[rtsan] Add fork/execve interceptors (#117198)

[llvm-project.git] / llvm / test / Transforms / AtomicExpand / AMDGPU / expand-atomic-i8.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -passes=atomic-expand %s | FileCheck %s --check-prefixes=CHECK,GCN
; RUN: opt -mtriple=r600-mesa-mesa3d -S -passes=atomic-expand %s | FileCheck %s --check-prefixes=CHECK,R600

define i8 @test_atomicrmw_xchg_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_xchg_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[TMP5:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[TMP6:%.*]] = or i32 [[TMP5]], [[VALOPERAND_SHIFTED]]
; GCN-NEXT:    [[TMP7:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP6]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP7]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP7]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_xchg_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[TMP5:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[TMP6:%.*]] = or i32 [[TMP5]], [[VALOPERAND_SHIFTED]]
; R600-NEXT:    [[TMP7:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP6]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP7]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP7]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_add_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_add_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; GCN-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; GCN-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_add_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; R600-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; R600-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_add_i8_global_agent_align2(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_add_i8_global_agent_align2(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; GCN-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; GCN-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_add_i8_global_agent_align2(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; R600-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; R600-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_add_i8_global_agent_align4(ptr addrspace(1) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i8_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i8 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP2]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[TMP1]]
; CHECK-NEXT:    [[TMP3:%.*]] = and i32 [[NEW]], 255
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[TMP5:%.*]] = or i32 [[TMP4]], [[TMP3]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[TMP5]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_sub_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_sub_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[NEW:%.*]] = sub i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; GCN-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; GCN-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_sub_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[NEW:%.*]] = sub i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; R600-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; R600-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw sub ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_and_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_and_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; GCN-NEXT:    [[TMP4:%.*]] = atomicrmw and ptr addrspace(1) [[ALIGNEDADDR]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; R600-NEXT:    [[TMP4:%.*]] = atomicrmw and ptr addrspace(1) [[ALIGNEDADDR]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw and ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_nand_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_nand_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[TMP5:%.*]] = and i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; GCN-NEXT:    [[NEW:%.*]] = xor i32 [[TMP5]], -1
; GCN-NEXT:    [[TMP6:%.*]] = and i32 [[NEW]], [[MASK]]
; GCN-NEXT:    [[TMP7:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[TMP8:%.*]] = or i32 [[TMP7]], [[TMP6]]
; GCN-NEXT:    [[TMP9:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP8]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP9]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP9]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_nand_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[TMP5:%.*]] = and i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; R600-NEXT:    [[NEW:%.*]] = xor i32 [[TMP5]], -1
; R600-NEXT:    [[TMP6:%.*]] = and i32 [[NEW]], [[MASK]]
; R600-NEXT:    [[TMP7:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[TMP8:%.*]] = or i32 [[TMP7]], [[TMP6]]
; R600-NEXT:    [[TMP9:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP8]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP9]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP9]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw nand ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_or_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_or_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = atomicrmw or ptr addrspace(1) [[ALIGNEDADDR]], i32 [[VALOPERAND_SHIFTED]] syncscope("agent") seq_cst, align 4
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_or_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = atomicrmw or ptr addrspace(1) [[ALIGNEDADDR]], i32 [[VALOPERAND_SHIFTED]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw or ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_xor_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_xor_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; GCN-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP4:%.*]] = atomicrmw xor ptr addrspace(1) [[ALIGNEDADDR]], i32 [[VALOPERAND_SHIFTED]] syncscope("agent") seq_cst, align 4
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_xor_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP4:%.*]] = atomicrmw xor ptr addrspace(1) [[ALIGNEDADDR]], i32 [[VALOPERAND_SHIFTED]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw xor ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_max_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_max_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = icmp sgt i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_max_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = icmp sgt i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw max ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_min_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_min_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = icmp sle i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_min_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = icmp sle i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw min ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_umax_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_umax_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_umax_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw umax ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_umin_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_umin_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = icmp ule i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_umin_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = icmp ule i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i8 [[EXTRACTED]], i8 [[VALUE]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw umin ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_cmpxchg_i8_global_agent(ptr addrspace(1) %out, i8 %in, i8 %old) {
; GCN-LABEL: @test_cmpxchg_i8_global_agent(
; GCN-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr addrspace(1) [[OUT:%.*]], i64 4
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[GEP]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[GEP]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = zext i8 [[IN:%.*]] to i32
; GCN-NEXT:    [[TMP4:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP5:%.*]] = zext i8 [[OLD:%.*]] to i32
; GCN-NEXT:    [[TMP6:%.*]] = shl i32 [[TMP5]], [[SHIFTAMT]]
; GCN-NEXT:    [[TMP7:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    [[TMP8:%.*]] = and i32 [[TMP7]], [[INV_MASK]]
; GCN-NEXT:    br label [[PARTWORD_CMPXCHG_LOOP:%.*]]
; GCN:       partword.cmpxchg.loop:
; GCN-NEXT:    [[TMP9:%.*]] = phi i32 [ [[TMP8]], [[TMP0:%.*]] ], [ [[TMP15:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ]
; GCN-NEXT:    [[TMP10:%.*]] = or i32 [[TMP9]], [[TMP4]]
; GCN-NEXT:    [[TMP11:%.*]] = or i32 [[TMP9]], [[TMP6]]
; GCN-NEXT:    [[TMP12:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[TMP11]], i32 [[TMP10]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[TMP13:%.*]] = extractvalue { i32, i1 } [[TMP12]], 0
; GCN-NEXT:    [[TMP14:%.*]] = extractvalue { i32, i1 } [[TMP12]], 1
; GCN-NEXT:    br i1 [[TMP14]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]]
; GCN:       partword.cmpxchg.failure:
; GCN-NEXT:    [[TMP15]] = and i32 [[TMP13]], [[INV_MASK]]
; GCN-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP9]], [[TMP15]]
; GCN-NEXT:    br i1 [[TMP16]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]]
; GCN:       partword.cmpxchg.end:
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP13]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP17:%.*]] = insertvalue { i8, i1 } poison, i8 [[EXTRACTED]], 0
; GCN-NEXT:    [[TMP18:%.*]] = insertvalue { i8, i1 } [[TMP17]], i1 [[TMP14]], 1
; GCN-NEXT:    [[EXTRACT:%.*]] = extractvalue { i8, i1 } [[TMP18]], 0
; GCN-NEXT:    ret i8 [[EXTRACT]]
;
; R600-LABEL: @test_cmpxchg_i8_global_agent(
; R600-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr addrspace(1) [[OUT:%.*]], i64 4
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[GEP]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[GEP]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i8 [[IN:%.*]] to i32
; R600-NEXT:    [[TMP4:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; R600-NEXT:    [[TMP5:%.*]] = zext i8 [[OLD:%.*]] to i32
; R600-NEXT:    [[TMP6:%.*]] = shl i32 [[TMP5]], [[TMP2]]
; R600-NEXT:    [[TMP7:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    [[TMP8:%.*]] = and i32 [[TMP7]], [[INV_MASK]]
; R600-NEXT:    br label [[PARTWORD_CMPXCHG_LOOP:%.*]]
; R600:       partword.cmpxchg.loop:
; R600-NEXT:    [[TMP9:%.*]] = phi i32 [ [[TMP8]], [[TMP0:%.*]] ], [ [[TMP15:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ]
; R600-NEXT:    [[TMP10:%.*]] = or i32 [[TMP9]], [[TMP4]]
; R600-NEXT:    [[TMP11:%.*]] = or i32 [[TMP9]], [[TMP6]]
; R600-NEXT:    [[TMP12:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[TMP11]], i32 [[TMP10]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[TMP13:%.*]] = extractvalue { i32, i1 } [[TMP12]], 0
; R600-NEXT:    [[TMP14:%.*]] = extractvalue { i32, i1 } [[TMP12]], 1
; R600-NEXT:    br i1 [[TMP14]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]]
; R600:       partword.cmpxchg.failure:
; R600-NEXT:    [[TMP15]] = and i32 [[TMP13]], [[INV_MASK]]
; R600-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP9]], [[TMP15]]
; R600-NEXT:    br i1 [[TMP16]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]]
; R600:       partword.cmpxchg.end:
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP13]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP17:%.*]] = insertvalue { i8, i1 } poison, i8 [[EXTRACTED]], 0
; R600-NEXT:    [[TMP18:%.*]] = insertvalue { i8, i1 } [[TMP17]], i1 [[TMP14]], 1
; R600-NEXT:    [[EXTRACT:%.*]] = extractvalue { i8, i1 } [[TMP18]], 0
; R600-NEXT:    ret i8 [[EXTRACT]]
;
  %gep = getelementptr i8, ptr addrspace(1) %out, i64 4
  %res = cmpxchg ptr addrspace(1) %gep, i8 %old, i8 %in syncscope("agent") seq_cst seq_cst
  %extract = extractvalue {i8, i1} %res, 0
  ret i8 %extract
}

define i8 @test_cmpxchg_i8_local_align2(ptr addrspace(3) %out, i8 %in, i8 %old) {
; CHECK-LABEL: @test_cmpxchg_i8_local_align2(
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i8, ptr addrspace(3) [[OUT:%.*]], i64 4
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[GEP]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[GEP]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i8 [[IN:%.*]] to i32
; CHECK-NEXT:    [[TMP4:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; CHECK-NEXT:    [[TMP5:%.*]] = zext i8 [[OLD:%.*]] to i32
; CHECK-NEXT:    [[TMP6:%.*]] = shl i32 [[TMP5]], [[TMP2]]
; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    [[TMP8:%.*]] = and i32 [[TMP7]], [[INV_MASK]]
; CHECK-NEXT:    br label [[PARTWORD_CMPXCHG_LOOP:%.*]]
; CHECK:       partword.cmpxchg.loop:
; CHECK-NEXT:    [[TMP9:%.*]] = phi i32 [ [[TMP8]], [[TMP0:%.*]] ], [ [[TMP15:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ]
; CHECK-NEXT:    [[TMP10:%.*]] = or i32 [[TMP9]], [[TMP4]]
; CHECK-NEXT:    [[TMP11:%.*]] = or i32 [[TMP9]], [[TMP6]]
; CHECK-NEXT:    [[TMP12:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[TMP11]], i32 [[TMP10]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[TMP13:%.*]] = extractvalue { i32, i1 } [[TMP12]], 0
; CHECK-NEXT:    [[TMP14:%.*]] = extractvalue { i32, i1 } [[TMP12]], 1
; CHECK-NEXT:    br i1 [[TMP14]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]]
; CHECK:       partword.cmpxchg.failure:
; CHECK-NEXT:    [[TMP15]] = and i32 [[TMP13]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP16:%.*]] = icmp ne i32 [[TMP9]], [[TMP15]]
; CHECK-NEXT:    br i1 [[TMP16]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]]
; CHECK:       partword.cmpxchg.end:
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP13]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    [[TMP17:%.*]] = insertvalue { i8, i1 } poison, i8 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP18:%.*]] = insertvalue { i8, i1 } [[TMP17]], i1 [[TMP14]], 1
; CHECK-NEXT:    [[EXTRACT:%.*]] = extractvalue { i8, i1 } [[TMP18]], 0
; CHECK-NEXT:    ret i8 [[EXTRACT]]
;
  %gep = getelementptr i8, ptr addrspace(3) %out, i64 4
  %res = cmpxchg ptr addrspace(3) %gep, i8 %old, i8 %in seq_cst seq_cst, align 2
  %extract = extractvalue {i8, i1} %res, 0
  ret i8 %extract
}

define i8 @test_atomicrmw_inc_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_inc_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_inc_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_global_agent_align2(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_inc_i8_global_agent_align2(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_inc_i8_global_agent_align2(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_global_agent_align4(ptr addrspace(1) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i8_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i8
; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP3]], i8 0, i8 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED1]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_local(ptr addrspace(3) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i8_local(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(3) %ptr, i8 %value seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_local_align2(ptr addrspace(3) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i8_local_align2(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(3) %ptr, i8 %value seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_local_align4(ptr addrspace(3) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i8_local_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(3) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i8
; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP3]], i8 0, i8 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED1]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(3) %ptr, i8 %value seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_flat_agent(ptr %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_inc_i8_flat_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_inc_i8_flat_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_flat_agent_align2(ptr %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_inc_i8_flat_agent_align2(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_inc_i8_flat_agent_align2(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = add i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 0, i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr %ptr, i8 %value syncscope("agent") seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_inc_i8_flat_agent_align4(ptr %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i8_flat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i8
; CHECK-NEXT:    [[TMP2:%.*]] = add i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp uge i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP3]], i8 0, i8 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP4]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP4]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED1]]
;
  %res = atomicrmw uinc_wrap ptr %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_global_agent(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_dec_i8_global_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; GCN-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_dec_i8_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; R600-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_global_agent_align2(ptr addrspace(1) %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_dec_i8_global_agent_align2(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; GCN-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_dec_i8_global_agent_align2(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i32(ptr addrspace(1) [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; R600-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_global_agent_align4(ptr addrspace(1) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i8_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i8
; CHECK-NEXT:    [[TMP2:%.*]] = sub i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 [[VALUE]], i8 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED1]]
;
  %res = atomicrmw udec_wrap ptr addrspace(1) %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_local(ptr addrspace(3) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i8_local(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr addrspace(3) %ptr, i8 %value seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_local_align2(ptr addrspace(3) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i8_local_align2(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(3) @llvm.ptrmask.p3.i32(ptr addrspace(3) [[PTR:%.*]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(3) [[PTR]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(3) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr addrspace(3) %ptr, i8 %value seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_local_align4(ptr addrspace(3) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i8_local_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr addrspace(3) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i8
; CHECK-NEXT:    [[TMP2:%.*]] = sub i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 [[VALUE]], i8 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED1]]
;
  %res = atomicrmw udec_wrap ptr addrspace(3) %ptr, i8 %value seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_flat_agent(ptr %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_dec_i8_flat_agent(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; GCN-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_dec_i8_flat_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; R600-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_flat_agent_align2(ptr %ptr, i8 %value) {
; GCN-LABEL: @test_atomicrmw_dec_i8_flat_agent_align2(
; GCN-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; GCN-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; GCN-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; GCN-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; GCN-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; GCN-NEXT:    [[MASK:%.*]] = shl i32 255, [[SHIFTAMT]]
; GCN-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; GCN-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; GCN-NEXT:    br label [[ATOMICRMW_START:%.*]]
; GCN:       atomicrmw.start:
; GCN-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; GCN-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; GCN-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; GCN-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; GCN-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; GCN-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; GCN-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; GCN-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; GCN-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; GCN-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; GCN-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; GCN-NEXT:    [[TMP8:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; GCN-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; GCN-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; GCN-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; GCN:       atomicrmw.end:
; GCN-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; GCN-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; GCN-NEXT:    ret i8 [[EXTRACTED3]]
;
; R600-LABEL: @test_atomicrmw_dec_i8_flat_agent_align2(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i32(ptr [[PTR:%.*]], i32 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i32
; R600-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; R600-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; R600-NEXT:    br label [[ATOMICRMW_START:%.*]]
; R600:       atomicrmw.start:
; R600-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[LOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; R600-NEXT:    [[TMP4:%.*]] = sub i8 [[EXTRACTED]], 1
; R600-NEXT:    [[TMP5:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; R600-NEXT:    [[TMP6:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; R600-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; R600-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i8 [[VALUE]], i8 [[TMP4]]
; R600-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; R600-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[TMP2]]
; R600-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; R600-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; R600-NEXT:    [[TMP8:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; R600-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; R600-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; R600-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; R600:       atomicrmw.end:
; R600-NEXT:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; R600-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i8
; R600-NEXT:    ret i8 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr %ptr, i8 %value syncscope("agent") seq_cst, align 2
  ret i8 %res
}

define i8 @test_atomicrmw_dec_i8_flat_agent_align4(ptr %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i8_flat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = load i32, ptr [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP1]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[LOADED]] to i8
; CHECK-NEXT:    [[TMP2:%.*]] = sub i8 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i8 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i8 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i8 [[VALUE]], i8 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i8 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED1]]
;
  %res = atomicrmw udec_wrap ptr %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_xchg_i8_buffer_fat_agent(ptr addrspace(7) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i8_buffer_fat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(7) @llvm.ptrmask.p7.i32(ptr addrspace(7) [[PTR:%.*]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(7) [[PTR]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(7) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP5:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP6:%.*]] = or i32 [[TMP5]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP7:%.*]] = cmpxchg ptr addrspace(7) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP6]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP7]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP7]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(7) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_xchg_i8_buffer_fat_agent_align4(ptr addrspace(7) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i8_buffer_fat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i8 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(7) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP2]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP3:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[TMP4:%.*]] = or i32 [[TMP3]], [[TMP1]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[LOADED]], i32 [[TMP4]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP5]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP5]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(7) %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}

define i8 @test_atomicrmw_add_i8_buffer_fat_agent(ptr addrspace(7) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i8_buffer_fat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(7) @llvm.ptrmask.p7.i32(ptr addrspace(7) [[PTR:%.*]], i32 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(7) [[PTR]] to i32
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i32 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[PTRLSB]], 3
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 255, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i8 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(7) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP5:%.*]] = and i32 [[NEW]], [[MASK]]
; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[TMP5]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(7) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP8]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(7) %ptr, i8 %value syncscope("agent") seq_cst
  ret i8 %res
}

define i8 @test_atomicrmw_add_i8_buffer_fat_agent_align4(ptr addrspace(7) %ptr, i8 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i8_buffer_fat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i8 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(7) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP2]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[NEW:%.*]] = add i32 [[LOADED]], [[TMP1]]
; CHECK-NEXT:    [[TMP3:%.*]] = and i32 [[NEW]], 255
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -256
; CHECK-NEXT:    [[TMP5:%.*]] = or i32 [[TMP4]], [[TMP3]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(7) [[PTR]], i32 [[LOADED]], i32 [[TMP5]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP6]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP6]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i8
; CHECK-NEXT:    ret i8 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(7) %ptr, i8 %value syncscope("agent") seq_cst, align 4
  ret i8 %res
}