; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
; Test moves between FPRs and GPRs.  The 32-bit cases test the z10
; implementation, which has no high-word support.
;
; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z10 | FileCheck %s

declare i64 @foo()
declare double @bar()
@dptr = external global double
@iptr = external global i64

; Test 32-bit moves from GPRs to FPRs.  The GPR must be moved into the high
; 32 bits of the FPR.
define float @f1(i32 %a) {
; CHECK-LABEL: f1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    # kill: def $r2l killed $r2l def $r2d
; CHECK-NEXT:    sllg %r0, %r2, 32
; CHECK-NEXT:    ldgr %f0, %r0
; CHECK-NEXT:    # kill: def $f0s killed $f0s killed $f0d
; CHECK-NEXT:    br %r14
  %res = bitcast i32 %a to float
  ret float %res
}

; Like f1, but create a situation where the shift can be folded with
; surrounding code.
define float @f2(i64 %big) {
; CHECK-LABEL: f2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    risbg %r0, %r2, 0, 159, 31
; CHECK-NEXT:    ldgr %f0, %r0
; CHECK-NEXT:    # kill: def $f0s killed $f0s killed $f0d
; CHECK-NEXT:    br %r14
  %shift = lshr i64 %big, 1
  %a = trunc i64 %shift to i32
  %res = bitcast i32 %a to float
  ret float %res
}

; Another example of the same thing.
define float @f3(i64 %big) {
; CHECK-LABEL: f3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    risbg %r0, %r2, 0, 159, 2
; CHECK-NEXT:    ldgr %f0, %r0
; CHECK-NEXT:    # kill: def $f0s killed $f0s killed $f0d
; CHECK-NEXT:    br %r14
  %shift = ashr i64 %big, 30
  %a = trunc i64 %shift to i32
  %res = bitcast i32 %a to float
  ret float %res
}

; Like f1, but the value to transfer is already in the high 32 bits.
define float @f4(i64 %big) {
; CHECK-LABEL: f4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    nilf %r2, 0
; CHECK-NEXT:    ldgr %f0, %r2
; CHECK-NEXT:    # kill: def $f0s killed $f0s killed $f0d
; CHECK-NEXT:    br %r14
  %shift = ashr i64 %big, 32
  %a = trunc i64 %shift to i32
  %res = bitcast i32 %a to float
  ret float %res
}

; Test 64-bit moves from GPRs to FPRs.
define double @f5(i64 %a) {
; CHECK-LABEL: f5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    ldgr %f0, %r2
; CHECK-NEXT:    br %r14
  %res = bitcast i64 %a to double
  ret double %res
}

; Test 128-bit moves from GPRs to FPRs.  i128 isn't a legitimate type,
; so this goes through memory.
define void @f6(ptr %a, ptr %b) {
; CHECK-LABEL: f6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lg %r0, 8(%r3)
; CHECK-NEXT:    lg %r1, 0(%r3)
; CHECK-NEXT:    stg %r0, 8(%r2)
; CHECK-NEXT:    stg %r1, 0(%r2)
; CHECK-NEXT:    br %r14
  %val = load i128, ptr %b
  %res = bitcast i128 %val to fp128
  store fp128 %res, ptr %a
  ret void
}

; Test 32-bit moves from FPRs to GPRs.  The high 32 bits of the FPR should
; be moved into the low 32 bits of the GPR.
define i32 @f7(float %a) {
; CHECK-LABEL: f7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    # kill: def $f0s killed $f0s def $f0d
; CHECK-NEXT:    lgdr %r0, %f0
; CHECK-NEXT:    srlg %r2, %r0, 32
; CHECK-NEXT:    # kill: def $r2l killed $r2l killed $r2d
; CHECK-NEXT:    br %r14
  %res = bitcast float %a to i32
  ret i32 %res
}

; Test 64-bit moves from FPRs to GPRs.
define i64 @f8(double %a) {
; CHECK-LABEL: f8:
; CHECK:       # %bb.0:
; CHECK-NEXT:    lgdr %r2, %f0
; CHECK-NEXT:    br %r14
  %res = bitcast double %a to i64
  ret i64 %res
}

