llvm/test/CodeGen/AMDGPU/global_atomics_optimizer_fp_no_rtn.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN:  opt -S -mtriple=amdgcn-- -mcpu=gfx906 -passes='amdgpu-atomic-optimizer<strategy=iterative>,verify<domtree>' %s | FileCheck --check-prefixes=IR,IR-ITERATIVE %s
; RUN:  opt -S -mtriple=amdgcn-- -mcpu=gfx906 -passes='amdgpu-atomic-optimizer<strategy=dpp>,verify<domtree>' %s | FileCheck --check-prefixes=IR,IR-DPP %s

; Tests various combinations of uniform/divergent address and uniform/divergent value inputs of various types for atomic operations.
; Optimization remains same for Iterative and DPP strategies when value in uniform. These different scan/reduction
; strategies are valid for only divergent values. This optimization is valid for divergent addresses. Test also covers different scopes.

define amdgpu_ps void @global_atomic_fadd_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, float inreg %val) #0 {
; IR-LABEL: @global_atomic_fadd_uni_address_uni_value_agent_scope_unsafe(
; IR-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR:       2:
; IR-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]])
; IR-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-NEXT:    [[TMP11:%.*]] = uitofp i32 [[TMP10]] to float
; IR-NEXT:    [[TMP12:%.*]] = fmul float [[VAL:%.*]], [[TMP11]]
; IR-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR:       14:
; IR-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] syncscope("agent") monotonic, align 4
; IR-NEXT:    br label [[TMP16]]
; IR:       16:
; IR-NEXT:    br label [[TMP17]]
; IR:       17:
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_uni_address_div_value_scope_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, float %val) #0 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_uni_address_div_value_scope_agent_scope_unsafe(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-ITERATIVE-NEXT:    br label [[COMPUTELOOP:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP19:%.*]] syncscope("agent") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12:%.*]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
; IR-ITERATIVE:       ComputeLoop:
; IR-ITERATIVE-NEXT:    [[ACCUMULATOR:%.*]] = phi float [ -0.000000e+00, [[TMP2]] ], [ [[TMP19]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[ACTIVEBITS:%.*]] = phi i64 [ [[TMP9]], [[TMP2]] ], [ [[TMP22:%.*]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[TMP14:%.*]] = call i64 @llvm.cttz.i64(i64 [[ACTIVEBITS]], i1 true)
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
; IR-ITERATIVE-NEXT:    [[TMP16:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-ITERATIVE-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP16]], i32 [[TMP15]])
; IR-ITERATIVE-NEXT:    [[TMP18:%.*]] = bitcast i32 [[TMP17]] to float
; IR-ITERATIVE-NEXT:    [[TMP19]] = fadd float [[ACCUMULATOR]], [[TMP18]]
; IR-ITERATIVE-NEXT:    [[TMP20:%.*]] = shl i64 1, [[TMP14]]
; IR-ITERATIVE-NEXT:    [[TMP21:%.*]] = xor i64 [[TMP20]], -1
; IR-ITERATIVE-NEXT:    [[TMP22]] = and i64 [[ACTIVEBITS]], [[TMP21]]
; IR-ITERATIVE-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[TMP22]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP23]], label [[COMPUTEEND:%.*]], label [[COMPUTELOOP]]
; IR-ITERATIVE:       ComputeEnd:
; IR-ITERATIVE-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP24]], label [[TMP10:%.*]], label [[TMP12]]
;
; IR-DPP-LABEL: @global_atomic_fadd_uni_address_div_value_scope_agent_scope_unsafe(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP33:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-DPP-NEXT:    [[TMP9:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-DPP-NEXT:    [[TMP10:%.*]] = call i32 @llvm.amdgcn.set.inactive.i32(i32 [[TMP9]], i32 -2147483648)
; IR-DPP-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP10]] to float
; IR-DPP-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP9]] to float
; IR-DPP-NEXT:    [[TMP13:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP11]], i32 273, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP14:%.*]] = fadd float [[TMP11]], [[TMP13]]
; IR-DPP-NEXT:    [[TMP15:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP14]], i32 274, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP16:%.*]] = fadd float [[TMP14]], [[TMP15]]
; IR-DPP-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP16]], i32 276, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP18:%.*]] = fadd float [[TMP16]], [[TMP17]]
; IR-DPP-NEXT:    [[TMP19:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP18]], i32 280, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP20:%.*]] = fadd float [[TMP18]], [[TMP19]]
; IR-DPP-NEXT:    [[TMP21:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP20]], i32 322, i32 10, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP22:%.*]] = fadd float [[TMP20]], [[TMP21]]
; IR-DPP-NEXT:    [[TMP23:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP22]], i32 323, i32 12, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP24:%.*]] = fadd float [[TMP22]], [[TMP23]]
; IR-DPP-NEXT:    [[TMP25:%.*]] = bitcast float [[TMP24]] to i32
; IR-DPP-NEXT:    [[TMP26:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP25]], i32 63)
; IR-DPP-NEXT:    [[TMP27:%.*]] = bitcast i32 [[TMP26]] to float
; IR-DPP-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.strict.wwm.f32(float [[TMP27]])
; IR-DPP-NEXT:    [[TMP29:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP29]], label [[TMP30:%.*]], label [[TMP32:%.*]]
; IR-DPP:       30:
; IR-DPP-NEXT:    [[TMP31:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP28]] syncscope("agent") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP32]]
; IR-DPP:       32:
; IR-DPP-NEXT:    br label [[TMP33]]
