX10-NEXT: scratch_load_dword v2, off, s0
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0x7c, v1 ; GFX10-NEXT: scratch_store_dword v0, v3, off
; GFX10-NEXT: scratch_store_dword v0, v2, off ; GFX10-NEXT: scratch_load_dword v0, v1, off offset:124
; GFX10-NEXT: scratch_load_dword v0, v1, off
; GFX10-NEXT: s_endpgm ; GFX10-NEXT: s_endpgm
bb: bb:
%padding = alloca [64 x i32], align 4, addrspace(5) %padding = alloca [64 x i32], align 4, addrspace(5)
%i = alloca [32 x float], align 4, addrspace(5) %i = alloca [32 x float], align 4, addrspace(5)
%pad_gep = getelementptr inbounds [64 x i32], [64 x i32] addrspace(5)* %padding, i32 0, i32 undef %pad_gep = getelementptr inbounds [64 x i32], [64 x i32] addrspace(5)* %padding, i32 0, i32 undef
%pad_load = load volatile i32, i32 addrspace(5)* %pad_gep, align 4 %pad_load = load volatile i32, i32 addrspace(5)* %pad_gep, align 4
%i1 = bitcast [32 x float] addrspace(5)* %i to i8 addrspace(5)* %i1 = bitcast [32 x float] addrspace(5)* %i to i8 addrspace(5)*
%i2 = tail call i32 @llvm.amdgcn.workitem.id.x() %i2 = tail call i32 @llvm.amdgcn.workitem.id.x()
▲ Show 20 Lines • Show All 125 Lines • ▼ Show 20 Lines
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s1, 0 ; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s1, 0
; GFX9-NEXT: s_add_u32 s0, 4, 0 ; GFX9-NEXT: s_add_u32 s0, 4, 0
; GFX9-NEXT: scratch_load_dword v1, off, s0 ; GFX9-NEXT: scratch_load_dword v1, off, s0
; GFX9-NEXT: s_waitcnt vmcnt(0) ; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v1, 2, v0
; GFX9-NEXT: v_sub_u32_e32 v0, 0, v0 ; GFX9-NEXT: v_sub_u32_e32 v0, 0, v0
; GFX9-NEXT: v_mov_b32_e32 v2, 0x4004 ; GFX9-NEXT: v_mov_b32_e32 v2, 0x4004
; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX9-NEXT: v_add_u32_e32 v0, v2, v0
; GFX9-NEXT: v_add_u32_e32 v1, v2, v1 ; GFX9-NEXT: v_add_u32_e32 v1, v2, v1
; GFX9-NEXT: v_mov_b32_e32 v3, 15 ; GFX9-NEXT: v_mov_b32_e32 v3, 15
; GFX9-NEXT: scratch_store_dword v1, v3, off ; GFX9-NEXT: scratch_store_dword v1, v3, off
; GFX9-NEXT: v_add_u32_e32 v0, 0x7c, v0 ; GFX9-NEXT: v_add_u32_e32 v0, v2, v0
; GFX9-NEXT: scratch_load_dword v0, v0, off ; GFX9-NEXT: scratch_load_dword v0, v0, off offset:124
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
; ;
; GFX10-LABEL: store_load_vindex_large_offset_kernel: ; GFX10-LABEL: store_load_vindex_large_offset_kernel:
; GFX10: ; %bb.0: ; %bb ; GFX10: ; %bb.0: ; %bb
; GFX10-NEXT: s_add_u32 s0, s0, s3 ; GFX10-NEXT: s_add_u32 s0, s0, s3
; GFX10-NEXT: s_addc_u32 s1, s1, 0 ; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
; GFX10-NEXT: v_sub_nc_u32_e32 v1, 0, v0 ; GFX10-NEXT: v_sub_nc_u32_e32 v1, 0, v0
; GFX10-NEXT: v_mov_b32_e32 v2, 0x4004
; GFX10-NEXT: v_lshlrev_b32_e32 v0, 2, v0 ; GFX10-NEXT: v_lshlrev_b32_e32 v0, 2, v0
; GFX10-NEXT: v_mov_b32_e32 v2, 0x4004
; GFX10-NEXT: v_mov_b32_e32 v3, 15
; GFX10-NEXT: s_add_u32 s0, 4, 0 ; GFX10-NEXT: s_add_u32 s0, 4, 0
; GFX10-NEXT: v_lshlrev_b32_e32 v1, 2, v1 ; GFX10-NEXT: v_lshlrev_b32_e32 v1, 2, v1
; GFX10-NEXT: scratch_load_dword v3, off, s0
; GFX10-NEXT: v_add_nc_u32_e32 v0, v2, v0 ; GFX10-NEXT: v_add_nc_u32_e32 v0, v2, v0
; GFX10-NEXT: v_add_nc_u32_e32 v1, v2, v1 ; GFX10-NEXT: v_add_nc_u32_e32 v1, v2, v1
; GFX10-NEXT: v_mov_b32_e32 v2, 15 ; GFX10-NEXT: scratch_load_dword v2, off, s0
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0x7c, v1 ; GFX10-NEXT: scratch_store_dword v0, v3, off
; GFX10-NEXT: scratch_store_dword v0, v2, off ; GFX10-NEXT: scratch_load_dword v0, v1, off offset:124
; GFX10-NEXT: scratch_load_dword v0, v1, off
; GFX10-NEXT: s_endpgm ; GFX10-NEXT: s_endpgm
bb: bb:
%padding = alloca [4096 x i32], align 4, addrspace(5) %padding = alloca [4096 x i32], align 4, addrspace(5)
%i = alloca [32 x float], align 4, addrspace(5) %i = alloca [32 x float], align 4, addrspace(5)
%pad_gep = getelementptr inbounds [4096 x i32], [4096 x i32] addrspace(5)* %padding, i32 0, i32 undef %pad_gep = getelementptr inbounds [4096 x i32], [4096 x i32] addrspace(5)* %padding, i32 0, i32 undef
%pad_load = load volatile i32, i32 addrspace(5)* %pad_gep, align 4 %pad_load = load volatile i32, i32 addrspace(5)* %pad_gep, align 4
%i1 = bitcast [32 x float] addrspace(5)* %i to i8 addrspace(5)* %i1 = bitcast [32 x float] addrspace(5)* %i to i8 addrspace(5)*
%i2 = tail call i32 @llvm.amdgcn.workitem.id.x() %i2 = tail call i32 @llvm.amdgcn.workitem.id.x()
▲ Show 20 Lines • Show All 150 Lines • ▼ Show 20 Lines
; GFX9: ; %bb.0: ; %bb ; GFX9: ; %bb.0: ; %bb
; GFX9-NEXT: s_load_dword s0, s[0:1], 0x24 ; GFX9-NEXT: s_load_dword s0, s[0:1], 0x24
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s2, s5 ; GFX9-NEXT: s_add_u32 flat_scratch_lo, s2, s5
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s3, 0 ; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s3, 0
; GFX9-NEXT: v_mov_b32_e32 v1, 15 ; GFX9-NEXT: v_mov_b32_e32 v1, 15
; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_add_lshl_u32 v0, s0, v0, 2 ; GFX9-NEXT: v_add_lshl_u32 v0, s0, v0, 2
; GFX9-NEXT: v_add_u32_e32 v0, 4, v0 ; GFX9-NEXT: v_add_u32_e32 v0, 4, v0
; GFX9-NEXT: v_add_u32_e32 v0, 0x400, v0 ; GFX9-NEXT: scratch_store_dword v0, v1, off offset:1024
; GFX9-NEXT: scratch_store_dword v0, v1, off ; GFX9-NEXT: scratch_load_dword v0, v0, off offset:1024
; GFX9-NEXT: scratch_load_dword v0, v0, off
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
; ;
; GFX10-LABEL: store_load_vidx_sidx_offset: ; GFX10-LABEL: store_load_vidx_sidx_offset:
; GFX10: ; %bb.0: ; %bb ; GFX10: ; %bb.0: ; %bb
; GFX10-NEXT: s_add_u32 s2, s2, s5 ; GFX10-NEXT: s_add_u32 s2, s2, s5
; GFX10-NEXT: s_addc_u32 s3, s3, 0 ; GFX10-NEXT: s_addc_u32 s3, s3, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3 ; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
; GFX10-NEXT: s_load_dword s0, s[0:1], 0x24 ; GFX10-NEXT: s_load_dword s0, s[0:1], 0x24
; GFX10-NEXT: v_mov_b32_e32 v1, 15 ; GFX10-NEXT: v_mov_b32_e32 v1, 15
; GFX10-NEXT: s_waitcnt lgkmcnt(0) ; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: v_add_lshl_u32 v0, s0, v0, 2 ; GFX10-NEXT: v_add_lshl_u32 v0, s0, v0, 2
; GFX10-NEXT: v_add_nc_u32_e32 v0, 4, v0 ; GFX10-NEXT: v_add_nc_u32_e32 v0, 4, v0
; GFX10-NEXT: v_add_nc_u32_e32 v0, 0x400, v0 ; GFX10-NEXT: scratch_store_dword v0, v1, off offset:1024
; GFX10-NEXT: scratch_store_dword v0, v1, off ; GFX10-NEXT: scratch_load_dword v0, v0, off offset:1024
; GFX10-NEXT: scratch_load_dword v0, v0, off
; GFX10-NEXT: s_endpgm ; GFX10-NEXT: s_endpgm
bb: bb:
%alloca = alloca [32 x i32], align 4, addrspace(5) %alloca = alloca [32 x i32], align 4, addrspace(5)
%vidx = tail call i32 @llvm.amdgcn.workitem.id.x() %vidx = tail call i32 @llvm.amdgcn.workitem.id.x()
%add1 = add nsw i32 %sidx, %vidx %add1 = add nsw i32 %sidx, %vidx
%add2 = add nsw i32 %add1, 256 %add2 = add nsw i32 %add1, 256
%gep = getelementptr inbounds [32 x i32], [32 x i32] addrspace(5)* %alloca, i32 0, i32 %add2 %gep = getelementptr inbounds [32 x i32], [32 x i32] addrspace(5)* %alloca, i32 0, i32 %add2
store volatile i32 15, i32 addrspace(5)* %gep, align 4 store volatile i32 15, i32 addrspace(5)* %gep, align 4
▲ Show 20 Lines • Show All 137 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.atomic.inc.ll

Show First 20 Lines • Show All 993 Lines • ▼ Show 20 Lines
; GCN-NEXT: flat_store_dword v[0:1], v2 ; GCN-NEXT: flat_store_dword v[0:1], v2
; GCN-NEXT: s_endpgm ; GCN-NEXT: s_endpgm
%result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %ptr, i32 42, i32 0, i32 0, i1 false) %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %ptr, i32 42, i32 0, i32 0, i1 false)
store i32 %result, i32* %out store i32 %result, i32* %out
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset(i32* %out, i32* %ptr) #0 { define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset(i32* %out, i32* %ptr) #0 {
; GCN-LABEL: flat_atomic_inc_ret_i32_offset: ; CI-LABEL: flat_atomic_inc_ret_i32_offset:
; GCN: ; %bb.0: ; CI: ; %bb.0:
; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 ; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v2, 42 ; CI-NEXT: v_mov_b32_e32 v2, 42
; GCN-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_add_u32 s2, s2, 16 ; CI-NEXT: s_add_u32 s2, s2, 16
; GCN-NEXT: s_addc_u32 s3, s3, 0 ; CI-NEXT: s_addc_u32 s3, s3, 0
; GCN-NEXT: v_mov_b32_e32 v0, s2 ; CI-NEXT: v_mov_b32_e32 v0, s2
; GCN-NEXT: v_mov_b32_e32 v1, s3 ; CI-NEXT: v_mov_b32_e32 v1, s3
; GCN-NEXT: flat_atomic_inc v2, v[0:1], v2 glc ; CI-NEXT: flat_atomic_inc v2, v[0:1], v2 glc
; GCN-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: v_mov_b32_e32 v1, s1
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN-NEXT: flat_store_dword v[0:1], v2 ; CI-NEXT: flat_store_dword v[0:1], v2
; GCN-NEXT: s_endpgm ; CI-NEXT: s_endpgm
;
; VI-LABEL: flat_atomic_inc_ret_i32_offset:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; VI-NEXT: v_mov_b32_e32 v2, 42
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_add_u32 s2, s2, 16
; VI-NEXT: s_addc_u32 s3, s3, 0
; VI-NEXT: v_mov_b32_e32 v0, s2
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: flat_atomic_inc v2, v[0:1], v2 glc
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: flat_atomic_inc_ret_i32_offset:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX9-NEXT: v_mov_b32_e32 v2, 42
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v1, s3
; GFX9-NEXT: flat_atomic_inc v2, v[0:1], v2 offset:16 glc
; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: flat_store_dword v[0:1], v2
; GFX9-NEXT: s_endpgm
%gep = getelementptr i32, i32* %ptr, i32 4 %gep = getelementptr i32, i32* %ptr, i32 4
%result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false) %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false)
store i32 %result, i32* %out store i32 %result, i32* %out
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_noret_i32(i32* %ptr) nounwind { define amdgpu_kernel void @flat_atomic_inc_noret_i32(i32* %ptr) nounwind {
; GCN-LABEL: flat_atomic_inc_noret_i32: ; GCN-LABEL: flat_atomic_inc_noret_i32:
; GCN: ; %bb.0: ; GCN: ; %bb.0:
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v2, 42 ; GCN-NEXT: v_mov_b32_e32 v2, 42
; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v1, s1
; GCN-NEXT: flat_atomic_inc v0, v[0:1], v2 glc ; GCN-NEXT: flat_atomic_inc v0, v[0:1], v2 glc
; GCN-NEXT: s_endpgm ; GCN-NEXT: s_endpgm
%result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %ptr, i32 42, i32 0, i32 0, i1 false) %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %ptr, i32 42, i32 0, i32 0, i1 false)
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset(i32* %ptr) nounwind { define amdgpu_kernel void @flat_atomic_inc_noret_i32_offset(i32* %ptr) nounwind {
; GCN-LABEL: flat_atomic_inc_noret_i32_offset: ; CI-LABEL: flat_atomic_inc_noret_i32_offset:
; GCN: ; %bb.0: ; CI: ; %bb.0:
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v2, 42 ; CI-NEXT: v_mov_b32_e32 v2, 42