; Test 128-bit moves from FPRs to GPRs, with the same restriction as f6.
define void @f9(ptr %a, ptr %b) {
; CHECK-LABEL: f9:
; CHECK:       # %bb.0:
; CHECK-NEXT:    ld %f0, 0(%r2)
; CHECK-NEXT:    ld %f2, 8(%r2)
; CHECK-NEXT:    std %f0, 0(%r3)
; CHECK-NEXT:    std %f2, 8(%r3)
; CHECK-NEXT:    br %r14
  %val = load fp128, ptr %a
  %res = bitcast fp128 %val to i128
  store i128 %res, ptr %b
  ret void
}

; Test cases where the destination of an LGDR needs to be spilled.
; We shouldn't have any integer stack stores or floating-point loads.
define void @f10(double %extra) {
; CHECK-LABEL: f10:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    stmg %r6, %r15, 48(%r15)
; CHECK-NEXT:    .cfi_offset %r6, -112
; CHECK-NEXT:    .cfi_offset %r7, -104
; CHECK-NEXT:    .cfi_offset %r8, -96
; CHECK-NEXT:    .cfi_offset %r9, -88
; CHECK-NEXT:    .cfi_offset %r10, -80
; CHECK-NEXT:    .cfi_offset %r11, -72
; CHECK-NEXT:    .cfi_offset %r12, -64
; CHECK-NEXT:    .cfi_offset %r13, -56
; CHECK-NEXT:    .cfi_offset %r14, -48
; CHECK-NEXT:    .cfi_offset %r15, -40
; CHECK-NEXT:    aghi %r15, -184
; CHECK-NEXT:    .cfi_def_cfa_offset 344
; CHECK-NEXT:    lgrl %r1, dptr@GOT
; CHECK-NEXT:    ldr %f1, %f0
; CHECK-NEXT:    adb %f1, 0(%r1)
; CHECK-NEXT:    ldr %f2, %f0
; CHECK-NEXT:    adb %f2, 0(%r1)
; CHECK-NEXT:    ldr %f3, %f0
; CHECK-NEXT:    adb %f3, 0(%r1)
; CHECK-NEXT:    std %f1, 176(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f2, 168(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f3, 160(%r15) # 8-byte Spill
; CHECK-NEXT:    ldr %f1, %f0
; CHECK-NEXT:    adb %f1, 0(%r1)
; CHECK-NEXT:    ldr %f2, %f0
; CHECK-NEXT:    adb %f2, 0(%r1)
; CHECK-NEXT:    ldr %f3, %f0
; CHECK-NEXT:    adb %f3, 0(%r1)
; CHECK-NEXT:    ldr %f4, %f0
; CHECK-NEXT:    adb %f4, 0(%r1)
; CHECK-NEXT:    lgdr %r10, %f1
; CHECK-NEXT:    lgdr %r9, %f2
; CHECK-NEXT:    lgdr %r8, %f3
; CHECK-NEXT:    lgdr %r7, %f4
; CHECK-NEXT:    ldr %f1, %f0
; CHECK-NEXT:    adb %f1, 0(%r1)
; CHECK-NEXT:    ldr %f2, %f0
; CHECK-NEXT:    adb %f2, 0(%r1)
; CHECK-NEXT:    adb %f0, 0(%r1)
; CHECK-NEXT:    lgrl %r6, iptr@GOT
; CHECK-NEXT:    lgdr %r13, %f1
; CHECK-NEXT:    lgdr %r12, %f2
; CHECK-NEXT:    lgdr %r11, %f0
; CHECK-NEXT:  .LBB9_1: # %loop
; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    brasl %r14, foo@PLT
; CHECK-NEXT:    lgr %r0, %r2
; CHECK-NEXT:    og %r0, 176(%r15) # 8-byte Folded Reload
; CHECK-NEXT:    og %r0, 168(%r15) # 8-byte Folded Reload
; CHECK-NEXT:    og %r0, 160(%r15) # 8-byte Folded Reload
; CHECK-NEXT:    ogr %r0, %r10
; CHECK-NEXT:    ogr %r0, %r9
; CHECK-NEXT:    ogr %r0, %r8
; CHECK-NEXT:    ogr %r0, %r7
; CHECK-NEXT:    ogr %r0, %r13
; CHECK-NEXT:    ogr %r0, %r12
; CHECK-NEXT:    ogr %r0, %r11
; CHECK-NEXT:    stg %r0, 0(%r6)
; CHECK-NEXT:    cgijlh %r2, 1, .LBB9_1
; CHECK-NEXT:  # %bb.2: # %exit
; CHECK-NEXT:    lmg %r6, %r15, 232(%r15)
; CHECK-NEXT:    br %r14
entry:
  %double0 = load volatile double, ptr@dptr
  %biased0 = fadd double %double0, %extra
  %int0 = bitcast double %biased0 to i64
  %double1 = load volatile double, ptr@dptr
  %biased1 = fadd double %double1, %extra
  %int1 = bitcast double %biased1 to i64
  %double2 = load volatile double, ptr@dptr
  %biased2 = fadd double %double2, %extra
  %int2 = bitcast double %biased2 to i64
  %double3 = load volatile double, ptr@dptr
  %biased3 = fadd double %double3, %extra
  %int3 = bitcast double %biased3 to i64
  %double4 = load volatile double, ptr@dptr
  %biased4 = fadd double %double4, %extra
  %int4 = bitcast double %biased4 to i64
  %double5 = load volatile double, ptr@dptr
  %biased5 = fadd double %double5, %extra
  %int5 = bitcast double %biased5 to i64
  %double6 = load volatile double, ptr@dptr
  %biased6 = fadd double %double6, %extra
  %int6 = bitcast double %biased6 to i64
  %double7 = load volatile double, ptr@dptr
  %biased7 = fadd double %double7, %extra
  %int7 = bitcast double %biased7 to i64
  %double8 = load volatile double, ptr@dptr
  %biased8 = fadd double %double8, %extra
  %int8 = bitcast double %biased8 to i64
  %double9 = load volatile double, ptr@dptr
  %biased9 = fadd double %double9, %extra
  %int9 = bitcast double %biased9 to i64
  br label %loop

loop:
  %start = call i64 @foo()
  %or0 = or i64 %start, %int0
  %or1 = or i64 %or0, %int1
  %or2 = or i64 %or1, %int2
  %or3 = or i64 %or2, %int3
  %or4 = or i64 %or3, %int4
  %or5 = or i64 %or4, %int5
  %or6 = or i64 %or5, %int6
  %or7 = or i64 %or6, %int7
  %or8 = or i64 %or7, %int8
  %or9 = or i64 %or8, %int9
  store i64 %or9, ptr@iptr
  %cont = icmp ne i64 %start, 1
  br i1 %cont, label %loop, label %exit

exit:
  ret void
}

; ...likewise LDGR, with the requirements the other way around.
define void @f11(i64 %mask) {
; CHECK-LABEL: f11:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    stmg %r12, %r15, 96(%r15)
; CHECK-NEXT:    .cfi_offset %r12, -64
; CHECK-NEXT:    .cfi_offset %r13, -56
; CHECK-NEXT:    .cfi_offset %r14, -48
; CHECK-NEXT:    .cfi_offset %r15, -40
; CHECK-NEXT:    aghi %r15, -240
; CHECK-NEXT:    .cfi_def_cfa_offset 400
; CHECK-NEXT:    std %f8, 232(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f9, 224(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f10, 216(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f11, 208(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f12, 200(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f13, 192(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f14, 184(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f15, 176(%r15) # 8-byte Spill
; CHECK-NEXT:    .cfi_offset %f8, -168
; CHECK-NEXT:    .cfi_offset %f9, -176
; CHECK-NEXT:    .cfi_offset %f10, -184
; CHECK-NEXT:    .cfi_offset %f11, -192
; CHECK-NEXT:    .cfi_offset %f12, -200
; CHECK-NEXT:    .cfi_offset %f13, -208
; CHECK-NEXT:    .cfi_offset %f14, -216
; CHECK-NEXT:    .cfi_offset %f15, -224
; CHECK-NEXT:    lgrl %r1, iptr@GOT
; CHECK-NEXT:    lgr %r0, %r2
; CHECK-NEXT:    ng %r0, 0(%r1)
; CHECK-NEXT:    lgr %r3, %r2
; CHECK-NEXT:    ng %r3, 0(%r1)
; CHECK-NEXT:    lgr %r4, %r2
; CHECK-NEXT:    ng %r4, 0(%r1)
; CHECK-NEXT:    stg %r0, 168(%r15) # 8-byte Spill
; CHECK-NEXT:    stg %r3, 160(%r15) # 8-byte Spill
; CHECK-NEXT:    lgr %r0, %r2
; CHECK-NEXT:    ng %r0, 0(%r1)
; CHECK-NEXT:    ldgr %f10, %r4
; CHECK-NEXT:    lgr %r3, %r2
; CHECK-NEXT:    ng %r3, 0(%r1)
; CHECK-NEXT:    lgr %r4, %r2
; CHECK-NEXT:    ng %r4, 0(%r1)
; CHECK-NEXT:    ldgr %f11, %r0
; CHECK-NEXT:    lgr %r0, %r2
; CHECK-NEXT:    ng %r0, 0(%r1)
; CHECK-NEXT:    ldgr %f12, %r3
; CHECK-NEXT:    ldgr %f13, %r4
; CHECK-NEXT:    lgr %r3, %r2
; CHECK-NEXT:    ng %r3, 0(%r1)
; CHECK-NEXT:    ldgr %f14, %r0
; CHECK-NEXT:    lgr %r0, %r2
; CHECK-NEXT:    ng %r0, 0(%r1)
; CHECK-NEXT:    ng %r2, 0(%r1)
