200 changes: 200 additions & 0 deletions llvm/test/CodeGen/AMDGPU/buffer-atomic-fadd.f64.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,200 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s

define amdgpu_ps void @buffer_atomic_fadd_f64_offset_no_rtn(double %val, <4 x i32> inreg %rsrc, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_offset_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_ADD_F64_OFFSET killed [[COPY7]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 0, i32 %soffset, i32 0)
ret void
}

define amdgpu_ps void @buffer_atomic_fadd_f64_offen_no_rtn(double %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_offen_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr2, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_ADD_F64_OFFEN killed [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}

define amdgpu_ps void @buffer_atomic_fadd_f64_idxen_no_rtn(double %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_idxen_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr2, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_ADD_F64_IDXEN killed [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
ret void
}

define amdgpu_ps void @buffer_atomic_fadd_f64_bothen_no_rtn(double %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_bothen_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr2, $vgpr3, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_ADD_F64_BOTHEN killed [[COPY9]], killed [[REG_SEQUENCE2]], killed [[REG_SEQUENCE]], [[COPY]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
ret void
}

define amdgpu_ps double @buffer_atomic_fadd_f64_offset_rtn(double %val, <4 x i32> inreg %rsrc, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_offset_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_OFFSET_RTN [[COPY7]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN]].sub0
; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFSET_RTN]].sub1
; GFX90A_GFX940-NEXT: $sgpr0 = COPY [[COPY8]]
; GFX90A_GFX940-NEXT: $sgpr1 = COPY [[COPY9]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 0, i32 %soffset, i32 0)
ret double %ret
}

define amdgpu_ps double @buffer_atomic_fadd_f64_offen_rtn(double %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_offen_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr2, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_OFFEN_RTN [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN]].sub0
; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_OFFEN_RTN]].sub1
; GFX90A_GFX940-NEXT: $sgpr0 = COPY [[COPY9]]
; GFX90A_GFX940-NEXT: $sgpr1 = COPY [[COPY10]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret double %ret
}

define amdgpu_ps double @buffer_atomic_fadd_f64_idxen_rtn(double %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_idxen_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr2, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY6]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_IDXEN_RTN [[COPY8]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN]].sub0
; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_IDXEN_RTN]].sub1
; GFX90A_GFX940-NEXT: $sgpr0 = COPY [[COPY9]]
; GFX90A_GFX940-NEXT: $sgpr1 = COPY [[COPY10]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
ret double %ret
}

define amdgpu_ps double @buffer_atomic_fadd_f64_bothen_rtn(double %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_f64_bothen_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr2, $vgpr3, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY8]], %subreg.sub0, [[COPY7]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY9:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN:%[0-9]+]]:vreg_64_align2 = BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN [[COPY9]], killed [[REG_SEQUENCE2]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s64), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN]].sub0
; GFX90A_GFX940-NEXT: [[COPY11:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_ATOMIC_ADD_F64_BOTHEN_RTN]].sub1
; GFX90A_GFX940-NEXT: $sgpr0 = COPY [[COPY10]]
; GFX90A_GFX940-NEXT: $sgpr1 = COPY [[COPY11]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0)
ret double %ret
}

declare double @llvm.amdgcn.raw.buffer.atomic.fadd.f64(double, <4 x i32>, i32, i32, i32 immarg)
declare double @llvm.amdgcn.struct.buffer.atomic.fadd.f64(double, <4 x i32>, i32, i32, i32, i32 immarg)
141 changes: 141 additions & 0 deletions llvm/test/CodeGen/AMDGPU/buffer-atomic-fadd.v2f16-no-rtn.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,141 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX908 %s
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s