; GCN-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_add_u32 s0, s0, 16 ; CI-NEXT: s_add_u32 s0, s0, 16
; GCN-NEXT: s_addc_u32 s1, s1, 0 ; CI-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: v_mov_b32_e32 v1, s1
; GCN-NEXT: flat_atomic_inc v0, v[0:1], v2 glc ; CI-NEXT: flat_atomic_inc v0, v[0:1], v2 glc
; GCN-NEXT: s_endpgm ; CI-NEXT: s_endpgm
;
; VI-LABEL: flat_atomic_inc_noret_i32_offset:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; VI-NEXT: v_mov_b32_e32 v2, 42
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_add_u32 s0, s0, 16
; VI-NEXT: s_addc_u32 s1, s1, 0
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_atomic_inc v0, v[0:1], v2 glc
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: flat_atomic_inc_noret_i32_offset:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX9-NEXT: v_mov_b32_e32 v2, 42
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: flat_atomic_inc v0, v[0:1], v2 offset:16 glc
; GFX9-NEXT: s_endpgm
%gep = getelementptr i32, i32* %ptr, i32 4 %gep = getelementptr i32, i32* %ptr, i32 4
%result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false) %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false)
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(i32* %out, i32* %ptr) #0 { define amdgpu_kernel void @flat_atomic_inc_ret_i32_offset_addr64(i32* %out, i32* %ptr) #0 {
; CI-LABEL: flat_atomic_inc_ret_i32_offset_addr64: ; CI-LABEL: flat_atomic_inc_ret_i32_offset_addr64:
; CI: ; %bb.0: ; CI: ; %bb.0:
Show All 35 Lines
; VI-NEXT: flat_atomic_inc v2, v[2:3], v4 glc ; VI-NEXT: flat_atomic_inc v2, v[2:3], v4 glc
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: flat_store_dword v[0:1], v2 ; VI-NEXT: flat_store_dword v[0:1], v2
; VI-NEXT: s_endpgm ; VI-NEXT: s_endpgm
; ;
; GFX9-LABEL: flat_atomic_inc_ret_i32_offset_addr64: ; GFX9-LABEL: flat_atomic_inc_ret_i32_offset_addr64:
; GFX9: ; %bb.0: ; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v4, 2, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s2 ; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v3, s1
; GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GFX9-NEXT: v_mov_b32_e32 v1, s3
; GFX9-NEXT: v_add_co_u32_e32 v3, vcc, v0, v2 ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v4
; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v1, vcc
; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 20, v3 ; GFX9-NEXT: v_mov_b32_e32 v2, s0
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v4, vcc ; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v4
; GFX9-NEXT: v_mov_b32_e32 v4, 42 ; GFX9-NEXT: v_mov_b32_e32 v4, 42
; GFX9-NEXT: flat_atomic_inc v2, v[2:3], v4 glc ; GFX9-NEXT: flat_atomic_inc v0, v[0:1], v4 offset:20 glc
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: flat_store_dword v[0:1], v2 ; GFX9-NEXT: flat_store_dword v[2:3], v0
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x() %id = call i32 @llvm.amdgcn.workitem.id.x()
%gep.tid = getelementptr i32, i32* %ptr, i32 %id %gep.tid = getelementptr i32, i32* %ptr, i32 %id
%out.gep = getelementptr i32, i32* %out, i32 %id %out.gep = getelementptr i32, i32* %out, i32 %id
%gep = getelementptr i32, i32* %gep.tid, i32 5 %gep = getelementptr i32, i32* %gep.tid, i32 5
%result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false) %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false)
store i32 %result, i32* %out.gep store i32 %result, i32* %out.gep
ret void ret void
Show All 34 Lines
; GFX9: ; %bb.0: ; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 2, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v2, 2, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s0 ; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2 ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
; GFX9-NEXT: v_mov_b32_e32 v1, s1 ; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 20, v0
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_mov_b32_e32 v2, 42 ; GFX9-NEXT: v_mov_b32_e32 v2, 42
; GFX9-NEXT: flat_atomic_inc v0, v[0:1], v2 glc ; GFX9-NEXT: flat_atomic_inc v0, v[0:1], v2 offset:20 glc
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x() %id = call i32 @llvm.amdgcn.workitem.id.x()
%gep.tid = getelementptr i32, i32* %ptr, i32 %id %gep.tid = getelementptr i32, i32* %ptr, i32 %id
%gep = getelementptr i32, i32* %gep.tid, i32 5 %gep = getelementptr i32, i32* %gep.tid, i32 5
%result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false) %result = call i32 @llvm.amdgcn.atomic.inc.i32.p0i32(i32* %gep, i32 42, i32 0, i32 0, i1 false)
ret void ret void
} }
▲ Show 20 Lines • Show All 74 Lines • ▼ Show 20 Lines
; GCN-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; GCN-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; GCN-NEXT: s_endpgm ; GCN-NEXT: s_endpgm
%result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %ptr, i64 42, i32 0, i32 0, i1 false) %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %ptr, i64 42, i32 0, i32 0, i1 false)
store i64 %result, i64* %out store i64 %result, i64* %out
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset(i64* %out, i64* %ptr) #0 { define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset(i64* %out, i64* %ptr) #0 {
; GCN-LABEL: flat_atomic_inc_ret_i64_offset: ; CI-LABEL: flat_atomic_inc_ret_i64_offset:
; GCN: ; %bb.0: ; CI: ; %bb.0:
