[SelectionDAG] change getConstant() to use the input SDLoc when building splat vectors

[spatel]

Summary

The change is simple enough (assuming it's what we want): instead of attaching an anonymous SDLoc to splatted vector constants, use the scalar constant's existing SDLoc since that is what is passed into getConstant() as a param. The changes in instruction scheduling seem harmless to me, but I'll explain why that happens.

The motivation for this patch starts near:
http://reviews.llvm.org/rL258833
...I made a note to myself that x86's getZeroVector() could be similarly cleaned up and thought it would be 'NFC' as well. But when I made that change locally, several x86 codegen tests wiggled.

It turns out that the lack of SDLoc consistency in getConstant() changes the way ScheduleDAGRRList behaves. This is because the SDLoc contains 'IROrder' and some DAG scheduler algorithms use IROrder for tie-breaking. I suppose this is working as intended, although it's a bit surprising to me that SDLoc's can have this effect on codegen.

[qcolombet]

Hi Sanjay,

The change makes sense to me.

Cheers,
-Quentin

llvm/trunk/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

else if (NewNodesMustHaveLegalTypes && VT.isVector() &&
     // perform this reversal because getConstant() is producing a vector
     // splat.
     // This situation occurs in MIPS MSA.
 
     SmallVector<SDValue, 8> Ops;
     for (unsigned i = 0; i < VT.getVectorNumElements(); ++i)
       Ops.insert(Ops.end(), EltParts.begin(), EltParts.end());
 
-    SDValue Result = getNode(ISD::BITCAST, SDLoc(), VT,
-                             getNode(ISD::BUILD_VECTOR, SDLoc(), ViaVecVT,
-                                     Ops));
+    SDValue Result = getNode(ISD::BITCAST, DL, VT,
+                             getNode(ISD::BUILD_VECTOR, DL, ViaVecVT, Ops));
     return Result;
   }
 
   assert(Elt->getBitWidth() == EltVT.getSizeInBits() &&
          "APInt size does not match type size!");
   unsigned Opc = isT ? ISD::TargetConstant : ISD::Constant;
   FoldingSetNodeID ID;
   AddNodeIDNode(ID, Opc, getVTList(EltVT), None);
if (!N) {
     CSEMap.InsertNode(N, IP);
     InsertNode(N);
   }
 
   SDValue Result(N, 0);
   if (VT.isVector()) {
     SmallVector<SDValue, 8> Ops;
     Ops.assign(VT.getVectorNumElements(), Result);
-    Result = getNode(ISD::BUILD_VECTOR, SDLoc(), VT, Ops);
+    Result = getNode(ISD::BUILD_VECTOR, DL, VT, Ops);
   }
   return Result;
 }
 
 SDValue SelectionDAG::getIntPtrConstant(uint64_t Val, SDLoc DL, bool isTarget) {
   return getConstant(Val, DL, TLI->getPointerTy(getDataLayout()), isTarget);
 }
 
if (!N) {
     CSEMap.InsertNode(N, IP);
     InsertNode(N);
   }
 
   SDValue Result(N, 0);
   if (VT.isVector()) {
     SmallVector<SDValue, 8> Ops;
     Ops.assign(VT.getVectorNumElements(), Result);
-    Result = getNode(ISD::BUILD_VECTOR, SDLoc(), VT, Ops);
+    Result = getNode(ISD::BUILD_VECTOR, DL, VT, Ops);
   }
   return Result;
 }
 
 SDValue SelectionDAG::getConstantFP(double Val, SDLoc DL, EVT VT,
                                     bool isTarget) {
   EVT EltVT = VT.getScalarType();
   if (EltVT == MVT::f32)

llvm/trunk/test/CodeGen/X86/sse41.ll

; X64-NEXT:    retq
   %vecinit2 = insertelement <4 x i32> %vecinit1, i32 0, i32 2
   %vecinit4 = shufflevector <4 x i32> %vecinit2, <4 x i32> %a, <4 x i32> <i32 0, i32 1, i32 2, i32 4>
   ret <4 x i32> %vecinit4
 }
 
