[SelectionDAG] Handle more opcodes in canCreateUndefOrPoison

Deal with these possibly poison generating operations in
SelectionDAG::canCreateUndefOrPoison:
  EXTRACT_VECTOR_ELT, SRL, SRA

Also handle these operations that only propagate poison/undef based
on the input operands:
  SELECT, SADDSAT, UADDSAT, SSUBSAT, USUBSAT, MULHU, MULHS, SMIN,
  SMAX, UMIN, UMAX

Also handle the integer comparison variant of SELECT_CC.

The goal here is to allow pushing freeze through these operations
when allowed, as well as letting analyses such as
isGuaranteedNotToBeUndefOrPoison to not break on such operations.

Author: bjope

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -15383,6 +15383,12 @@ SDValue DAGCombiner::visitFREEZE(SDNode *N) {
   if (DAG.isGuaranteedNotToBeUndefOrPoison(N0, /*PoisonOnly*/ false))
     return N0;
 
+  // There is a reverse transform in visitEXTRACT_VECTOR_ELT, so we need to
+  // avoid inifite looping by not pulling freeze through EXTRACT_VECTOR_ELT
+  // here.
+  if (N0.getOpcode() == ISD::EXTRACT_VECTOR_ELT)
+    return SDValue();
+
   // Fold freeze(op(x, ...)) -> op(freeze(x), ...).
   // Try to push freeze through instructions that propagate but don't produce
   // poison as far as possible. If an operand of freeze follows three

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

@@ -5073,6 +5073,16 @@ bool SelectionDAG::canCreateUndefOrPoison(SDValue Op, const APInt &DemandedElts,
   case ISD::FREEZE:
   case ISD::CONCAT_VECTORS:
   case ISD::INSERT_SUBVECTOR:
+  case ISD::SADDSAT:
+  case ISD::UADDSAT:
+  case ISD::SSUBSAT:
+  case ISD::USUBSAT:
+  case ISD::MULHU:
+  case ISD::MULHS:
+  case ISD::SMIN:
+  case ISD::SMAX:
+  case ISD::UMIN:
+  case ISD::UMAX:
   case ISD::AND:
   case ISD::XOR:
   case ISD::ROTL:
@@ -5091,8 +5101,18 @@ bool SelectionDAG::canCreateUndefOrPoison(SDValue Op, const APInt &DemandedElts,
   case ISD::BITCAST:
   case ISD::BUILD_VECTOR:
   case ISD::BUILD_PAIR:
+  case ISD::SELECT:
     return false;
 
+  case ISD::SELECT_CC: {
+    // Integer select_cc cannot create undef or poison.
+    if (Op.getOperand(0).getValueType().isInteger())
+      return false;
+
+    // FIXME: Handle FP compares.
+    return true;
+  }
+
   case ISD::SETCC: {
     // Integer setcc cannot create undef or poison.
     if (Op.getOperand(0).getValueType().isInteger())
@@ -5123,23 +5143,36 @@ bool SelectionDAG::canCreateUndefOrPoison(SDValue Op, const APInt &DemandedElts,
     return ConsiderFlags && (Op->getFlags().hasNoSignedWrap() ||
                              Op->getFlags().hasNoUnsignedWrap());
 
-  case ISD::SHL:
+  case ISD::SHL: {
     // If the max shift amount isn't in range, then the shift can create poison.
     if (!getValidMaximumShiftAmountConstant(Op, DemandedElts))
       return true;
 
     // Matches hasPoisonGeneratingFlags().
     return ConsiderFlags && (Op->getFlags().hasNoSignedWrap() ||
                              Op->getFlags().hasNoUnsignedWrap());
+  }
+
+  case ISD::SRL:
+  case ISD::SRA: {
+    // If the max shift amount isn't in range, then the shift can create poison.
+    if (!getValidMaximumShiftAmountConstant(Op, DemandedElts))
+      return true;
+
+    // Matches hasPoisonGeneratingFlags().
+    return ConsiderFlags && Op->getFlags().hasExact();
+  }
 
   // Matches hasPoisonGeneratingFlags().
   case ISD::OR:
     return ConsiderFlags && Op->getFlags().hasDisjoint();
 