; GCN-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 ; CI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v2, 42 ; CI-NEXT: v_mov_b32_e32 v2, 42
; GCN-NEXT: v_mov_b32_e32 v3, 0 ; CI-NEXT: v_mov_b32_e32 v3, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_add_u32 s2, s2, 32 ; CI-NEXT: s_add_u32 s2, s2, 32
; GCN-NEXT: s_addc_u32 s3, s3, 0 ; CI-NEXT: s_addc_u32 s3, s3, 0
; GCN-NEXT: v_mov_b32_e32 v0, s2 ; CI-NEXT: v_mov_b32_e32 v0, s2
; GCN-NEXT: v_mov_b32_e32 v1, s3 ; CI-NEXT: v_mov_b32_e32 v1, s3
; GCN-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc ; CI-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc
; GCN-NEXT: v_mov_b32_e32 v3, s1 ; CI-NEXT: v_mov_b32_e32 v3, s1
; GCN-NEXT: v_mov_b32_e32 v2, s0 ; CI-NEXT: v_mov_b32_e32 v2, s0
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; CI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GCN-NEXT: flat_store_dwordx2 v[2:3], v[0:1] ; CI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; GCN-NEXT: s_endpgm ; CI-NEXT: s_endpgm
;
; VI-LABEL: flat_atomic_inc_ret_i64_offset:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; VI-NEXT: v_mov_b32_e32 v2, 42
; VI-NEXT: v_mov_b32_e32 v3, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_add_u32 s2, s2, 32
; VI-NEXT: s_addc_u32 s3, s3, 0
; VI-NEXT: v_mov_b32_e32 v0, s2
; VI-NEXT: v_mov_b32_e32 v1, s3
; VI-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc
; VI-NEXT: v_mov_b32_e32 v3, s1
; VI-NEXT: v_mov_b32_e32 v2, s0
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: flat_atomic_inc_ret_i64_offset:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX9-NEXT: v_mov_b32_e32 v0, 42
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v2, s2
; GFX9-NEXT: v_mov_b32_e32 v3, s3
; GFX9-NEXT: flat_atomic_inc_x2 v[0:1], v[2:3], v[0:1] offset:32 glc
; GFX9-NEXT: v_mov_b32_e32 v3, s1
; GFX9-NEXT: v_mov_b32_e32 v2, s0
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; GFX9-NEXT: s_endpgm
%gep = getelementptr i64, i64* %ptr, i32 4 %gep = getelementptr i64, i64* %ptr, i32 4
%result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false) %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false)
store i64 %result, i64* %out store i64 %result, i64* %out
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_noret_i64(i64* %ptr) nounwind { define amdgpu_kernel void @flat_atomic_inc_noret_i64(i64* %ptr) nounwind {
; GCN-LABEL: flat_atomic_inc_noret_i64: ; GCN-LABEL: flat_atomic_inc_noret_i64:
; GCN: ; %bb.0: ; GCN: ; %bb.0:
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v2, 42 ; GCN-NEXT: v_mov_b32_e32 v2, 42
; GCN-NEXT: v_mov_b32_e32 v3, 0 ; GCN-NEXT: v_mov_b32_e32 v3, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: v_mov_b32_e32 v0, s0 ; GCN-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: v_mov_b32_e32 v1, s1 ; GCN-NEXT: v_mov_b32_e32 v1, s1
; GCN-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc ; GCN-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc
; GCN-NEXT: s_endpgm ; GCN-NEXT: s_endpgm
%result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %ptr, i64 42, i32 0, i32 0, i1 false) %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %ptr, i64 42, i32 0, i32 0, i1 false)
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset(i64* %ptr) nounwind { define amdgpu_kernel void @flat_atomic_inc_noret_i64_offset(i64* %ptr) nounwind {
; GCN-LABEL: flat_atomic_inc_noret_i64_offset: ; CI-LABEL: flat_atomic_inc_noret_i64_offset:
; GCN: ; %bb.0: ; CI: ; %bb.0:
; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; CI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GCN-NEXT: v_mov_b32_e32 v2, 42 ; CI-NEXT: v_mov_b32_e32 v2, 42
; GCN-NEXT: v_mov_b32_e32 v3, 0 ; CI-NEXT: v_mov_b32_e32 v3, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0) ; CI-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_add_u32 s0, s0, 32 ; CI-NEXT: s_add_u32 s0, s0, 32
; GCN-NEXT: s_addc_u32 s1, s1, 0 ; CI-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: v_mov_b32_e32 v0, s0 ; CI-NEXT: v_mov_b32_e32 v0, s0
; GCN-NEXT: v_mov_b32_e32 v1, s1 ; CI-NEXT: v_mov_b32_e32 v1, s1
; GCN-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc ; CI-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc
; GCN-NEXT: s_endpgm ; CI-NEXT: s_endpgm
;
; VI-LABEL: flat_atomic_inc_noret_i64_offset:
; VI: ; %bb.0:
; VI-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; VI-NEXT: v_mov_b32_e32 v2, 42
; VI-NEXT: v_mov_b32_e32 v3, 0
; VI-NEXT: s_waitcnt lgkmcnt(0)
; VI-NEXT: s_add_u32 s0, s0, 32
; VI-NEXT: s_addc_u32 s1, s1, 0
; VI-NEXT: v_mov_b32_e32 v0, s0
; VI-NEXT: v_mov_b32_e32 v1, s1
; VI-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc
; VI-NEXT: s_endpgm
;
; GFX9-LABEL: flat_atomic_inc_noret_i64_offset:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX9-NEXT: v_mov_b32_e32 v0, 42
; GFX9-NEXT: v_mov_b32_e32 v1, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v3, s1
; GFX9-NEXT: v_mov_b32_e32 v2, s0
; GFX9-NEXT: flat_atomic_inc_x2 v[0:1], v[2:3], v[0:1] offset:32 glc
; GFX9-NEXT: s_endpgm
%gep = getelementptr i64, i64* %ptr, i32 4 %gep = getelementptr i64, i64* %ptr, i32 4
%result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false) %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false)
ret void ret void
} }
define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(i64* %out, i64* %ptr) #0 { define amdgpu_kernel void @flat_atomic_inc_ret_i64_offset_addr64(i64* %out, i64* %ptr) #0 {
