caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
9eddc8b9bf4e4e0b01e2ecc90a71c4b3b4e9c8af
clang-p2996/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32.ll
Carl Ritson 86627149f6 [AMDGPU] Mitigate GFX12 VALU read SGPR hazard (#100067) 
Any SGPR read by a VALU can potentially obscure SALU writes to the same
register.
Insert s_wait_alu instructions to mitigate the hazard on affected paths.

Compute a global cache of SGPRs with any VALU reads and use this to
avoid inserting mitigation for SGPRs never accessed by VALUs.

To avoid excessive search when compile time is priority implement
secondary mode where all SALU writes are mitigated.

Co-authored-by: Shilei Tian <shilei.tian@amd.com>
2024-09-04 12:15:20 +09:00

743 lines
35 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti < %s | FileCheck -check-prefix=GFX6 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii < %s | FileCheck -check-prefix=GFX7 %s
; Not supported in gfx8 or gfx9
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 < %s | FileCheck -check-prefix=GFX11 %s
; RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1200 < %s | FileCheck -check-prefix=GFX12 %s
define float @struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen glc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen glc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen glc
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[1:2], s[0:3], s6 idxen offen glc
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[1:2], s[0:3], s6 idxen offen th:TH_ATOMIC_RETURN
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
ret float %ret
}
define float @struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen offset:256 glc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen offset:256 glc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen offset:256 glc
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[1:2], s[0:3], s6 idxen offen offset:256 glc
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[1:2], s[0:3], s6 idxen offen offset:256 th:TH_ATOMIC_RETURN
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: s_setpc_b64 s[30:31]
%voffset.add = add i32 %voffset, 256
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset.add, i32 %soffset, i32 0)
ret float %ret
}
define float @struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v1, s[8:11], s18 idxen glc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[8:11], s18 idxen glc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[8:11], s18 idxen glc
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v1, s[0:3], s6 idxen glc
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v1, s[0:3], s6 idxen th:TH_ATOMIC_RETURN
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
ret float %ret
}
define float @struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen glc slc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen glc slc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen glc slc
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[1:2], s[0:3], s6 idxen offen glc slc
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[1:2], s[0:3], s6 idxen offen th:TH_ATOMIC_NT_RETURN
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
ret float %ret
}
define void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[1:2], s[0:3], s6 idxen offen
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[1:2], s[0:3], s6 idxen offen
; GFX12-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
ret void
}
define void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen offset:256
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen offset:256
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen offset:256
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[1:2], s[0:3], s6 idxen offen offset:256
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[1:2], s[0:3], s6 idxen offen offset:256
; GFX12-NEXT: s_setpc_b64 s[30:31]
%voffset.add = add i32 %voffset, 256
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset.add, i32 %soffset, i32 0)
ret void
}
; Natural mapping, no voffset
define void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v1, s[8:11], s18 idxen
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v1, s[8:11], s18 idxen
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v1, s[8:11], s18 idxen
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v1, s[0:3], s6 idxen
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__0_voffset__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v1, s[0:3], s6 idxen