define amdgpu_ps void @buffer_atomic_fadd_v2f16_offset_no_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 inreg %soffset) {
; GFX908-LABEL: name: buffer_atomic_fadd_v2f16_offset_no_rtn
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX908-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX908-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX908-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
; GFX908-NEXT: BUFFER_ATOMIC_PK_ADD_F16_OFFSET [[COPY5]], killed [[REG_SEQUENCE]], [[COPY]], 4095, 0, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_offset_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_PK_ADD_F16_OFFSET [[COPY5]], killed [[REG_SEQUENCE]], [[COPY]], 4095, 0, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 4095, i32 %soffset, i32 0)
ret void
}

define amdgpu_ps void @buffer_atomic_fadd_v2f16_offen_no_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; GFX908-LABEL: name: buffer_atomic_fadd_v2f16_offen_no_rtn
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX908-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX908-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX908-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX908-NEXT: BUFFER_ATOMIC_PK_ADD_F16_OFFEN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_offen_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_PK_ADD_F16_OFFEN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret void
}

define amdgpu_ps void @buffer_atomic_fadd_v2f16_idxen_no_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
; GFX908-LABEL: name: buffer_atomic_fadd_v2f16_idxen_no_rtn
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX908-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX908-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX908-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX908-NEXT: BUFFER_ATOMIC_PK_ADD_F16_IDXEN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_idxen_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_PK_ADD_F16_IDXEN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
ret void
}

define amdgpu_ps void @buffer_atomic_fadd_v2f16_bothen_no_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX908-LABEL: name: buffer_atomic_fadd_v2f16_bothen_no_rtn
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $vgpr2, $sgpr4
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX908-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX908-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX908-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX908-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
; GFX908-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908-NEXT: BUFFER_ATOMIC_PK_ADD_F16_BOTHEN [[COPY7]], killed [[REG_SEQUENCE1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_bothen_no_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $vgpr2, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: BUFFER_ATOMIC_PK_ADD_F16_BOTHEN [[COPY7]], killed [[REG_SEQUENCE1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 2, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
ret void
}

declare <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half>, <4 x i32>, i32, i32, i32 immarg)
declare <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half>, <4 x i32>, i32, i32, i32, i32 immarg)
87 changes: 87 additions & 0 deletions llvm/test/CodeGen/AMDGPU/buffer-atomic-fadd.v2f16-rtn.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,87 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s

define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_offset_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_offset_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY3]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY1]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN [[COPY5]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_OFFSET_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 0, i32 %soffset, i32 0)
ret <2 x half> %ret
}

define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_offen_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_offen_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_OFFEN_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %voffset, i32 %soffset, i32 0)
ret <2 x half> %ret
}

define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_idxen_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_idxen_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY5]], %subreg.sub0, [[COPY4]], %subreg.sub1, [[COPY3]], %subreg.sub2, [[COPY2]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN [[COPY6]], [[COPY1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_IDXEN_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 0, i32 %soffset, i32 0)
ret <2 x half> %ret
}

define amdgpu_ps <2 x half> @buffer_atomic_fadd_v2f16_bothen_rtn(<2 x half> %val, <4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) {
; GFX90A_GFX940-LABEL: name: buffer_atomic_fadd_v2f16_bothen_rtn
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr1, $vgpr2, $sgpr4
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr4
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY $sgpr3
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:sgpr_32 = COPY $sgpr2
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY4]], %subreg.sub2, [[COPY3]], %subreg.sub3
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN:%[0-9]+]]:vgpr_32 = BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN [[COPY7]], killed [[REG_SEQUENCE1]], killed [[REG_SEQUENCE]], [[COPY]], 0, 3, implicit $exec :: (volatile dereferenceable load store (s32), align 1, addrspace 4)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[BUFFER_ATOMIC_PK_ADD_F16_BOTHEN_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half> %val, <4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 2)
ret <2 x half> %ret
}

declare <2 x half> @llvm.amdgcn.raw.buffer.atomic.fadd.v2f16(<2 x half>, <4 x i32>, i32, i32, i32 immarg)
declare <2 x half> @llvm.amdgcn.struct.buffer.atomic.fadd.v2f16(<2 x half>, <4 x i32>, i32, i32, i32, i32 immarg)
162 changes: 162 additions & 0 deletions llvm/test/CodeGen/AMDGPU/flat-atomic-fadd.f32.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,162 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX11 %s