; CI-LABEL: flat_atomic_inc_ret_i64_offset_addr64: ; CI-LABEL: flat_atomic_inc_ret_i64_offset_addr64:
; CI: ; %bb.0: ; CI: ; %bb.0:
Show All 37 Lines
; VI-NEXT: flat_atomic_inc_x2 v[2:3], v[2:3], v[4:5] glc ; VI-NEXT: flat_atomic_inc_x2 v[2:3], v[2:3], v[4:5] glc
; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; VI-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3] ; VI-NEXT: flat_store_dwordx2 v[0:1], v[2:3]
; VI-NEXT: s_endpgm ; VI-NEXT: s_endpgm
; ;
; GFX9-LABEL: flat_atomic_inc_ret_i64_offset_addr64: ; GFX9-LABEL: flat_atomic_inc_ret_i64_offset_addr64:
; GFX9: ; %bb.0: ; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 ; GFX9-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 3, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v4, 3, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s2 ; GFX9-NEXT: v_mov_b32_e32 v0, s2
; GFX9-NEXT: v_mov_b32_e32 v3, s1
; GFX9-NEXT: v_mov_b32_e32 v1, s3 ; GFX9-NEXT: v_mov_b32_e32 v1, s3
; GFX9-NEXT: v_add_co_u32_e32 v3, vcc, v0, v2 ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v4
; GFX9-NEXT: v_addc_co_u32_e32 v4, vcc, 0, v1, vcc
; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, 40, v3 ; GFX9-NEXT: v_mov_b32_e32 v2, s0
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v4, vcc ; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v4
; GFX9-NEXT: v_mov_b32_e32 v4, 42 ; GFX9-NEXT: v_mov_b32_e32 v4, 42
; GFX9-NEXT: v_mov_b32_e32 v5, 0 ; GFX9-NEXT: v_mov_b32_e32 v5, 0
; GFX9-NEXT: flat_atomic_inc_x2 v[2:3], v[2:3], v[4:5] glc ; GFX9-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[4:5] offset:40 glc
; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v3, vcc
; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) ; GFX9-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0)
; GFX9-NEXT: flat_store_dwordx2 v[0:1], v[2:3] ; GFX9-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x() %id = call i32 @llvm.amdgcn.workitem.id.x()
%gep.tid = getelementptr i64, i64* %ptr, i32 %id %gep.tid = getelementptr i64, i64* %ptr, i32 %id
%out.gep = getelementptr i64, i64* %out, i32 %id %out.gep = getelementptr i64, i64* %out, i32 %id
%gep = getelementptr i64, i64* %gep.tid, i32 5 %gep = getelementptr i64, i64* %gep.tid, i32 5
%result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false) %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false)
store i64 %result, i64* %out.gep store i64 %result, i64* %out.gep
ret void ret void
Show All 35 Lines
; GFX9-LABEL: flat_atomic_inc_noret_i64_offset_addr64: ; GFX9-LABEL: flat_atomic_inc_noret_i64_offset_addr64:
; GFX9: ; %bb.0: ; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; GFX9-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX9-NEXT: v_lshlrev_b32_e32 v2, 3, v0 ; GFX9-NEXT: v_lshlrev_b32_e32 v2, 3, v0
; GFX9-NEXT: s_waitcnt lgkmcnt(0) ; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: v_mov_b32_e32 v0, s0 ; GFX9-NEXT: v_mov_b32_e32 v0, s0
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2 ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, v0, v2
; GFX9-NEXT: v_mov_b32_e32 v1, s1 ; GFX9-NEXT: v_mov_b32_e32 v1, s1
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 40, v0
; GFX9-NEXT: v_mov_b32_e32 v2, 42 ; GFX9-NEXT: v_mov_b32_e32 v2, 42
; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc ; GFX9-NEXT: v_addc_co_u32_e32 v1, vcc, 0, v1, vcc
; GFX9-NEXT: v_mov_b32_e32 v3, 0 ; GFX9-NEXT: v_mov_b32_e32 v3, 0
; GFX9-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] glc ; GFX9-NEXT: flat_atomic_inc_x2 v[0:1], v[0:1], v[2:3] offset:40 glc
; GFX9-NEXT: s_endpgm ; GFX9-NEXT: s_endpgm
%id = call i32 @llvm.amdgcn.workitem.id.x() %id = call i32 @llvm.amdgcn.workitem.id.x()
%gep.tid = getelementptr i64, i64* %ptr, i32 %id %gep.tid = getelementptr i64, i64* %ptr, i32 %id
%gep = getelementptr i64, i64* %gep.tid, i32 5 %gep = getelementptr i64, i64* %gep.tid, i32 5
%result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false) %result = call i64 @llvm.amdgcn.atomic.inc.i64.p0i64(i64* %gep, i64 42, i32 0, i32 0, i1 false)
ret void ret void
} }
▲ Show 20 Lines • Show All 67 Lines • Show Last 20 Lines

llvm/test/CodeGen/AMDGPU/GlobalISel/llvm.amdgcn.global.atomic.fadd.ll

Show All 10 Lines; GFX908-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %ptr, float %data) %ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %ptr, float %data)
ret void ret void
} }
define void @global_atomic_fadd_f32_off_2048(float addrspace(1)* %ptr, float %data) { define void @global_atomic_fadd_f32_off_2048(float addrspace(1)* %ptr, float %data) {
; GFX908-LABEL: global_atomic_fadd_f32_off_2048: ; GFX908-LABEL: global_atomic_fadd_f32_off_2048:
; GFX908: ; %bb.0: ; GFX908: ; %bb.0:
; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX908-NEXT: s_movk_i32 s4, 0x800 ; GFX908-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2048
; GFX908-NEXT: s_mov_b32 s5, 0
; GFX908-NEXT: v_mov_b32_e32 v3, s4
; GFX908-NEXT: v_mov_b32_e32 v4, s5
; GFX908-NEXT: v_add_co_u32_e32 v0, vcc, v0, v3
; GFX908-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v4, vcc
; GFX908-NEXT: global_atomic_add_f32 v[0:1], v2, off
; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0)
; GFX908-NEXT: s_setpc_b64 s[30:31] ; GFX908-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr float, float addrspace(1)* %ptr, i64 512 %gep = getelementptr float, float addrspace(1)* %ptr, i64 512