+  case ISD::EXTRACT_VECTOR_ELT:
   case ISD::INSERT_VECTOR_ELT:{
     // Ensure that the element index is in bounds.
     EVT VecVT = Op.getOperand(0).getValueType();
-    KnownBits KnownIdx = computeKnownBits(Op.getOperand(2), Depth + 1);
+    unsigned IdxOp = Opcode == ISD::INSERT_VECTOR_ELT ? 2 : 1;
+    KnownBits KnownIdx = computeKnownBits(Op.getOperand(IdxOp), Depth + 1);
     return KnownIdx.getMaxValue().uge(VecVT.getVectorMinNumElements());
   }
 

llvm/test/CodeGen/AMDGPU/div_i128.ll

@@ -283,21 +283,21 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    ; kill: def $vgpr15 killed $vgpr15 def $vgpr15_vgpr16 killed $exec
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v16, v1
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v9, v15
-; GFX9-O0-NEXT:    v_mov_b32_e32 v1, v16
+; GFX9-O0-NEXT:    v_mov_b32_e32 v5, v16
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v10, v13
-; GFX9-O0-NEXT:    v_mov_b32_e32 v5, v14
+; GFX9-O0-NEXT:    v_mov_b32_e32 v1, v14
 ; GFX9-O0-NEXT:    v_sub_co_u32_e32 v9, vcc, v9, v4
-; GFX9-O0-NEXT:    v_subb_co_u32_e32 v1, vcc, v1, v6, vcc
-; GFX9-O0-NEXT:    v_subb_co_u32_e32 v13, vcc, v10, v4, vcc
 ; GFX9-O0-NEXT:    v_subb_co_u32_e32 v5, vcc, v5, v6, vcc
+; GFX9-O0-NEXT:    v_subb_co_u32_e32 v13, vcc, v10, v4, vcc
+; GFX9-O0-NEXT:    v_subb_co_u32_e32 v1, vcc, v1, v6, vcc
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
-; GFX9-O0-NEXT:    ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14 killed $exec
-; GFX9-O0-NEXT:    v_mov_b32_e32 v14, v5
+; GFX9-O0-NEXT:    ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
+; GFX9-O0-NEXT:    v_mov_b32_e32 v10, v5
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
-; GFX9-O0-NEXT:    ; kill: def $vgpr9 killed $vgpr9 def $vgpr9_vgpr10 killed $exec
-; GFX9-O0-NEXT:    v_mov_b32_e32 v10, v1
+; GFX9-O0-NEXT:    ; kill: def $vgpr13 killed $vgpr13 def $vgpr13_vgpr14 killed $exec
+; GFX9-O0-NEXT:    v_mov_b32_e32 v14, v1
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v1, v3
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v5, v12
 ; GFX9-O0-NEXT:    v_xor_b32_e64 v1, v5, v1
@@ -313,21 +313,21 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    ; kill: def $vgpr7 killed $vgpr7 def $vgpr7_vgpr8 killed $exec
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v8, v1
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v1, v7
-; GFX9-O0-NEXT:    v_mov_b32_e32 v7, v8
-; GFX9-O0-NEXT:    v_mov_b32_e32 v8, v11
+; GFX9-O0-NEXT:    ; kill: def $vgpr8 killed $vgpr8 killed $vgpr7_vgpr8 killed $exec
+; GFX9-O0-NEXT:    v_mov_b32_e32 v7, v11
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v2, v12
 ; GFX9-O0-NEXT:    v_sub_co_u32_e32 v1, vcc, v1, v3
-; GFX9-O0-NEXT:    v_subb_co_u32_e32 v7, vcc, v7, v5, vcc
-; GFX9-O0-NEXT:    v_subb_co_u32_e32 v11, vcc, v8, v3, vcc
-; GFX9-O0-NEXT:    v_subb_co_u32_e32 v2, vcc, v2, v5, vcc
+; GFX9-O0-NEXT:    v_subb_co_u32_e32 v8, vcc, v8, v5, vcc
+; GFX9-O0-NEXT:    v_subb_co_u32_e32 v11, vcc, v7, v3, vcc
+; GFX9-O0-NEXT:    v_subb_co_u32_e32 v7, vcc, v2, v5, vcc
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
-; GFX9-O0-NEXT:    ; kill: def $vgpr11 killed $vgpr11 def $vgpr11_vgpr12 killed $exec
-; GFX9-O0-NEXT:    v_mov_b32_e32 v12, v2
+; GFX9-O0-NEXT:    ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
+; GFX9-O0-NEXT:    v_mov_b32_e32 v2, v8
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
 ; GFX9-O0-NEXT:    ; implicit-def: $sgpr4
-; GFX9-O0-NEXT:    ; kill: def $vgpr1 killed $vgpr1 def $vgpr1_vgpr2 killed $exec
-; GFX9-O0-NEXT:    v_mov_b32_e32 v2, v7
+; GFX9-O0-NEXT:    ; kill: def $vgpr11 killed $vgpr11 def $vgpr11_vgpr12 killed $exec
+; GFX9-O0-NEXT:    v_mov_b32_e32 v12, v7
 ; GFX9-O0-NEXT:    v_xor_b32_e64 v5, v5, v6
 ; GFX9-O0-NEXT:    v_xor_b32_e64 v3, v3, v4
 ; GFX9-O0-NEXT:    ; kill: def $vgpr3 killed $vgpr3 def $vgpr3_vgpr4 killed $exec
@@ -340,18 +340,26 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(0)
 ; GFX9-O0-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
-; GFX9-O0-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    v_mov_b32_e32 v3, v11
+; GFX9-O0-NEXT:    v_mov_b32_e32 v4, v12
+; GFX9-O0-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
-; GFX9-O0-NEXT:    buffer_store_dword v11, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    v_mov_b32_e32 v4, v2
+; GFX9-O0-NEXT:    v_mov_b32_e32 v3, v1
+; GFX9-O0-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT:    buffer_store_dword v12, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
-; GFX9-O0-NEXT:    buffer_store_dword v9, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    v_mov_b32_e32 v3, v13
+; GFX9-O0-NEXT:    v_mov_b32_e32 v4, v14
+; GFX9-O0-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT:    buffer_store_dword v10, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
-; GFX9-O0-NEXT:    buffer_store_dword v13, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    v_mov_b32_e32 v3, v9
