PowerPC: Add emergency stack spill slots for SPE

[jhibbits]

Summary

The powerpcspe 64-bit load/store only allows a 8-bit delta (32 64-bit
words), so if the stack size is any larger than that, we need extra
spill slots for doing indexing.

[jhibbits]

Ping?

[sfertile]

llvm/lib/Target/PowerPC/PPCFrameLowering.cpp

2071

Should probably extend this sentence to include (or 8-bit immediate for powerpcspe).

llvm/test/CodeGen/PowerPC/spe-spills.ll

8

Is it possible to handwrite a simple test to expose the issue? If its too difficult we can stick with this test but I think something simpler which showcases the problem would be ideal.

[jhibbits]

llvm/lib/Target/PowerPC/PPCFrameLowering.cpp

2071

Good point, will do.

llvm/test/CodeGen/PowerPC/spe-spills.ll

8

I'll see if I can reduce it to something simple. This is actually the post-creduce result that exhibited the issue, any more reduction and it doesn't seem to trip it.

[jhibbits]

llvm/test/CodeGen/PowerPC/spe-spills.ll

8

This is really odd, I am unable to reproduce the failure now, even rolling back to origin/release/10.x. The change needed is obviously correct, but being unable to test it is a problem.

[jhibbits]

Funny thing, I'm able to reproduce it *only* while running natively on a powerpcspe based device, not on any other device I've tested. I'll try to reduce the testcase even further, since it really is unwieldy.

[jhibbits]

Simplify the test a little.

The problem with trying to write a simplified test by hand is it requires all
registers to be in use simultaneously, and all variables to be in scope
together, to exert register pressure. The best I can do is this test that's
slightly simplified (saves over 200 lines over the previous test) vs the
previous test. I hope it's sufficient.

[jhibbits]

As mentioned before, I can only seem to reproduce this running natively on powerpcspe; running the test on powerpc64, even without this patch, succeeds. This is extremely bizarre, but this fix gets the native build to produce the exact same text as running the test on powerpc64.

llvm/test/CodeGen/PowerPC/spe-spills.ll

424–460

One thing to note: these flags are required to reproduce the problem, along with the !tbaa annotations in the code. Without them, even the otherwise failing llc would succeed.

[jhibbits]

Adding a couple explicit reviewers.

[jhibbits]

Superseded by D151711 .

llvm/lib/Target/PowerPC/PPCFrameLowering.cpp

PPCFrameLowering::addScavengingSpillSlot(MachineFunction &MF,
   // a large stack, which will require scavenging a register to materialize a
   // large offset.
 
   // We need to have a scavenger spill slot for spills if the frame size is
   // large. In case there is no free register for large-offset addressing,
   // this slot is used for the necessary emergency spill. Also, we need the
   // slot for dynamic stack allocations.
 
   // The scavenger might be invoked if the frame offset does not fit into
-  // the 16-bit immediate. We don't know the complete frame size here
-  // because we've not yet computed callee-saved register spills or the
-  // needed alignment padding.
+  // the 16-bit immediate (or 8-bit in the case of SPE). We don't know the
+  // complete frame size here because we've not yet computed callee-saved
+  // register spills or the needed alignment padding.
   unsigned StackSize = determineFrameLayout(MF, true);
   MachineFrameInfo &MFI = MF.getFrameInfo();
   if (MFI.hasVarSizedObjects() || spillsCR(MF) || spillsVRSAVE(MF) ||
-      hasNonRISpills(MF) || (hasSpills(MF) && !isInt<16>(StackSize))) {
+      hasNonRISpills(MF) || (hasSpills(MF) && !isInt<16>(StackSize)) ||
+      (Subtarget.hasSPE() && hasSpills(MF) && !isInt<8>(StackSize))) {
     const TargetRegisterClass &GPRC = PPC::GPRCRegClass;
     const TargetRegisterClass &G8RC = PPC::G8RCRegClass;
     const TargetRegisterClass &RC = Subtarget.isPPC64() ? G8RC : GPRC;
     const TargetRegisterInfo &TRI = *Subtarget.getRegisterInfo();
     unsigned Size = TRI.getSpillSize(RC);
     unsigned Align = TRI.getSpillAlignment(RC);
     RS->addScavengingFrameIndex(MFI.CreateStackObject(Size, Align, false));
 

llvm/test/CodeGen/PowerPC/spe-spills.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -verify-machineinstrs < %s -mtriple=powerpcspe-unknown-linux-gnu \
+; RUN:     -mattr=+spe | FileCheck %s
+; Test that two or more spill slots are allocated for stacks larger than 256
+; bytes on powerpcspe targets
+%struct.gammapweights = type { %struct.anon, %struct.anon.0, %struct.anon.1, %struct.anon.2, %struct.anon.3, %struct.anon.4 }
+%struct.anon = type { %struct.g, double, double }
+%struct.g = type { double, double, double }
+%struct.anon.0 = type { double, double, double }
+%struct.anon.1 = type { double, double, double, double, double, double, double, double, double }
+%struct.anon.2 = type { double, double, double }
+%struct.anon.3 = type { double, double }
+%struct.anon.4 = type { double }
+
+; Function Attrs: nofree norecurse nounwind uwtable
+define dso_local i32 @r(%struct.gammapweights* %dst, %struct.gammapweights* nocapture readonly %s, double %t, %struct.gammapweights* readonly %u, double %v) local_unnamed_addr #0 {
+; CHECK-LABEL: r:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    stwu 1, -144(1)
+; CHECK-NEXT:    .cfi_def_cfa_offset 144
+; CHECK-NEXT:    .cfi_offset r22, -40
+; CHECK-NEXT:    .cfi_offset r23, -36
+; CHECK-NEXT:    .cfi_offset r24, -32
+; CHECK-NEXT:    .cfi_offset r25, -28
+; CHECK-NEXT:    .cfi_offset r26, -24
+; CHECK-NEXT:    .cfi_offset r27, -20
+; CHECK-NEXT:    .cfi_offset r28, -16
+; CHECK-NEXT:    .cfi_offset r29, -12
+; CHECK-NEXT:    .cfi_offset r30, -8
+; CHECK-NEXT:    .cfi_offset r22, -128
+; CHECK-NEXT:    .cfi_offset r23, -120
+; CHECK-NEXT:    .cfi_offset r24, -112
+; CHECK-NEXT:    .cfi_offset r25, -104
+; CHECK-NEXT:    .cfi_offset r26, -96
+; CHECK-NEXT:    .cfi_offset r27, -88
+; CHECK-NEXT:    .cfi_offset r28, -80
+; CHECK-NEXT:    .cfi_offset r29, -72
+; CHECK-NEXT:    .cfi_offset r30, -64
+; CHECK-NEXT:    li 0, 16
+; CHECK-NEXT:    stw 22, 104(1) # 4-byte Folded Spill
+; CHECK-NEXT:    li 11, 8
+; CHECK-NEXT:    stw 23, 108(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 24, 112(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 25, 116(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 26, 120(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 27, 124(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 28, 128(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 29, 132(1) # 4-byte Folded Spill
+; CHECK-NEXT:    stw 30, 136(1) # 4-byte Folded Spill
+; CHECK-NEXT:    evstdd 22, 16(1) # 8-byte Folded Spill
+; CHECK-NEXT:    evstdd 23, 24(1) # 8-byte Folded Spill
+; CHECK-NEXT:    li 23, 48
+; CHECK-NEXT:    evstdd 24, 32(1) # 8-byte Folded Spill
+; CHECK-NEXT:    evstdd 25, 40(1) # 8-byte Folded Spill
+; CHECK-NEXT:    li 25, 40
+; CHECK-NEXT:    evstdd 26, 48(1) # 8-byte Folded Spill
+; CHECK-NEXT:    evstdd 27, 56(1) # 8-byte Folded Spill
+; CHECK-NEXT:    li 27, 32
+; CHECK-NEXT:    evstdd 28, 64(1) # 8-byte Folded Spill
+; CHECK-NEXT:    evstdd 29, 72(1) # 8-byte Folded Spill
+; CHECK-NEXT:    li 29, 24
+; CHECK-NEXT:    evstdd 30, 80(1) # 8-byte Folded Spill
+; CHECK-NEXT:    evmergelo 8, 9, 10
+; CHECK-NEXT:    evmergelo 5, 5, 6
+; CHECK-NEXT:    evlddx 30, 7, 0
+; CHECK-NEXT:    evlddx 10, 4, 11
+; CHECK-NEXT:    evlddx 12, 7, 11
+; CHECK-NEXT:    efdmul 30, 30, 8
+; CHECK-NEXT:    evlddx 28, 7, 29
+; CHECK-NEXT:    efdmul 10, 10, 5
+; CHECK-NEXT:    evlddx 26, 7, 27
+; CHECK-NEXT:    efdmul 10, 10, 12
+; CHECK-NEXT:    evlddx 22, 4, 23
+; CHECK-NEXT:    efdmul 12, 28, 8
+; CHECK-NEXT:    evstddx 30, 3, 0
+; CHECK-NEXT:    efdadd 12, 12, 5
+; CHECK-NEXT:    evlddx 0, 7, 23
+; CHECK-NEXT:    evlddx 24, 7, 25
+; CHECK-NEXT:    evstddx 12, 3, 29
+; CHECK-NEXT:    efdmul 29, 26, 8
+; CHECK-NEXT:    efdmul 26, 22, 5
+; CHECK-NEXT:    efdmul 0, 0, 8
+; CHECK-NEXT:    evstddx 10, 3, 11
+; CHECK-NEXT:    li 10, 64
+; CHECK-NEXT:    efdadd 29, 29, 5
+; CHECK-NEXT:    efdmul 30, 24, 8
+; CHECK-NEXT:    evlddx 11, 4, 10
+; CHECK-NEXT:    efdadd 0, 26, 0
+; CHECK-NEXT:    evlddx 12, 7, 10
+; CHECK-NEXT:    evstddx 29, 3, 27
+; CHECK-NEXT:    li 29, 72
+; CHECK-NEXT:    evstddx 30, 3, 25
+; CHECK-NEXT:    li 30, 56
+; CHECK-NEXT:    evlddx 28, 4, 29
+; CHECK-NEXT:    efdmul 11, 11, 5
+; CHECK-NEXT:    efdmul 12, 12, 8
+; CHECK-NEXT:    evlddx 27, 7, 29
+; CHECK-NEXT:    efdadd 11, 11, 12
+; CHECK-NEXT:    evstddx 0, 3, 23
+; CHECK-NEXT:    li 0, 80
+; CHECK-NEXT:    evlddx 25, 4, 30
+; CHECK-NEXT:    efdmul 28, 28, 5
+; CHECK-NEXT:    efdmul 27, 27, 8
+; CHECK-NEXT:    evlddx 24, 7, 30
+; CHECK-NEXT:    efdadd 28, 28, 27
+; CHECK-NEXT:    evlddx 26, 4, 0
+; CHECK-NEXT:    efdmul 25, 25, 5
+; CHECK-NEXT:    evlddx 23, 7, 0
+; CHECK-NEXT:    efdmul 24, 24, 8
+; CHECK-NEXT:    evstddx 11, 3, 10
+; CHECK-NEXT:    li 10, 96
+; CHECK-NEXT:    efdmul 26, 26, 5
+; CHECK-NEXT:    efdmul 23, 23, 8
+; CHECK-NEXT:    efdadd 25, 25, 24
+; CHECK-NEXT:    evlddx 11, 4, 10
+; CHECK-NEXT:    efdadd 26, 26, 23
+; CHECK-NEXT:    evlddx 12, 7, 10
+; CHECK-NEXT:    evstddx 28, 3, 29
+; CHECK-NEXT:    li 29, 104
+; CHECK-NEXT:    evstddx 25, 3, 30
+; CHECK-NEXT:    li 30, 88
+; CHECK-NEXT:    evlddx 28, 4, 29
+; CHECK-NEXT:    efdmul 11, 11, 5
+; CHECK-NEXT:    efdmul 12, 12, 8
+; CHECK-NEXT:    evlddx 27, 7, 29
+; CHECK-NEXT:    efdadd 11, 11, 12
+; CHECK-NEXT:    evstddx 26, 3, 0
+; CHECK-NEXT:    li 0, 112
+; CHECK-NEXT:    evlddx 25, 4, 30
+; CHECK-NEXT:    efdmul 28, 28, 5
+; CHECK-NEXT:    efdmul 27, 27, 8
+; CHECK-NEXT:    evlddx 24, 7, 30
+; CHECK-NEXT:    efdadd 28, 28, 27
+; CHECK-NEXT:    evlddx 26, 4, 0
+; CHECK-NEXT:    efdmul 25, 25, 5
+; CHECK-NEXT:    evlddx 23, 7, 0
+; CHECK-NEXT:    efdmul 24, 24, 8
+; CHECK-NEXT:    evstddx 11, 3, 10
+; CHECK-NEXT:    li 10, 128
+; CHECK-NEXT:    efdmul 26, 26, 5
+; CHECK-NEXT:    efdmul 23, 23, 8
+; CHECK-NEXT:    efdadd 25, 25, 24
+; CHECK-NEXT:    evlddx 11, 4, 10
+; CHECK-NEXT:    efdadd 26, 26, 23
+; CHECK-NEXT:    evlddx 12, 7, 10
+; CHECK-NEXT:    evstddx 28, 3, 29
+; CHECK-NEXT:    li 29, 136
+; CHECK-NEXT:    evstddx 25, 3, 30
+; CHECK-NEXT:    li 30, 120
+; CHECK-NEXT:    evlddx 28, 4, 29
+; CHECK-NEXT:    efdmul 11, 11, 5
+; CHECK-NEXT:    efdmul 12, 12, 8
+; CHECK-NEXT:    evlddx 27, 7, 29
+; CHECK-NEXT:    efdadd 11, 11, 12
+; CHECK-NEXT:    evstddx 26, 3, 0
+; CHECK-NEXT:    li 0, 144
+; CHECK-NEXT:    evlddx 25, 4, 30
+; CHECK-NEXT:    efdmul 28, 28, 5
+; CHECK-NEXT:    efdmul 27, 27, 8
+; CHECK-NEXT:    evlddx 24, 7, 30
+; CHECK-NEXT:    efdadd 28, 28, 27
+; CHECK-NEXT:    evlddx 26, 4, 0
+; CHECK-NEXT:    efdmul 25, 25, 5
+; CHECK-NEXT:    evlddx 23, 7, 0
+; CHECK-NEXT:    efdmul 24, 24, 8
+; CHECK-NEXT:    evstddx 11, 3, 10
+; CHECK-NEXT:    li 10, 160
+; CHECK-NEXT:    efdmul 27, 26, 5
+; CHECK-NEXT:    efdmul 26, 23, 8
+; CHECK-NEXT:    efdadd 25, 25, 24
+; CHECK-NEXT:    evlddx 11, 4, 10
+; CHECK-NEXT:    efdadd 27, 27, 26
+; CHECK-NEXT:    evlddx 12, 7, 10
+; CHECK-NEXT:    evstddx 25, 3, 30
+; CHECK-NEXT:    li 30, 152
+; CHECK-NEXT:    evstddx 28, 3, 29
+; CHECK-NEXT:    li 29, 168
+; CHECK-NEXT:    evstddx 27, 3, 0
+; CHECK-NEXT:    li 27, 176
+; CHECK-NEXT:    evldd 9, 0(4)
+; CHECK-NEXT:    efdmul 11, 11, 5
+; CHECK-NEXT:    efdmul 12, 12, 8
+; CHECK-NEXT:    evldd 6, 0(7)
+; CHECK-NEXT:    efdadd 11, 11, 12
+; CHECK-NEXT:    evlddx 25, 4, 30
+; CHECK-NEXT:    efdmul 9, 9, 5
+; CHECK-NEXT:    evlddx 24, 7, 30
+; CHECK-NEXT:    efdmul 6, 9, 6
+; CHECK-NEXT:    evlddx 28, 4, 29
+; CHECK-NEXT:    efdmul 26, 25, 5
+; CHECK-NEXT:    evlddx 0, 7, 29
+; CHECK-NEXT:    efdmul 25, 24, 8
+; CHECK-NEXT:    evlddx 4, 4, 27
+; CHECK-NEXT:    efdadd 26, 26, 25
+; CHECK-NEXT:    evlddx 7, 7, 27
+; CHECK-NEXT:    evstddx 11, 3, 10
+; CHECK-NEXT:    efdmul 10, 28, 5
+; CHECK-NEXT:    efdmul 4, 4, 5
+; CHECK-NEXT:    efdmul 5, 0, 8
+; CHECK-NEXT:    efdmul 7, 7, 8
+; CHECK-NEXT:    efdadd 5, 10, 5
+; CHECK-NEXT:    efdadd 4, 4, 7
+; CHECK-NEXT:    evstddx 26, 3, 30
+; CHECK-NEXT:    evstddx 5, 3, 29
+; CHECK-NEXT:    evstddx 4, 3, 27
+; CHECK-NEXT:    evldd 30, 80(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 29, 72(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 28, 64(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 27, 56(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 26, 48(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 25, 40(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 24, 32(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 23, 24(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evldd 22, 16(1) # 8-byte Folded Reload
+; CHECK-NEXT:    evstdd 6, 0(3)
+; CHECK-NEXT:    lwz 30, 136(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 29, 132(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 28, 128(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 27, 124(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 26, 120(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 25, 116(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 24, 112(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 23, 108(1) # 4-byte Folded Reload
+; CHECK-NEXT:    lwz 22, 104(1) # 4-byte Folded Reload
+; CHECK-NEXT:    addi 1, 1, 144
+; CHECK-NEXT:    blr
+entry:
+  %b = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 0, i32 0, i32 0
+  %0 = load double, double* %b, align 8, !tbaa !2
+  %mul = fmul double %0, %t
+  %b3 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 0, i32 0, i32 0
+  %1 = load double, double* %b3, align 8, !tbaa !2
+  %mul4 = fmul double %mul, %1
+  %b7 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 0, i32 0, i32 0
+  store double %mul4, double* %b7, align 8, !tbaa !2
+  %c10 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 0, i32 0, i32 1
+  %2 = load double, double* %c10, align 8, !tbaa !12
+  %mul11 = fmul double %2, %t
+  %c14 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 0, i32 0, i32 1
+  %3 = load double, double* %c14, align 8, !tbaa !12
+  %mul15 = fmul double %mul11, %3
+  %c18 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 0, i32 0, i32 1
+  store double %mul15, double* %c18, align 8, !tbaa !12
+  %e = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 0, i32 0, i32 2
+  %4 = load double, double* %e, align 8, !tbaa !13
+  %mul21 = fmul double %4, %v
+  %e24 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 0, i32 0, i32 2
+  store double %mul21, double* %e24, align 8, !tbaa !13
+  %i = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 0, i32 1
+  %5 = load double, double* %i, align 8, !tbaa !14
+  %mul26 = fmul double %5, %v
+  %add = fadd double %mul26, %t
+  %i28 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 0, i32 1
+  store double %add, double* %i28, align 8, !tbaa !14
+  %cx = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 0, i32 2
+  %6 = load double, double* %cx, align 8, !tbaa !15
+  %mul30 = fmul double %6, %v
+  %add31 = fadd double %mul30, %t
+  %cx33 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 0, i32 2
+  store double %add31, double* %cx33, align 8, !tbaa !15
+  %j = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 1, i32 0
+  %7 = load double, double* %j, align 8, !tbaa !16
+  %mul35 = fmul double %7, %v
+  %j37 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 1, i32 0
+  store double %mul35, double* %j37, align 8, !tbaa !16
+  %e39 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 1, i32 1
+  %8 = load double, double* %e39, align 8, !tbaa !17
+  %mul40 = fmul double %8, %t
+  %e42 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 1, i32 1
+  %9 = load double, double* %e42, align 8, !tbaa !17
+  %mul43 = fmul double %9, %v
+  %add44 = fadd double %mul40, %mul43
+  %e46 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 1, i32 1
+  store double %add44, double* %e46, align 8, !tbaa !17
+  %b48 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 1, i32 2
+  %10 = load double, double* %b48, align 8, !tbaa !18
+  %mul49 = fmul double %10, %t
+  %b51 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 1, i32 2
+  %11 = load double, double* %b51, align 8, !tbaa !18
+  %mul52 = fmul double %11, %v
+  %add53 = fadd double %mul49, %mul52
+  %b55 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 1, i32 2
+  store double %add53, double* %b55, align 8, !tbaa !18
+  %j56 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 0
+  %12 = load double, double* %j56, align 8, !tbaa !19
+  %mul57 = fmul double %12, %t
+  %j59 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 0
+  %13 = load double, double* %j59, align 8, !tbaa !19
+  %mul60 = fmul double %13, %v
+  %add61 = fadd double %mul57, %mul60
+  %j63 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 0
+  store double %add61, double* %j63, align 8, !tbaa !19
+  %e65 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 1
+  %14 = load double, double* %e65, align 8, !tbaa !20
+  %mul66 = fmul double %14, %t
+  %e68 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 1
+  %15 = load double, double* %e68, align 8, !tbaa !20
+  %mul69 = fmul double %15, %v
+  %add70 = fadd double %mul66, %mul69
+  %e72 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 1
+  store double %add70, double* %e72, align 8, !tbaa !20
+  %k = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 2
+  %16 = load double, double* %k, align 8, !tbaa !21
+  %mul74 = fmul double %16, %t
+  %k76 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 2
+  %17 = load double, double* %k76, align 8, !tbaa !21
+  %mul77 = fmul double %17, %v
+  %add78 = fadd double %mul74, %mul77
+  %k80 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 2
+  store double %add78, double* %k80, align 8, !tbaa !21
+  %gl = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 3
+  %18 = load double, double* %gl, align 8, !tbaa !22
+  %mul82 = fmul double %18, %t
+  %gl84 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 3
+  %19 = load double, double* %gl84, align 8, !tbaa !22
+  %mul85 = fmul double %19, %v
+  %add86 = fadd double %mul82, %mul85
+  %gl88 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 3
