RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 02:33:06 +01:00
Code Issues Projects Releases Wiki Activity

[AMDGPU] Init scratch only if necessary

Browse Source 
If no scratch or flat instructions are used, we do not need to
initialize the flat scratch hardware register.

Differential Revision: https://reviews.llvm.org/D105920
This commit is contained in:
Sebastian Neubauer 2021-07-14 10:03:54 +02:00
parent 5a48184106
commit 811bfa448e
15 changed files with 52 additions and 150 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
lib/Target/AMDGPU/SIFrameLowering.cpp
View File

@ -413,6 +413,7 @@ void SIFrameLowering::emitEntryFunctionPrologue(MachineFunction &MF,
const SIRegisterInfo *TRI = &TII->getRegisterInfo();
MachineRegisterInfo &MRI = MF.getRegInfo();
const Function &F = MF.getFunction();
MachineFrameInfo &FrameInfo = MF.getFrameInfo();
assert(MFI->isEntryFunction());
@ -490,7 +491,7 @@ void SIFrameLowering::emitEntryFunctionPrologue(MachineFunction &MF,
Register SPReg = MFI->getStackPtrOffsetReg();
assert(SPReg != AMDGPU::SP_REG);
BuildMI(MBB, I, DL, TII->get(AMDGPU::S_MOV_B32), SPReg)
.addImm(MF.getFrameInfo().getStackSize() * getScratchScaleFactor(ST));
.addImm(FrameInfo.getStackSize() * getScratchScaleFactor(ST));
}
if (hasFP(MF)) {
@ -499,13 +500,18 @@ void SIFrameLowering::emitEntryFunctionPrologue(MachineFunction &MF,
BuildMI(MBB, I, DL, TII->get(AMDGPU::S_MOV_B32), FPReg).addImm(0);
}
if ((MFI->hasFlatScratchInit() || ScratchRsrcReg) &&
bool NeedsFlatScratchInit =
MFI->hasFlatScratchInit() &&
(MRI.isPhysRegUsed(AMDGPU::FLAT_SCR) || FrameInfo.hasCalls() ||
(!allStackObjectsAreDead(FrameInfo) && ST.enableFlatScratch()));
if ((NeedsFlatScratchInit || ScratchRsrcReg) &&
!ST.flatScratchIsArchitected()) {
MRI.addLiveIn(PreloadedScratchWaveOffsetReg);
MBB.addLiveIn(PreloadedScratchWaveOffsetReg);
}
if (MFI->hasFlatScratchInit()) {
if (NeedsFlatScratchInit) {
emitEntryFunctionFlatScratchInit(MF, MBB, I, DL, ScratchWaveOffsetReg);
}

10
lib/Target/AMDGPU/SIMachineFunctionInfo.cpp
View File

@ -178,14 +178,14 @@ SIMachineFunctionInfo::SIMachineFunctionInfo(const MachineFunction &MF)
if (UseFixedABI || F.hasFnAttribute("amdgpu-kernarg-segment-ptr"))
KernargSegmentPtr = true;
// TODO: This could be refined a lot. The attribute is a poor way of
// detecting calls or stack objects that may require it before argument
// lowering.
if (ST.hasFlatAddressSpace() && isEntryFunction() &&
(isAmdHsaOrMesa || ST.enableFlatScratch()) &&
(HasCalls || HasStackObjects || ST.enableFlatScratch()) &&
!ST.flatScratchIsArchitected()) {
// TODO: This could be refined a lot. The attribute is a poor way of
// detecting calls or stack objects that may require it before argument
// lowering.
if (HasCalls || HasStackObjects || ST.enableFlatScratch())
FlatScratchInit = true;
FlatScratchInit = true;
}
Attribute A = F.getFnAttribute("amdgpu-git-ptr-high");