+; GFX9-O0-NEXT:    v_mov_b32_e32 v4, v10
+; GFX9-O0-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(0)
-; GFX9-O0-NEXT:    buffer_store_dword v14, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GFX9-O0-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v7, v12
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v8, v2
 ; GFX9-O0-NEXT:    v_or_b32_e64 v3, v8, v7
@@ -404,7 +412,8 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    ; kill: def $vgpr8 killed $vgpr8 def $vgpr8_vgpr9 killed $exec
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v9, v5
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v5, v9
-; GFX9-O0-NEXT:    v_cmp_ne_u64_e64 s[12:13], v[11:12], s[6:7]
+; GFX9-O0-NEXT:    s_mov_b64 s[12:13], s[6:7]
+; GFX9-O0-NEXT:    v_cmp_ne_u64_e64 s[12:13], v[11:12], s[12:13]
 ; GFX9-O0-NEXT:    v_cndmask_b32_e64 v5, v5, v10, s[12:13]
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v7, v6
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v6, v8
@@ -440,7 +449,8 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    ; kill: def $vgpr11 killed $vgpr11 def $vgpr11_vgpr12 killed $exec
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v12, v5
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v5, v12
-; GFX9-O0-NEXT:    v_cmp_ne_u64_e64 s[8:9], v[13:14], s[6:7]
+; GFX9-O0-NEXT:    s_mov_b64 s[8:9], s[6:7]
+; GFX9-O0-NEXT:    v_cmp_ne_u64_e64 s[8:9], v[13:14], s[8:9]
 ; GFX9-O0-NEXT:    v_cndmask_b32_e64 v5, v5, v8, s[8:9]
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v7, v6
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v6, v11
@@ -691,10 +701,10 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    buffer_load_dword v28, off, s[0:3], s32 offset:260 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v25, off, s[0:3], s32 offset:264 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v26, off, s[0:3], s32 offset:268 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v19, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v20, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v21, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v22, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v19, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v20, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v21, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v22, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v14, off, s[0:3], s32 offset:272 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v15, off, s[0:3], s32 offset:276 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v11, off, s[0:3], s32 offset:280 ; 4-byte Folded Reload
@@ -904,14 +914,14 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    s_or_saveexec_b64 s[18:19], -1
 ; GFX9-O0-NEXT:    buffer_load_dword v16, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    s_mov_b64 exec, s[18:19]
-; GFX9-O0-NEXT:    buffer_load_dword v17, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v18, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v13, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v14, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v19, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v20, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v21, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v22, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v17, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v18, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v13, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v14, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v19, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v20, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v21, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v22, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(9)
 ; GFX9-O0-NEXT:    v_mov_b32_e32 v4, v10
 ; GFX9-O0-NEXT:    s_waitcnt vmcnt(0)
@@ -1029,10 +1039,10 @@ define i128 @v_sdiv_i128_vv(i128 %lhs, i128 %rhs) {
 ; GFX9-O0-NEXT:    s_or_saveexec_b64 s[18:19], -1
 ; GFX9-O0-NEXT:    buffer_load_dword v0, off, s[0:3], s32 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    s_mov_b64 exec, s[18:19]
-; GFX9-O0-NEXT:    buffer_load_dword v7, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v8, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v11, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
-; GFX9-O0-NEXT:    buffer_load_dword v12, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v7, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v8, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v11, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GFX9-O0-NEXT:    buffer_load_dword v12, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v5, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v6, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
 ; GFX9-O0-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload