; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 5
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -amdgpu-perf-hint < %s | FileCheck -check-prefix=CHECK %s
; RUN: opt -S -mtriple=amdgcn-amd-amdhsa -passes=amdgpu-perf-hint < %s | FileCheck -check-prefix=CHECK %s
; RUN: llc -mtriple=amdgcn < %s | FileCheck -check-prefix=GCN %s

; GCN-LABEL: {{^}}test_membound:
; GCN: MemoryBound: 1
; GCN: WaveLimiterHint : 1
define amdgpu_kernel void @test_membound(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
; CHECK-LABEL: define amdgpu_kernel void @test_membound(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]]) #[[ATTR0:[0-9]+]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP3]], align 16
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP2]]
; CHECK-NEXT:    store <4 x i32> [[TMP4]], ptr addrspace(1) [[TMP5]], align 16
; CHECK-NEXT:    [[TMP6:%.*]] = add nuw nsw i64 [[TMP2]], 1
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP8:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP7]], align 16
; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP6]]
; CHECK-NEXT:    store <4 x i32> [[TMP8]], ptr addrspace(1) [[TMP9]], align 16
; CHECK-NEXT:    ret void
;
bb:
  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
  %tmp2 = zext i32 %tmp to i64
  %tmp3 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp2
  %tmp4 = load <4 x i32>, ptr addrspace(1) %tmp3, align 16
  %tmp5 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp2
  store <4 x i32> %tmp4, ptr addrspace(1) %tmp5, align 16
  %tmp6 = add nuw nsw i64 %tmp2, 1
  %tmp7 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp6
  %tmp8 = load <4 x i32>, ptr addrspace(1) %tmp7, align 16
  %tmp9 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp6
  store <4 x i32> %tmp8, ptr addrspace(1) %tmp9, align 16
  ret void
}

; GCN-LABEL: {{^}}test_membound_1:
; GCN: MemoryBound: 1
define amdgpu_kernel void @test_membound_1(ptr addrspace(1) nocapture readonly %ptr.0,
; CHECK-LABEL: define amdgpu_kernel void @test_membound_1(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[PTR_0:%.*]], ptr addrspace(1) captures(none) [[PTR_1:%.*]], <2 x double> [[ARG_0:%.*]], i32 [[ARG_1:%.*]], <4 x double> [[ARG_2:%.*]]) #[[ATTR1:[0-9]+]] {
; CHECK-NEXT:  [[BB_ENTRY:.*:]]
; CHECK-NEXT:    [[ID_32:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    [[ID_0:%.*]] = zext i32 [[ID_32]] to i64
; CHECK-NEXT:    [[GEP_0:%.*]] = getelementptr inbounds <2 x double>, ptr addrspace(1) [[PTR_0]], i64 [[ID_0]]
; CHECK-NEXT:    [[LD_0:%.*]] = load <2 x double>, ptr addrspace(1) [[GEP_0]], align 16
; CHECK-NEXT:    [[ADD_0:%.*]] = fadd <2 x double> [[ARG_0]], [[LD_0]]
; CHECK-NEXT:    [[ID_1:%.*]] = add nuw nsw i64 [[ID_0]], 1
; CHECK-NEXT:    [[GEP_1:%.*]] = getelementptr inbounds <2 x double>, ptr addrspace(1) [[PTR_0]], i64 [[ID_1]]
; CHECK-NEXT:    [[LD_1:%.*]] = load <2 x double>, ptr addrspace(1) [[GEP_1]], align 16
; CHECK-NEXT:    [[ADD_1:%.*]] = fadd <2 x double> [[ADD_0]], [[LD_1]]
; CHECK-NEXT:    [[ID_2:%.*]] = add nuw nsw i64 [[ID_0]], 2
; CHECK-NEXT:    [[GEP_2:%.*]] = getelementptr inbounds <2 x double>, ptr addrspace(1) [[PTR_0]], i64 [[ID_2]]
; CHECK-NEXT:    [[LD_2:%.*]] = load <2 x double>, ptr addrspace(1) [[GEP_2]], align 16
; CHECK-NEXT:    [[ADD_2:%.*]] = fadd <2 x double> [[ADD_1]], [[LD_2]]
; CHECK-NEXT:    [[ID_3:%.*]] = add nuw nsw i64 [[ID_0]], 3
; CHECK-NEXT:    [[GEP_3:%.*]] = getelementptr inbounds <2 x double>, ptr addrspace(1) [[PTR_0]], i64 [[ID_3]]
; CHECK-NEXT:    [[LD_3:%.*]] = load <2 x double>, ptr addrspace(1) [[GEP_3]], align 16
; CHECK-NEXT:    [[ADD_3:%.*]] = fadd <2 x double> [[ADD_2]], [[LD_3]]
; CHECK-NEXT:    [[ID_4:%.*]] = add nuw nsw i64 [[ID_0]], 4
; CHECK-NEXT:    [[GEP_4:%.*]] = getelementptr inbounds <2 x double>, ptr addrspace(1) [[PTR_0]], i64 [[ID_4]]
; CHECK-NEXT:    [[LD_4:%.*]] = load <2 x double>, ptr addrspace(1) [[GEP_4]], align 16
; CHECK-NEXT:    [[ADD_4:%.*]] = fadd <2 x double> [[ADD_3]], [[LD_4]]
; CHECK-NEXT:    store <2 x double> [[ADD_4]], ptr addrspace(1) [[PTR_1]], align 16
; CHECK-NEXT:    [[COND:%.*]] = icmp eq i32 [[ARG_1]], 0
; CHECK-NEXT:    br i1 [[COND]], label %[[BB_TRUE:.*]], label %[[BB_RET:.*]]
; CHECK:       [[BB_TRUE]]:
; CHECK-NEXT:    [[I0_ARG_0:%.*]] = extractelement <2 x double> [[ARG_0]], i32 0
; CHECK-NEXT:    [[I1_ARG_0:%.*]] = extractelement <2 x double> [[ARG_0]], i32 1
; CHECK-NEXT:    [[ADD_1_0:%.*]] = fadd double [[I0_ARG_0]], [[I1_ARG_0]]
; CHECK-NEXT:    [[I0_ARG_2:%.*]] = extractelement <4 x double> [[ARG_2]], i32 0
; CHECK-NEXT:    [[I1_ARG_2:%.*]] = extractelement <4 x double> [[ARG_2]], i32 1
; CHECK-NEXT:    [[ADD_1_1:%.*]] = fadd double [[I0_ARG_2]], [[I1_ARG_2]]
; CHECK-NEXT:    [[ADD_1_2:%.*]] = fadd double [[ADD_1_0]], [[ADD_1_1]]
; CHECK-NEXT:    [[I2_ARG_2:%.*]] = extractelement <4 x double> [[ARG_2]], i32 2
; CHECK-NEXT:    [[I3_ARG_2:%.*]] = extractelement <4 x double> [[ARG_2]], i32 3
; CHECK-NEXT:    [[ADD_1_3:%.*]] = fadd double [[I2_ARG_2]], [[I3_ARG_2]]
; CHECK-NEXT:    [[ADD_1_4:%.*]] = fadd double [[ADD_1_2]], [[ADD_1_3]]
; CHECK-NEXT:    [[I0_ADD_0:%.*]] = extractelement <2 x double> [[ADD_0]], i32 0
; CHECK-NEXT:    [[I1_ADD_0:%.*]] = extractelement <2 x double> [[ADD_0]], i32 1
; CHECK-NEXT:    [[ADD_1_5:%.*]] = fadd double [[I0_ADD_0]], [[I1_ADD_0]]
; CHECK-NEXT:    [[ADD_1_6:%.*]] = fadd double [[ADD_1_4]], [[ADD_1_5]]
; CHECK-NEXT:    [[I0_ADD_1:%.*]] = extractelement <2 x double> [[ADD_1]], i32 0
; CHECK-NEXT:    [[I1_ADD_1:%.*]] = extractelement <2 x double> [[ADD_1]], i32 1
; CHECK-NEXT:    [[ADD_1_7:%.*]] = fadd double [[I0_ADD_1]], [[I1_ADD_1]]
; CHECK-NEXT:    [[ADD_1_8:%.*]] = fadd double [[ADD_1_6]], [[ADD_1_7]]
; CHECK-NEXT:    [[I0_ADD_2:%.*]] = extractelement <2 x double> [[ADD_2]], i32 0
; CHECK-NEXT:    [[I1_ADD_2:%.*]] = extractelement <2 x double> [[ADD_2]], i32 1
; CHECK-NEXT:    [[ADD_1_9:%.*]] = fadd double [[I0_ADD_2]], [[I1_ADD_2]]
; CHECK-NEXT:    [[ADD_1_10:%.*]] = fadd double [[ADD_1_8]], [[ADD_1_9]]
; CHECK-NEXT:    store double [[ADD_1_8]], ptr addrspace(1) [[PTR_1]], align 8
; CHECK-NEXT:    br label %[[BB_RET]]
; CHECK:       [[BB_RET]]:
; CHECK-NEXT:    ret void
;
  ptr addrspace(1) nocapture %ptr.1,
  <2 x double> %arg.0, i32 %arg.1, <4 x double> %arg.2) {
bb.entry:
  %id.32 = tail call i32 @llvm.amdgcn.workitem.id.x()
  %id.0 = zext i32 %id.32 to i64
  %gep.0 = getelementptr inbounds <2 x double>, ptr addrspace(1) %ptr.0, i64 %id.0
  %ld.0 = load <2 x double>, ptr addrspace(1) %gep.0, align 16
  %add.0 = fadd <2 x double> %arg.0, %ld.0

  %id.1 = add nuw nsw i64 %id.0, 1
  %gep.1 = getelementptr inbounds <2 x double>, ptr addrspace(1) %ptr.0, i64 %id.1
  %ld.1 = load <2 x double>, ptr addrspace(1) %gep.1, align 16
  %add.1 = fadd <2 x double> %add.0, %ld.1

  %id.2 = add nuw nsw i64 %id.0, 2
  %gep.2 = getelementptr inbounds <2 x double>, ptr addrspace(1) %ptr.0, i64 %id.2
  %ld.2 = load <2 x double>, ptr addrspace(1) %gep.2, align 16
  %add.2 = fadd <2 x double> %add.1, %ld.2

  %id.3 = add nuw nsw i64 %id.0, 3
  %gep.3= getelementptr inbounds <2 x double>, ptr addrspace(1) %ptr.0, i64 %id.3
  %ld.3 = load <2 x double>, ptr addrspace(1) %gep.3, align 16
  %add.3 = fadd <2 x double> %add.2, %ld.3

  %id.4 = add nuw nsw i64 %id.0, 4
  %gep.4= getelementptr inbounds <2 x double>, ptr addrspace(1) %ptr.0, i64 %id.4
  %ld.4 = load <2 x double>, ptr addrspace(1) %gep.4, align 16
  %add.4 = fadd <2 x double> %add.3, %ld.4

  store <2 x double> %add.4, ptr addrspace(1) %ptr.1, align 16
  %cond = icmp eq i32 %arg.1, 0
  br i1 %cond, label %bb.true, label %bb.ret

bb.true:
  %i0.arg.0 = extractelement <2 x double> %arg.0, i32 0
  %i1.arg.0 = extractelement <2 x double> %arg.0, i32 1
  %add.1.0 = fadd double %i0.arg.0, %i1.arg.0
  %i0.arg.2 = extractelement <4 x double> %arg.2, i32 0
  %i1.arg.2 = extractelement <4 x double> %arg.2, i32 1
  %add.1.1 = fadd double %i0.arg.2, %i1.arg.2
  %add.1.2 = fadd double %add.1.0, %add.1.1
  %i2.arg.2 = extractelement <4 x double> %arg.2, i32 2
  %i3.arg.2 = extractelement <4 x double> %arg.2, i32 3
  %add.1.3 = fadd double %i2.arg.2, %i3.arg.2
  %add.1.4 = fadd double %add.1.2, %add.1.3
  %i0.add.0 = extractelement <2 x double> %add.0, i32 0
  %i1.add.0 = extractelement <2 x double> %add.0, i32 1
  %add.1.5 = fadd double %i0.add.0, %i1.add.0
  %add.1.6 = fadd double %add.1.4, %add.1.5
  %i0.add.1 = extractelement <2 x double> %add.1, i32 0
  %i1.add.1 = extractelement <2 x double> %add.1, i32 1
  %add.1.7 = fadd double %i0.add.1, %i1.add.1
  %add.1.8 = fadd double %add.1.6, %add.1.7
  %i0.add.2 = extractelement <2 x double> %add.2, i32 0
  %i1.add.2 = extractelement <2 x double> %add.2, i32 1
  %add.1.9 = fadd double %i0.add.2, %i1.add.2
  %add.1.10 = fadd double %add.1.8, %add.1.9

  store double %add.1.8, ptr addrspace(1) %ptr.1, align 8
  br label %bb.ret

bb.ret:
  ret void
}

; GCN-LABEL: {{^}}test_large_stride:
; GCN: MemoryBound: 0
; GCN: WaveLimiterHint : 1
define amdgpu_kernel void @test_large_stride(ptr addrspace(1) nocapture %arg) {
; CHECK-LABEL: define amdgpu_kernel void @test_large_stride(
; CHECK-SAME: ptr addrspace(1) captures(none) [[ARG:%.*]]) #[[ATTR2:[0-9]+]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 4096
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(1) [[TMP]], align 4
; CHECK-NEXT:    [[MUL1:%.*]] = mul i32 [[TMP1]], [[TMP1]]
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 1
; CHECK-NEXT:    store i32 [[MUL1]], ptr addrspace(1) [[TMP2]], align 4
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 8192
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[TMP3]], align 4
; CHECK-NEXT:    [[MUL4:%.*]] = mul i32 [[TMP4]], [[TMP4]]
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 2
; CHECK-NEXT:    store i32 [[MUL4]], ptr addrspace(1) [[TMP5]], align 4
; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 12288
; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr addrspace(1) [[TMP6]], align 4
; CHECK-NEXT:    [[MUL7:%.*]] = mul i32 [[TMP7]], [[TMP7]]
; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 3
; CHECK-NEXT:    store i32 [[MUL7]], ptr addrspace(1) [[TMP8]], align 4
; CHECK-NEXT:    ret void
;
bb:
  %tmp = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 4096
  %tmp1 = load i32, ptr addrspace(1) %tmp, align 4
  %mul1 = mul i32 %tmp1, %tmp1
  %tmp2 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 1
  store i32 %mul1, ptr addrspace(1) %tmp2, align 4
  %tmp3 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 8192
  %tmp4 = load i32, ptr addrspace(1) %tmp3, align 4
  %mul4 = mul i32 %tmp4, %tmp4
  %tmp5 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 2
  store i32 %mul4, ptr addrspace(1) %tmp5, align 4
  %tmp6 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 12288
  %tmp7 = load i32, ptr addrspace(1) %tmp6, align 4
  %mul7 = mul i32 %tmp7, %tmp7
  %tmp8 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 3
  store i32 %mul7, ptr addrspace(1) %tmp8, align 4
  ret void
}