; GFX12-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
ret void
}
define void @struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc(float %val, ptr addrspace(8) inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s11, s17
; GFX6-NEXT: s_mov_b32 s10, s16
; GFX6-NEXT: s_mov_b32 s9, s7
; GFX6-NEXT: s_mov_b32 s8, s6
; GFX6-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen slc
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b32 s11, s17
; GFX7-NEXT: s_mov_b32 s10, s16
; GFX7-NEXT: s_mov_b32 s9, s7
; GFX7-NEXT: s_mov_b32 s8, s6
; GFX7-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen slc
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s11, s17
; GFX10-NEXT: s_mov_b32 s10, s16
; GFX10-NEXT: s_mov_b32 s9, s7
; GFX10-NEXT: s_mov_b32 s8, s6
; GFX10-NEXT: buffer_atomic_fmax v0, v[1:2], s[8:11], s18 idxen offen slc
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[1:2], s[0:3], s6 idxen offen slc
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_noret__vgpr_val__sgpr_rsrc__vgpr_voffset__sgpr_soffset_slc:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[1:2], s[0:3], s6 idxen offen th:TH_ATOMIC_NT
; GFX12-NEXT: s_setpc_b64 s[30:31]
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
ret void
}
; Test waterfall loop on resource
define float @struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__vgpr_rsrc__vgpr_voffset_add__sgpr_soffset(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__vgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b64 s[12:13], exec
; GFX6-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1
; GFX6-NEXT: v_readfirstlane_b32 s8, v1
; GFX6-NEXT: v_readfirstlane_b32 s9, v2
; GFX6-NEXT: v_readfirstlane_b32 s10, v3
; GFX6-NEXT: v_readfirstlane_b32 s11, v4
; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, s[8:9], v[1:2]
; GFX6-NEXT: v_cmp_eq_u64_e64 s[4:5], s[10:11], v[3:4]
; GFX6-NEXT: s_and_b64 s[4:5], vcc, s[4:5]
; GFX6-NEXT: s_and_saveexec_b64 s[4:5], s[4:5]
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: buffer_atomic_fmax v0, v[5:6], s[8:11], s6 idxen offen offset:256 glc
; GFX6-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX6-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX6-NEXT: s_xor_b64 exec, exec, s[4:5]
; GFX6-NEXT: s_cbranch_execnz .LBB8_1
; GFX6-NEXT: ; %bb.2:
; GFX6-NEXT: s_mov_b64 exec, s[12:13]
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__vgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b64 s[12:13], exec
; GFX7-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT: v_readfirstlane_b32 s8, v1
; GFX7-NEXT: v_readfirstlane_b32 s9, v2
; GFX7-NEXT: v_readfirstlane_b32 s10, v3
; GFX7-NEXT: v_readfirstlane_b32 s11, v4
; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, s[8:9], v[1:2]
; GFX7-NEXT: v_cmp_eq_u64_e64 s[4:5], s[10:11], v[3:4]
; GFX7-NEXT: s_and_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT: s_and_saveexec_b64 s[4:5], s[4:5]
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: buffer_atomic_fmax v0, v[5:6], s[8:11], s6 idxen offen offset:256 glc
; GFX7-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX7-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX7-NEXT: s_xor_b64 exec, exec, s[4:5]
; GFX7-NEXT: s_cbranch_execnz .LBB8_1
; GFX7-NEXT: ; %bb.2:
; GFX7-NEXT: s_mov_b64 exec, s[12:13]
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__vgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s5, exec_lo
; GFX10-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1
; GFX10-NEXT: v_readfirstlane_b32 s8, v1
; GFX10-NEXT: v_readfirstlane_b32 s9, v2
; GFX10-NEXT: v_readfirstlane_b32 s10, v3
; GFX10-NEXT: v_readfirstlane_b32 s11, v4
; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[1:2]
; GFX10-NEXT: v_cmp_eq_u64_e64 s4, s[10:11], v[3:4]
; GFX10-NEXT: s_and_b32 s4, vcc_lo, s4
; GFX10-NEXT: s_and_saveexec_b32 s4, s4
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: buffer_atomic_fmax v0, v[5:6], s[8:11], s6 idxen offen offset:256 glc
; GFX10-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX10-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX10-NEXT: s_waitcnt_depctr 0xffe3
; GFX10-NEXT: s_xor_b32 exec_lo, exec_lo, s4
; GFX10-NEXT: s_cbranch_execnz .LBB8_1
; GFX10-NEXT: ; %bb.2:
; GFX10-NEXT: s_mov_b32 exec_lo, s5
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__vgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s2, exec_lo
; GFX11-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1
; GFX11-NEXT: v_readfirstlane_b32 s4, v1
; GFX11-NEXT: v_readfirstlane_b32 s5, v2
; GFX11-NEXT: v_readfirstlane_b32 s6, v3