42
test/CodeGen/AMDGPU/GlobalISel/dynamic-alloca-uniform.ll
View File

@ -8,17 +8,15 @@ define amdgpu_kernel void @kernel_dynamic_stackalloc_sgpr_align4(i32 %n) {
; GFX9-LABEL: kernel_dynamic_stackalloc_sgpr_align4:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dword s4, s[4:5], 0x0
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; GFX9-NEXT: s_add_u32 s0, s0, s9
; GFX9-NEXT: s_addc_u32 s1, s1, 0
; GFX9-NEXT: s_movk_i32 s32, 0x400
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_lshl2_add_u32 s4, s4, 15
; GFX9-NEXT: s_and_b32 s4, s4, -16
; GFX9-NEXT: s_movk_i32 s32, 0x400
; GFX9-NEXT: s_lshl_b32 s4, s4, 6
; GFX9-NEXT: s_add_u32 s4, s32, s4
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: s_mov_b32 s33, 0
; GFX9-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen
@ -26,16 +24,12 @@ define amdgpu_kernel void @kernel_dynamic_stackalloc_sgpr_align4(i32 %n) {
;
; GFX10-LABEL: kernel_dynamic_stackalloc_sgpr_align4:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_add_u32 s6, s6, s9
; GFX10-NEXT: s_movk_i32 s32, 0x200
; GFX10-NEXT: s_mov_b32 s33, 0
; GFX10-NEXT: s_addc_u32 s7, s7, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s6
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s7
; GFX10-NEXT: s_load_dword s4, s[4:5], 0x0
; GFX10-NEXT: s_add_u32 s0, s0, s9
; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: s_movk_i32 s32, 0x200
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_mov_b32 s33, 0
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: s_lshl2_add_u32 s4, s4, 15
; GFX10-NEXT: s_and_b32 s4, s4, -16
@ -110,17 +104,15 @@ define amdgpu_kernel void @kernel_dynamic_stackalloc_sgpr_align16(i32 %n) {
; GFX9-LABEL: kernel_dynamic_stackalloc_sgpr_align16:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dword s4, s[4:5], 0x0
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; GFX9-NEXT: s_add_u32 s0, s0, s9
; GFX9-NEXT: s_addc_u32 s1, s1, 0
; GFX9-NEXT: s_movk_i32 s32, 0x400
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_lshl2_add_u32 s4, s4, 15
; GFX9-NEXT: s_and_b32 s4, s4, -16
; GFX9-NEXT: s_movk_i32 s32, 0x400
; GFX9-NEXT: s_lshl_b32 s4, s4, 6
; GFX9-NEXT: s_add_u32 s4, s32, s4
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: s_mov_b32 s33, 0
; GFX9-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen
@ -128,16 +120,12 @@ define amdgpu_kernel void @kernel_dynamic_stackalloc_sgpr_align16(i32 %n) {
;
; GFX10-LABEL: kernel_dynamic_stackalloc_sgpr_align16:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_add_u32 s6, s6, s9
; GFX10-NEXT: s_movk_i32 s32, 0x200
; GFX10-NEXT: s_mov_b32 s33, 0
; GFX10-NEXT: s_addc_u32 s7, s7, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s6
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s7
; GFX10-NEXT: s_load_dword s4, s[4:5], 0x0
; GFX10-NEXT: s_add_u32 s0, s0, s9
; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: s_movk_i32 s32, 0x200
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_mov_b32 s33, 0
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: s_lshl2_add_u32 s4, s4, 15
; GFX10-NEXT: s_and_b32 s4, s4, -16
@ -212,18 +200,16 @@ define amdgpu_kernel void @kernel_dynamic_stackalloc_sgpr_align32(i32 %n) {
; GFX9-LABEL: kernel_dynamic_stackalloc_sgpr_align32:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_load_dword s4, s[4:5], 0x0
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; GFX9-NEXT: s_add_u32 s0, s0, s9
; GFX9-NEXT: s_addc_u32 s1, s1, 0
; GFX9-NEXT: s_movk_i32 s32, 0x800
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: s_waitcnt lgkmcnt(0)
; GFX9-NEXT: s_lshl2_add_u32 s4, s4, 15
; GFX9-NEXT: s_and_b32 s4, s4, -16
; GFX9-NEXT: s_movk_i32 s32, 0x800
; GFX9-NEXT: s_lshl_b32 s4, s4, 6
; GFX9-NEXT: s_add_u32 s4, s32, s4
; GFX9-NEXT: s_and_b32 s4, s4, 0xfffff800
; GFX9-NEXT: v_mov_b32_e32 v0, 0
; GFX9-NEXT: v_mov_b32_e32 v1, s4
; GFX9-NEXT: s_mov_b32 s33, 0
; GFX9-NEXT: buffer_store_dword v0, v1, s[0:3], 0 offen
@ -231,16 +217,12 @@ define amdgpu_kernel void @kernel_dynamic_stackalloc_sgpr_align32(i32 %n) {
;
; GFX10-LABEL: kernel_dynamic_stackalloc_sgpr_align32:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_add_u32 s6, s6, s9
; GFX10-NEXT: s_movk_i32 s32, 0x400
; GFX10-NEXT: s_mov_b32 s33, 0
; GFX10-NEXT: s_addc_u32 s7, s7, 0
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s6
; GFX10-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s7
; GFX10-NEXT: s_load_dword s4, s[4:5], 0x0
; GFX10-NEXT: s_add_u32 s0, s0, s9
; GFX10-NEXT: s_addc_u32 s1, s1, 0
; GFX10-NEXT: s_movk_i32 s32, 0x400
; GFX10-NEXT: v_mov_b32_e32 v0, 0
; GFX10-NEXT: s_mov_b32 s33, 0
; GFX10-NEXT: s_waitcnt lgkmcnt(0)
; GFX10-NEXT: s_lshl2_add_u32 s4, s4, 15
; GFX10-NEXT: s_and_b32 s4, s4, -16

5
test/CodeGen/AMDGPU/GlobalISel/flat-scratch-init.ll
View File

@ -28,8 +28,6 @@ define amdgpu_kernel void @stack_object_addrspacecast_in_kernel_no_calls() {
; TODO: Could optimize out in this case
; GCN-LABEL: {{^}}stack_object_in_kernel_no_calls:
; RW-FLAT: s_add_u32 flat_scratch_lo, s4, s7
; RW-FLAT: s_addc_u32 flat_scratch_hi, s5, 0
; RO-FLAT-NOT: flat_scratch
; RW-FLAT: buffer_store_dword
; RO-FLAT: scratch_store_dword
@ -41,7 +39,8 @@ define amdgpu_kernel void @stack_object_addrspacecast_in_kernel_no_calls() {
; RW-FLAT-NOT: .amdhsa_enable_private_segment
; RO-FLAT-NOT: .amdhsa_system_sgpr_private_segment_wavefront_offset
; RO-FLAT: .amdhsa_enable_private_segment 1
; GCN-NOT: .amdhsa_reserve_flat_scratch
; RW-FLAT: .amdhsa_reserve_flat_scratch 0
; RO-FLAT-NOT: .amdhsa_reserve_flat_scratch
; GCN: COMPUTE_PGM_RSRC2:SCRATCH_EN: 1
; RW-FLAT: COMPUTE_PGM_RSRC2:USER_SGPR: 6
; RO-FLAT: COMPUTE_PGM_RSRC2:USER_SGPR: 0

8
test/CodeGen/AMDGPU/GlobalISel/non-entry-alloca.ll
View File

@ -13,15 +13,13 @@
define amdgpu_kernel void @kernel_non_entry_block_static_alloca_uniformly_reached_align4(i32 addrspace(1)* %out, i32 %arg.cond0, i32 %arg.cond1, i32 %in) {
; GCN-LABEL: kernel_non_entry_block_static_alloca_uniformly_reached_align4:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; GCN-NEXT: s_load_dword s6, s[4:5], 0x8
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; GCN-NEXT: s_add_u32 s0, s0, s9
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_movk_i32 s32, 0x400
; GCN-NEXT: s_mov_b32 s33, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_cmp_lg_u32 s6, 0
; GCN-NEXT: s_mov_b32 s33, 0
; GCN-NEXT: s_cbranch_scc1 BB0_3
; GCN-NEXT: ; %bb.1: ; %bb.0
; GCN-NEXT: s_load_dword s6, s[4:5], 0xc
@ -86,15 +84,13 @@ bb.2:
define amdgpu_kernel void @kernel_non_entry_block_static_alloca_uniformly_reached_align64(i32 addrspace(1)* %out, i32 %arg.cond, i32 %in) {
; GCN-LABEL: kernel_non_entry_block_static_alloca_uniformly_reached_align64:
; GCN: ; %bb.0: ; %entry
; GCN-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; GCN-NEXT: s_load_dword s6, s[4:5], 0x8
; GCN-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; GCN-NEXT: s_add_u32 s0, s0, s9
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_movk_i32 s32, 0x1000
; GCN-NEXT: s_mov_b32 s33, 0
; GCN-NEXT: s_waitcnt lgkmcnt(0)
; GCN-NEXT: s_cmp_lg_u32 s6, 0
; GCN-NEXT: s_mov_b32 s33, 0
; GCN-NEXT: s_cbranch_scc1 BB1_2
; GCN-NEXT: ; %bb.1: ; %bb.0
; GCN-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x0

9
test/CodeGen/AMDGPU/call-constant.ll
View File

@ -1,12 +1,7 @@
; RUN: llc -global-isel=0 -amdgpu-fixed-function-abi=0 -mtriple=amdgcn-amd-amdhsa < %s | FileCheck -check-prefixes=GCN,SDAG %s
; RUN: llc -global-isel=1 -amdgpu-fixed-function-abi=1 -mtriple=amdgcn-amd-amdhsa < %s | FileCheck -check-prefixes=GCN,GISEL %s
; FIXME: Emitting unnecessary flat_scratch setup
; GCN-LABEL: {{^}}test_call_undef:
; SDAG: s_mov_b32 flat_scratch_lo, s13
; SDAG: s_add_i32 s12, s12, s17
; SDAG: s_lshr_b32
; GCN: s_endpgm
define amdgpu_kernel void @test_call_undef() #0 {
%val = call i32 undef(i32 1)
@ -26,10 +21,6 @@ define i32 @test_tail_call_undef() #0 {
}
; GCN-LABEL: {{^}}test_call_null:
; SDAG: s_mov_b32 flat_scratch_lo, s13
; SDAG: s_add_i32 s12, s12, s17
; SDAG: s_lshr_b32
; GISEL: s_swappc_b64 s{{\[[0-9]+:[0-9]+\]}}, 0{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @test_call_null() #0 {