; IR-DPP:       33:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_uni_address_uni_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, float inreg %val) #1 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_uni_address_uni_value_one_as_scope_unsafe_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7:[0-9]+]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP12:%.*]] = call float @llvm.experimental.constrained.fmul.f32(float [[VAL:%.*]], float [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-ITERATIVE:       14:
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] syncscope("one-as") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP16]]
; IR-ITERATIVE:       16:
; IR-ITERATIVE-NEXT:    br label [[TMP17]]
; IR-ITERATIVE:       17:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fadd_uni_address_uni_value_one_as_scope_unsafe_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8:[0-9]+]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-DPP-NEXT:    [[TMP11:%.*]] = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP12:%.*]] = call float @llvm.experimental.constrained.fmul.f32(float [[VAL:%.*]], float [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-DPP:       14:
; IR-DPP-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] syncscope("one-as") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP16]]
; IR-DPP:       16:
; IR-DPP-NEXT:    br label [[TMP17]]
; IR-DPP:       17:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_uni_address_div_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, float %val) #1 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_uni_address_div_value_one_as_scope_unsafe_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br label [[COMPUTELOOP:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP19:%.*]] syncscope("one-as") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12:%.*]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
; IR-ITERATIVE:       ComputeLoop:
; IR-ITERATIVE-NEXT:    [[ACCUMULATOR:%.*]] = phi float [ -0.000000e+00, [[TMP2]] ], [ [[TMP19]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[ACTIVEBITS:%.*]] = phi i64 [ [[TMP9]], [[TMP2]] ], [ [[TMP22:%.*]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[TMP14:%.*]] = call i64 @llvm.cttz.i64(i64 [[ACTIVEBITS]], i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
; IR-ITERATIVE-NEXT:    [[TMP16:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-ITERATIVE-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP16]], i32 [[TMP15]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP18:%.*]] = bitcast i32 [[TMP17]] to float
; IR-ITERATIVE-NEXT:    [[TMP19]] = call float @llvm.experimental.constrained.fadd.f32(float [[ACCUMULATOR]], float [[TMP18]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP20:%.*]] = shl i64 1, [[TMP14]]
; IR-ITERATIVE-NEXT:    [[TMP21:%.*]] = xor i64 [[TMP20]], -1
; IR-ITERATIVE-NEXT:    [[TMP22]] = and i64 [[ACTIVEBITS]], [[TMP21]]
; IR-ITERATIVE-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[TMP22]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP23]], label [[COMPUTEEND:%.*]], label [[COMPUTELOOP]]
; IR-ITERATIVE:       ComputeEnd:
; IR-ITERATIVE-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP24]], label [[TMP10:%.*]], label [[TMP12]]
;
; IR-DPP-LABEL: @global_atomic_fadd_uni_address_div_value_one_as_scope_unsafe_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP33:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-DPP-NEXT:    [[TMP10:%.*]] = call i32 @llvm.amdgcn.set.inactive.i32(i32 [[TMP9]], i32 -2147483648) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP10]] to float
; IR-DPP-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP9]] to float
; IR-DPP-NEXT:    [[TMP13:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP11]], i32 273, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP14:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP11]], float [[TMP13]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP15:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP14]], i32 274, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP16:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP14]], float [[TMP15]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP16]], i32 276, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP18:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP16]], float [[TMP17]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP19:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP18]], i32 280, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP20:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP18]], float [[TMP19]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP21:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP20]], i32 322, i32 10, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP22:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP20]], float [[TMP21]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP23:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP22]], i32 323, i32 12, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP24:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP22]], float [[TMP23]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP25:%.*]] = bitcast float [[TMP24]] to i32
; IR-DPP-NEXT:    [[TMP26:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP25]], i32 63) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP27:%.*]] = bitcast i32 [[TMP26]] to float
; IR-DPP-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.strict.wwm.f32(float [[TMP27]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP29:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP29]], label [[TMP30:%.*]], label [[TMP32:%.*]]
; IR-DPP:       30:
; IR-DPP-NEXT:    [[TMP31:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP28]] syncscope("one-as") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP32]]
; IR-DPP:       32:
; IR-DPP-NEXT:    br label [[TMP33]]
; IR-DPP:       33:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_uni_address_uni_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, float inreg %val) #2 {