define amdgpu_ps void @flat_atomic_fadd_f32_no_rtn_intrinsic(float* %ptr, float %data) {
; GFX940-LABEL: name: flat_atomic_fadd_f32_no_rtn_intrinsic
; GFX940: bb.0 (%ir-block.0):
; GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX940-NEXT: {{ $}}
; GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX940-NEXT: FLAT_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s32) on %ir.ptr)
; GFX940-NEXT: S_ENDPGM 0
; GFX11-LABEL: name: flat_atomic_fadd_f32_no_rtn_intrinsic
; GFX11: bb.0 (%ir-block.0):
; GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX11-NEXT: FLAT_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s32) on %ir.ptr)
; GFX11-NEXT: S_ENDPGM 0
%ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float* %ptr, float %data)
ret void
}

define amdgpu_ps float @flat_atomic_fadd_f32_rtn_intrinsic(float* %ptr, float %data) {
; GFX940-LABEL: name: flat_atomic_fadd_f32_rtn_intrinsic
; GFX940: bb.0 (%ir-block.0):
; GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX940-NEXT: {{ $}}
; GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX940-NEXT: [[FLAT_ATOMIC_ADD_F32_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_F32_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s32) on %ir.ptr)
; GFX940-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_F32_RTN]]
; GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
; GFX11-LABEL: name: flat_atomic_fadd_f32_rtn_intrinsic
; GFX11: bb.0 (%ir-block.0):
; GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX11-NEXT: [[FLAT_ATOMIC_ADD_F32_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_F32_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s32) on %ir.ptr)
; GFX11-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_F32_RTN]]
; GFX11-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float* %ptr, float %data)
ret float %ret
}

define amdgpu_ps void @flat_atomic_fadd_f32_no_rtn_atomicrmw(float* %ptr, float %data) #0 {
; GFX940-LABEL: name: flat_atomic_fadd_f32_no_rtn_atomicrmw
; GFX940: bb.0 (%ir-block.0):
; GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX940-NEXT: {{ $}}
; GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX940-NEXT: FLAT_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr)
; GFX940-NEXT: S_ENDPGM 0
; GFX11-LABEL: name: flat_atomic_fadd_f32_no_rtn_atomicrmw
; GFX11: bb.0 (%ir-block.0):
; GFX11-NEXT: successors: %bb.1(0x80000000)
; GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX11-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY4]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %ir.ptr)
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.1.atomicrmw.start:
; GFX11-NEXT: successors: %bb.2(0x04000000), %bb.1(0x7c000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[PHI:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_]], %bb.0, %4, %bb.1
; GFX11-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[FLAT_LOAD_DWORD]], %bb.0, %3, %bb.1
; GFX11-NEXT: %13:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[PHI1]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE killed %13, %subreg.sub0, [[PHI1]], %subreg.sub1
; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
; GFX11-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY3]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic monotonic (s32) on %ir.2)
; GFX11-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_EQ_U32_e64 [[FLAT_ATOMIC_CMPSWAP_RTN]], [[PHI1]], implicit $exec
; GFX11-NEXT: [[SI_IF_BREAK:%[0-9]+]]:sreg_32 = SI_IF_BREAK killed [[V_CMP_EQ_U32_e64_]], [[PHI]], implicit-def dead $scc
; GFX11-NEXT: SI_LOOP [[SI_IF_BREAK]], %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
; GFX11-NEXT: S_BRANCH %bb.2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.2.atomicrmw.end:
; GFX11-NEXT: [[PHI2:%[0-9]+]]:sreg_32 = PHI [[SI_IF_BREAK]], %bb.1
; GFX11-NEXT: SI_END_CF [[PHI2]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec
; GFX11-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd float* %ptr, float %data syncscope("wavefront") monotonic
ret void
}