45
test/CodeGen/AMDGPU/cc-update.ll
View File

@ -22,20 +22,15 @@ entry:
define amdgpu_kernel void @test_kern_stack() local_unnamed_addr #0 {
; GFX803-LABEL: test_kern_stack:
; GFX803: ; %bb.0: ; %entry
; GFX803-NEXT: s_add_i32 s4, s4, s7
; GFX803-NEXT: s_lshr_b32 flat_scratch_hi, s4, 8
; GFX803-NEXT: s_add_u32 s0, s0, s7
; GFX803-NEXT: s_addc_u32 s1, s1, 0
; GFX803-NEXT: v_mov_b32_e32 v0, 0
; GFX803-NEXT: s_mov_b32 flat_scratch_lo, s5
; GFX803-NEXT: buffer_store_dword v0, off, s[0:3], 0 offset:4
; GFX803-NEXT: s_waitcnt vmcnt(0)
; GFX803-NEXT: s_endpgm
;
; GFX900-LABEL: test_kern_stack:
; GFX900: ; %bb.0: ; %entry
; GFX900-NEXT: s_add_u32 flat_scratch_lo, s4, s7
; GFX900-NEXT: s_addc_u32 flat_scratch_hi, s5, 0
; GFX900-NEXT: s_add_u32 s0, s0, s7
; GFX900-NEXT: s_addc_u32 s1, s1, 0
; GFX900-NEXT: v_mov_b32_e32 v0, 0
@ -45,10 +40,6 @@ define amdgpu_kernel void @test_kern_stack() local_unnamed_addr #0 {
;
; GFX1010-LABEL: test_kern_stack:
; GFX1010: ; %bb.0: ; %entry
; GFX1010-NEXT: s_add_u32 s4, s4, s7
; GFX1010-NEXT: s_addc_u32 s5, s5, 0
; GFX1010-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
; GFX1010-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
; GFX1010-NEXT: v_mov_b32_e32 v0, 0
; GFX1010-NEXT: s_add_u32 s0, s0, s7
; GFX1010-NEXT: s_addc_u32 s1, s1, 0
@ -188,21 +179,16 @@ entry:
define amdgpu_kernel void @test_force_fp_kern_stack() local_unnamed_addr #2 {
; GFX803-LABEL: test_force_fp_kern_stack:
; GFX803: ; %bb.0: ; %entry
; GFX803-NEXT: s_add_i32 s4, s4, s7
; GFX803-NEXT: s_lshr_b32 flat_scratch_hi, s4, 8
; GFX803-NEXT: s_add_u32 s0, s0, s7
; GFX803-NEXT: s_mov_b32 s33, 0
; GFX803-NEXT: s_addc_u32 s1, s1, 0
; GFX803-NEXT: v_mov_b32_e32 v0, 0
; GFX803-NEXT: s_mov_b32 flat_scratch_lo, s5
; GFX803-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:4
; GFX803-NEXT: s_waitcnt vmcnt(0)
; GFX803-NEXT: s_endpgm
;
; GFX900-LABEL: test_force_fp_kern_stack:
; GFX900: ; %bb.0: ; %entry
; GFX900-NEXT: s_add_u32 flat_scratch_lo, s4, s7
; GFX900-NEXT: s_addc_u32 flat_scratch_hi, s5, 0
; GFX900-NEXT: s_add_u32 s0, s0, s7
; GFX900-NEXT: s_mov_b32 s33, 0
; GFX900-NEXT: s_addc_u32 s1, s1, 0
@ -213,13 +199,9 @@ define amdgpu_kernel void @test_force_fp_kern_stack() local_unnamed_addr #2 {
;
; GFX1010-LABEL: test_force_fp_kern_stack:
; GFX1010: ; %bb.0: ; %entry
; GFX1010-NEXT: s_add_u32 s4, s4, s7
; GFX1010-NEXT: s_mov_b32 s33, 0
; GFX1010-NEXT: s_addc_u32 s5, s5, 0
; GFX1010-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
; GFX1010-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
; GFX1010-NEXT: v_mov_b32_e32 v0, 0
; GFX1010-NEXT: s_add_u32 s0, s0, s7
; GFX1010-NEXT: s_mov_b32 s33, 0
; GFX1010-NEXT: s_addc_u32 s1, s1, 0
; GFX1010-NEXT: buffer_store_dword v0, off, s[0:3], s33 offset:4
; GFX1010-NEXT: s_waitcnt_vscnt null, 0x0
@ -344,14 +326,11 @@ entry:
define amdgpu_kernel void @test_sgpr_offset_kernel() #1 {
; GFX803-LABEL: test_sgpr_offset_kernel:
; GFX803: ; %bb.0: ; %entry
; GFX803-NEXT: s_add_i32 s4, s4, s7
; GFX803-NEXT: s_lshr_b32 flat_scratch_hi, s4, 8
; GFX803-NEXT: s_add_u32 s0, s0, s7
; GFX803-NEXT: s_addc_u32 s1, s1, 0
; GFX803-NEXT: buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
; GFX803-NEXT: s_waitcnt vmcnt(0)
; GFX803-NEXT: s_mov_b32 s4, 0x40000
; GFX803-NEXT: s_mov_b32 flat_scratch_lo, s5
; GFX803-NEXT: buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; GFX803-NEXT: ;;#ASMSTART
; GFX803-NEXT: ;;#ASMEND
@ -364,18 +343,16 @@ define amdgpu_kernel void @test_sgpr_offset_kernel() #1 {
;
; GFX900-LABEL: test_sgpr_offset_kernel:
; GFX900: ; %bb.0: ; %entry
; GFX900-NEXT: s_add_u32 flat_scratch_lo, s4, s7
; GFX900-NEXT: s_addc_u32 flat_scratch_hi, s5, 0
; GFX900-NEXT: s_add_u32 s0, s0, s7
; GFX900-NEXT: s_addc_u32 s1, s1, 0
; GFX900-NEXT: buffer_load_dword v0, off, s[0:3], 0 offset:8 glc
; GFX900-NEXT: s_waitcnt vmcnt(0)
; GFX900-NEXT: s_mov_b32 s6, 0x40000
; GFX900-NEXT: buffer_store_dword v0, off, s[0:3], s6 ; 4-byte Folded Spill
; GFX900-NEXT: s_mov_b32 s4, 0x40000
; GFX900-NEXT: buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; GFX900-NEXT: ;;#ASMSTART
; GFX900-NEXT: ;;#ASMEND
; GFX900-NEXT: s_mov_b32 s6, 0x40000
; GFX900-NEXT: buffer_load_dword v0, off, s[0:3], s6 ; 4-byte Folded Reload
; GFX900-NEXT: s_mov_b32 s4, 0x40000
; GFX900-NEXT: buffer_load_dword v0, off, s[0:3], s4 ; 4-byte Folded Reload
; GFX900-NEXT: s_waitcnt vmcnt(0)
; GFX900-NEXT: buffer_store_dword v0, off, s[0:3], 0 offset:8
; GFX900-NEXT: s_waitcnt vmcnt(0)
@ -383,21 +360,17 @@ define amdgpu_kernel void @test_sgpr_offset_kernel() #1 {
;
; GFX1010-LABEL: test_sgpr_offset_kernel:
; GFX1010: ; %bb.0: ; %entry
; GFX1010-NEXT: s_add_u32 s4, s4, s7
; GFX1010-NEXT: s_addc_u32 s5, s5, 0
; GFX1010-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s4
; GFX1010-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s5
; GFX1010-NEXT: s_add_u32 s0, s0, s7
; GFX1010-NEXT: s_addc_u32 s1, s1, 0
; GFX1010-NEXT: s_mov_b32 s6, 0x20000
; GFX1010-NEXT: s_mov_b32 s4, 0x20000
; GFX1010-NEXT: buffer_load_dword v0, off, s[0:3], 0 offset:8 glc dlc
; GFX1010-NEXT: s_waitcnt vmcnt(0)
; GFX1010-NEXT: buffer_store_dword v0, off, s[0:3], s6 ; 4-byte Folded Spill
; GFX1010-NEXT: buffer_store_dword v0, off, s[0:3], s4 ; 4-byte Folded Spill
; GFX1010-NEXT: s_waitcnt_depctr 0xffe3
; GFX1010-NEXT: s_mov_b32 s6, 0x20000
; GFX1010-NEXT: s_mov_b32 s4, 0x20000
; GFX1010-NEXT: ;;#ASMSTART
; GFX1010-NEXT: ;;#ASMEND
; GFX1010-NEXT: buffer_load_dword v0, off, s[0:3], s6 ; 4-byte Folded Reload
; GFX1010-NEXT: buffer_load_dword v0, off, s[0:3], s4 ; 4-byte Folded Reload
; GFX1010-NEXT: s_waitcnt vmcnt(0)
; GFX1010-NEXT: buffer_store_dword v0, off, s[0:3], 0 offset:8
; GFX1010-NEXT: s_waitcnt_vscnt null, 0x0

