Automerge: [slp][profcheck] Mark selects as having unknown profile (#162960)

There are 2 cases:

- either the `select`​ condition is a vector of bools, case in which we don't currently have a way to represent the per-element branch probabilities anyway;
- or the select condition is a scalar, for example from a `llvm.vector.reduce`​. We could potentially try and do more here - if the reduced vector contained conditions from other selects, for instance

In either case, IIUC, chances are the `select`​ doesn't get lowered to a branch, at least I'm not seeing any evidence of that in an internal complex application (CSFDO + ThinLTO). Seems sufficient to mark the selects are unknown (for profiled functions); since that metadata carries with it the pass name (`DEBUG_TYPE`​) that marked it as such, we can revisit this if we detect later lowerings of these selects that would have required an actual profile.

Issue #147390

Author: mtrofin

llvm/include/llvm/IR/IRBuilder.h

@@ -2548,6 +2548,11 @@ class IRBuilderBase {
       std::optional<RoundingMode> Rounding = std::nullopt,
       std::optional<fp::ExceptionBehavior> Except = std::nullopt);
 
+  LLVM_ABI Value *CreateSelectWithUnknownProfile(Value *C, Value *True,
+                                                 Value *False,
+                                                 StringRef PassName,
+                                                 const Twine &Name = "");
+
   LLVM_ABI Value *CreateSelect(Value *C, Value *True, Value *False,
                                const Twine &Name = "",
                                Instruction *MDFrom = nullptr);

llvm/lib/IR/IRBuilder.cpp

@@ -25,6 +25,7 @@
 #include "llvm/IR/Module.h"
 #include "llvm/IR/NoFolder.h"
 #include "llvm/IR/Operator.h"
+#include "llvm/IR/ProfDataUtils.h"
 #include "llvm/IR/Statepoint.h"
 #include "llvm/IR/Type.h"
 #include "llvm/IR/Value.h"
@@ -1002,6 +1003,18 @@ CallInst *IRBuilderBase::CreateConstrainedFPCall(
   return C;
 }
 
+Value *IRBuilderBase::CreateSelectWithUnknownProfile(Value *C, Value *True,
+                                                     Value *False,
+                                                     StringRef PassName,
+                                                     const Twine &Name) {
+  Value *Ret = CreateSelectFMF(C, True, False, {}, Name);
+  if (auto *SI = dyn_cast<SelectInst>(Ret)) {
+    setExplicitlyUnknownBranchWeightsIfProfiled(
+        *SI, *SI->getParent()->getParent(), PassName);
+  }
+  return Ret;
+}
+
 Value *IRBuilderBase::CreateSelect(Value *C, Value *True, Value *False,
                                    const Twine &Name, Instruction *MDFrom) {
   return CreateSelectFMF(C, True, False, {}, Name, MDFrom);

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -19460,7 +19460,8 @@ Value *BoUpSLP::vectorizeTree(TreeEntry *E) {
       }
       assert(getNumElements(Cond->getType()) == TrueNumElements &&
              "Cannot vectorize Instruction::Select");
-      Value *V = Builder.CreateSelect(Cond, True, False);
+      Value *V =
+          Builder.CreateSelectWithUnknownProfile(Cond, True, False, DEBUG_TYPE);
       V = FinalShuffle(V, E);
 
