shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/flat-atomic-fadd.f64.ll
Chris Jackson 7d0203b39f
[AMDGPU] Prevent generation of unused SGPR IMPLICIT_DEF assignments (#155241) 
Dead VGPR->SGPR copies were converted to IMPLICIT_DEF assignments that
were unused. Prevent these from being created and update the numerous
affected tests.
2025-08-27 13:18:18 +01:00

143 lines
10 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
; RUN: llc -mtriple=amdgcn -mcpu=gfx90a -stop-after=finalize-isel < %s | FileCheck -check-prefix=GFX90A_GFX942 %s
; RUN: llc -mtriple=amdgcn -mcpu=gfx942 -stop-after=finalize-isel < %s | FileCheck -check-prefix=GFX90A_GFX942 %s
define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_intrinsic(ptr %ptr, double %data) {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_no_rtn_intrinsic
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:av_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-NEXT: FLAT_ATOMIC_ADD_F64 killed [[COPY4]], killed [[COPY5]], 0, 0, implicit $exec, implicit $flat_scr :: (load store syncscope("agent") seq_cst (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: S_ENDPGM 0
%ret = call double @llvm.amdgcn.flat.atomic.fadd.f64.p1.f64(ptr %ptr, double %data)
ret void
}
define amdgpu_ps double @flat_atomic_fadd_f64_rtn_intrinsic(ptr %ptr, double %data) {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_rtn_intrinsic
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-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("agent") seq_cst (s64) on %ir.ptr)
; GFX90A_GFX942-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY6]], implicit $exec
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY7]], implicit $exec
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = call double @llvm.amdgcn.flat.atomic.fadd.f64.p1.f64(ptr %ptr, double %data)
ret double %ret
}
define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_atomicrmw(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_no_rtn_atomicrmw
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:av_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-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_GFX942-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret void
}
define amdgpu_ps void @flat_atomic_fadd_f64_no_rtn_atomicrmw_noprivate(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_no_rtn_atomicrmw_noprivate
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:av_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-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_GFX942-NEXT: S_ENDPGM 0
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret void
}
define amdgpu_ps double @flat_atomic_fadd_f64_rtn_atomicrmw(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_rtn_atomicrmw
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-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_GFX942-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY6]], implicit $exec
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY7]], implicit $exec
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret double %ret
}
define amdgpu_ps double @flat_atomic_fadd_f64_rtn_atomicrmw__noprivate(ptr %ptr, double %data) #0 {
; GFX90A_GFX942-LABEL: name: flat_atomic_fadd_f64_rtn_atomicrmw__noprivate
; GFX90A_GFX942: bb.0 (%ir-block.0):
; GFX90A_GFX942-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
; GFX90A_GFX942-NEXT: {{ $}}
; GFX90A_GFX942-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr3
; GFX90A_GFX942-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr2
; GFX90A_GFX942-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr1
; GFX90A_GFX942-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr0
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY1]], %subreg.sub0, [[COPY]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY3]], %subreg.sub0, [[COPY2]], %subreg.sub1
; GFX90A_GFX942-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE1]]
; GFX90A_GFX942-NEXT: [[COPY5:%[0-9]+]]:vreg_64_align2 = COPY [[REG_SEQUENCE]]
; GFX90A_GFX942-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_GFX942-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub0
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY6]], implicit $exec
; GFX90A_GFX942-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[FLAT_ATOMIC_ADD_F64_RTN]].sub1
; GFX90A_GFX942-NEXT: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 killed [[COPY7]], implicit $exec
; GFX90A_GFX942-NEXT: $sgpr0 = COPY [[V_READFIRSTLANE_B32_]]
; GFX90A_GFX942-NEXT: $sgpr1 = COPY [[V_READFIRSTLANE_B32_1]]
; GFX90A_GFX942-NEXT: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1
%ret = atomicrmw fadd ptr %ptr, double %data syncscope("wavefront") monotonic, !noalias.addrspace !1, !amdgpu.no.fine.grained.memory !0
ret double %ret
}
declare double @llvm.amdgcn.flat.atomic.fadd.f64.p1.f64(ptr, double)
attributes #0 = { nounwind }
!0 = !{}
!1 = !{i32 5, i32 6}
Reference in New Issue View Git Blame Copy Permalink