[ARM] Add MVE addressing to isLegalT2AddressImmediate

[dmgreen]

Summary

Now with MVE being added, we can add the vector addressing modes for it. These are generally imm7 multiplied by the size of the type being loaded/stored.

[samparker]

llvm/lib/Target/ARM/ARMISelLowering.cpp

13305 ↗

(On Diff #203394)

I think it would be cleaner to exit early for floats without support, like we do currently at the start of the function. Then we can have a single switch statement here.

[dmgreen]

llvm/lib/Target/ARM/ARMISelLowering.cpp

13305 ↗

(On Diff #203394)

OK. yeah. I see what you mean. That sounds like a good idea.

[dmgreen]

Cleanup and added mve (nofp) to the tests.

Worth noting that I'm pretty sure we do have sensible vector load instructions for floats without mve.fp (they are the same as integer loads). We don't generate them at the moment, and this more accurately models the mess that we do create (i.e. it's expensive).

[samparker]

LGTM.

llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp

if ((V & (Scale - 1)) != 0)
     return false;
   return isUInt<5>(V / Scale);
 }
 
 static bool isLegalT2AddressImmediate(int64_t V, EVT VT,
                                       const ARMSubtarget *Subtarget) {
   if (!VT.isInteger() && !VT.isFloatingPoint())
     return false;
-  if (Subtarget->hasNEON() && VT.isVector())
+  if (VT.isVector() && Subtarget->hasNEON())
+    return false;
+  if (VT.isVector() && VT.isFloatingPoint() && Subtarget->hasMVEIntegerOps() &&
+      !Subtarget->hasMVEFloatOps())
     return false;
 
   bool IsNeg = false;
   if (V < 0) {
     IsNeg = true;
     V = -V;
   }
 
   unsigned NumBytes = std::max(VT.getSizeInBits() / 8, 1U);
 
+  // MVE: size * imm7
+  if (VT.isVector() && Subtarget->hasMVEIntegerOps()) {
+    switch (VT.getSimpleVT().getVectorElementType().SimpleTy) {
+    case MVT::i32:
+    case MVT::f32:
+      return isShiftedUInt<7,2>(V);
+    case MVT::i16:
+    case MVT::f16:
+      return isShiftedUInt<7,1>(V);
+    case MVT::i8:
+      return isUInt<7>(V);
+    default:
+      return false;
+    }
+  }
+
   // half VLDR: 2 * imm8
   if (VT.isFloatingPoint() && NumBytes == 2 && Subtarget->hasFPRegs16())
     return isShiftedUInt<8, 1>(V);
   // VLDR and LDRD: 4 * imm8
   if ((VT.isFloatingPoint() && Subtarget->hasVFP2Base()) || NumBytes == 8)
     return isShiftedUInt<8, 2>(V);
 
   if (NumBytes == 1 || NumBytes == 2 || NumBytes == 4) {

llvm/trunk/test/Analysis/CostModel/ARM/gep.ll

 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 0
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 0
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 0
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 0
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 0
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 0
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 1
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 1
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 1
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 1
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 1
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 1
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 1
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 1
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 1
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 4
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 4
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 4
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 4
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 4
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 4
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 4
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 4
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 4
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 31
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 31
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 31
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 31
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 31
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 32
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 32
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 32
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 32
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 32
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 -31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 -31
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 -31
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 -31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 -31
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -31
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 -31
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 -31
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 -32
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 -32
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 -32
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 -32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 -32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 -32
-; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -32
+; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 -32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 -32
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 %i
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %d0 = getelementptr inbounds i8, i8* undef, i32 -1
 ; CHECK-MVE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
 ; CHECK-MVEFP-LABEL: 'testvecs'
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %a13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 0
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 1
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 1
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 1
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 1
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 1
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 1
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 1
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %b13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 1
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %b13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 1
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 4
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 4
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 4
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 4
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 4
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 4
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 4
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %o13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 4
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %o13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 4
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 31
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 31
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 31
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %p13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 31
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 32
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %p13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 31
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 32
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 32
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %q12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %q13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 -31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 -31
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 -31
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 -31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 -31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -31
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 -31
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %r13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 -31
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 -32
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 -32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %r13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 -31
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 -32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 -32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 -32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 -32
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -32
-; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %s12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 -32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 -32
+; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 -32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %s13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 -32
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c7 = getelementptr inbounds <4 x i8>, <4 x i8>* undef, i32 %i
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c8 = getelementptr inbounds <4 x i16>, <4 x i16>* undef, i32 %i
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c9 = getelementptr inbounds <4 x i32>, <4 x i32>* undef, i32 %i
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c10 = getelementptr inbounds <4 x i64>, <4 x i64>* undef, i32 %i
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c11 = getelementptr inbounds <4 x half>, <4 x half>* undef, i32 %i
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c12 = getelementptr inbounds <4 x float>, <4 x float>* undef, i32 %i
 ; CHECK-MVEFP-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %c13 = getelementptr inbounds <4 x double>, <4 x double>* undef, i32 %i