6
test/CodeGen/AMDGPU/chain-hi-to-lo.ll
View File

@ -357,8 +357,6 @@ bb:
define amdgpu_kernel void @vload2_private(i16 addrspace(1)* nocapture readonly %in, <2 x i16> addrspace(1)* nocapture %out) #0 {
; GFX900-LABEL: vload2_private:
; GFX900: ; %bb.0: ; %entry
; GFX900-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; GFX900-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; GFX900-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x0
; GFX900-NEXT: v_mov_b32_e32 v2, 0
; GFX900-NEXT: s_add_u32 s0, s0, s9
@ -420,10 +418,6 @@ define amdgpu_kernel void @vload2_private(i16 addrspace(1)* nocapture readonly %
;
; GFX10_DEFAULT-LABEL: vload2_private:
; GFX10_DEFAULT: ; %bb.0: ; %entry
; GFX10_DEFAULT-NEXT: s_add_u32 s6, s6, s9
; GFX10_DEFAULT-NEXT: s_addc_u32 s7, s7, 0
; GFX10_DEFAULT-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s6
; GFX10_DEFAULT-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s7
; GFX10_DEFAULT-NEXT: s_load_dwordx4 s[4:7], s[4:5], 0x0
; GFX10_DEFAULT-NEXT: v_mov_b32_e32 v2, 0
; GFX10_DEFAULT-NEXT: s_add_u32 s0, s0, s9

6
test/CodeGen/AMDGPU/flat-scratch-init.ll
View File

@ -181,7 +181,6 @@ define amdgpu_kernel void @test(i32 addrspace(1)* %out, i32 %in) {
; GCN-NEXT: v_mov_b32_e32 v0, vcc_lo
; GCN-NEXT: ;;#ASMSTART
; GCN-NEXT: ;;#ASMEND
; GCN-NEXT: v_mov_b32_e32 v1, 0
; GCN-NEXT: s_mov_b32 s2, exec_lo
; GCN-NEXT: s_mov_b32 exec_lo, 3
; GCN-NEXT: s_mov_b32 s3, 0
@ -198,6 +197,7 @@ define amdgpu_kernel void @test(i32 addrspace(1)* %out, i32 %in) {
; GCN-NEXT: s_waitcnt vmcnt(0)
; GCN-NEXT: s_waitcnt_depctr 0xffe3
; GCN-NEXT: s_mov_b32 exec_lo, s2
; GCN-NEXT: v_mov_b32_e32 v1, 0
; GCN-NEXT: global_store_dword v1, v0, s[0:1]
; GCN-NEXT: s_endpgm
call void asm sideeffect "", "~{s[0:7]}" ()
@ -254,10 +254,6 @@ define amdgpu_kernel void @test(i32 addrspace(1)* %out, i32 %in) {
define amdgpu_kernel void @kernel_no_calls_no_stack() {
; GCN-LABEL: kernel_no_calls_no_stack:
; GCN: ; %bb.0:
; GCN-NEXT: s_add_u32 s0, s0, s3
; GCN-NEXT: s_addc_u32 s1, s1, 0
; GCN-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s0
; GCN-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s1
; GCN-NEXT: s_endpgm
ret void
}