; GCN-LABEL: {{^}}test_indirect:
; GCN: MemoryBound: 1
; GCN: WaveLimiterHint : 1
define amdgpu_kernel void @test_indirect(ptr addrspace(1) nocapture %arg) {
; CHECK-LABEL: define amdgpu_kernel void @test_indirect(
; CHECK-SAME: ptr addrspace(1) captures(none) [[ARG:%.*]]) #[[ATTR2]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 1
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 2
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 3
; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i32>, ptr addrspace(1) [[ARG]], align 4
; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i32> [[TMP4]], i32 0
; CHECK-NEXT:    [[TMP6:%.*]] = sext i32 [[TMP5]] to i64
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP8:%.*]] = load i32, ptr addrspace(1) [[TMP7]], align 4
; CHECK-NEXT:    store i32 [[TMP8]], ptr addrspace(1) [[ARG]], align 4
; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i32> [[TMP4]], i32 1
; CHECK-NEXT:    [[TMP10:%.*]] = sext i32 [[TMP9]] to i64
; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 [[TMP10]]
; CHECK-NEXT:    [[TMP12:%.*]] = load i32, ptr addrspace(1) [[TMP11]], align 4
; CHECK-NEXT:    store i32 [[TMP12]], ptr addrspace(1) [[TMP]], align 4
; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <4 x i32> [[TMP4]], i32 2
; CHECK-NEXT:    [[TMP14:%.*]] = sext i32 [[TMP13]] to i64
; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 [[TMP14]]
; CHECK-NEXT:    [[TMP16:%.*]] = load i32, ptr addrspace(1) [[TMP15]], align 4
; CHECK-NEXT:    store i32 [[TMP16]], ptr addrspace(1) [[TMP1]], align 4
; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <4 x i32> [[TMP4]], i32 3
; CHECK-NEXT:    [[TMP18:%.*]] = sext i32 [[TMP17]] to i64
; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds i32, ptr addrspace(1) [[ARG]], i64 [[TMP18]]
; CHECK-NEXT:    [[TMP20:%.*]] = load i32, ptr addrspace(1) [[TMP19]], align 4
; CHECK-NEXT:    store i32 [[TMP20]], ptr addrspace(1) [[TMP2]], align 4
; CHECK-NEXT:    ret void
;
bb:
  %tmp = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 1
  %tmp1 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 2
  %tmp2 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 3
  %tmp4 = load <4 x i32>, ptr addrspace(1) %arg, align 4
  %tmp5 = extractelement <4 x i32> %tmp4, i32 0
  %tmp6 = sext i32 %tmp5 to i64
  %tmp7 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 %tmp6
  %tmp8 = load i32, ptr addrspace(1) %tmp7, align 4
  store i32 %tmp8, ptr addrspace(1) %arg, align 4
  %tmp9 = extractelement <4 x i32> %tmp4, i32 1
  %tmp10 = sext i32 %tmp9 to i64
  %tmp11 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 %tmp10
  %tmp12 = load i32, ptr addrspace(1) %tmp11, align 4
  store i32 %tmp12, ptr addrspace(1) %tmp, align 4
  %tmp13 = extractelement <4 x i32> %tmp4, i32 2
  %tmp14 = sext i32 %tmp13 to i64
  %tmp15 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 %tmp14
  %tmp16 = load i32, ptr addrspace(1) %tmp15, align 4
  store i32 %tmp16, ptr addrspace(1) %tmp1, align 4
  %tmp17 = extractelement <4 x i32> %tmp4, i32 3
  %tmp18 = sext i32 %tmp17 to i64
  %tmp19 = getelementptr inbounds i32, ptr addrspace(1) %arg, i64 %tmp18
  %tmp20 = load i32, ptr addrspace(1) %tmp19, align 4
  store i32 %tmp20, ptr addrspace(1) %tmp2, align 4
  ret void
}

