; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -mtriple=amdgcn-amd-amdhsa -S -passes='require<libcall-lowering-info>,atomic-expand' %s | FileCheck --check-prefixes=CHECK,GCN %s

define i16 @test_atomicrmw_xchg_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_xchg_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i16_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[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:    [[TMP3:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[TMP4:%.*]] = or i32 [[TMP3]], [[TMP1]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_add_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_add_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i16_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[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]], 65535
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_sub_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_sub_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = sub 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(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw sub ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[NEW]] 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:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; 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]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; 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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

; Drop unknown metadata and noundef
define i16 @test_atomicrmw_and_i16_global_agent_drop_md(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_drop_md(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[NEW]] 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:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_drop_md(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; 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]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !noundef !0, !some.unknown.md !0
  ret i16 %res
}

; Drop unknown metadata
define i16 @test_atomicrmw_and_i16_global_agent_align4_drop_md(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4_drop_md(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; 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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4_drop_md(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !noundef !0, !some.unknown.md !0
  ret i16 %res
}

; Drop noundef, preserve mmra
define i16 @test_atomicrmw_and_i16_global_agent_preserve_mmra(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_preserve_mmra(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4, !mmra [[META0:![0-9]+]]
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[NEW:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4, !mmra [[META0]]
; 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:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_preserve_mmra(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; 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, !mmra [[META0:![0-9]+]]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !noundef !0, !mmra !1
  ret i16 %res
}

; Drop noundef, preserve mmra
define i16 @test_atomicrmw_and_i16_global_agent_align4_preserve_mmra(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_mmra(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; CHECK-NEXT:    [[TMP2:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4, !mmra [[META0]]
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP2]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[NEW:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4, !mmra [[META0]]
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_mmra(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !mmra [[META0]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !noundef !0, !mmra !1
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4_preserve_alias_scope(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_alias_scope(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; 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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4, !alias.scope [[META1:![0-9]+]]
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_alias_scope(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !alias.scope [[META1:![0-9]+]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !alias.scope !2
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4_preserve_noalias(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_noalias(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; 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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4, !noalias [[META1]]
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_noalias(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !noalias [[META1]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !noalias !2
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4_preserve_tbaa_struct(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_tbaa_struct(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; 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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4, !tbaa.struct [[TBAA_STRUCT4:![0-9]+]]
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_tbaa_struct(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !tbaa.struct [[TBAA_STRUCT4:![0-9]+]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !tbaa.struct !5
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4_preserve_tbaa(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_tbaa(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; 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:%.*]] = and i32 [[LOADED]], [[ANDOPERAND]]
; CHECK-NEXT:    [[TMP3:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[NEW]] syncscope("agent") seq_cst seq_cst, align 4, !tbaa [[TBAA5:![0-9]+]]
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP3]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP3]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4_preserve_tbaa(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !tbaa [[TBAA5:![0-9]+]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !tbaa !6
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent__amdgpu_no_remote_memory(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP4:%.*]] = atomicrmw and ptr addrspace(1) [[ALIGNEDADDR]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory [[META8:![0-9]+]]
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent__amdgpu_no_remote_memory(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; 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, !amdgpu.no.remote.memory [[META8:![0-9]+]]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !amdgpu.no.remote.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4__amdgpu_no_remote_memory(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; CHECK-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory [[META8]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4__amdgpu_no_remote_memory(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory [[META8]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !amdgpu.no.remote.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent__amdgpu_no_fine_grained_memory(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[VALOPERAND_SHIFTED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP4:%.*]] = atomicrmw and ptr addrspace(1) [[ALIGNEDADDR]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META8]]
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent__amdgpu_no_fine_grained_memory(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; 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, !amdgpu.no.fine.grained.memory [[META8]]
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_and_i16_global_agent_align4__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_and_i16_global_agent_align4__amdgpu_no_fine_grained_memory(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; CHECK-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META8]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_and_i16_global_agent_align4__amdgpu_no_fine_grained_memory(
; R600-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[ANDOPERAND:%.*]] = or i32 [[TMP1]], -65536
; R600-NEXT:    [[TMP2:%.*]] = atomicrmw and ptr addrspace(1) [[PTR:%.*]], i32 [[ANDOPERAND]] syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META8]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw and ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4, !amdgpu.no.fine.grained.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_nand_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_nand_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[NEW:%.*]] = xor i32 [[TMP5]], -1
; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[NEW]], [[MASK]]
; CHECK-NEXT:    [[TMP7:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP8:%.*]] = or i32 [[TMP7]], [[TMP6]]
; CHECK-NEXT:    [[TMP9:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP8]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP9]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP9]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw nand ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_or_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_or_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = or i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[NEW]] 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:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_or_i16_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; R600-NEXT:    [[TMP4:%.*]] = atomicrmw or ptr addrspace(1) [[ALIGNEDADDR]], i32 [[VALOPERAND_SHIFTED]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw or ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_xor_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xor_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = xor i32 [[LOADED]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[NEW]] 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:    [[SHIFTED:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
; R600-LABEL: @test_atomicrmw_xor_i16_global_agent(
; R600-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; R600-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; R600-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; R600-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; R600-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; R600-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; R600-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; R600-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; R600-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; R600-NEXT:    [[TMP4:%.*]] = atomicrmw xor ptr addrspace(1) [[ALIGNEDADDR]], i32 [[VALOPERAND_SHIFTED]] syncscope("agent") seq_cst, align 4
; R600-NEXT:    [[SHIFTED:%.*]] = lshr i32 [[TMP4]], [[SHIFTAMT]]
; R600-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; R600-NEXT:    ret i16 [[EXTRACTED]]
  %res = atomicrmw xor ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_max_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_max_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp sgt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[EXTRACTED]], i16 [[VALUE]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw max ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_min_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_min_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp sle i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[EXTRACTED]], i16 [[VALUE]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw min ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_umax_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_umax_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[EXTRACTED]], i16 [[VALUE]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw umax ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_umin_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_umin_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ule i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[EXTRACTED]], i16 [[VALUE]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP5:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw umin ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_cmpxchg_i16_global_agent(ptr addrspace(1) %out, i16 %in, i16 %old) {
; CHECK-LABEL: @test_cmpxchg_i16_global_agent(
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i16, ptr addrspace(1) [[OUT:%.*]], i64 4
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[GEP]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[GEP]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[IN:%.*]] to i32
; CHECK-NEXT:    [[TMP4:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP5:%.*]] = zext i16 [[OLD:%.*]] to i32
; CHECK-NEXT:    [[TMP6:%.*]] = shl i32 [[TMP5]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP7:%.*]] = load i32, ptr addrspace(1) [[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(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP17:%.*]] = insertvalue { i16, i1 } poison, i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP18:%.*]] = insertvalue { i16, i1 } [[TMP17]], i1 [[TMP14]], 1
; CHECK-NEXT:    [[EXTRACT:%.*]] = extractvalue { i16, i1 } [[TMP18]], 0
; CHECK-NEXT:    ret i16 [[EXTRACT]]
;
  %gep = getelementptr i16, ptr addrspace(1) %out, i64 4
  %res = cmpxchg ptr addrspace(1) %gep, i16 %old, i16 %in seq_cst seq_cst
  %extract = extractvalue {i16, i1} %res, 0
  ret i16 %extract
}

define i16 @test_cmpxchg_i16_global_agent_align4(ptr addrspace(1) %out, i16 %in, i16 %old) {
; CHECK-LABEL: @test_cmpxchg_i16_global_agent_align4(
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i16, ptr addrspace(1) [[OUT:%.*]], i64 4
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[IN:%.*]] to i32
; CHECK-NEXT:    [[TMP2:%.*]] = zext i16 [[OLD:%.*]] to i32
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[GEP]], align 4
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[TMP3]], -65536
; CHECK-NEXT:    br label [[PARTWORD_CMPXCHG_LOOP:%.*]]
; CHECK:       partword.cmpxchg.loop:
; CHECK-NEXT:    [[TMP5:%.*]] = phi i32 [ [[TMP4]], [[TMP0:%.*]] ], [ [[TMP11:%.*]], [[PARTWORD_CMPXCHG_FAILURE:%.*]] ]
; CHECK-NEXT:    [[TMP6:%.*]] = or i32 [[TMP5]], [[TMP1]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i32 [[TMP5]], [[TMP2]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[GEP]], i32 [[TMP7]], i32 [[TMP6]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[TMP9:%.*]] = extractvalue { i32, i1 } [[TMP8]], 0
; CHECK-NEXT:    [[TMP10:%.*]] = extractvalue { i32, i1 } [[TMP8]], 1
; CHECK-NEXT:    br i1 [[TMP10]], label [[PARTWORD_CMPXCHG_END:%.*]], label [[PARTWORD_CMPXCHG_FAILURE]]
; CHECK:       partword.cmpxchg.failure:
; CHECK-NEXT:    [[TMP11]] = and i32 [[TMP9]], -65536
; CHECK-NEXT:    [[TMP12:%.*]] = icmp ne i32 [[TMP5]], [[TMP11]]
; CHECK-NEXT:    br i1 [[TMP12]], label [[PARTWORD_CMPXCHG_LOOP]], label [[PARTWORD_CMPXCHG_END]]
; CHECK:       partword.cmpxchg.end:
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP9]] to i16
; CHECK-NEXT:    [[TMP13:%.*]] = insertvalue { i16, i1 } poison, i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP14:%.*]] = insertvalue { i16, i1 } [[TMP13]], i1 [[TMP10]], 1
; CHECK-NEXT:    [[EXTRACT:%.*]] = extractvalue { i16, i1 } [[TMP14]], 0
; CHECK-NEXT:    ret i16 [[EXTRACT]]
;
  %gep = getelementptr i16, ptr addrspace(1) %out, i64 4
  %res = cmpxchg ptr addrspace(1) %gep, i16 %old, i16 %in seq_cst seq_cst, align 4
  %extract = extractvalue {i16, i1} %res, 0
  ret i16 %extract
}

define i16 @test_atomicrmw_xchg_i16_local(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i16_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 65535, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(3) [[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(3) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP6]] 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(3) %ptr, i16 %value seq_cst
  ret i16 %res
}

define i16 @test_cmpxchg_i16_local(ptr addrspace(3) %out, i16 %in, i16 %old) {
; CHECK-LABEL: @test_cmpxchg_i16_local(
; CHECK-NEXT:    [[GEP:%.*]] = getelementptr i16, 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 65535, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[IN:%.*]] to i32
; CHECK-NEXT:    [[TMP4:%.*]] = shl i32 [[TMP3]], [[TMP2]]
; CHECK-NEXT:    [[TMP5:%.*]] = zext i16 [[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 i16
; CHECK-NEXT:    [[TMP17:%.*]] = insertvalue { i16, i1 } poison, i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP18:%.*]] = insertvalue { i16, i1 } [[TMP17]], i1 [[TMP14]], 1
; CHECK-NEXT:    [[EXTRACT:%.*]] = extractvalue { i16, i1 } [[TMP18]], 0
; CHECK-NEXT:    ret i16 [[EXTRACT]]
;
  %gep = getelementptr i16, ptr addrspace(3) %out, i64 4
  %res = cmpxchg ptr addrspace(3) %gep, i16 %old, i16 %in seq_cst seq_cst
  %extract = extractvalue {i16, i1} %res, 0
  ret i16 %extract
}

define i16 @test_atomicrmw_xor_i16_local_align4(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xor_i16_local_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[TMP2:%.*]] = atomicrmw xor ptr addrspace(3) [[PTR:%.*]], i32 [[TMP1]] seq_cst, align 4
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[TMP2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw xor ptr addrspace(3) %ptr, i16 %value seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_inc_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = add i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i16 0, i16 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_inc_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = add i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP3]], i16 0, i16 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_inc_i16_local(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i16_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 65535, [[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 i16
; CHECK-NEXT:    [[TMP4:%.*]] = add i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i16 0, i16 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[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]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(3) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_inc_i16_local_align4(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = add i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP3]], i16 0, i16 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw uinc_wrap ptr addrspace(3) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_inc_i16_flat_agent(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i16_flat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = add i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i16 0, i16 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw uinc_wrap ptr %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_inc_i16_flat_agent_align4(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_inc_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = add i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP3]], i16 0, i16 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw uinc_wrap ptr %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_dec_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = sub i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i16 [[VALUE]], i16 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_dec_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = sub i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i16 [[VALUE]], i16 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw udec_wrap ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_dec_i16_local(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i16_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 65535, [[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 i16
; CHECK-NEXT:    [[TMP4:%.*]] = sub i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i16 [[VALUE]], i16 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr addrspace(3) %ptr, i16 %value seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_dec_i16_local_align4(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = sub i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i16 [[VALUE]], i16 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw udec_wrap ptr addrspace(3) %ptr, i16 %value seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_dec_i16_flat_agent(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i16_flat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = sub i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP6:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i1 [[TMP5]], [[TMP6]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP7]], i16 [[VALUE]], i16 [[TMP4]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[INSERTED]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw udec_wrap ptr %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_dec_i16_flat_agent_align4(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_dec_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = sub i16 [[EXTRACTED]], 1
; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i16 [[EXTRACTED]], 0
; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = or i1 [[TMP3]], [[TMP4]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP5]], i16 [[VALUE]], i16 [[TMP2]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw udec_wrap ptr %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define half @test_atomicrmw_xchg_f16_global_agent(ptr addrspace(1) %ptr, half %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_f16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = bitcast half [[VALUE:%.*]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = zext i16 [[TMP3]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP4]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP5]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i16 [[EXTRACTED]] to half
; CHECK-NEXT:    ret half [[TMP9]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, half %value syncscope("agent") seq_cst
  ret half %res
}

define half @test_atomicrmw_xchg_f16_global_agent_align4(ptr addrspace(1) %ptr, half %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_f16_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = bitcast half [[VALUE:%.*]] to i16
; CHECK-NEXT:    [[TMP2:%.*]] = zext i16 [[TMP1]] to i32
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[TMP5:%.*]] = or i32 [[TMP4]], [[TMP2]]
; 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 i16
; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i16 [[EXTRACTED]] to half
; CHECK-NEXT:    ret half [[TMP7]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, half %value syncscope("agent") seq_cst, align 4
  ret half %res
}

define half @test_atomicrmw_xchg_f16_flat_agent(ptr %ptr, half %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_f16_flat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = bitcast half [[VALUE:%.*]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = zext i16 [[TMP3]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP4]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP5]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i16 [[EXTRACTED]] to half
; CHECK-NEXT:    ret half [[TMP9]]
;
  %res = atomicrmw xchg ptr %ptr, half %value syncscope("agent") seq_cst
  ret half %res
}

define half @test_atomicrmw_xchg_f16_flat_agent_align4(ptr %ptr, half %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_f16_flat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = bitcast half [[VALUE:%.*]] to i16
; CHECK-NEXT:    [[TMP2:%.*]] = zext i16 [[TMP1]] to i32
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[TMP5:%.*]] = or i32 [[TMP4]], [[TMP2]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[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 i16
; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i16 [[EXTRACTED]] to half
; CHECK-NEXT:    ret half [[TMP7]]
;
  %res = atomicrmw xchg ptr %ptr, half %value syncscope("agent") seq_cst, align 4
  ret half %res
}

define bfloat @test_atomicrmw_xchg_bf16_global_agent(ptr addrspace(1) %ptr, bfloat %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_bf16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = bitcast bfloat [[VALUE:%.*]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = zext i16 [[TMP3]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP4]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP5:%.*]] = load i32, ptr addrspace(1) [[ALIGNEDADDR]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP5]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP6:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[TMP7:%.*]] = or i32 [[TMP6]], [[VALOPERAND_SHIFTED]]
; CHECK-NEXT:    [[TMP8:%.*]] = cmpxchg ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP9:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat
; CHECK-NEXT:    ret bfloat [[TMP9]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, bfloat %value syncscope("agent") seq_cst
  ret bfloat %res
}

define bfloat @test_atomicrmw_xchg_bf16_global_agent_align4(ptr addrspace(1) %ptr, bfloat %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_bf16_global_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = bitcast bfloat [[VALUE:%.*]] to i16
; CHECK-NEXT:    [[TMP2:%.*]] = zext i16 [[TMP1]] to i32
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[PTR:%.*]], align 4
; CHECK-NEXT:    br label [[ATOMICRMW_START:%.*]]
; CHECK:       atomicrmw.start:
; CHECK-NEXT:    [[LOADED:%.*]] = phi i32 [ [[TMP3]], [[TMP0:%.*]] ], [ [[NEWLOADED:%.*]], [[ATOMICRMW_START]] ]
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[TMP5:%.*]] = or i32 [[TMP4]], [[TMP2]]
; 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 i16
; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i16 [[EXTRACTED]] to bfloat
; CHECK-NEXT:    ret bfloat [[TMP7]]
;
  %res = atomicrmw xchg ptr addrspace(1) %ptr, bfloat %value syncscope("agent") seq_cst, align 4
  ret bfloat %res
}

define i16 @test_atomicrmw_xchg_i16_buffer_fat_agent(ptr addrspace(7) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i16_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 65535, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(7) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_xchg_i16_buffer_fat_agent_align4(ptr addrspace(7) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_xchg_i16_buffer_fat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[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]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw xchg ptr addrspace(7) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_add_i16_buffer_fat_agent(ptr addrspace(7) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i16_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 65535, [[TMP2]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(7) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_add_i16_buffer_fat_agent_align4(ptr addrspace(7) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_add_i16_buffer_fat_agent_align4(
; CHECK-NEXT:    [[TMP1:%.*]] = zext i16 [[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]], 65535
; CHECK-NEXT:    [[TMP4:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw add ptr addrspace(7) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_sub_i16_global_agent__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_sub_i16_global_agent__amdgpu_no_remote_memory(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = sub 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(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META8]]
; 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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw sub ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !amdgpu.no.remote.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_sub_i16_global_agent__amdgpu_no_fine_grained_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_sub_i16_global_agent__amdgpu_no_fine_grained_memory(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = sub 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(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.fine.grained.memory [[META8]]
; 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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw sub ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_sub_i16_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_sub_i16_global_agent__amdgpu_no_fine_grained_memory__amdgpu_no_remote_memory(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = zext i16 [[VALUE:%.*]] to i32
; CHECK-NEXT:    [[VALOPERAND_SHIFTED:%.*]] = shl i32 [[TMP3]], [[SHIFTAMT]]
; CHECK-NEXT:    [[TMP4:%.*]] = load i32, ptr addrspace(1) [[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:%.*]] = sub 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(1) [[ALIGNEDADDR]], i32 [[LOADED]], i32 [[TMP7]] syncscope("agent") seq_cst seq_cst, align 4, !amdgpu.no.remote.memory [[META8]], !amdgpu.no.fine.grained.memory [[META8]]
; 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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED]]
;
  %res = atomicrmw sub ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, !amdgpu.no.fine.grained.memory !0, !amdgpu.no.remote.memory !0
  ret i16 %res
}

define i16 @test_atomicrmw_usub_cond_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_cond_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = sub i16 [[EXTRACTED]], [[VALUE]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[TMP5]], i16 [[EXTRACTED]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw usub_cond ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_usub_cond_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_cond_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP3:%.*]] = sub i16 [[EXTRACTED]], [[VALUE]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP2]], i16 [[TMP3]], i16 [[EXTRACTED]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw usub_cond ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_usub_cond_i16_local(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_cond_i16_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 65535, [[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 i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = sub i16 [[EXTRACTED]], [[VALUE]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[TMP5]], i16 [[EXTRACTED]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[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]] 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw usub_cond ptr addrspace(3) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_usub_cond_i16_local_align4(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_cond_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP3:%.*]] = sub i16 [[EXTRACTED]], [[VALUE]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP2]], i16 [[TMP3]], i16 [[EXTRACTED]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw usub_cond ptr addrspace(3) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_usub_cond_i16_flat_agent(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_cond_i16_flat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[TMP4:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP5:%.*]] = sub i16 [[EXTRACTED]], [[VALUE]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP4]], i16 [[TMP5]], i16 [[EXTRACTED]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP6:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw usub_cond ptr %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_usub_cond_i16_flat_agent_align4(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_cond_i16_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 i16
; CHECK-NEXT:    [[TMP2:%.*]] = icmp uge i16 [[EXTRACTED]], [[VALUE:%.*]]
; CHECK-NEXT:    [[TMP3:%.*]] = sub i16 [[EXTRACTED]], [[VALUE]]
; CHECK-NEXT:    [[NEW:%.*]] = select i1 [[TMP2]], i16 [[TMP3]], i16 [[EXTRACTED]]
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; 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 i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw usub_cond ptr %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_usub_sat_i16_global_agent(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_sat_i16_global_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr addrspace(1) @llvm.ptrmask.p1.i64(ptr addrspace(1) [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr addrspace(1) [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr addrspace(1) [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[NEW:%.*]] = call i16 @llvm.usub.sat.i16(i16 [[EXTRACTED]], i16 [[VALUE:%.*]])
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr addrspace(1) [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw usub_sat ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_usub_sat_i16_global_agent_align4(ptr addrspace(1) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_sat_i16_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 i16
; CHECK-NEXT:    [[NEW:%.*]] = call i16 @llvm.usub.sat.i16(i16 [[EXTRACTED]], i16 [[VALUE:%.*]])
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP2:%.*]] = cmpxchg ptr addrspace(1) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP2]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP2]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw usub_sat ptr addrspace(1) %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_usub_sat_i16_local(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_sat_i16_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 65535, [[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 i16
; CHECK-NEXT:    [[NEW:%.*]] = call i16 @llvm.usub.sat.i16(i16 [[EXTRACTED]], i16 [[VALUE:%.*]])
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[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:    [[TMP4:%.*]] = cmpxchg ptr addrspace(3) [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[TMP2]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw usub_sat ptr addrspace(3) %ptr, i16 %value seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_usub_sat_i16_local_align4(ptr addrspace(3) %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_sat_i16_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 i16
; CHECK-NEXT:    [[NEW:%.*]] = call i16 @llvm.usub.sat.i16(i16 [[EXTRACTED]], i16 [[VALUE:%.*]])
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP2:%.*]] = cmpxchg ptr addrspace(3) [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP2]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP2]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw usub_sat ptr addrspace(3) %ptr, i16 %value seq_cst, align 4
  ret i16 %res
}

define i16 @test_atomicrmw_usub_sat_i16_flat_agent(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_sat_i16_flat_agent(
; CHECK-NEXT:    [[ALIGNEDADDR:%.*]] = call ptr @llvm.ptrmask.p0.i64(ptr [[PTR:%.*]], i64 -4)
; CHECK-NEXT:    [[TMP1:%.*]] = ptrtoint ptr [[PTR]] to i64
; CHECK-NEXT:    [[PTRLSB:%.*]] = and i64 [[TMP1]], 3
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[PTRLSB]], 3
; CHECK-NEXT:    [[SHIFTAMT:%.*]] = trunc i64 [[TMP2]] to i32
; CHECK-NEXT:    [[MASK:%.*]] = shl i32 65535, [[SHIFTAMT]]
; CHECK-NEXT:    [[INV_MASK:%.*]] = xor i32 [[MASK]], -1
; CHECK-NEXT:    [[TMP3:%.*]] = load i32, ptr [[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]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED:%.*]] = trunc i32 [[SHIFTED]] to i16
; CHECK-NEXT:    [[NEW:%.*]] = call i16 @llvm.usub.sat.i16(i16 [[EXTRACTED]], i16 [[VALUE:%.*]])
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[SHIFTED1:%.*]] = shl nuw i32 [[EXTENDED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], [[INV_MASK]]
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[SHIFTED1]]
; CHECK-NEXT:    [[TMP4:%.*]] = cmpxchg ptr [[ALIGNEDADDR]], 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:    [[SHIFTED2:%.*]] = lshr i32 [[NEWLOADED]], [[SHIFTAMT]]
; CHECK-NEXT:    [[EXTRACTED3:%.*]] = trunc i32 [[SHIFTED2]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED3]]
;
  %res = atomicrmw usub_sat ptr %ptr, i16 %value syncscope("agent") seq_cst
  ret i16 %res
}

define i16 @test_atomicrmw_usub_sat_i16_flat_agent_align4(ptr %ptr, i16 %value) {
; CHECK-LABEL: @test_atomicrmw_usub_sat_i16_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 i16
; CHECK-NEXT:    [[NEW:%.*]] = call i16 @llvm.usub.sat.i16(i16 [[EXTRACTED]], i16 [[VALUE:%.*]])
; CHECK-NEXT:    [[EXTENDED:%.*]] = zext i16 [[NEW]] to i32
; CHECK-NEXT:    [[UNMASKED:%.*]] = and i32 [[LOADED]], -65536
; CHECK-NEXT:    [[INSERTED:%.*]] = or i32 [[UNMASKED]], [[EXTENDED]]
; CHECK-NEXT:    [[TMP2:%.*]] = cmpxchg ptr [[PTR]], i32 [[LOADED]], i32 [[INSERTED]] syncscope("agent") seq_cst seq_cst, align 4
; CHECK-NEXT:    [[SUCCESS:%.*]] = extractvalue { i32, i1 } [[TMP2]], 1
; CHECK-NEXT:    [[NEWLOADED]] = extractvalue { i32, i1 } [[TMP2]], 0
; CHECK-NEXT:    br i1 [[SUCCESS]], label [[ATOMICRMW_END:%.*]], label [[ATOMICRMW_START]]
; CHECK:       atomicrmw.end:
; CHECK-NEXT:    [[EXTRACTED1:%.*]] = trunc i32 [[NEWLOADED]] to i16
; CHECK-NEXT:    ret i16 [[EXTRACTED1]]
;
  %res = atomicrmw usub_sat ptr %ptr, i16 %value syncscope("agent") seq_cst, align 4
  ret i16 %res
}

!0 = !{}
!1 = !{!"foo", !"bar"}
!2 = !{!3}
!3 = distinct !{!3, !4}
!4 = distinct !{!4}
!5 = !{i64 0, i64 4, !1, i64 8, i64 4}
!6 = !{!7, !7, i64 0}
!7 = !{!"omnipotent char", !8, i64 0}
!8 = !{!"Simple C/C++ TBAA"}

;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
; GCN: {{.*}}