8
test/CodeGen/AMDGPU/large-alloca-compute.ll
View File

@ -3,7 +3,7 @@
; RUN: llc -march=amdgcn -mcpu=gfx900 --show-mc-encoding < %s | FileCheck --check-prefixes=GCN,GFX9,ALL %s
; RUN: llc -march=amdgcn -mcpu=bonaire -mtriple=amdgcn-unknown-amdhsa --amdhsa-code-object-version=4 < %s -mattr=-flat-for-global | FileCheck --check-prefixes=GCNHSA,ALL %s
; RUN: llc -march=amdgcn -mcpu=carrizo -mtriple=amdgcn-unknown-amdhsa --amdhsa-code-object-version=4 -mattr=-flat-for-global < %s | FileCheck --check-prefixes=GCNHSA,ALL %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 -mtriple=amdgcn-unknown-amdhsa --amdhsa-code-object-version=4 -mattr=-flat-for-global < %s | FileCheck --check-prefixes=GCNHSA,GFX10HSA,ALL %s
; RUN: llc -march=amdgcn -mcpu=gfx1010 -mtriple=amdgcn-unknown-amdhsa --amdhsa-code-object-version=4 -mattr=-flat-for-global < %s | FileCheck --check-prefixes=GCNHSA,ALL %s
; FIXME: align on alloca seems to be ignored for private_segment_alignment
@ -18,12 +18,6 @@
; VI-DAG: s_mov_b32 s{{[0-9]+}}, 0xe80000
; GFX9-DAG: s_mov_b32 s{{[0-9]+}}, 0xe00000
; GFX10HSA: s_add_u32 [[FLAT_SCR_LO:s[0-9]+]], s{{[0-9]+}}, s{{[0-9]+}}
; GFX10HSA-DAG: s_addc_u32 [[FLAT_SCR_HI:s[0-9]+]], s{{[0-9]+}}, 0
; GFX10HSA-DAG: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), [[FLAT_SCR_LO]]
; GFX10HSA-DAG: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), [[FLAT_SCR_HI]]
; GCNHSA: buffer_store_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], 0 offen
; GCNHSA: buffer_load_dword {{v[0-9]+}}, {{v[0-9]+}}, s[0:3], 0 offen

4
test/CodeGen/AMDGPU/local-stack-alloc-block-sp-reference.ll
View File

@ -20,8 +20,6 @@ define amdgpu_kernel void @local_stack_offset_uses_sp(i64 addrspace(1)* %out) {
; MUBUF-LABEL: local_stack_offset_uses_sp:
; MUBUF: ; %bb.0: ; %entry
; MUBUF-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
; MUBUF-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; MUBUF-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; MUBUF-NEXT: s_add_u32 s0, s0, s9
; MUBUF-NEXT: v_mov_b32_e32 v1, 0x3000
; MUBUF-NEXT: s_addc_u32 s1, s1, 0
@ -204,8 +202,6 @@ define amdgpu_kernel void @local_stack_offset_uses_sp_flat(<3 x i64> addrspace(1
; MUBUF-LABEL: local_stack_offset_uses_sp_flat:
; MUBUF: ; %bb.0: ; %entry
; MUBUF-NEXT: s_load_dwordx2 s[4:5], s[4:5], 0x0
; MUBUF-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; MUBUF-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; MUBUF-NEXT: s_add_u32 s0, s0, s9
; MUBUF-NEXT: s_addc_u32 s1, s1, 0
; MUBUF-NEXT: v_mov_b32_e32 v0, 0x4000

12
test/CodeGen/AMDGPU/memory_clause.ll
View File

@ -25,10 +25,6 @@ define amdgpu_kernel void @vector_clause(<4 x i32> addrspace(1)* noalias nocaptu
;
; GCN-SCRATCH-LABEL: vector_clause:
; GCN-SCRATCH: ; %bb.0: ; %bb
; GCN-SCRATCH-NEXT: s_add_u32 s2, s2, s5
; GCN-SCRATCH-NEXT: s_addc_u32 s3, s3, 0
; GCN-SCRATCH-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
; GCN-SCRATCH-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
; GCN-SCRATCH-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN-SCRATCH-NEXT: v_lshlrev_b32_e32 v16, 4, v0
; GCN-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x2c
@ -108,10 +104,6 @@ define amdgpu_kernel void @scalar_clause(<4 x i32> addrspace(1)* noalias nocaptu
;
; GCN-SCRATCH-LABEL: scalar_clause:
; GCN-SCRATCH: ; %bb.0: ; %bb
; GCN-SCRATCH-NEXT: s_add_u32 s2, s2, s5
; GCN-SCRATCH-NEXT: s_addc_u32 s3, s3, 0
; GCN-SCRATCH-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
; GCN-SCRATCH-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
; GCN-SCRATCH-NEXT: s_clause 0x1
; GCN-SCRATCH-NEXT: s_load_dwordx2 s[12:13], s[0:1], 0x24
; GCN-SCRATCH-NEXT: s_load_dwordx2 s[16:17], s[0:1], 0x2c
@ -277,10 +269,6 @@ define amdgpu_kernel void @vector_clause_indirect(i64 addrspace(1)* noalias noca
;
; GCN-SCRATCH-LABEL: vector_clause_indirect:
; GCN-SCRATCH: ; %bb.0: ; %bb
; GCN-SCRATCH-NEXT: s_add_u32 s2, s2, s5
; GCN-SCRATCH-NEXT: s_addc_u32 s3, s3, 0
; GCN-SCRATCH-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_LO), s2
; GCN-SCRATCH-NEXT: s_setreg_b32 hwreg(HW_REG_FLAT_SCR_HI), s3
; GCN-SCRATCH-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
; GCN-SCRATCH-NEXT: v_lshlrev_b32_e32 v0, 3, v0
; GCN-SCRATCH-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x34