; CHECK-NEXT:    ldgr %f15, %r3
; CHECK-NEXT:    lgrl %r13, dptr@GOT
; CHECK-NEXT:    ldgr %f8, %r0
; CHECK-NEXT:    ldgr %f9, %r2
; CHECK-NEXT:    larl %r12, .LCPI10_0
; CHECK-NEXT:  .LBB10_1: # %loop
; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    brasl %r14, bar@PLT
; CHECK-NEXT:    ldr %f1, %f0
; CHECK-NEXT:    adb %f1, 168(%r15) # 8-byte Folded Reload
; CHECK-NEXT:    adb %f1, 160(%r15) # 8-byte Folded Reload
; CHECK-NEXT:    adbr %f1, %f10
; CHECK-NEXT:    adbr %f1, %f11
; CHECK-NEXT:    adbr %f1, %f12
; CHECK-NEXT:    adbr %f1, %f13
; CHECK-NEXT:    adbr %f1, %f14
; CHECK-NEXT:    adbr %f1, %f15
; CHECK-NEXT:    adbr %f1, %f8
; CHECK-NEXT:    adbr %f1, %f9
; CHECK-NEXT:    cdb %f0, 0(%r12)
; CHECK-NEXT:    std %f1, 0(%r13)
; CHECK-NEXT:    jlh .LBB10_1
; CHECK-NEXT:  # %bb.2: # %exit
; CHECK-NEXT:    ld %f8, 232(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f9, 224(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f10, 216(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f11, 208(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f12, 200(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f13, 192(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f14, 184(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f15, 176(%r15) # 8-byte Reload
; CHECK-NEXT:    lmg %r12, %r15, 336(%r15)
; CHECK-NEXT:    br %r14
entry:
  %int0 = load volatile i64, ptr@iptr
  %masked0 = and i64 %int0, %mask
  %double0 = bitcast i64 %masked0 to double
  %int1 = load volatile i64, ptr@iptr
  %masked1 = and i64 %int1, %mask
  %double1 = bitcast i64 %masked1 to double
  %int2 = load volatile i64, ptr@iptr
  %masked2 = and i64 %int2, %mask
  %double2 = bitcast i64 %masked2 to double
  %int3 = load volatile i64, ptr@iptr
  %masked3 = and i64 %int3, %mask
  %double3 = bitcast i64 %masked3 to double
  %int4 = load volatile i64, ptr@iptr
  %masked4 = and i64 %int4, %mask
  %double4 = bitcast i64 %masked4 to double
  %int5 = load volatile i64, ptr@iptr
  %masked5 = and i64 %int5, %mask
  %double5 = bitcast i64 %masked5 to double
  %int6 = load volatile i64, ptr@iptr
  %masked6 = and i64 %int6, %mask
  %double6 = bitcast i64 %masked6 to double
  %int7 = load volatile i64, ptr@iptr
  %masked7 = and i64 %int7, %mask
  %double7 = bitcast i64 %masked7 to double
  %int8 = load volatile i64, ptr@iptr
  %masked8 = and i64 %int8, %mask
  %double8 = bitcast i64 %masked8 to double
  %int9 = load volatile i64, ptr@iptr
  %masked9 = and i64 %int9, %mask
  %double9 = bitcast i64 %masked9 to double
  br label %loop

loop:
  %start = call double @bar()
  %add0 = fadd double %start, %double0
  %add1 = fadd double %add0, %double1
  %add2 = fadd double %add1, %double2
  %add3 = fadd double %add2, %double3
  %add4 = fadd double %add3, %double4
  %add5 = fadd double %add4, %double5
  %add6 = fadd double %add5, %double6
  %add7 = fadd double %add6, %double7
  %add8 = fadd double %add7, %double8
  %add9 = fadd double %add8, %double9
  store double %add9, ptr@dptr
  %cont = fcmp one double %start, 1.0
  br i1 %cont, label %loop, label %exit

exit:
  ret void
}

; Test cases where the source of an LDGR needs to be spilled.
; We shouldn't have any integer stack stores or floating-point loads.
define void @f12() {
; CHECK-LABEL: f12:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    stmg %r6, %r15, 48(%r15)
; CHECK-NEXT:    .cfi_offset %r6, -112
; CHECK-NEXT:    .cfi_offset %r7, -104
; CHECK-NEXT:    .cfi_offset %r8, -96
; CHECK-NEXT:    .cfi_offset %r9, -88
; CHECK-NEXT:    .cfi_offset %r10, -80
; CHECK-NEXT:    .cfi_offset %r11, -72
; CHECK-NEXT:    .cfi_offset %r12, -64
; CHECK-NEXT:    .cfi_offset %r13, -56
; CHECK-NEXT:    .cfi_offset %r14, -48
; CHECK-NEXT:    .cfi_offset %r15, -40
; CHECK-NEXT:    aghi %r15, -176
; CHECK-NEXT:    .cfi_def_cfa_offset 336
; CHECK-NEXT:    lghi %r12, 0
; CHECK-NEXT:    lghi %r13, 0
; CHECK-NEXT:    lghi %r6, 0
; CHECK-NEXT:    lghi %r7, 0
; CHECK-NEXT:    lghi %r8, 0
; CHECK-NEXT:    lghi %r9, 0
; CHECK-NEXT:    lghi %r10, 0
; CHECK-NEXT:    lghi %r11, 0
; CHECK-NEXT:    mvghi 160(%r15), 0 # 8-byte Folded Spill
; CHECK-NEXT:    mvghi 168(%r15), 0 # 8-byte Folded Spill
; CHECK-NEXT:  .LBB11_1: # %loop
; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    brasl %r14, foo@PLT
; CHECK-NEXT:    agr %r12, %r2
; CHECK-NEXT:    agr %r13, %r2
; CHECK-NEXT:    agr %r6, %r2
; CHECK-NEXT:    agr %r7, %r2
; CHECK-NEXT:    agr %r8, %r2
; CHECK-NEXT:    agr %r9, %r2
; CHECK-NEXT:    agr %r10, %r2
; CHECK-NEXT:    agr %r11, %r2
; CHECK-NEXT:    lg %r0, 160(%r15) # 8-byte Reload
; CHECK-NEXT:    agr %r0, %r2
; CHECK-NEXT:    stg %r0, 160(%r15) # 8-byte Spill
; CHECK-NEXT:    lg %r0, 168(%r15) # 8-byte Reload
; CHECK-NEXT:    agr %r0, %r2
; CHECK-NEXT:    stg %r0, 168(%r15) # 8-byte Spill
; CHECK-NEXT:    cgijlh %r2, 1, .LBB11_1
; CHECK-NEXT:  # %bb.2: # %exit
; CHECK-NEXT:    brasl %r14, foo@PLT
; CHECK-NEXT:    lgrl %r1, dptr@GOT
; CHECK-NEXT:    ld %f0, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r12
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r13
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r6
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r7
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r8
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r9
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r10
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ldgr %f1, %r11
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ld %f1, 160(%r15) # 8-byte Reload
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    ld %f1, 168(%r15) # 8-byte Reload
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 0(%r1)
; CHECK-NEXT:    brasl %r14, foo@PLT
; CHECK-NEXT:    lmg %r6, %r15, 224(%r15)
; CHECK-NEXT:    br %r14
entry:
  br label %loop

loop:
  %int0 = phi i64 [ 0, %entry ], [ %add0, %loop ]
  %int1 = phi i64 [ 0, %entry ], [ %add1, %loop ]
  %int2 = phi i64 [ 0, %entry ], [ %add2, %loop ]
  %int3 = phi i64 [ 0, %entry ], [ %add3, %loop ]
  %int4 = phi i64 [ 0, %entry ], [ %add4, %loop ]
  %int5 = phi i64 [ 0, %entry ], [ %add5, %loop ]
  %int6 = phi i64 [ 0, %entry ], [ %add6, %loop ]
  %int7 = phi i64 [ 0, %entry ], [ %add7, %loop ]
  %int8 = phi i64 [ 0, %entry ], [ %add8, %loop ]
  %int9 = phi i64 [ 0, %entry ], [ %add9, %loop ]

  %bias = call i64 @foo()
  %add0 = add i64 %int0, %bias
  %add1 = add i64 %int1, %bias
  %add2 = add i64 %int2, %bias
  %add3 = add i64 %int3, %bias
  %add4 = add i64 %int4, %bias
  %add5 = add i64 %int5, %bias
  %add6 = add i64 %int6, %bias
  %add7 = add i64 %int7, %bias
  %add8 = add i64 %int8, %bias
  %add9 = add i64 %int9, %bias
  %cont = icmp ne i64 %bias, 1
  br i1 %cont, label %loop, label %exit

exit:
  %unused1 = call i64 @foo()
  %factor = load volatile double, ptr@dptr

  %conv0 = bitcast i64 %add0 to double
  %mul0 = fmul double %conv0, %factor
  store volatile double %mul0, ptr@dptr
  %conv1 = bitcast i64 %add1 to double
  %mul1 = fmul double %conv1, %factor
  store volatile double %mul1, ptr@dptr
  %conv2 = bitcast i64 %add2 to double
  %mul2 = fmul double %conv2, %factor
  store volatile double %mul2, ptr@dptr
  %conv3 = bitcast i64 %add3 to double
  %mul3 = fmul double %conv3, %factor
  store volatile double %mul3, ptr@dptr
  %conv4 = bitcast i64 %add4 to double
  %mul4 = fmul double %conv4, %factor
  store volatile double %mul4, ptr@dptr
  %conv5 = bitcast i64 %add5 to double
  %mul5 = fmul double %conv5, %factor
  store volatile double %mul5, ptr@dptr
  %conv6 = bitcast i64 %add6 to double
  %mul6 = fmul double %conv6, %factor
  store volatile double %mul6, ptr@dptr
  %conv7 = bitcast i64 %add7 to double
  %mul7 = fmul double %conv7, %factor
  store volatile double %mul7, ptr@dptr
  %conv8 = bitcast i64 %add8 to double
  %mul8 = fmul double %conv8, %factor
  store volatile double %mul8, ptr@dptr
  %conv9 = bitcast i64 %add9 to double
  %mul9 = fmul double %conv9, %factor
  store volatile double %mul9, ptr@dptr