; GCN-LABEL: {{^}}test_indirect_through_phi:
; GCN: MemoryBound: 0
; GCN: WaveLimiterHint : 0
define amdgpu_kernel void @test_indirect_through_phi(ptr addrspace(1) %arg) {
; CHECK-LABEL: define amdgpu_kernel void @test_indirect_through_phi(
; CHECK-SAME: ptr addrspace(1) [[ARG:%.*]]) {
; CHECK-NEXT:  [[BB:.*]]:
; CHECK-NEXT:    [[LOAD:%.*]] = load float, ptr addrspace(1) [[ARG]], align 8
; CHECK-NEXT:    [[LOAD_F:%.*]] = bitcast float [[LOAD]] to i32
; CHECK-NEXT:    [[N:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    br label %[[BB1:.*]]
; CHECK:       [[BB1]]:
; CHECK-NEXT:    [[PHI:%.*]] = phi i32 [ [[LOAD_F]], %[[BB]] ], [ [[AND2:%.*]], %[[BB1]] ]
; CHECK-NEXT:    [[IND:%.*]] = phi i32 [ 0, %[[BB]] ], [ [[INC2:%.*]], %[[BB1]] ]
; CHECK-NEXT:    [[AND1:%.*]] = and i32 [[PHI]], [[N]]
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds float, ptr addrspace(1) [[ARG]], i32 [[AND1]]
; CHECK-NEXT:    store float [[LOAD]], ptr addrspace(1) [[GEP]], align 4
; CHECK-NEXT:    [[INC1:%.*]] = add nsw i32 [[PHI]], 1310720
; CHECK-NEXT:    [[AND2]] = and i32 [[INC1]], [[N]]
; CHECK-NEXT:    [[INC2]] = add nuw nsw i32 [[IND]], 1
; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[INC2]], 1024
; CHECK-NEXT:    br i1 [[CMP]], label %[[BB2:.*]], label %[[BB1]]
; CHECK:       [[BB2]]:
; CHECK-NEXT:    ret void
;
bb:
  %load = load float, ptr addrspace(1) %arg, align 8
  %load.f = bitcast float %load to i32
  %n = tail call i32 @llvm.amdgcn.workitem.id.x()
  br label %bb1

bb1:                                              ; preds = %bb1, %bb
  %phi = phi i32 [ %load.f, %bb ], [ %and2, %bb1 ]
  %ind = phi i32 [ 0, %bb ], [ %inc2, %bb1 ]
  %and1 = and i32 %phi, %n
  %gep = getelementptr inbounds float, ptr addrspace(1) %arg, i32 %and1
  store float %load, ptr addrspace(1) %gep, align 4
  %inc1 = add nsw i32 %phi, 1310720
  %and2 = and i32 %inc1, %n
  %inc2 = add nuw nsw i32 %ind, 1
  %cmp = icmp eq i32 %inc2, 1024
  br i1 %cmp, label %bb2, label %bb1

bb2:                                              ; preds = %bb1
  ret void
}

define void @test_membound_func(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
; CHECK-LABEL: define void @test_membound_func(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP3]], align 16
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP2]]
; CHECK-NEXT:    store <4 x i32> [[TMP4]], ptr addrspace(1) [[TMP5]], align 16
; CHECK-NEXT:    [[TMP6:%.*]] = add nuw nsw i64 [[TMP2]], 1
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP8:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP7]], align 16
; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP6]]
; CHECK-NEXT:    store <4 x i32> [[TMP8]], ptr addrspace(1) [[TMP9]], align 16
; CHECK-NEXT:    ret void
;
bb:
  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
  %tmp2 = zext i32 %tmp to i64
  %tmp3 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp2
  %tmp4 = load <4 x i32>, ptr addrspace(1) %tmp3, align 16
  %tmp5 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp2
  store <4 x i32> %tmp4, ptr addrspace(1) %tmp5, align 16
  %tmp6 = add nuw nsw i64 %tmp2, 1
  %tmp7 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp6
  %tmp8 = load <4 x i32>, ptr addrspace(1) %tmp7, align 16
  %tmp9 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp6
  store <4 x i32> %tmp8, ptr addrspace(1) %tmp9, align 16
  ret void
}