4
test/CodeGen/AMDGPU/non-entry-alloca.ll
View File

@ -15,8 +15,6 @@
define amdgpu_kernel void @kernel_non_entry_block_static_alloca_uniformly_reached_align4(i32 addrspace(1)* %out, i32 %arg.cond0, i32 %arg.cond1, i32 %in) {
; MUBUF-LABEL: kernel_non_entry_block_static_alloca_uniformly_reached_align4:
; MUBUF: ; %bb.0: ; %entry
; MUBUF-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; MUBUF-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; MUBUF-NEXT: s_add_u32 s0, s0, s9
; MUBUF-NEXT: s_load_dwordx4 s[8:11], s[4:5], 0x8
; MUBUF-NEXT: s_addc_u32 s1, s1, 0
@ -119,8 +117,6 @@ bb.2:
define amdgpu_kernel void @kernel_non_entry_block_static_alloca_uniformly_reached_align64(i32 addrspace(1)* %out, i32 %arg.cond, i32 %in) {
; MUBUF-LABEL: kernel_non_entry_block_static_alloca_uniformly_reached_align64:
; MUBUF: ; %bb.0: ; %entry
; MUBUF-NEXT: s_add_u32 flat_scratch_lo, s6, s9
; MUBUF-NEXT: s_addc_u32 flat_scratch_hi, s7, 0
; MUBUF-NEXT: s_load_dwordx2 s[6:7], s[4:5], 0x8
; MUBUF-NEXT: s_add_u32 s0, s0, s9
; MUBUF-NEXT: s_addc_u32 s1, s1, 0

10
test/CodeGen/AMDGPU/spill-offset-calculation.ll
View File

@ -37,8 +37,8 @@ entry:
%aptr = getelementptr i32, i32 addrspace(5)* %buf, i32 1
; 0x40000 / 64 = 4096 (for wave64)
; MUBUF: s_mov_b32 s6, 0x40000
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 ; 4-byte Folded Spill
; MUBUF: s_mov_b32 s4, 0x40000
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 ; 4-byte Folded Spill
; FLATSCR: s_movk_i32 s2, 0x1000
; FLATSCR: scratch_store_dword off, v{{[0-9]+}}, s2 ; 4-byte Folded Spill
%a = load volatile i32, i32 addrspace(5)* %aptr
@ -149,9 +149,9 @@ entry:
%bufv2 = bitcast i8 addrspace(5)* %alloca to <2 x i32> addrspace(5)*
; 0x3ff00 / 64 = 4092 (for wave64)
; MUBUF: s_mov_b32 s6, 0x3ff00
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 ; 4-byte Folded Spill
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s6 offset:4 ; 4-byte Folded Spill
; MUBUF: s_mov_b32 s4, 0x3ff00
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 ; 4-byte Folded Spill
; MUBUF: buffer_store_dword v{{[0-9]+}}, off, s[{{[0-9]+:[0-9]+}}], s4 offset:4 ; 4-byte Folded Spill
; FLATSCR: s_movk_i32 [[SOFF:s[0-9]+]], 0xffc
; FLATSCR: scratch_store_dwordx2 off, v[{{[0-9:]+}}], [[SOFF]] ; 8-byte Folded Spill
%aptr = getelementptr <2 x i32>, <2 x i32> addrspace(5)* %bufv2, i32 1