  %unused2 = call i64 @foo()

  ret void
}

; ...likewise LGDR, with the requirements the other way around.
define void @f13() {
; CHECK-LABEL: f13:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    stmg %r13, %r15, 104(%r15)
; CHECK-NEXT:    .cfi_offset %r13, -56
; CHECK-NEXT:    .cfi_offset %r14, -48
; CHECK-NEXT:    .cfi_offset %r15, -40
; CHECK-NEXT:    aghi %r15, -240
; CHECK-NEXT:    .cfi_def_cfa_offset 400
; CHECK-NEXT:    std %f8, 232(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f9, 224(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f10, 216(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f11, 208(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f12, 200(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f13, 192(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f14, 184(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f15, 176(%r15) # 8-byte Spill
; CHECK-NEXT:    .cfi_offset %f8, -168
; CHECK-NEXT:    .cfi_offset %f9, -176
; CHECK-NEXT:    .cfi_offset %f10, -184
; CHECK-NEXT:    .cfi_offset %f11, -192
; CHECK-NEXT:    .cfi_offset %f12, -200
; CHECK-NEXT:    .cfi_offset %f13, -208
; CHECK-NEXT:    .cfi_offset %f14, -216
; CHECK-NEXT:    .cfi_offset %f15, -224
; CHECK-NEXT:    larl %r13, .LCPI12_0
; CHECK-NEXT:    ld %f8, 0(%r13)
; CHECK-NEXT:    ldr %f9, %f8
; CHECK-NEXT:    ldr %f15, %f8
; CHECK-NEXT:    ldr %f14, %f8
; CHECK-NEXT:    ldr %f13, %f8
; CHECK-NEXT:    ldr %f12, %f8
; CHECK-NEXT:    ldr %f11, %f8
; CHECK-NEXT:    ldr %f10, %f8
; CHECK-NEXT:    std %f8, 160(%r15) # 8-byte Spill
; CHECK-NEXT:    std %f8, 168(%r15) # 8-byte Spill
; CHECK-NEXT:  .LBB12_1: # %loop
; CHECK-NEXT:    # =>This Inner Loop Header: Depth=1
; CHECK-NEXT:    brasl %r14, bar@PLT
; CHECK-NEXT:    mdbr %f8, %f0
; CHECK-NEXT:    mdbr %f9, %f0
; CHECK-NEXT:    mdbr %f15, %f0
; CHECK-NEXT:    mdbr %f14, %f0
; CHECK-NEXT:    mdbr %f13, %f0
; CHECK-NEXT:    mdbr %f12, %f0
; CHECK-NEXT:    cdb %f0, 0(%r13)
; CHECK-NEXT:    mdbr %f11, %f0
; CHECK-NEXT:    mdbr %f10, %f0
; CHECK-NEXT:    ld %f1, 160(%r15) # 8-byte Reload
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 160(%r15) # 8-byte Spill
; CHECK-NEXT:    ld %f1, 168(%r15) # 8-byte Reload
; CHECK-NEXT:    mdbr %f1, %f0
; CHECK-NEXT:    std %f1, 168(%r15) # 8-byte Spill
; CHECK-NEXT:    jlh .LBB12_1
; CHECK-NEXT:  # %bb.2: # %exit
; CHECK-NEXT:    brasl %r14, foo@PLT
; CHECK-NEXT:    lgrl %r1, iptr@GOT
; CHECK-NEXT:    lg %r0, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f8
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f9
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f15
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f14
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f13
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f12
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f11
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lgdr %r2, %f10
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lg %r2, 160(%r15) # 8-byte Reload
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    lg %r2, 168(%r15) # 8-byte Reload
; CHECK-NEXT:    agr %r2, %r0
; CHECK-NEXT:    stg %r2, 0(%r1)
; CHECK-NEXT:    brasl %r14, foo@PLT
; CHECK-NEXT:    ld %f8, 232(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f9, 224(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f10, 216(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f11, 208(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f12, 200(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f13, 192(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f14, 184(%r15) # 8-byte Reload
; CHECK-NEXT:    ld %f15, 176(%r15) # 8-byte Reload
; CHECK-NEXT:    lmg %r13, %r15, 344(%r15)
; CHECK-NEXT:    br %r14
entry:
  br label %loop