define amdgpu_ps float @flat_atomic_fadd_f32_rtn_atomicrmw(float* %ptr, float %data) #0 {
; GFX940-LABEL: name: flat_atomic_fadd_f32_rtn_atomicrmw
; GFX940: bb.0 (%ir-block.0):
; GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX940-NEXT: {{ $}}
; GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX940-NEXT: [[FLAT_ATOMIC_ADD_F32_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_ADD_F32_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr)
; GFX940-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_ADD_F32_RTN]]
; GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
; GFX11-LABEL: name: flat_atomic_fadd_f32_rtn_atomicrmw
; GFX11: bb.0 (%ir-block.0):
; GFX11-NEXT: successors: %bb.1(0x80000000)
; GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX11-NEXT: [[COPY4:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX11-NEXT: [[FLAT_LOAD_DWORD:%[0-9]+]]:vgpr_32 = FLAT_LOAD_DWORD [[COPY4]], 0, 0, implicit $exec, implicit $flat_scr :: (load (s32) from %ir.ptr)
; GFX11-NEXT: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.1.atomicrmw.start:
; GFX11-NEXT: successors: %bb.2(0x04000000), %bb.1(0x7c000000)
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: [[PHI:%[0-9]+]]:sreg_32 = PHI [[S_MOV_B32_]], %bb.0, %4, %bb.1
; GFX11-NEXT: [[PHI1:%[0-9]+]]:vgpr_32 = PHI [[FLAT_LOAD_DWORD]], %bb.0, %3, %bb.1
; GFX11-NEXT: %14:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[PHI1]], 0, [[COPY]], 0, 0, implicit $mode, implicit $exec
; GFX11-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE killed %14, %subreg.sub0, [[PHI1]], %subreg.sub1
; GFX11-NEXT: [[COPY5:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE1]]
; GFX11-NEXT: [[FLAT_ATOMIC_CMPSWAP_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_CMPSWAP_RTN [[COPY3]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic monotonic (s32) on %ir.2)
; GFX11-NEXT: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_32 = V_CMP_EQ_U32_e64 [[FLAT_ATOMIC_CMPSWAP_RTN]], [[PHI1]], implicit $exec
; GFX11-NEXT: [[SI_IF_BREAK:%[0-9]+]]:sreg_32 = SI_IF_BREAK killed [[V_CMP_EQ_U32_e64_]], [[PHI]], implicit-def dead $scc
; GFX11-NEXT: SI_LOOP [[SI_IF_BREAK]], %bb.1, implicit-def dead $exec, implicit-def dead $scc, implicit $exec
; GFX11-NEXT: S_BRANCH %bb.2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: bb.2.atomicrmw.end:
; GFX11-NEXT: [[PHI2:%[0-9]+]]:vgpr_32 = PHI [[FLAT_ATOMIC_CMPSWAP_RTN]], %bb.1
; GFX11-NEXT: [[PHI3:%[0-9]+]]:sreg_32 = PHI [[SI_IF_BREAK]], %bb.1
; GFX11-NEXT: SI_END_CF [[PHI3]], implicit-def dead $exec, implicit-def dead $scc, implicit $exec
; GFX11-NEXT: $vgpr0 = COPY [[PHI2]]
; GFX11-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = atomicrmw fadd float* %ptr, float %data syncscope("wavefront") monotonic
ret float %ret
}

declare float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float*, float)

attributes #0 = {"amdgpu-unsafe-fp-atomics"="true" }
91 changes: 91 additions & 0 deletions llvm/test/CodeGen/AMDGPU/flat-atomic-fadd.f64.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,91 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s

define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_intrinsic(double* %ptr, double %data) {
; GFX90A_GFX940-LABEL: name: flat_atomic_fadd_f64_no_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: FLAT_ATOMIC_ADD_F64 killed [[COPY4]], killed [[COPY5]], 0, 0, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s64) on %ir.ptr)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.flat.atomic.fadd.f64.p1f64.f64(double* %ptr, double %data)
ret void
}