; IR-ITERATIVE-LABEL: @global_atomic_fsub_uni_address_uni_value_agent_scope_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP12:%.*]] = call float @llvm.experimental.constrained.fmul.f32(float [[VAL:%.*]], float [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-ITERATIVE:       14:
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] syncscope("agent") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP16]]
; IR-ITERATIVE:       16:
; IR-ITERATIVE-NEXT:    br label [[TMP17]]
; IR-ITERATIVE:       17:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fsub_uni_address_uni_value_agent_scope_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-DPP-NEXT:    [[TMP11:%.*]] = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP12:%.*]] = call float @llvm.experimental.constrained.fmul.f32(float [[VAL:%.*]], float [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-DPP:       14:
; IR-DPP-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] syncscope("agent") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP16]]
; IR-DPP:       16:
; IR-DPP-NEXT:    br label [[TMP17]]
; IR-DPP:       17:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_uni_address_div_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, float %val) #2 {
; IR-ITERATIVE-LABEL: @global_atomic_fsub_uni_address_div_value_agent_scope_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br label [[COMPUTELOOP:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], float [[TMP19:%.*]] syncscope("agent") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12:%.*]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
; IR-ITERATIVE:       ComputeLoop:
; IR-ITERATIVE-NEXT:    [[ACCUMULATOR:%.*]] = phi float [ -0.000000e+00, [[TMP2]] ], [ [[TMP19]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[ACTIVEBITS:%.*]] = phi i64 [ [[TMP9]], [[TMP2]] ], [ [[TMP22:%.*]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[TMP14:%.*]] = call i64 @llvm.cttz.i64(i64 [[ACTIVEBITS]], i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
; IR-ITERATIVE-NEXT:    [[TMP16:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-ITERATIVE-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP16]], i32 [[TMP15]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP18:%.*]] = bitcast i32 [[TMP17]] to float
; IR-ITERATIVE-NEXT:    [[TMP19]] = call float @llvm.experimental.constrained.fadd.f32(float [[ACCUMULATOR]], float [[TMP18]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP20:%.*]] = shl i64 1, [[TMP14]]
; IR-ITERATIVE-NEXT:    [[TMP21:%.*]] = xor i64 [[TMP20]], -1
; IR-ITERATIVE-NEXT:    [[TMP22]] = and i64 [[ACTIVEBITS]], [[TMP21]]
; IR-ITERATIVE-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[TMP22]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP23]], label [[COMPUTEEND:%.*]], label [[COMPUTELOOP]]
; IR-ITERATIVE:       ComputeEnd:
; IR-ITERATIVE-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP24]], label [[TMP10:%.*]], label [[TMP12]]
;
; IR-DPP-LABEL: @global_atomic_fsub_uni_address_div_value_agent_scope_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP33:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-DPP-NEXT:    [[TMP10:%.*]] = call i32 @llvm.amdgcn.set.inactive.i32(i32 [[TMP9]], i32 -2147483648) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP10]] to float
; IR-DPP-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP9]] to float
; IR-DPP-NEXT:    [[TMP13:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP11]], i32 273, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP14:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP11]], float [[TMP13]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP15:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP14]], i32 274, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP16:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP14]], float [[TMP15]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP16]], i32 276, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP18:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP16]], float [[TMP17]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP19:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP18]], i32 280, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP20:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP18]], float [[TMP19]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP21:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP20]], i32 322, i32 10, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP22:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP20]], float [[TMP21]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP23:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP22]], i32 323, i32 12, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP24:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP22]], float [[TMP23]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP25:%.*]] = bitcast float [[TMP24]] to i32
; IR-DPP-NEXT:    [[TMP26:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP25]], i32 63) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP27:%.*]] = bitcast i32 [[TMP26]] to float
; IR-DPP-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.strict.wwm.f32(float [[TMP27]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP29:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP29]], label [[TMP30:%.*]], label [[TMP32:%.*]]
; IR-DPP:       30:
; IR-DPP-NEXT:    [[TMP31:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], float [[TMP28]] syncscope("agent") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP32]]
; IR-DPP:       32:
; IR-DPP-NEXT:    br label [[TMP33]]
; IR-DPP:       33:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fsub ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, float inreg %val) #0 {
; IR-LABEL: @global_atomic_fmin_uni_address_uni_value_agent_scope_unsafe(
; IR-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR:       2:
; IR-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-NEXT:    br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
; IR:       10:
; IR-NEXT:    [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    br label [[TMP12]]
; IR:       12:
; IR-NEXT:    br label [[TMP13]]
; IR:       13:
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, float %val) #0 {
; IR-ITERATIVE-LABEL: @global_atomic_fmin_uni_address_div_value_agent_scope_unsafe(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-ITERATIVE-NEXT:    br label [[COMPUTELOOP:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], float [[TMP19:%.*]] syncscope("agent") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12:%.*]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
; IR-ITERATIVE:       ComputeLoop:
; IR-ITERATIVE-NEXT:    [[ACCUMULATOR:%.*]] = phi float [ 0x7FF0000000000000, [[TMP2]] ], [ [[TMP19]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[ACTIVEBITS:%.*]] = phi i64 [ [[TMP9]], [[TMP2]] ], [ [[TMP22:%.*]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[TMP14:%.*]] = call i64 @llvm.cttz.i64(i64 [[ACTIVEBITS]], i1 true)
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
; IR-ITERATIVE-NEXT:    [[TMP16:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-ITERATIVE-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP16]], i32 [[TMP15]])
; IR-ITERATIVE-NEXT:    [[TMP18:%.*]] = bitcast i32 [[TMP17]] to float
; IR-ITERATIVE-NEXT:    [[TMP19]] = call float @llvm.minnum.f32(float [[ACCUMULATOR]], float [[TMP18]])
; IR-ITERATIVE-NEXT:    [[TMP20:%.*]] = shl i64 1, [[TMP14]]
; IR-ITERATIVE-NEXT:    [[TMP21:%.*]] = xor i64 [[TMP20]], -1
; IR-ITERATIVE-NEXT:    [[TMP22]] = and i64 [[ACTIVEBITS]], [[TMP21]]
; IR-ITERATIVE-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[TMP22]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP23]], label [[COMPUTEEND:%.*]], label [[COMPUTELOOP]]
; IR-ITERATIVE:       ComputeEnd:
; IR-ITERATIVE-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP24]], label [[TMP10:%.*]], label [[TMP12]]
;
; IR-DPP-LABEL: @global_atomic_fmin_uni_address_div_value_agent_scope_unsafe(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP33:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-DPP-NEXT:    [[TMP9:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-DPP-NEXT:    [[TMP10:%.*]] = call i32 @llvm.amdgcn.set.inactive.i32(i32 [[TMP9]], i32 2139095040)
; IR-DPP-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP10]] to float
; IR-DPP-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP9]] to float
; IR-DPP-NEXT:    [[TMP13:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0x7FF0000000000000, float [[TMP11]], i32 273, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP14:%.*]] = call float @llvm.minnum.f32(float [[TMP11]], float [[TMP13]])
; IR-DPP-NEXT:    [[TMP15:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0x7FF0000000000000, float [[TMP14]], i32 274, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP16:%.*]] = call float @llvm.minnum.f32(float [[TMP14]], float [[TMP15]])
; IR-DPP-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0x7FF0000000000000, float [[TMP16]], i32 276, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP18:%.*]] = call float @llvm.minnum.f32(float [[TMP16]], float [[TMP17]])
; IR-DPP-NEXT:    [[TMP19:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0x7FF0000000000000, float [[TMP18]], i32 280, i32 15, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP20:%.*]] = call float @llvm.minnum.f32(float [[TMP18]], float [[TMP19]])
; IR-DPP-NEXT:    [[TMP21:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0x7FF0000000000000, float [[TMP20]], i32 322, i32 10, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP22:%.*]] = call float @llvm.minnum.f32(float [[TMP20]], float [[TMP21]])
; IR-DPP-NEXT:    [[TMP23:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0x7FF0000000000000, float [[TMP22]], i32 323, i32 12, i32 15, i1 false)
; IR-DPP-NEXT:    [[TMP24:%.*]] = call float @llvm.minnum.f32(float [[TMP22]], float [[TMP23]])
; IR-DPP-NEXT:    [[TMP25:%.*]] = bitcast float [[TMP24]] to i32
; IR-DPP-NEXT:    [[TMP26:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP25]], i32 63)
; IR-DPP-NEXT:    [[TMP27:%.*]] = bitcast i32 [[TMP26]] to float
; IR-DPP-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.strict.wwm.f32(float [[TMP27]])
; IR-DPP-NEXT:    [[TMP29:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP29]], label [[TMP30:%.*]], label [[TMP32:%.*]]
; IR-DPP:       30:
; IR-DPP-NEXT:    [[TMP31:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], float [[TMP28]] syncscope("agent") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP32]]
; IR-DPP:       32:
; IR-DPP-NEXT:    br label [[TMP33]]
; IR-DPP:       33:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_uni_address_uni_value_agent_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, float inreg %val) #1{
; IR-ITERATIVE-LABEL: @global_atomic_fmax_uni_address_uni_value_agent_scope_unsafe_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fmax_uni_address_uni_value_agent_scope_unsafe_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
; IR-DPP:       10:
; IR-DPP-NEXT:    [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP12]]
; IR-DPP:       12:
; IR-DPP-NEXT:    br label [[TMP13]]
; IR-DPP:       13:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_uni_address_div_value_agent_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, float %val) #1{
; IR-ITERATIVE-LABEL: @global_atomic_fmax_uni_address_div_value_agent_scope_unsafe_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br label [[COMPUTELOOP:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], float [[TMP19:%.*]] syncscope("agent") monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12:%.*]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
; IR-ITERATIVE:       ComputeLoop:
; IR-ITERATIVE-NEXT:    [[ACCUMULATOR:%.*]] = phi float [ 0xFFF0000000000000, [[TMP2]] ], [ [[TMP19]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[ACTIVEBITS:%.*]] = phi i64 [ [[TMP9]], [[TMP2]] ], [ [[TMP22:%.*]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[TMP14:%.*]] = call i64 @llvm.cttz.i64(i64 [[ACTIVEBITS]], i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