; GFX11-NEXT: v_readfirstlane_b32 s7, v4
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[1:2]
; GFX11-NEXT: v_cmp_eq_u64_e64 s1, s[6:7], v[3:4]
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_and_b32 s1, vcc_lo, s1
; GFX11-NEXT: s_and_saveexec_b32 s1, s1
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[5:6], s[4:7], s0 idxen offen offset:256 glc
; GFX11-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX11-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX11-NEXT: s_xor_b32 exec_lo, exec_lo, s1
; GFX11-NEXT: s_cbranch_execnz .LBB8_1
; GFX11-NEXT: ; %bb.2:
; GFX11-NEXT: s_mov_b32 exec_lo, s2
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__vgpr_rsrc__vgpr_voffset_add__sgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: s_mov_b32 s2, exec_lo
; GFX12-NEXT: .LBB8_1: ; =>This Inner Loop Header: Depth=1
; GFX12-NEXT: v_readfirstlane_b32 s4, v1
; GFX12-NEXT: v_readfirstlane_b32 s5, v2
; GFX12-NEXT: v_readfirstlane_b32 s6, v3
; GFX12-NEXT: v_readfirstlane_b32 s7, v4
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[1:2]
; GFX12-NEXT: v_cmp_eq_u64_e64 s1, s[6:7], v[3:4]
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: s_and_b32 s1, vcc_lo, s1
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_and_saveexec_b32 s1, s1
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[5:6], s[4:7], s0 idxen offen offset:256 th:TH_ATOMIC_RETURN
; GFX12-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX12-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_xor_b32 exec_lo, exec_lo, s1
; GFX12-NEXT: s_cbranch_execnz .LBB8_1
; GFX12-NEXT: ; %bb.2:
; GFX12-NEXT: s_mov_b32 exec_lo, s2
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_setpc_b64 s[30:31]
%voffset.add = add i32 %voffset, 256
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset.add, i32 %soffset, i32 0)
ret float %ret
}
; Test waterfall loop on soffset
define float @struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__vgpr_soffset(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset, i32 %soffset) {
; GFX6-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__vgpr_soffset:
; GFX6: ; %bb.0:
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX6-NEXT: s_mov_b64 s[6:7], exec
; GFX6-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1
; GFX6-NEXT: v_readfirstlane_b32 s8, v1
; GFX6-NEXT: v_readfirstlane_b32 s9, v2
; GFX6-NEXT: v_readfirstlane_b32 s10, v3
; GFX6-NEXT: v_readfirstlane_b32 s11, v4
; GFX6-NEXT: v_cmp_eq_u64_e32 vcc, s[8:9], v[1:2]
; GFX6-NEXT: v_cmp_eq_u64_e64 s[4:5], s[10:11], v[3:4]
; GFX6-NEXT: v_readfirstlane_b32 s12, v7
; GFX6-NEXT: s_and_b64 s[4:5], vcc, s[4:5]
; GFX6-NEXT: v_cmp_eq_u32_e32 vcc, s12, v7
; GFX6-NEXT: s_and_b64 s[4:5], s[4:5], vcc
; GFX6-NEXT: s_and_saveexec_b64 s[4:5], s[4:5]
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: buffer_atomic_fmax v0, v[5:6], s[8:11], s12 idxen offen offset:256 glc
; GFX6-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX6-NEXT: ; implicit-def: $vgpr7
; GFX6-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX6-NEXT: s_xor_b64 exec, exec, s[4:5]
; GFX6-NEXT: s_cbranch_execnz .LBB9_1
; GFX6-NEXT: ; %bb.2:
; GFX6-NEXT: s_mov_b64 exec, s[6:7]
; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; GFX6-NEXT: s_setpc_b64 s[30:31]
;
; GFX7-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__vgpr_soffset:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX7-NEXT: s_mov_b64 s[6:7], exec
; GFX7-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1
; GFX7-NEXT: v_readfirstlane_b32 s8, v1
; GFX7-NEXT: v_readfirstlane_b32 s9, v2
; GFX7-NEXT: v_readfirstlane_b32 s10, v3
; GFX7-NEXT: v_readfirstlane_b32 s11, v4
; GFX7-NEXT: v_cmp_eq_u64_e32 vcc, s[8:9], v[1:2]
; GFX7-NEXT: v_cmp_eq_u64_e64 s[4:5], s[10:11], v[3:4]
; GFX7-NEXT: v_readfirstlane_b32 s12, v7
; GFX7-NEXT: s_and_b64 s[4:5], vcc, s[4:5]
; GFX7-NEXT: v_cmp_eq_u32_e32 vcc, s12, v7
; GFX7-NEXT: s_and_b64 s[4:5], s[4:5], vcc
; GFX7-NEXT: s_and_saveexec_b64 s[4:5], s[4:5]
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: buffer_atomic_fmax v0, v[5:6], s[8:11], s12 idxen offen offset:256 glc
; GFX7-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX7-NEXT: ; implicit-def: $vgpr7
; GFX7-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX7-NEXT: s_xor_b64 exec, exec, s[4:5]
; GFX7-NEXT: s_cbranch_execnz .LBB9_1
; GFX7-NEXT: ; %bb.2:
; GFX7-NEXT: s_mov_b64 exec, s[6:7]
; GFX7-NEXT: s_waitcnt vmcnt(0)
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__vgpr_soffset:
; GFX10: ; %bb.0:
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: s_mov_b32 s6, exec_lo