define amdgpu_ps double @flat_atomic_fadd_f64_rtn_intrinsic(double* %ptr, double %data) {
; GFX90A_GFX940-LABEL: name: flat_atomic_fadd_f64_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[FLAT_ATOMIC_ADD_F64_RTN:%[0-9]+]]:vreg_64_align2 = FLAT_ATOMIC_ADD_F64_RTN killed [[COPY4]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s64) on %ir.ptr)
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX940-NEXT: $sgpr0 = COPY [[COPY6]]
; GFX90A_GFX940-NEXT: $sgpr1 = COPY [[COPY7]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.flat.atomic.fadd.f64.p1f64.f64(double* %ptr, double %data)
ret double %ret
}

define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_atomicrmw(double* %ptr, double %data) #0 {
; GFX90A_GFX940-LABEL: name: flat_atomic_fadd_f64_no_rtn_atomicrmw
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: FLAT_ATOMIC_ADD_F64 killed [[COPY4]], killed [[COPY5]], 0, 0, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s64) on %ir.ptr)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd double* %ptr, double %data syncscope("wavefront") monotonic
ret void
}

define amdgpu_ps double @flat_atomic_fadd_f64_rtn_atomicrmw(double* %ptr, double %data) #0 {
; GFX90A_GFX940-LABEL: name: flat_atomic_fadd_f64_rtn_atomicrmw
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX940-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[FLAT_ATOMIC_ADD_F64_RTN:%[0-9]+]]:vreg_64_align2 = FLAT_ATOMIC_ADD_F64_RTN killed [[COPY4]], killed [[COPY5]], 0, 1, implicit $exec, implicit $flat_scr :: (load store syncscope("wavefront") monotonic (s64) on %ir.ptr)
; GFX90A_GFX940-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX940-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX940-NEXT: $sgpr0 = COPY [[COPY6]]
; GFX90A_GFX940-NEXT: $sgpr1 = COPY [[COPY7]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = atomicrmw fadd double* %ptr, double %data syncscope("wavefront") monotonic
ret double %ret
}

declare double @llvm.amdgcn.flat.atomic.fadd.f64.p1f64.f64(double*, double)

attributes #0 = {"amdgpu-unsafe-fp-atomics"="true" }
37 changes: 37 additions & 0 deletions llvm/test/CodeGen/AMDGPU/flat-atomic-fadd.v2f16.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,37 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX940 %s

define amdgpu_ps void @flat_atomic_fadd_v2f16_no_rtn_intrinsic(<2 x half>* %ptr, <2 x half> %data) {
; GFX940-LABEL: name: flat_atomic_fadd_v2f16_no_rtn_intrinsic
; GFX940: bb.0 (%ir-block.0):
; GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX940-NEXT: {{ $}}
; GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX940-NEXT: FLAT_ATOMIC_PK_ADD_F16 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s32) on %ir.ptr)
; GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half>* %ptr, <2 x half> %data)
ret void
}

define amdgpu_ps <2 x half> @flat_atomic_fadd_v2f16_rtn_intrinsic(<2 x half>* %ptr, <2 x half> %data) {
; GFX940-LABEL: name: flat_atomic_fadd_v2f16_rtn_intrinsic
; GFX940: bb.0 (%ir-block.0):
; GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX940-NEXT: {{ $}}
; GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX940-NEXT: [[FLAT_ATOMIC_PK_ADD_F16_RTN:%[0-9]+]]:vgpr_32 = FLAT_ATOMIC_PK_ADD_F16_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec, implicit $flat_scr :: (volatile dereferenceable load store (s32) on %ir.ptr)
; GFX940-NEXT: $vgpr0 = COPY [[FLAT_ATOMIC_PK_ADD_F16_RTN]]
; GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half>* %ptr, <2 x half> %data)
ret <2 x half> %ret
}

declare <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half>*, <2 x half>)
172 changes: 172 additions & 0 deletions llvm/test/CodeGen/AMDGPU/global-atomic-fadd.f32-no-rtn.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,172 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX908_GFX11 %s
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX908_GFX11 %s

define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_intrinsic(float addrspace(1)* %ptr, float %data) {
; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_no_rtn_intrinsic
; GFX908_GFX11: bb.0 (%ir-block.0):
; GFX908_GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX908_GFX11-NEXT: {{ $}}
; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908_GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908_GFX11-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_no_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %ptr, float %data)
ret void
}

define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_intrinsic(float addrspace(1)* inreg %ptr, float %data) {
; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_intrinsic
; GFX908_GFX11: bb.0 (%ir-block.0):
; GFX908_GFX11-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX908_GFX11-NEXT: {{ $}}
; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908_GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908_GFX11-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* inreg %ptr, float %data)
ret void
}

define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_flat_intrinsic(float addrspace(1)* %ptr, float %data) {
; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_no_rtn_flat_intrinsic
; GFX908_GFX11: bb.0 (%ir-block.0):
; GFX908_GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX908_GFX11-NEXT: {{ $}}
; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908_GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908_GFX11-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_no_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %ptr, float %data)
ret void
}

define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_flat_intrinsic(float addrspace(1)* inreg %ptr, float %data) {
; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_flat_intrinsic
; GFX908_GFX11: bb.0 (%ir-block.0):
; GFX908_GFX11-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX908_GFX11-NEXT: {{ $}}
; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908_GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908_GFX11-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float addrspace(1)* inreg %ptr, float %data)
ret void
}

define amdgpu_ps void @global_atomic_fadd_f32_no_rtn_atomicrmw(float addrspace(1)* %ptr, float %data) #0 {
; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_no_rtn_atomicrmw
; GFX908_GFX11: bb.0 (%ir-block.0):
; GFX908_GFX11-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX908_GFX11-NEXT: {{ $}}
; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908_GFX11-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
; GFX908_GFX11-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_no_rtn_atomicrmw
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd float addrspace(1)* %ptr, float %data syncscope("wavefront") monotonic
ret void
}

define amdgpu_ps void @global_atomic_fadd_f32_saddr_no_rtn_atomicrmw(float addrspace(1)* inreg %ptr, float %data) #0 {
; GFX908_GFX11-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_atomicrmw
; GFX908_GFX11: bb.0 (%ir-block.0):
; GFX908_GFX11-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX908_GFX11-NEXT: {{ $}}
; GFX908_GFX11-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908_GFX11-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908_GFX11-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908_GFX11-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908_GFX11-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX908_GFX11-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
; GFX908_GFX11-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_no_rtn_atomicrmw
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_ADD_F32_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd float addrspace(1)* %ptr, float %data syncscope("wavefront") monotonic
ret void
}

attributes #0 = {"amdgpu-unsafe-fp-atomics"="true" }

declare float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)*, float)
declare float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float addrspace(1)*, float)
113 changes: 113 additions & 0 deletions llvm/test/CodeGen/AMDGPU/global-atomic-fadd.f32-rtn.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,113 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: not --crash llc -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs -stop-after=amdgpu-isel < %s 2>&1 | FileCheck -check-prefix=GFX11 %s

; GFX11: error: {{.*}} return versions of fp atomics not supported

define amdgpu_ps float @global_atomic_fadd_f32_rtn_intrinsic(float addrspace(1)* %ptr, float %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_ADD_F32_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_ADD_F32_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_ADD_F32_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %ptr, float %data)
ret float %ret
}

define amdgpu_ps float @global_atomic_fadd_f32_saddr_rtn_intrinsic(float addrspace(1)* inreg %ptr, float %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_ADD_F32_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_ADD_F32_SADDR_RTN killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_ADD_F32_SADDR_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)* inreg %ptr, float %data)
ret float %ret
}

define amdgpu_ps float @global_atomic_fadd_f32_rtn_flat_intrinsic(float addrspace(1)* %ptr, float %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_ADD_F32_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_ADD_F32_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_ADD_F32_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float addrspace(1)* %ptr, float %data)
ret float %ret
}

define amdgpu_ps float @global_atomic_fadd_f32_saddr_rtn_flat_intrinsic(float addrspace(1)* inreg %ptr, float %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_ADD_F32_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_ADD_F32_SADDR_RTN killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_ADD_F32_SADDR_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float addrspace(1)* inreg %ptr, float %data)
ret float %ret
}

define amdgpu_ps float @global_atomic_fadd_f32_rtn_atomicrmw(float addrspace(1)* %ptr, float %data) #0 {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_rtn_atomicrmw
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_ADD_F32_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_ADD_F32_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_ADD_F32_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = atomicrmw fadd float addrspace(1)* %ptr, float %data syncscope("wavefront") monotonic
ret float %ret
}

define amdgpu_ps float @global_atomic_fadd_f32_saddr_rtn_atomicrmw(float addrspace(1)* inreg %ptr, float %data) #0 {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_f32_saddr_rtn_atomicrmw
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_ADD_F32_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_ADD_F32_SADDR_RTN killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 1, implicit $exec :: (load store syncscope("wavefront") monotonic (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_ADD_F32_SADDR_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = atomicrmw fadd float addrspace(1)* %ptr, float %data syncscope("wavefront") monotonic
ret float %ret
}

declare float @llvm.amdgcn.global.atomic.fadd.f32.p1f32.f32(float addrspace(1)*, float)
declare float @llvm.amdgcn.flat.atomic.fadd.f32.p1f32.f32(float addrspace(1)*, float)

attributes #0 = {"amdgpu-unsafe-fp-atomics"="true" }
260 changes: 260 additions & 0 deletions llvm/test/CodeGen/AMDGPU/global-atomic-fadd.f64.ll
View file

Large diffs are not rendered by default.

115 changes: 115 additions & 0 deletions llvm/test/CodeGen/AMDGPU/global-atomic-fadd.v2f16-no-rtn.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,115 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx908 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX908 %s
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s

define amdgpu_ps void @global_atomic_fadd_v2f16_no_rtn_intrinsic(<2 x half> addrspace(1)* %ptr, <2 x half> %data) {
; GFX908-LABEL: name: global_atomic_fadd_v2f16_no_rtn_intrinsic
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX908-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX908-NEXT: GLOBAL_ATOMIC_PK_ADD_F16 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_no_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_PK_ADD_F16 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%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
}

define amdgpu_ps void @global_atomic_fadd_v2f16_saddr_no_rtn_intrinsic(<2 x half> addrspace(1)* inreg %ptr, <2 x half> %data) {
; GFX908-LABEL: name: global_atomic_fadd_v2f16_saddr_no_rtn_intrinsic
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX908-NEXT: GLOBAL_ATOMIC_PK_ADD_F16_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_saddr_no_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_PK_ADD_F16_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%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
}

define amdgpu_ps void @global_atomic_fadd_v2f16_no_rtn_flat_intrinsic(<2 x half> addrspace(1)* %ptr, <2 x half> %data) {
; GFX908-LABEL: name: global_atomic_fadd_v2f16_no_rtn_flat_intrinsic
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX908-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX908-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908-NEXT: [[COPY3:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]]
; GFX908-NEXT: GLOBAL_ATOMIC_PK_ADD_F16 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_no_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_PK_ADD_F16 killed [[COPY3]], [[COPY]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret void
}

define amdgpu_ps void @global_atomic_fadd_v2f16_saddr_no_rtn_flat_intrinsic(<2 x half> addrspace(1)* inreg %ptr, <2 x half> %data) {
; GFX908-LABEL: name: global_atomic_fadd_v2f16_saddr_no_rtn_flat_intrinsic
; GFX908: bb.0 (%ir-block.0):
; GFX908-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX908-NEXT: {{ $}}
; GFX908-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX908-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX908-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX908-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX908-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX908-NEXT: GLOBAL_ATOMIC_PK_ADD_F16_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX908-NEXT: S_ENDPGM 0
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_saddr_no_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: GLOBAL_ATOMIC_PK_ADD_F16_SADDR killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 0, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: S_ENDPGM 0
%ret = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret void
}