; IR-ITERATIVE-NEXT:    [[TMP16:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-ITERATIVE-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP16]], i32 [[TMP15]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP18:%.*]] = bitcast i32 [[TMP17]] to float
; IR-ITERATIVE-NEXT:    [[TMP19]] = call float @llvm.experimental.constrained.maxnum.f32(float [[ACCUMULATOR]], float [[TMP18]], metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP20:%.*]] = shl i64 1, [[TMP14]]
; IR-ITERATIVE-NEXT:    [[TMP21:%.*]] = xor i64 [[TMP20]], -1
; IR-ITERATIVE-NEXT:    [[TMP22]] = and i64 [[ACTIVEBITS]], [[TMP21]]
; IR-ITERATIVE-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[TMP22]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP23]], label [[COMPUTEEND:%.*]], label [[COMPUTELOOP]]
; IR-ITERATIVE:       ComputeEnd:
; IR-ITERATIVE-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP24]], label [[TMP10:%.*]], label [[TMP12]]
;
; IR-DPP-LABEL: @global_atomic_fmax_uni_address_div_value_agent_scope_unsafe_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP33:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-DPP-NEXT:    [[TMP10:%.*]] = call i32 @llvm.amdgcn.set.inactive.i32(i32 [[TMP9]], i32 -8388608) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP10]] to float
; IR-DPP-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP9]] to float
; IR-DPP-NEXT:    [[TMP13:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0xFFF0000000000000, float [[TMP11]], i32 273, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP14:%.*]] = call float @llvm.experimental.constrained.maxnum.f32(float [[TMP11]], float [[TMP13]], metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP15:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0xFFF0000000000000, float [[TMP14]], i32 274, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP16:%.*]] = call float @llvm.experimental.constrained.maxnum.f32(float [[TMP14]], float [[TMP15]], metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0xFFF0000000000000, float [[TMP16]], i32 276, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP18:%.*]] = call float @llvm.experimental.constrained.maxnum.f32(float [[TMP16]], float [[TMP17]], metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP19:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0xFFF0000000000000, float [[TMP18]], i32 280, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP20:%.*]] = call float @llvm.experimental.constrained.maxnum.f32(float [[TMP18]], float [[TMP19]], metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP21:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0xFFF0000000000000, float [[TMP20]], i32 322, i32 10, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP22:%.*]] = call float @llvm.experimental.constrained.maxnum.f32(float [[TMP20]], float [[TMP21]], metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP23:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float 0xFFF0000000000000, float [[TMP22]], i32 323, i32 12, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP24:%.*]] = call float @llvm.experimental.constrained.maxnum.f32(float [[TMP22]], float [[TMP23]], metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP25:%.*]] = bitcast float [[TMP24]] to i32
; IR-DPP-NEXT:    [[TMP26:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP25]], i32 63) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP27:%.*]] = bitcast i32 [[TMP26]] to float
; IR-DPP-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.strict.wwm.f32(float [[TMP27]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP29:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP29]], label [[TMP30:%.*]], label [[TMP32:%.*]]
; IR-DPP:       30:
; IR-DPP-NEXT:    [[TMP31:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], float [[TMP28]] syncscope("agent") monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP32]]
; IR-DPP:       32:
; IR-DPP-NEXT:    br label [[TMP33]]
; IR-DPP:       33:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_uni_address_uni_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, float inreg %val) #2 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_uni_address_uni_value_system_scope_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP12:%.*]] = call float @llvm.experimental.constrained.fmul.f32(float [[VAL:%.*]], float [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-ITERATIVE:       14:
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP16]]
; IR-ITERATIVE:       16:
; IR-ITERATIVE-NEXT:    br label [[TMP17]]
; IR-ITERATIVE:       17:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fadd_uni_address_uni_value_system_scope_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-DPP-NEXT:    [[TMP11:%.*]] = call float @llvm.experimental.constrained.uitofp.f32.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP12:%.*]] = call float @llvm.experimental.constrained.fmul.f32(float [[VAL:%.*]], float [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-DPP:       14:
; IR-DPP-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP12]] monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP16]]
; IR-DPP:       16:
; IR-DPP-NEXT:    br label [[TMP17]]
; IR-DPP:       17:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_uni_address_div_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, float %val) #2 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_uni_address_div_value_system_scope_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br label [[COMPUTELOOP:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP19:%.*]] monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP12:%.*]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
; IR-ITERATIVE:       ComputeLoop:
; IR-ITERATIVE-NEXT:    [[ACCUMULATOR:%.*]] = phi float [ -0.000000e+00, [[TMP2]] ], [ [[TMP19]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[ACTIVEBITS:%.*]] = phi i64 [ [[TMP9]], [[TMP2]] ], [ [[TMP22:%.*]], [[COMPUTELOOP]] ]
; IR-ITERATIVE-NEXT:    [[TMP14:%.*]] = call i64 @llvm.cttz.i64(i64 [[ACTIVEBITS]], i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = trunc i64 [[TMP14]] to i32
; IR-ITERATIVE-NEXT:    [[TMP16:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-ITERATIVE-NEXT:    [[TMP17:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP16]], i32 [[TMP15]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP18:%.*]] = bitcast i32 [[TMP17]] to float
; IR-ITERATIVE-NEXT:    [[TMP19]] = call float @llvm.experimental.constrained.fadd.f32(float [[ACCUMULATOR]], float [[TMP18]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP20:%.*]] = shl i64 1, [[TMP14]]
; IR-ITERATIVE-NEXT:    [[TMP21:%.*]] = xor i64 [[TMP20]], -1
; IR-ITERATIVE-NEXT:    [[TMP22]] = and i64 [[ACTIVEBITS]], [[TMP21]]
; IR-ITERATIVE-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[TMP22]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP23]], label [[COMPUTEEND:%.*]], label [[COMPUTELOOP]]
; IR-ITERATIVE:       ComputeEnd:
; IR-ITERATIVE-NEXT:    [[TMP24:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP24]], label [[TMP10:%.*]], label [[TMP12]]
;
; IR-DPP-LABEL: @global_atomic_fadd_uni_address_div_value_system_scope_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP33:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = bitcast float [[VAL:%.*]] to i32
; IR-DPP-NEXT:    [[TMP10:%.*]] = call i32 @llvm.amdgcn.set.inactive.i32(i32 [[TMP9]], i32 -2147483648) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP11:%.*]] = bitcast i32 [[TMP10]] to float
; IR-DPP-NEXT:    [[TMP12:%.*]] = bitcast i32 [[TMP9]] to float
; IR-DPP-NEXT:    [[TMP13:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP11]], i32 273, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP14:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP11]], float [[TMP13]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP15:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP14]], i32 274, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP16:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP14]], float [[TMP15]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP17:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP16]], i32 276, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP18:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP16]], float [[TMP17]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP19:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP18]], i32 280, i32 15, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP20:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP18]], float [[TMP19]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP21:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP20]], i32 322, i32 10, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP22:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP20]], float [[TMP21]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP23:%.*]] = call float @llvm.amdgcn.update.dpp.f32(float -0.000000e+00, float [[TMP22]], i32 323, i32 12, i32 15, i1 false) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP24:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[TMP22]], float [[TMP23]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP25:%.*]] = bitcast float [[TMP24]] to i32
; IR-DPP-NEXT:    [[TMP26:%.*]] = call i32 @llvm.amdgcn.readlane(i32 [[TMP25]], i32 63) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP27:%.*]] = bitcast i32 [[TMP26]] to float
; IR-DPP-NEXT:    [[TMP28:%.*]] = call float @llvm.amdgcn.strict.wwm.f32(float [[TMP27]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP29:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP29]], label [[TMP30:%.*]], label [[TMP32:%.*]]
; IR-DPP:       30:
; IR-DPP-NEXT:    [[TMP31:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[TMP28]] monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP32]]
; IR-DPP:       32:
; IR-DPP-NEXT:    br label [[TMP33]]
; IR-DPP:       33:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_div_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr, float inreg %val) #0 {
; IR-LABEL: @global_atomic_fadd_div_address_uni_value_agent_scope_unsafe(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_div_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr, float %val) #0 {
; IR-LABEL: @global_atomic_fadd_div_address_div_value_agent_scope_unsafe(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_div_address_uni_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) %ptr, float inreg %val) #1 {
; IR-LABEL: @global_atomic_fadd_div_address_uni_value_one_as_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("one-as") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_div_address_div_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) %ptr, float %val) #1 {
; IR-LABEL: @global_atomic_fadd_div_address_div_value_one_as_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("one-as") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_div_address_uni_value_agent_scope_strictfp(ptr addrspace(1) %ptr, float inreg %val) #2 {
; IR-LABEL: @global_atomic_fsub_div_address_uni_value_agent_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_div_address_div_value_agent_scope_strictfp(ptr addrspace(1) %ptr, float %val) #2 {
; IR-LABEL: @global_atomic_fsub_div_address_div_value_agent_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fsub ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_div_address_uni_value_agent_scope(ptr addrspace(1) %ptr, float inreg %val) #0 {
; IR-LABEL: @global_atomic_fmin_div_address_uni_value_agent_scope(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_div_address_div_value_agent_scope(ptr addrspace(1) %ptr, float %val) #0 {
; IR-LABEL: @global_atomic_fmin_div_address_div_value_agent_scope(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_div_address_uni_value_agent_scope_unsafe_strictfp(ptr addrspace(1) %ptr, float inreg %val) #1{
; IR-LABEL: @global_atomic_fmax_div_address_uni_value_agent_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_div_address_div_value_agent_scope_unsafe_strictfp(ptr addrspace(1) %ptr, float %val) #1{
; IR-LABEL: @global_atomic_fmax_div_address_div_value_agent_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, float %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_div_address_uni_value_system_scope_strictfp(ptr addrspace(1) %ptr, float inreg %val) #2 {
; IR-LABEL: @global_atomic_fadd_div_address_uni_value_system_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_div_address_div_value_system_scope_strictfp(ptr addrspace(1) %ptr, float %val) #2 {
; IR-LABEL: @global_atomic_fadd_div_address_div_value_system_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], float [[VAL:%.*]] monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, float %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double inreg %val) #0 {
; IR-LABEL: @global_atomic_fadd_double_uni_address_uni_value_agent_scope_unsafe(
; IR-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR:       2:
; IR-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]])
; IR-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-NEXT:    [[TMP11:%.*]] = uitofp i32 [[TMP10]] to double
; IR-NEXT:    [[TMP12:%.*]] = fmul double [[VAL:%.*]], [[TMP11]]
; IR-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR:       14:
; IR-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 4
; IR-NEXT:    br label [[TMP16]]
; IR:       16:
; IR-NEXT:    br label [[TMP17]]
; IR:       17:
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double %val) #0 {
; IR-LABEL: @global_atomic_fadd_double_uni_address_div_value_scope_agent_scope_unsafe(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #1 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-ITERATIVE:       14:
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("one-as") monotonic, align 8
; IR-ITERATIVE-NEXT:    br label [[TMP16]]
; IR-ITERATIVE:       16:
; IR-ITERATIVE-NEXT:    br label [[TMP17]]
; IR-ITERATIVE:       17:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_one_as_scope_unsafe_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-DPP-NEXT:    [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-DPP:       14:
; IR-DPP-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("one-as") monotonic, align 8
; IR-DPP-NEXT:    br label [[TMP16]]
; IR-DPP:       16:
; IR-DPP-NEXT:    br label [[TMP17]]
; IR-DPP:       17:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, double %val) #1 {
; IR-LABEL: @global_atomic_fadd_double_uni_address_div_value_one_as_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #2 {
; IR-ITERATIVE-LABEL: @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-ITERATIVE:       14:
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 8
; IR-ITERATIVE-NEXT:    br label [[TMP16]]
; IR-ITERATIVE:       16:
; IR-ITERATIVE-NEXT:    br label [[TMP17]]
; IR-ITERATIVE:       17:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fsub_double_uni_address_uni_value_agent_scope_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-DPP-NEXT:    [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-DPP:       14:
; IR-DPP-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] syncscope("agent") monotonic, align 8
; IR-DPP-NEXT:    br label [[TMP16]]
; IR-DPP:       16:
; IR-DPP-NEXT:    br label [[TMP17]]
; IR-DPP:       17:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(ptr addrspace(1) inreg %ptr, double %val) #2 {
; IR-LABEL: @global_atomic_fsub_double_uni_address_div_value_agent_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fsub ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double inreg %val) #0 {
; IR-LABEL: @global_atomic_fmin_double_uni_address_uni_value_agent_scope_unsafe(
; IR-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live()
; IR-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR:       2:
; IR-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true)
; IR-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0)
; IR-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]])
; IR-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-NEXT:    br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
; IR:       10:
; IR-NEXT:    [[TMP11:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    br label [[TMP12]]
; IR:       12:
; IR-NEXT:    br label [[TMP13]]
; IR:       13:
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(ptr addrspace(1) inreg %ptr, double %val) #0 {
; IR-LABEL: @global_atomic_fmin_double_uni_address_div_value_agent_scope_unsafe(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #1{
; IR-ITERATIVE-LABEL: @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
; IR-ITERATIVE:       10:
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-ITERATIVE-NEXT:    br label [[TMP12]]
; IR-ITERATIVE:       12:
; IR-ITERATIVE-NEXT:    br label [[TMP13]]
; IR-ITERATIVE:       13:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fmax_double_uni_address_uni_value_agent_scope_unsafe_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP13:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP9]], label [[TMP10:%.*]], label [[TMP12:%.*]]
; IR-DPP:       10:
; IR-DPP-NEXT:    [[TMP11:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-DPP-NEXT:    br label [[TMP12]]
; IR-DPP:       12:
; IR-DPP-NEXT:    br label [[TMP13]]
; IR-DPP:       13:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe_strictfp(ptr addrspace(1) inreg %ptr, double %val) #1{
; IR-LABEL: @global_atomic_fmax_double_uni_address_div_value_agent_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, double inreg %val) #2 {
; IR-ITERATIVE-LABEL: @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(
; IR-ITERATIVE-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR7]]
; IR-ITERATIVE-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-ITERATIVE:       2:
; IR-ITERATIVE-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-ITERATIVE-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-ITERATIVE-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-ITERATIVE-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-ITERATIVE-NEXT:    [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR7]]
; IR-ITERATIVE-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-ITERATIVE-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-ITERATIVE:       14:
; IR-ITERATIVE-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] monotonic, align 4
; IR-ITERATIVE-NEXT:    br label [[TMP16]]
; IR-ITERATIVE:       16:
; IR-ITERATIVE-NEXT:    br label [[TMP17]]
; IR-ITERATIVE:       17:
; IR-ITERATIVE-NEXT:    ret void
;
; IR-DPP-LABEL: @global_atomic_fadd_double_uni_address_uni_value_system_scope_strictfp(
; IR-DPP-NEXT:    [[TMP1:%.*]] = call i1 @llvm.amdgcn.ps.live() #[[ATTR8]]
; IR-DPP-NEXT:    br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP17:%.*]]
; IR-DPP:       2:
; IR-DPP-NEXT:    [[TMP3:%.*]] = call i64 @llvm.amdgcn.ballot.i64(i1 true) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP4:%.*]] = trunc i64 [[TMP3]] to i32
; IR-DPP-NEXT:    [[TMP5:%.*]] = lshr i64 [[TMP3]], 32
; IR-DPP-NEXT:    [[TMP6:%.*]] = trunc i64 [[TMP5]] to i32
; IR-DPP-NEXT:    [[TMP7:%.*]] = call i32 @llvm.amdgcn.mbcnt.lo(i32 [[TMP4]], i32 0) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP8:%.*]] = call i32 @llvm.amdgcn.mbcnt.hi(i32 [[TMP6]], i32 [[TMP7]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP9:%.*]] = call i64 @llvm.ctpop.i64(i64 [[TMP3]]) #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP10:%.*]] = trunc i64 [[TMP9]] to i32
; IR-DPP-NEXT:    [[TMP11:%.*]] = call double @llvm.experimental.constrained.uitofp.f64.i32(i32 [[TMP10]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP12:%.*]] = call double @llvm.experimental.constrained.fmul.f64(double [[VAL:%.*]], double [[TMP11]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR8]]
; IR-DPP-NEXT:    [[TMP13:%.*]] = icmp eq i32 [[TMP8]], 0
; IR-DPP-NEXT:    br i1 [[TMP13]], label [[TMP14:%.*]], label [[TMP16:%.*]]
; IR-DPP:       14:
; IR-DPP-NEXT:    [[TMP15:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[TMP12]] monotonic, align 4
; IR-DPP-NEXT:    br label [[TMP16]]
; IR-DPP:       16:
; IR-DPP-NEXT:    br label [[TMP17]]
; IR-DPP:       17:
; IR-DPP-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(ptr addrspace(1) inreg %ptr, double %val) #2 {
; IR-LABEL: @global_atomic_fadd_double_uni_address_div_value_system_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(ptr addrspace(1) %ptr, double inreg %val) #0 {
; IR-LABEL: @global_atomic_fadd_double_div_address_uni_value_agent_scope_unsafe(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(ptr addrspace(1) %ptr, double %val) #0 {
; IR-LABEL: @global_atomic_fadd_double_div_address_div_value_agent_scope_unsafe(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) %ptr, double inreg %val) #1 {
; IR-LABEL: @global_atomic_fadd_double_div_address_uni_value_one_as_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_strictfp(ptr addrspace(1) %ptr, double %val) #1 {
; IR-LABEL: @global_atomic_fadd_double_div_address_div_value_one_as_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("one-as") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("one-as") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(ptr addrspace(1) %ptr, double inreg %val) #2 {
; IR-LABEL: @global_atomic_fsub_double_div_address_uni_value_agent_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(ptr addrspace(1) %ptr, double %val) #2 {
; IR-LABEL: @global_atomic_fsub_double_div_address_div_value_agent_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fsub ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fsub ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_double_div_address_uni_value_agent_scope(ptr addrspace(1) %ptr, double inreg %val) #0 {
; IR-LABEL: @global_atomic_fmin_double_div_address_uni_value_agent_scope(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmin_double_div_address_div_value_agent_scope(ptr addrspace(1) %ptr, double %val) #0 {
; IR-LABEL: @global_atomic_fmin_double_div_address_div_value_agent_scope(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmin ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fmin ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_double_div_address_uni_value_agent_scope_unsafe_strictfp(ptr addrspace(1) %ptr, double inreg %val) #1{
; IR-LABEL: @global_atomic_fmax_double_div_address_uni_value_agent_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fmax_double_div_address_div_value_agent_scope_unsafe_strictfp(ptr addrspace(1) %ptr, double %val) #1{
; IR-LABEL: @global_atomic_fmax_double_div_address_div_value_agent_scope_unsafe_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fmax ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] syncscope("agent") monotonic, align 8
; IR-NEXT:    ret void
;
  %result = atomicrmw fmax ptr addrspace(1) %ptr, double %val syncscope("agent") monotonic
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(ptr addrspace(1) %ptr, double inreg %val) #2 {
; IR-LABEL: @global_atomic_fadd_double_div_address_uni_value_system_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
  ret void
}

define amdgpu_ps void @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(ptr addrspace(1) %ptr, double %val) #2 {
; IR-LABEL: @global_atomic_fadd_double_div_address_div_value_system_scope_strictfp(
; IR-NEXT:    [[RESULT:%.*]] = atomicrmw fadd ptr addrspace(1) [[PTR:%.*]], double [[VAL:%.*]] monotonic, align 4
; IR-NEXT:    ret void
;
  %result = atomicrmw fadd ptr addrspace(1) %ptr, double %val monotonic, align 4
  ret void
}

attributes #0 = { "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #1 = { strictfp "denormal-fp-math-f32"="preserve-sign,preserve-sign" "amdgpu-unsafe-fp-atomics"="true" }
attributes #2 = { strictfp }