; GFX10-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1
; GFX10-NEXT: v_readfirstlane_b32 s8, v1
; GFX10-NEXT: v_readfirstlane_b32 s9, v2
; GFX10-NEXT: v_readfirstlane_b32 s10, v3
; GFX10-NEXT: v_readfirstlane_b32 s11, v4
; GFX10-NEXT: v_readfirstlane_b32 s7, v7
; GFX10-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[8:9], v[1:2]
; GFX10-NEXT: v_cmp_eq_u64_e64 s4, s[10:11], v[3:4]
; GFX10-NEXT: v_cmp_eq_u32_e64 s5, s7, v7
; GFX10-NEXT: s_and_b32 s4, vcc_lo, s4
; GFX10-NEXT: s_and_b32 s4, s4, s5
; GFX10-NEXT: s_and_saveexec_b32 s4, s4
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: buffer_atomic_fmax v0, v[5:6], s[8:11], s7 idxen offen offset:256 glc
; GFX10-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX10-NEXT: ; implicit-def: $vgpr7
; GFX10-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX10-NEXT: s_waitcnt_depctr 0xffe3
; GFX10-NEXT: s_xor_b32 exec_lo, exec_lo, s4
; GFX10-NEXT: s_cbranch_execnz .LBB9_1
; GFX10-NEXT: ; %bb.2:
; GFX10-NEXT: s_mov_b32 exec_lo, s6
; GFX10-NEXT: s_waitcnt vmcnt(0)
; GFX10-NEXT: s_setpc_b64 s[30:31]
;
; GFX11-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__vgpr_soffset:
; GFX11: ; %bb.0:
; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX11-NEXT: s_mov_b32 s2, exec_lo
; GFX11-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1
; GFX11-NEXT: v_readfirstlane_b32 s4, v1
; GFX11-NEXT: v_readfirstlane_b32 s5, v2
; GFX11-NEXT: v_readfirstlane_b32 s6, v3
; GFX11-NEXT: v_readfirstlane_b32 s7, v4
; GFX11-NEXT: v_readfirstlane_b32 s3, v7
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX11-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[1:2]
; GFX11-NEXT: v_cmp_eq_u64_e64 s0, s[6:7], v[3:4]
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-NEXT: v_cmp_eq_u32_e64 s1, s3, v7
; GFX11-NEXT: s_and_b32 s0, vcc_lo, s0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
; GFX11-NEXT: s_and_b32 s0, s0, s1
; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
; GFX11-NEXT: s_and_saveexec_b32 s0, s0
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: buffer_atomic_max_f32 v0, v[5:6], s[4:7], s3 idxen offen offset:256 glc
; GFX11-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX11-NEXT: ; implicit-def: $vgpr7
; GFX11-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX11-NEXT: s_xor_b32 exec_lo, exec_lo, s0
; GFX11-NEXT: s_cbranch_execnz .LBB9_1
; GFX11-NEXT: ; %bb.2:
; GFX11-NEXT: s_mov_b32 exec_lo, s2
; GFX11-NEXT: s_waitcnt vmcnt(0)
; GFX11-NEXT: s_setpc_b64 s[30:31]
;
; GFX12-LABEL: struct_ptr_buffer_atomic_add_f32_ret__vgpr_val__sgpr_rsrc__vgpr_voffset_add__vgpr_soffset:
; GFX12: ; %bb.0:
; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0
; GFX12-NEXT: s_wait_expcnt 0x0
; GFX12-NEXT: s_wait_samplecnt 0x0
; GFX12-NEXT: s_wait_bvhcnt 0x0
; GFX12-NEXT: s_wait_kmcnt 0x0
; GFX12-NEXT: s_mov_b32 s2, exec_lo
; GFX12-NEXT: .LBB9_1: ; =>This Inner Loop Header: Depth=1
; GFX12-NEXT: v_readfirstlane_b32 s4, v1
; GFX12-NEXT: v_readfirstlane_b32 s5, v2
; GFX12-NEXT: v_readfirstlane_b32 s6, v3
; GFX12-NEXT: v_readfirstlane_b32 s7, v4
; GFX12-NEXT: v_readfirstlane_b32 s3, v7
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_3)
; GFX12-NEXT: v_cmp_eq_u64_e32 vcc_lo, s[4:5], v[1:2]
; GFX12-NEXT: v_cmp_eq_u64_e64 s0, s[6:7], v[3:4]
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX12-NEXT: v_cmp_eq_u32_e64 s1, s3, v7
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_and_b32 s0, vcc_lo, s0
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX12-NEXT: s_and_b32 s0, s0, s1
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_and_saveexec_b32 s0, s0
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: buffer_atomic_max_num_f32 v0, v[5:6], s[4:7], s3 idxen offen offset:256 th:TH_ATOMIC_RETURN
; GFX12-NEXT: ; implicit-def: $vgpr1_vgpr2_vgpr3_vgpr4
; GFX12-NEXT: ; implicit-def: $vgpr7
; GFX12-NEXT: ; implicit-def: $vgpr5_vgpr6
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_xor_b32 exec_lo, exec_lo, s0
; GFX12-NEXT: s_cbranch_execnz .LBB9_1
; GFX12-NEXT: ; %bb.2:
; GFX12-NEXT: s_mov_b32 exec_lo, s2
; GFX12-NEXT: s_wait_loadcnt 0x0
; GFX12-NEXT: s_wait_alu 0xfffe
; GFX12-NEXT: s_setpc_b64 s[30:31]
%voffset.add = add i32 %voffset, 256
%ret = call float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float %val, ptr addrspace(8) %rsrc, i32 %vindex, i32 %voffset.add, i32 %soffset, i32 0)
ret float %ret
}
declare float @llvm.amdgcn.struct.ptr.buffer.atomic.fmax.f32(float, ptr addrspace(8), i32, i32, i32, i32 immarg)
Reference in New Issue View Git Blame Copy Permalink