declare <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)*, <2 x half>)
declare <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)*, <2 x half>)
74 changes: 74 additions & 0 deletions llvm/test/CodeGen/AMDGPU/global-atomic-fadd.v2f16-rtn.ll
View file
Original file line number Diff line number Diff line change
@@ -0,0 +1,74 @@
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -march=amdgcn -mcpu=gfx90a -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s
; RUN: llc -march=amdgcn -mcpu=gfx940 -verify-machineinstrs -stop-after=amdgpu-isel < %s | FileCheck -check-prefix=GFX90A_GFX940 %s

define amdgpu_ps <2 x half> @global_atomic_fadd_v2f16_rtn_intrinsic(<2 x half> addrspace(1)* %ptr, <2 x half> %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_PK_ADD_F16_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_PK_ADD_F16_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_PK_ADD_F16_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret <2 x half> %ret
}

define amdgpu_ps <2 x half> @global_atomic_fadd_v2f16_saddr_rtn_intrinsic(<2 x half> addrspace(1)* inreg %ptr, <2 x half> %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_saddr_rtn_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_PK_ADD_F16_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_PK_ADD_F16_SADDR_RTN killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_PK_ADD_F16_SADDR_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret <2 x half> %ret
}

define amdgpu_ps <2 x half> @global_atomic_fadd_v2f16_rtn_flat_intrinsic(<2 x half> addrspace(1)* %ptr, <2 x half> %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[COPY3:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_PK_ADD_F16_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_PK_ADD_F16_RTN killed [[COPY3]], [[COPY]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_PK_ADD_F16_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret <2 x half> %ret
}

define amdgpu_ps <2 x half> @global_atomic_fadd_v2f16_saddr_rtn_flat_intrinsic(<2 x half> addrspace(1)* inreg %ptr, <2 x half> %data) {
; GFX90A_GFX940-LABEL: name: global_atomic_fadd_v2f16_saddr_rtn_flat_intrinsic
; GFX90A_GFX940: bb.0 (%ir-block.0):
; GFX90A_GFX940-NEXT: liveins: $sgpr0, $sgpr1, $vgpr0
; GFX90A_GFX940-NEXT: {{ $}}
; GFX90A_GFX940-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX940-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; GFX90A_GFX940-NEXT: [[COPY2:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; GFX90A_GFX940-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sgpr_64 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY1]], %subreg.sub1
; GFX90A_GFX940-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
; GFX90A_GFX940-NEXT: [[GLOBAL_ATOMIC_PK_ADD_F16_SADDR_RTN:%[0-9]+]]:vgpr_32 = GLOBAL_ATOMIC_PK_ADD_F16_SADDR_RTN killed [[V_MOV_B32_e32_]], [[COPY]], killed [[REG_SEQUENCE]], 0, 1, implicit $exec :: (volatile dereferenceable load store (s32) on %ir.ptr, addrspace 1)
; GFX90A_GFX940-NEXT: $vgpr0 = COPY [[GLOBAL_ATOMIC_PK_ADD_F16_SADDR_RTN]]
; GFX90A_GFX940-NEXT: SI_RETURN_TO_EPILOG $vgpr0
%ret = call <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)* %ptr, <2 x half> %data)
ret <2 x half> %ret
}

declare <2 x half> @llvm.amdgcn.global.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)*, <2 x half>)
declare <2 x half> @llvm.amdgcn.flat.atomic.fadd.v2f16.p1v2f16.v2f16(<2 x half> addrspace(1)*, <2 x half>)
99 changes: 0 additions & 99 deletions llvm/test/CodeGen/AMDGPU/llvm.amdgcn.atomic.fadd.rtn_no-rtn.ll
View file

This file was deleted.

Loading