 define <4 x i32> @i32_shuf_X00X(<4 x i32> %x, <4 x i32> %a) {
 ; X32-LABEL: i32_shuf_X00X:
 ; X32:       ## BB#0:
-; X32-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,2,0]
-; X32-NEXT:    pxor %xmm0, %xmm0
-; X32-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
+; X32-NEXT:    pxor %xmm1, %xmm1
+; X32-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,0]
+; X32-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: i32_shuf_X00X:
 ; X64:       ## BB#0:
-; X64-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,1,2,0]
-; X64-NEXT:    pxor %xmm0, %xmm0
-; X64-NEXT:    pblendw {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3,4,5],xmm1[6,7]
+; X64-NEXT:    pxor %xmm1, %xmm1
+; X64-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,0]
+; X64-NEXT:    pblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3,4,5],xmm0[6,7]
 ; X64-NEXT:    retq
   %vecext = extractelement <4 x i32> %x, i32 0
   %vecinit = insertelement <4 x i32> undef, i32 %vecext, i32 0
   %vecinit1 = insertelement <4 x i32> %vecinit, i32 0, i32 1
   %vecinit2 = insertelement <4 x i32> %vecinit1, i32 0, i32 2
   %vecinit4 = shufflevector <4 x i32> %vecinit2, <4 x i32> %x, <4 x i32> <i32 0, i32 1, i32 2, i32 4>
   ret <4 x i32> %vecinit4
 }

llvm/trunk/test/CodeGen/X86/vec_ctbits.ll

; CHECK-NEXT:    retq
   ret <2 x i32> %c
 
 }
 
 define <2 x i32> @prompop(<2 x i32> %a) nounwind {
 ; CHECK-LABEL: prompop:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm0
+; CHECK-NEXT:    pxor %xmm2, %xmm2
 ; CHECK-NEXT:    movdqa %xmm0, %xmm1
 ; CHECK-NEXT:    psrlq $1, %xmm1
 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
 ; CHECK-NEXT:    psubq %xmm1, %xmm0
 ; CHECK-NEXT:    movdqa {{.*#+}} xmm1 = [3689348814741910323,3689348814741910323]
-; CHECK-NEXT:    movdqa %xmm0, %xmm2
-; CHECK-NEXT:    pand %xmm1, %xmm2
+; CHECK-NEXT:    movdqa %xmm0, %xmm3
+; CHECK-NEXT:    pand %xmm1, %xmm3
 ; CHECK-NEXT:    psrlq $2, %xmm0
 ; CHECK-NEXT:    pand %xmm1, %xmm0
-; CHECK-NEXT:    paddq %xmm2, %xmm0
+; CHECK-NEXT:    paddq %xmm3, %xmm0
 ; CHECK-NEXT:    movdqa %xmm0, %xmm1
 ; CHECK-NEXT:    psrlq $4, %xmm1
 ; CHECK-NEXT:    paddq %xmm0, %xmm1
 ; CHECK-NEXT:    pand {{.*}}(%rip), %xmm1
-; CHECK-NEXT:    pxor %xmm0, %xmm0
-; CHECK-NEXT:    psadbw %xmm0, %xmm1
+; CHECK-NEXT:    psadbw %xmm2, %xmm1
 ; CHECK-NEXT:    movdqa %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %c = call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a)
   ret <2 x i32> %c
 }

llvm/trunk/test/CodeGen/X86/vector-shuffle-v1.ll

; VL_BW_DQ-NEXT:    retq
   %d = bitcast <8 x i1>%c to i8
   ret i8 %d
 }
 
 define i8 @shuf8i1__9_6_1_10_3_7_7_1(i8 %a) {
 ; AVX512F-LABEL: shuf8i1__9_6_1_10_3_7_7_1:
 ; AVX512F:       # BB#0:
 ; AVX512F-NEXT:    kmovw %edi, %k1
-; AVX512F-NEXT:    movb $51, %al
-; AVX512F-NEXT:    kmovw %eax, %k2
 ; AVX512F-NEXT:    movq {{.*}}(%rip), %rax
+; AVX512F-NEXT:    movb $51, %cl
+; AVX512F-NEXT:    kmovw %ecx, %k2
 ; AVX512F-NEXT:    vpbroadcastq %rax, %zmm0 {%k2} {z}
 ; AVX512F-NEXT:    vpbroadcastq %rax, %zmm1 {%k1} {z}
 ; AVX512F-NEXT:    vmovdqa64 {{.*#+}} zmm2 = [9,6,1,0,3,7,7,1]
 ; AVX512F-NEXT:    vpermt2q %zmm1, %zmm2, %zmm0
 ; AVX512F-NEXT:    vpsllq $63, %zmm0, %zmm0
 ; AVX512F-NEXT:    vptestmq %zmm0, %zmm0, %k0
 ; AVX512F-NEXT:    kmovw %k0, %eax
 ; AVX512F-NEXT:    retq