; GCN-LABEL: {{^}}kernel_call_test_membound_func:
; GCN: MemoryBound: 1
; GCN: WaveLimiterHint : 1
define amdgpu_kernel void @kernel_call_test_membound_func(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
; CHECK-LABEL: define amdgpu_kernel void @kernel_call_test_membound_func(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    call void @test_membound_func(ptr addrspace(1) readonly captures(none) [[ARG]], ptr addrspace(1) captures(none) [[ARG1]])
; CHECK-NEXT:    ret void
;
  call void @test_membound_func(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1)
  ret void
}

; TODO: Probably should assume yes?
; GCN-LABEL: {{^}}kernel_indirect_call:
; GCN: MemoryBound: 0
; GCN: WaveLimiterHint : 0
define amdgpu_kernel void @kernel_indirect_call(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1, ptr %fptr) {
; CHECK-LABEL: define amdgpu_kernel void @kernel_indirect_call(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]], ptr [[FPTR:%.*]]) {
; CHECK-NEXT:    call void [[FPTR]](ptr addrspace(1) readonly captures(none) [[ARG]], ptr addrspace(1) captures(none) [[ARG1]])
; CHECK-NEXT:    ret void
;
  call void %fptr(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1)
  ret void
}

declare void @extern()

define void @maybe_recursive_test_membound_func(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
; CHECK-LABEL: define void @maybe_recursive_test_membound_func(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:  [[BB:.*:]]
; CHECK-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP3]], align 16
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP2]]
; CHECK-NEXT:    store <4 x i32> [[TMP4]], ptr addrspace(1) [[TMP5]], align 16
; CHECK-NEXT:    [[TMP6:%.*]] = add nuw nsw i64 [[TMP2]], 1
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP8:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP7]], align 16
; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP6]]
; CHECK-NEXT:    store <4 x i32> [[TMP8]], ptr addrspace(1) [[TMP9]], align 16
; CHECK-NEXT:    call void @extern()
; CHECK-NEXT:    ret void
;
bb:
  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
  %tmp2 = zext i32 %tmp to i64
  %tmp3 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp2
  %tmp4 = load <4 x i32>, ptr addrspace(1) %tmp3, align 16
  %tmp5 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp2
  store <4 x i32> %tmp4, ptr addrspace(1) %tmp5, align 16
  %tmp6 = add nuw nsw i64 %tmp2, 1
  %tmp7 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp6
  %tmp8 = load <4 x i32>, ptr addrspace(1) %tmp7, align 16
  %tmp9 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp6
  store <4 x i32> %tmp8, ptr addrspace(1) %tmp9, align 16
  call void @extern()
  ret void
}

; GCN-LABEL: {{^}}kernel_call_maybe_recursive_test_membound_func:
; GCN: MemoryBound: 1
; GCN: WaveLimiterHint : 1
define amdgpu_kernel void @kernel_call_maybe_recursive_test_membound_func(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1, ptr %fptr) {
; CHECK-LABEL: define amdgpu_kernel void @kernel_call_maybe_recursive_test_membound_func(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]], ptr [[FPTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    call void @maybe_recursive_test_membound_func(ptr addrspace(1) readonly captures(none) [[ARG]], ptr addrspace(1) captures(none) [[ARG1]])
; CHECK-NEXT:    ret void
;
  call void @maybe_recursive_test_membound_func(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1)
  ret void
}