21
test/CodeGen/AMDGPU/stack-realign-kernel.ll
View File

@ -6,12 +6,9 @@
define amdgpu_kernel void @max_alignment_128() #0 {
; VI-LABEL: max_alignment_128:
; VI: ; %bb.0:
; VI-NEXT: s_add_i32 s4, s4, s7
; VI-NEXT: s_lshr_b32 flat_scratch_hi, s4, 8
; VI-NEXT: s_add_u32 s0, s0, s7
; VI-NEXT: s_addc_u32 s1, s1, 0
; VI-NEXT: v_mov_b32_e32 v0, 9
; VI-NEXT: s_mov_b32 flat_scratch_lo, s5
; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 offset:128
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: s_endpgm
@ -37,6 +34,7 @@ define amdgpu_kernel void @max_alignment_128() #0 {
; VI-NEXT: .amdhsa_next_free_vgpr 1
; VI-NEXT: .amdhsa_next_free_sgpr 8
; VI-NEXT: .amdhsa_reserve_vcc 0
; VI-NEXT: .amdhsa_reserve_flat_scratch 0
; VI-NEXT: .amdhsa_float_round_mode_32 0
; VI-NEXT: .amdhsa_float_round_mode_16_64 0
; VI-NEXT: .amdhsa_float_denorm_mode_32 3
@ -55,8 +53,6 @@ define amdgpu_kernel void @max_alignment_128() #0 {
;
; GFX9-LABEL: max_alignment_128:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s4, s7
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s5, 0
; GFX9-NEXT: s_add_u32 s0, s0, s7
; GFX9-NEXT: s_addc_u32 s1, s1, 0
; GFX9-NEXT: v_mov_b32_e32 v0, 9
@ -85,6 +81,7 @@ define amdgpu_kernel void @max_alignment_128() #0 {
; GFX9-NEXT: .amdhsa_next_free_vgpr 1
; GFX9-NEXT: .amdhsa_next_free_sgpr 8
; GFX9-NEXT: .amdhsa_reserve_vcc 0
; GFX9-NEXT: .amdhsa_reserve_flat_scratch 0
; GFX9-NEXT: .amdhsa_reserve_xnack_mask 1
; GFX9-NEXT: .amdhsa_float_round_mode_32 0
; GFX9-NEXT: .amdhsa_float_round_mode_16_64 0
@ -110,12 +107,9 @@ define amdgpu_kernel void @max_alignment_128() #0 {
define amdgpu_kernel void @stackrealign_attr() #1 {
; VI-LABEL: stackrealign_attr:
; VI: ; %bb.0:
; VI-NEXT: s_add_i32 s4, s4, s7
; VI-NEXT: s_lshr_b32 flat_scratch_hi, s4, 8
; VI-NEXT: s_add_u32 s0, s0, s7
; VI-NEXT: s_addc_u32 s1, s1, 0
; VI-NEXT: v_mov_b32_e32 v0, 9
; VI-NEXT: s_mov_b32 flat_scratch_lo, s5
; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 offset:4
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: s_endpgm
@ -141,6 +135,7 @@ define amdgpu_kernel void @stackrealign_attr() #1 {
; VI-NEXT: .amdhsa_next_free_vgpr 1
; VI-NEXT: .amdhsa_next_free_sgpr 8
; VI-NEXT: .amdhsa_reserve_vcc 0
; VI-NEXT: .amdhsa_reserve_flat_scratch 0
; VI-NEXT: .amdhsa_float_round_mode_32 0
; VI-NEXT: .amdhsa_float_round_mode_16_64 0
; VI-NEXT: .amdhsa_float_denorm_mode_32 3
@ -159,8 +154,6 @@ define amdgpu_kernel void @stackrealign_attr() #1 {
;
; GFX9-LABEL: stackrealign_attr:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s4, s7
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s5, 0
; GFX9-NEXT: s_add_u32 s0, s0, s7
; GFX9-NEXT: s_addc_u32 s1, s1, 0
; GFX9-NEXT: v_mov_b32_e32 v0, 9
@ -189,6 +182,7 @@ define amdgpu_kernel void @stackrealign_attr() #1 {
; GFX9-NEXT: .amdhsa_next_free_vgpr 1
; GFX9-NEXT: .amdhsa_next_free_sgpr 8
; GFX9-NEXT: .amdhsa_reserve_vcc 0
; GFX9-NEXT: .amdhsa_reserve_flat_scratch 0
; GFX9-NEXT: .amdhsa_reserve_xnack_mask 1
; GFX9-NEXT: .amdhsa_float_round_mode_32 0
; GFX9-NEXT: .amdhsa_float_round_mode_16_64 0
@ -214,12 +208,9 @@ define amdgpu_kernel void @stackrealign_attr() #1 {
define amdgpu_kernel void @alignstack_attr() #2 {
; VI-LABEL: alignstack_attr:
; VI: ; %bb.0:
; VI-NEXT: s_add_i32 s4, s4, s7
; VI-NEXT: s_lshr_b32 flat_scratch_hi, s4, 8
; VI-NEXT: s_add_u32 s0, s0, s7
; VI-NEXT: s_addc_u32 s1, s1, 0
; VI-NEXT: v_mov_b32_e32 v0, 9
; VI-NEXT: s_mov_b32 flat_scratch_lo, s5
; VI-NEXT: buffer_store_dword v0, off, s[0:3], 0 offset:4
; VI-NEXT: s_waitcnt vmcnt(0)
; VI-NEXT: s_endpgm
@ -245,6 +236,7 @@ define amdgpu_kernel void @alignstack_attr() #2 {
; VI-NEXT: .amdhsa_next_free_vgpr 1
; VI-NEXT: .amdhsa_next_free_sgpr 8
; VI-NEXT: .amdhsa_reserve_vcc 0
; VI-NEXT: .amdhsa_reserve_flat_scratch 0
; VI-NEXT: .amdhsa_float_round_mode_32 0
; VI-NEXT: .amdhsa_float_round_mode_16_64 0
; VI-NEXT: .amdhsa_float_denorm_mode_32 3
@ -263,8 +255,6 @@ define amdgpu_kernel void @alignstack_attr() #2 {
;
; GFX9-LABEL: alignstack_attr:
; GFX9: ; %bb.0:
; GFX9-NEXT: s_add_u32 flat_scratch_lo, s4, s7
; GFX9-NEXT: s_addc_u32 flat_scratch_hi, s5, 0
; GFX9-NEXT: s_add_u32 s0, s0, s7
; GFX9-NEXT: s_addc_u32 s1, s1, 0
; GFX9-NEXT: v_mov_b32_e32 v0, 9
@ -293,6 +283,7 @@ define amdgpu_kernel void @alignstack_attr() #2 {
; GFX9-NEXT: .amdhsa_next_free_vgpr 1
; GFX9-NEXT: .amdhsa_next_free_sgpr 8
; GFX9-NEXT: .amdhsa_reserve_vcc 0
; GFX9-NEXT: .amdhsa_reserve_flat_scratch 0
; GFX9-NEXT: .amdhsa_reserve_xnack_mask 1
; GFX9-NEXT: .amdhsa_float_round_mode_32 0
; GFX9-NEXT: .amdhsa_float_round_mode_16_64 0