+  store double %add86, double* %gl88, align 8, !tbaa !22
+  %bl = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 4
+  %20 = load double, double* %bl, align 8, !tbaa !23
+  %mul90 = fmul double %20, %t
+  %bl92 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 4
+  %21 = load double, double* %bl92, align 8, !tbaa !23
+  %mul93 = fmul double %21, %v
+  %add94 = fadd double %mul90, %mul93
+  %bl96 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 4
+  store double %add94, double* %bl96, align 8, !tbaa !23
+  %l = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 5
+  %22 = load double, double* %l, align 8, !tbaa !24
+  %mul98 = fmul double %22, %t
+  %l100 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 5
+  %23 = load double, double* %l100, align 8, !tbaa !24
+  %mul101 = fmul double %23, %v
+  %add102 = fadd double %mul98, %mul101
+  %l104 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 5
+  store double %add102, double* %l104, align 8, !tbaa !24
+  %blpow = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 6
+  %24 = load double, double* %blpow, align 8, !tbaa !25
+  %mul106 = fmul double %24, %t
+  %blpow108 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 6
+  %25 = load double, double* %blpow108, align 8, !tbaa !25
+  %mul109 = fmul double %25, %v
+  %add110 = fadd double %mul106, %mul109
+  %blpow112 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 6
+  store double %add110, double* %blpow112, align 8, !tbaa !25
+  %lxpow = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 7
+  %26 = load double, double* %lxpow, align 8, !tbaa !26
+  %mul114 = fmul double %26, %t
+  %lxpow116 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 7
+  %27 = load double, double* %lxpow116, align 8, !tbaa !26
+  %mul117 = fmul double %27, %v
+  %add118 = fadd double %mul114, %mul117
+  %lxpow120 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 7
+  store double %add118, double* %lxpow120, align 8, !tbaa !26
+  %lxthr = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 2, i32 8
+  %28 = load double, double* %lxthr, align 8, !tbaa !27
+  %mul122 = fmul double %28, %t
+  %lxthr124 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 2, i32 8
+  %29 = load double, double* %lxthr124, align 8, !tbaa !27
+  %mul125 = fmul double %29, %v
+  %add126 = fadd double %mul122, %mul125
+  %lxthr128 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 2, i32 8
+  store double %add126, double* %lxthr128, align 8, !tbaa !27
+  %m = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 3, i32 0
+  %30 = load double, double* %m, align 8, !tbaa !28
+  %mul129 = fmul double %30, %t
+  %m131 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 3, i32 0
+  %31 = load double, double* %m131, align 8, !tbaa !28
+  %mul132 = fmul double %31, %v
+  %add133 = fadd double %mul129, %mul132
+  %m135 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 3, i32 0
+  store double %add133, double* %m135, align 8, !tbaa !28
+  %n = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 3, i32 1
+  %32 = load double, double* %n, align 8, !tbaa !29
+  %mul137 = fmul double %32, %t
+  %n139 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 3, i32 1
+  %33 = load double, double* %n139, align 8, !tbaa !29
+  %mul140 = fmul double %33, %v
+  %add141 = fadd double %mul137, %mul140
+  %n143 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 3, i32 1
+  store double %add141, double* %n143, align 8, !tbaa !29
+  %dsm = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 3, i32 2
+  %34 = load double, double* %dsm, align 8, !tbaa !30
+  %mul145 = fmul double %34, %t
+  %dsm147 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 3, i32 2
+  %35 = load double, double* %dsm147, align 8, !tbaa !30
+  %mul148 = fmul double %35, %v
+  %add149 = fadd double %mul145, %mul148
+  %dsm151 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 3, i32 2
+  store double %add149, double* %dsm151, align 8, !tbaa !30
+  %co = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 4, i32 0
+  %36 = load double, double* %co, align 8, !tbaa !31
+  %mul152 = fmul double %36, %t
+  %co154 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 4, i32 0
+  %37 = load double, double* %co154, align 8, !tbaa !31
+  %mul155 = fmul double %37, %v
+  %add156 = fadd double %mul152, %mul155
+  %co158 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 4, i32 0
+  store double %add156, double* %co158, align 8, !tbaa !31
+  %p = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 4, i32 1
+  %38 = load double, double* %p, align 8, !tbaa !32
+  %mul160 = fmul double %38, %t
+  %p162 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 4, i32 1
+  %39 = load double, double* %p162, align 8, !tbaa !32
+  %mul163 = fmul double %39, %v
+  %add164 = fadd double %mul160, %mul163
+  %p166 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 4, i32 1
+  store double %add164, double* %p166, align 8, !tbaa !32
+  %q = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %s, i32 0, i32 5, i32 0
+  %40 = load double, double* %q, align 8, !tbaa !33
+  %mul167 = fmul double %40, %t
+  %q169 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %u, i32 0, i32 5, i32 0
+  %41 = load double, double* %q169, align 8, !tbaa !33
+  %mul170 = fmul double %41, %v
+  %add171 = fadd double %mul167, %mul170
+  %q173 = getelementptr inbounds %struct.gammapweights, %struct.gammapweights* %dst, i32 0, i32 5, i32 0
+  store double %add171, double* %q173, align 8, !tbaa !33
+  ret i32 undef
+}
+
+attributes #0 = { nofree norecurse nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="none" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="ppc" "target-features"="+secure-plt,+spe,-altivec,-bpermd,-crypto,-direct-move,-extdiv,-htm,-power8-vector,-power9-vector,-qpx,-vsx" "unsafe-fp-math"="false" "use-soft-float"="false" }
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{!"FreeBSD clang version 10.0.1 ([e-mail]:llvm/llvm-project.git llvmorg-10.0.1-rc1-0-gf79cd71e145)"}
+!2 = !{!3, !6, i64 0}
+!3 = !{!"", !4, i64 0, !5, i64 40, !9, i64 64, !5, i64 136, !10, i64 160, !11, i64 176}
+!4 = !{!"", !5, i64 0, !6, i64 24, !6, i64 32}
+!5 = !{!"", !6, i64 0, !6, i64 8, !6, i64 16}
+!6 = !{!"double", !7, i64 0}
+!7 = !{!"omnipotent char", !8, i64 0}
+!8 = !{!"Simple C/C++ TBAA"}
+!9 = !{!"", !6, i64 0, !6, i64 8, !6, i64 16, !6, i64 24, !6, i64 32, !6, i64 40, !6, i64 48, !6, i64 56, !6, i64 64}
+!10 = !{!"", !6, i64 0, !6, i64 8}
+!11 = !{!"", !6, i64 0}
+!12 = !{!3, !6, i64 8}
+!13 = !{!3, !6, i64 16}
+!14 = !{!3, !6, i64 24}
+!15 = !{!3, !6, i64 32}
+!16 = !{!3, !6, i64 40}
+!17 = !{!3, !6, i64 48}
+!18 = !{!3, !6, i64 56}
+!19 = !{!3, !6, i64 64}
+!20 = !{!3, !6, i64 72}
+!21 = !{!3, !6, i64 80}
+!22 = !{!3, !6, i64 88}
+!23 = !{!3, !6, i64 96}
+!24 = !{!3, !6, i64 104}
+!25 = !{!3, !6, i64 112}
+!26 = !{!3, !6, i64 120}
+!27 = !{!3, !6, i64 128}
+!28 = !{!3, !6, i64 136}
+!29 = !{!3, !6, i64 144}
+!30 = !{!3, !6, i64 152}
+!31 = !{!3, !6, i64 160}
+!32 = !{!3, !6, i64 168}
+!33 = !{!3, !6, i64 176}