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %gep, float %data) %ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %gep, float %data)
ret void ret void
} }
define void @global_atomic_fadd_f32_off_neg2047(float addrspace(1)* %ptr, float %data) { define void @global_atomic_fadd_f32_off_neg2047(float addrspace(1)* %ptr, float %data) {
; GFX908-LABEL: global_atomic_fadd_f32_off_neg2047: ; GFX908-LABEL: global_atomic_fadd_f32_off_neg2047:
; GFX908: ; %bb.0: ; GFX908: ; %bb.0:
; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX908-NEXT: s_mov_b32 s4, 0xfffff804 ; GFX908-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:-2044
; GFX908-NEXT: s_mov_b32 s5, -1
; GFX908-NEXT: v_mov_b32_e32 v3, s4
; GFX908-NEXT: v_mov_b32_e32 v4, s5
; GFX908-NEXT: v_add_co_u32_e32 v0, vcc, v0, v3
; GFX908-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v4, vcc
; GFX908-NEXT: global_atomic_add_f32 v[0:1], v2, off
; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0)
; GFX908-NEXT: s_setpc_b64 s[30:31] ; GFX908-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr float, float addrspace(1)* %ptr, i64 -511 %gep = getelementptr float, float addrspace(1)* %ptr, i64 -511
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %gep, float %data) %ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %gep, float %data)
ret void ret void
} }
define amdgpu_kernel void @global_atomic_fadd_f32_off_ss(float addrspace(1)* %ptr, float %data) { define amdgpu_kernel void @global_atomic_fadd_f32_off_ss(float addrspace(1)* %ptr, float %data) {
; GFX908-LABEL: global_atomic_fadd_f32_off_ss: ; GFX908-LABEL: global_atomic_fadd_f32_off_ss:
; GFX908: ; %bb.0: ; GFX908: ; %bb.0:
; GFX908-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0 ; GFX908-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x0
; GFX908-NEXT: s_load_dword s2, s[4:5], 0x8 ; GFX908-NEXT: s_load_dword s2, s[4:5], 0x8
; GFX908-NEXT: s_waitcnt lgkmcnt(0) ; GFX908-NEXT: s_waitcnt lgkmcnt(0)
; GFX908-NEXT: s_add_u32 s0, s0, 0x800
; GFX908-NEXT: s_addc_u32 s1, s1, 0
; GFX908-NEXT: v_mov_b32_e32 v0, s0 ; GFX908-NEXT: v_mov_b32_e32 v0, s0
; GFX908-NEXT: v_mov_b32_e32 v1, s1
; GFX908-NEXT: v_mov_b32_e32 v2, s2 ; GFX908-NEXT: v_mov_b32_e32 v2, s2
; GFX908-NEXT: global_atomic_add_f32 v[0:1], v2, off ; GFX908-NEXT: v_mov_b32_e32 v1, s1
; GFX908-NEXT: global_atomic_add_f32 v[0:1], v2, off offset:2048
; GFX908-NEXT: s_endpgm ; GFX908-NEXT: s_endpgm
%gep = getelementptr float, float addrspace(1)* %ptr, i64 512 %gep = getelementptr float, float addrspace(1)* %ptr, i64 512
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %gep, float %data) %ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %gep, float %data)
ret void ret void
} }
define void @global_atomic_fadd_v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data) { define void @global_atomic_fadd_v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data) {
; GFX908-LABEL: global_atomic_fadd_v2f16: ; GFX908-LABEL: global_atomic_fadd_v2f16:
; GFX908: ; %bb.0: ; GFX908: ; %bb.0:
; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX908-NEXT: global_atomic_pk_add_f16 v[0:1], v2, off ; GFX908-NEXT: global_atomic_pk_add_f16 v[0:1], v2, off
; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0)
; GFX908-NEXT: s_setpc_b64 s[30:31] ; GFX908-NEXT: s_setpc_b64 s[30:31]
%ret = call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data) %ret = call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret void ret void
} }
define void @global_atomic_fadd_v2f16_off_neg2047(<2 x half> addrspace(1)* %ptr, <2 x half> %data) { define void @global_atomic_fadd_v2f16_off_neg2047(<2 x half> addrspace(1)* %ptr, <2 x half> %data) {
; GFX908-LABEL: global_atomic_fadd_v2f16_off_neg2047: ; GFX908-LABEL: global_atomic_fadd_v2f16_off_neg2047:
; GFX908: ; %bb.0: ; GFX908: ; %bb.0:
; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX908-NEXT: s_mov_b32 s4, 0xfffff804 ; GFX908-NEXT: global_atomic_pk_add_f16 v[0:1], v2, off offset:-2044
; GFX908-NEXT: s_mov_b32 s5, -1
; GFX908-NEXT: v_mov_b32_e32 v3, s4
; GFX908-NEXT: v_mov_b32_e32 v4, s5
; GFX908-NEXT: v_add_co_u32_e32 v0, vcc, v0, v3
; GFX908-NEXT: v_addc_co_u32_e32 v1, vcc, v1, v4, vcc
; GFX908-NEXT: global_atomic_pk_add_f16 v[0:1], v2, off
; GFX908-NEXT: s_waitcnt vmcnt(0) ; GFX908-NEXT: s_waitcnt vmcnt(0)
; GFX908-NEXT: s_setpc_b64 s[30:31] ; GFX908-NEXT: s_setpc_b64 s[30:31]
%gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %ptr, i64 -511 %gep = getelementptr <2 x half>, <2 x half> addrspace(1)* %ptr, i64 -511
%ret = call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %gep, <2 x half> %data) %ret = call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %gep, <2 x half> %data)
ret void ret void
} }
declare float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* nocapture, float) #0 declare float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* nocapture, float) #0
declare <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* nocapture, <2 x half>) #0 declare <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* nocapture, <2 x half>) #0
attributes #0 = { argmemonly nounwind willreturn } attributes #0 = { argmemonly nounwind willreturn }

llvm/test/CodeGen/AMDGPU/GlobalISel/load-constant.96.ll

Show All 12 Lines
; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-UNALIGNED-NEXT: global_load_dwordx3 v[0:2], v[0:1], off ; GFX9-UNALIGNED-NEXT: global_load_dwordx3 v[0:2], v[0:1], off
; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0) ; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0)
; GFX9-UNALIGNED-NEXT: s_setpc_b64 s[30:31] ; GFX9-UNALIGNED-NEXT: s_setpc_b64 s[30:31]
; ;
; GFX9-NOUNALIGNED-LABEL: v_load_constant_v3i32_align1: ; GFX9-NOUNALIGNED-LABEL: v_load_constant_v3i32_align1:
; GFX9-NOUNALIGNED: ; %bb.0: ; GFX9-NOUNALIGNED: ; %bb.0:
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NOUNALIGNED-NEXT: v_add_co_u32_e32 v2, vcc, 11, v0 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v2, v[0:1], off
; GFX9-NOUNALIGNED-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v1, vcc ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v3, v[0:1], off offset:1
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v0, v[0:1], off ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v4, v[0:1], off offset:2
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v1, v[2:3], off offset:-10 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v5, v[0:1], off offset:3
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v4, v[2:3], off offset:-9 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v6, v[0:1], off offset:4
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v5, v[2:3], off offset:-8 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v7, v[0:1], off offset:5
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v6, v[2:3], off offset:-7 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v8, v[0:1], off offset:6
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v7, v[2:3], off offset:-6 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v9, v[0:1], off offset:7
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v8, v[2:3], off offset:-5 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v10, v[0:1], off offset:8
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v9, v[2:3], off offset:-4 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v11, v[0:1], off offset:9
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v10, v[2:3], off offset:-3 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v12, v[0:1], off offset:10
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v11, v[2:3], off offset:-2 ; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v0, v[0:1], off offset:11
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v12, v[2:3], off offset:-1 ; GFX9-NOUNALIGNED-NEXT: v_mov_b32_e32 v1, 0xff
; GFX9-NOUNALIGNED-NEXT: global_load_ubyte v2, v[2:3], off
; GFX9-NOUNALIGNED-NEXT: v_mov_b32_e32 v3, 0xff
; GFX9-NOUNALIGNED-NEXT: s_movk_i32 s4, 0xff ; GFX9-NOUNALIGNED-NEXT: s_movk_i32 s4, 0xff
; GFX9-NOUNALIGNED-NEXT: v_mov_b32_e32 v13, 8
; GFX9-NOUNALIGNED-NEXT: s_mov_b32 s5, 8 ; GFX9-NOUNALIGNED-NEXT: s_mov_b32 s5, 8
; GFX9-NOUNALIGNED-NEXT: v_mov_b32_e32 v13, 8
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(10) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(10)
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_sdwa v1, s5, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_sdwa v3, s5, v3 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(9) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(9)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v4, s4, v4 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v4, s4, v4
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(8) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(8)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v5, s4, v5 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v5, s4, v5
; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v0, v0, s4, v1 ; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v2, v2, s4, v3
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(6) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(6)
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_sdwa v7, v13, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_sdwa v7, s5, v7 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(5) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(5)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v8, v8, v3 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v8, v8, v1
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(4) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(4)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v9, v9, v3 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v9, v9, v1
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v1, 16, v4 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v3, 16, v4
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(2) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(2)
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_sdwa v11, v13, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_sdwa v11, v13, v11 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(1) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(1)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v12, v12, v3 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v12, v12, v1
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v2, v2, v3 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v0, v0, v1
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v4, 24, v5 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v4, 24, v5
; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v5, v6, v3, v7 ; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v5, v6, s4, v7