define void @mutually_recursive_test_membound_func_0(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
; CHECK-LABEL: define void @mutually_recursive_test_membound_func_0(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP]] to i64
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP3]], align 16
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP2]]
; CHECK-NEXT:    store <4 x i32> [[TMP4]], ptr addrspace(1) [[TMP5]], align 16
; CHECK-NEXT:    call void @mutually_recursive_test_membound_func_0(ptr addrspace(1) readonly captures(none) [[ARG]], ptr addrspace(1) captures(none) [[ARG1]])
; CHECK-NEXT:    ret void
;
  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
  %tmp2 = zext i32 %tmp to i64
  %tmp3 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp2
  %tmp4 = load <4 x i32>, ptr addrspace(1) %tmp3, align 16
  %tmp5 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp2
  store <4 x i32> %tmp4, ptr addrspace(1) %tmp5, align 16
  call void @mutually_recursive_test_membound_func_0(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1)
  ret void
}

define void @mutually_recursive_test_membound_func_1(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1) {
; CHECK-LABEL: define void @mutually_recursive_test_membound_func_1(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT:    [[TMP:%.*]] = tail call i32 @llvm.amdgcn.workitem.id.x()
; CHECK-NEXT:    [[TMP2:%.*]] = zext i32 [[TMP]] to i64
; CHECK-NEXT:    [[TMP6:%.*]] = add nuw nsw i64 [[TMP2]], 1
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG]], i64 [[TMP6]]
; CHECK-NEXT:    [[TMP8:%.*]] = load <4 x i32>, ptr addrspace(1) [[TMP7]], align 16
; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds <4 x i32>, ptr addrspace(1) [[ARG1]], i64 [[TMP6]]
; CHECK-NEXT:    store <4 x i32> [[TMP8]], ptr addrspace(1) [[TMP9]], align 16
; CHECK-NEXT:    call void @mutually_recursive_test_membound_func_1(ptr addrspace(1) readonly captures(none) [[ARG]], ptr addrspace(1) captures(none) [[ARG1]])
; CHECK-NEXT:    ret void
;
  %tmp = tail call i32 @llvm.amdgcn.workitem.id.x()
  %tmp2 = zext i32 %tmp to i64
  %tmp6 = add nuw nsw i64 %tmp2, 1
  %tmp7 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg, i64 %tmp6
  %tmp8 = load <4 x i32>, ptr addrspace(1) %tmp7, align 16
  %tmp9 = getelementptr inbounds <4 x i32>, ptr addrspace(1) %arg1, i64 %tmp6
  store <4 x i32> %tmp8, ptr addrspace(1) %tmp9, align 16
  call void @mutually_recursive_test_membound_func_1(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1)
  ret void
}

; GCN-LABEL: {{^}}kernel_call_mutually_recursive_test_membound_func_0:
; GCN: MemoryBound: 1
; GCN: WaveLimiterHint : 1
define amdgpu_kernel void @kernel_call_mutually_recursive_test_membound_func_0(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1, ptr %fptr) {
; CHECK-LABEL: define amdgpu_kernel void @kernel_call_mutually_recursive_test_membound_func_0(
; CHECK-SAME: ptr addrspace(1) readonly captures(none) [[ARG:%.*]], ptr addrspace(1) captures(none) [[ARG1:%.*]], ptr [[FPTR:%.*]]) #[[ATTR0]] {
; CHECK-NEXT:    call void @mutually_recursive_test_membound_func_0(ptr addrspace(1) readonly captures(none) [[ARG]], ptr addrspace(1) captures(none) [[ARG1]])
; CHECK-NEXT:    ret void
;
  call void @mutually_recursive_test_membound_func_0(ptr addrspace(1) nocapture readonly %arg, ptr addrspace(1) nocapture %arg1)
  ret void
}

declare i32 @llvm.amdgcn.workitem.id.x()


;.
; CHECK: attributes #[[ATTR0]] = { "amdgpu-memory-bound"="true" "amdgpu-wave-limiter"="true" }
; CHECK: attributes #[[ATTR1]] = { "amdgpu-memory-bound"="true" }
; CHECK: attributes #[[ATTR2]] = { "amdgpu-wave-limiter"="true" }
; CHECK: attributes #[[ATTR3:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) }
;.