loop:
  %double0 = phi double [ 1.0, %entry ], [ %mul0, %loop ]
  %double1 = phi double [ 1.0, %entry ], [ %mul1, %loop ]
  %double2 = phi double [ 1.0, %entry ], [ %mul2, %loop ]
  %double3 = phi double [ 1.0, %entry ], [ %mul3, %loop ]
  %double4 = phi double [ 1.0, %entry ], [ %mul4, %loop ]
  %double5 = phi double [ 1.0, %entry ], [ %mul5, %loop ]
  %double6 = phi double [ 1.0, %entry ], [ %mul6, %loop ]
  %double7 = phi double [ 1.0, %entry ], [ %mul7, %loop ]
  %double8 = phi double [ 1.0, %entry ], [ %mul8, %loop ]
  %double9 = phi double [ 1.0, %entry ], [ %mul9, %loop ]

  %factor = call double @bar()
  %mul0 = fmul double %double0, %factor
  %mul1 = fmul double %double1, %factor
  %mul2 = fmul double %double2, %factor
  %mul3 = fmul double %double3, %factor
  %mul4 = fmul double %double4, %factor
  %mul5 = fmul double %double5, %factor
  %mul6 = fmul double %double6, %factor
  %mul7 = fmul double %double7, %factor
  %mul8 = fmul double %double8, %factor
  %mul9 = fmul double %double9, %factor
  %cont = fcmp one double %factor, 1.0
  br i1 %cont, label %loop, label %exit

exit:
  %unused1 = call i64 @foo()
  %bias = load volatile i64, ptr@iptr

  %conv0 = bitcast double %mul0 to i64
  %add0 = add i64 %conv0, %bias
  store volatile i64 %add0, ptr@iptr
  %conv1 = bitcast double %mul1 to i64
  %add1 = add i64 %conv1, %bias
  store volatile i64 %add1, ptr@iptr
  %conv2 = bitcast double %mul2 to i64
  %add2 = add i64 %conv2, %bias
  store volatile i64 %add2, ptr@iptr
  %conv3 = bitcast double %mul3 to i64
  %add3 = add i64 %conv3, %bias
  store volatile i64 %add3, ptr@iptr
  %conv4 = bitcast double %mul4 to i64
  %add4 = add i64 %conv4, %bias
  store volatile i64 %add4, ptr@iptr
  %conv5 = bitcast double %mul5 to i64
  %add5 = add i64 %conv5, %bias
  store volatile i64 %add5, ptr@iptr
  %conv6 = bitcast double %mul6 to i64
  %add6 = add i64 %conv6, %bias
  store volatile i64 %add6, ptr@iptr
  %conv7 = bitcast double %mul7 to i64
  %add7 = add i64 %conv7, %bias
  store volatile i64 %add7, ptr@iptr
  %conv8 = bitcast double %mul8 to i64
  %add8 = add i64 %conv8, %bias
  store volatile i64 %add8, ptr@iptr
  %conv9 = bitcast double %mul9 to i64
  %add9 = add i64 %conv9, %bias
  store volatile i64 %add9, ptr@iptr

  %unused2 = call i64 @foo()

  ret void
}