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v6, 16, v8 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v6, 16, v8
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v7, 24, v9 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v7, 24, v9
; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v3, v10, v3, v11 ; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v8, v10, v1, v11
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v8, 16, v12 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v10, 24, v0
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v2, 24, v2 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v9, 16, v12
; GFX9-NOUNALIGNED-NEXT: v_or3_b32 v0, v0, v1, v4 ; GFX9-NOUNALIGNED-NEXT: v_or3_b32 v0, v2, v3, v4
; GFX9-NOUNALIGNED-NEXT: v_or3_b32 v1, v5, v6, v7 ; GFX9-NOUNALIGNED-NEXT: v_or3_b32 v1, v5, v6, v7
; GFX9-NOUNALIGNED-NEXT: v_or3_b32 v2, v3, v8, v2 ; GFX9-NOUNALIGNED-NEXT: v_or3_b32 v2, v8, v9, v10
; GFX9-NOUNALIGNED-NEXT: s_setpc_b64 s[30:31] ; GFX9-NOUNALIGNED-NEXT: s_setpc_b64 s[30:31]
; ;
; GFX7-UNALIGNED-LABEL: v_load_constant_v3i32_align1: ; GFX7-UNALIGNED-LABEL: v_load_constant_v3i32_align1:
; GFX7-UNALIGNED: ; %bb.0: ; GFX7-UNALIGNED: ; %bb.0:
; GFX7-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-UNALIGNED-NEXT: s_mov_b32 s6, 0 ; GFX7-UNALIGNED-NEXT: s_mov_b32 s6, 0
; GFX7-UNALIGNED-NEXT: s_mov_b32 s7, 0xf000 ; GFX7-UNALIGNED-NEXT: s_mov_b32 s7, 0xf000
; GFX7-UNALIGNED-NEXT: s_mov_b64 s[4:5], 0 ; GFX7-UNALIGNED-NEXT: s_mov_b64 s[4:5], 0
▲ Show 20 Lines • Show All 74 Lines • ▼ Show 20 Lines
; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-UNALIGNED-NEXT: global_load_dwordx3 v[0:2], v[0:1], off ; GFX9-UNALIGNED-NEXT: global_load_dwordx3 v[0:2], v[0:1], off
; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0) ; GFX9-UNALIGNED-NEXT: s_waitcnt vmcnt(0)
; GFX9-UNALIGNED-NEXT: s_setpc_b64 s[30:31] ; GFX9-UNALIGNED-NEXT: s_setpc_b64 s[30:31]
; ;
; GFX9-NOUNALIGNED-LABEL: v_load_constant_v3i32_align2: ; GFX9-NOUNALIGNED-LABEL: v_load_constant_v3i32_align2:
; GFX9-NOUNALIGNED: ; %bb.0: ; GFX9-NOUNALIGNED: ; %bb.0:
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX9-NOUNALIGNED-NEXT: v_add_co_u32_e32 v2, vcc, 10, v0 ; GFX9-NOUNALIGNED-NEXT: global_load_ushort v2, v[0:1], off
; GFX9-NOUNALIGNED-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v1, vcc ; GFX9-NOUNALIGNED-NEXT: global_load_ushort v3, v[0:1], off offset:2
; GFX9-NOUNALIGNED-NEXT: global_load_ushort v0, v[0:1], off ; GFX9-NOUNALIGNED-NEXT: global_load_ushort v4, v[0:1], off offset:4
; GFX9-NOUNALIGNED-NEXT: global_load_ushort v1, v[2:3], off offset:-8 ; GFX9-NOUNALIGNED-NEXT: global_load_ushort v5, v[0:1], off offset:6
; GFX9-NOUNALIGNED-NEXT: global_load_ushort v4, v[2:3], off offset:-6 ; GFX9-NOUNALIGNED-NEXT: global_load_ushort v6, v[0:1], off offset:8
; GFX9-NOUNALIGNED-NEXT: global_load_ushort v5, v[2:3], off offset:-4 ; GFX9-NOUNALIGNED-NEXT: global_load_ushort v0, v[0:1], off offset:10
; GFX9-NOUNALIGNED-NEXT: global_load_ushort v6, v[2:3], off offset:-2
; GFX9-NOUNALIGNED-NEXT: global_load_ushort v2, v[2:3], off
; GFX9-NOUNALIGNED-NEXT: v_mov_b32_e32 v3, 0xffff
; GFX9-NOUNALIGNED-NEXT: s_mov_b32 s4, 0xffff ; GFX9-NOUNALIGNED-NEXT: s_mov_b32 s4, 0xffff
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(4) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(4)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v1, s4, v1 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v1, s4, v3
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v1, 16, v1 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v1, 16, v1
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(2) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(2)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v5, v5, v3 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v3, s4, v5
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v5, 16, v5 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v3, 16, v3
; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0) ; GFX9-NOUNALIGNED-NEXT: s_waitcnt vmcnt(0)
; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v2, v2, v3 ; GFX9-NOUNALIGNED-NEXT: v_and_b32_e32 v0, s4, v0
; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v2, 16, v2 ; GFX9-NOUNALIGNED-NEXT: v_lshlrev_b32_e32 v5, 16, v0
; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v0, v0, s4, v1 ; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v0, v2, s4, v1
; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v1, v4, v3, v5 ; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v1, v4, s4, v3
; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v2, v6, v3, v2 ; GFX9-NOUNALIGNED-NEXT: v_and_or_b32 v2, v6, s4, v5
; GFX9-NOUNALIGNED-NEXT: s_setpc_b64 s[30:31] ; GFX9-NOUNALIGNED-NEXT: s_setpc_b64 s[30:31]
; ;
; GFX7-UNALIGNED-LABEL: v_load_constant_v3i32_align2: ; GFX7-UNALIGNED-LABEL: v_load_constant_v3i32_align2:
; GFX7-UNALIGNED: ; %bb.0: ; GFX7-UNALIGNED: ; %bb.0:
; GFX7-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-UNALIGNED-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-UNALIGNED-NEXT: s_mov_b32 s6, 0 ; GFX7-UNALIGNED-NEXT: s_mov_b32 s6, 0
; GFX7-UNALIGNED-NEXT: s_mov_b32 s7, 0xf000 ; GFX7-UNALIGNED-NEXT: s_mov_b32 s7, 0xf000
; GFX7-UNALIGNED-NEXT: s_mov_b64 s[4:5], 0 ; GFX7-UNALIGNED-NEXT: s_mov_b64 s[4:5], 0
▲ Show 20 Lines • Show All 568 Lines • Show Last 20 Lines