       E->VectorizedValue = V;
@@ -23580,18 +23581,19 @@ class HorizontalReduction {
     switch (Kind) {
     case RecurKind::Or: {
       if (UseSelect && OpTy == CmpInst::makeCmpResultType(OpTy))
-        return Builder.CreateSelect(
+        return Builder.CreateSelectWithUnknownProfile(
             LHS, ConstantInt::getAllOnesValue(CmpInst::makeCmpResultType(OpTy)),
-            RHS, Name);
+            RHS, DEBUG_TYPE, Name);
       unsigned RdxOpcode = RecurrenceDescriptor::getOpcode(Kind);
       return Builder.CreateBinOp((Instruction::BinaryOps)RdxOpcode, LHS, RHS,
                                  Name);
     }
     case RecurKind::And: {
       if (UseSelect && OpTy == CmpInst::makeCmpResultType(OpTy))
-        return Builder.CreateSelect(
+        return Builder.CreateSelectWithUnknownProfile(
             LHS, RHS,
-            ConstantInt::getNullValue(CmpInst::makeCmpResultType(OpTy)), Name);
+            ConstantInt::getNullValue(CmpInst::makeCmpResultType(OpTy)),
+            DEBUG_TYPE, Name);
       unsigned RdxOpcode = RecurrenceDescriptor::getOpcode(Kind);
       return Builder.CreateBinOp((Instruction::BinaryOps)RdxOpcode, LHS, RHS,
                                  Name);
@@ -23612,7 +23614,8 @@ class HorizontalReduction {
       if (UseSelect) {
         CmpInst::Predicate Pred = llvm::getMinMaxReductionPredicate(Kind);
         Value *Cmp = Builder.CreateICmp(Pred, LHS, RHS, Name);
-        return Builder.CreateSelect(Cmp, LHS, RHS, Name);
+        return Builder.CreateSelectWithUnknownProfile(Cmp, LHS, RHS, DEBUG_TYPE,
+                                                      Name);
       }
       [[fallthrough]];
     case RecurKind::FMax:

llvm/utils/profcheck-xfail.txt

@@ -1312,82 +1312,6 @@ Transforms/SimpleLoopUnswitch/pr60736.ll
 Transforms/SimpleLoopUnswitch/trivial-unswitch-freeze-individual-conditions.ll
 Transforms/SimpleLoopUnswitch/trivial-unswitch.ll
 Transforms/SimpleLoopUnswitch/trivial-unswitch-logical-and-or.ll
-Transforms/SLPVectorizer/AArch64/gather-root.ll
-Transforms/SLPVectorizer/AArch64/horizontal.ll
-Transforms/SLPVectorizer/AArch64/loadi8.ll
-Transforms/SLPVectorizer/AArch64/phi-node-bitwidt-op-not.ll
-Transforms/SLPVectorizer/AArch64/uselistorder.ll
-Transforms/SLPVectorizer/AArch64/vec3-reorder-reshuffle.ll
-Transforms/SLPVectorizer/AArch64/vectorizable-selects-min-max.ll
-Transforms/SLPVectorizer/AArch64/vectorizable-selects-uniform-cmps.ll
-Transforms/SLPVectorizer/AMDGPU/horizontal-store.ll
-Transforms/SLPVectorizer/bool-logical-op-reduction-with-poison.ll
-Transforms/SLPVectorizer/call-arg-reduced-by-minbitwidth.ll
-Transforms/SLPVectorizer/const-bool-logical-or-reduction.ll
-Transforms/SLPVectorizer/extracts-with-undefs.ll
-Transforms/SLPVectorizer/freeze-signedness-missed.ll
-Transforms/SLPVectorizer/gathered-consecutive-loads-different-types.ll
-Transforms/SLPVectorizer/gather_extract_from_vectorbuild.ll
-Transforms/SLPVectorizer/insert-element-build-vector-const.ll
-Transforms/SLPVectorizer/insert-element-build-vector-inseltpoison.ll
-Transforms/SLPVectorizer/insert-element-build-vector.ll
-Transforms/SLPVectorizer/logical-ops-poisonous-repeated.ll
-Transforms/SLPVectorizer/minbitwidth-node-with-multi-users.ll
-Transforms/SLPVectorizer/minbitwidth-user-not-min.ll
-Transforms/SLPVectorizer/partial-register-extract.ll
-Transforms/SLPVectorizer/reduction-gather-non-scheduled-extracts.ll
-Transforms/SLPVectorizer/reorder-node.ll
-Transforms/SLPVectorizer/reused-buildvector-matching-vectorized-node.ll
-Transforms/SLPVectorizer/revec.ll
-Transforms/SLPVectorizer/RISCV/remarks_cmp_sel_min_max.ll
-Transforms/SLPVectorizer/RISCV/remarks-insert-into-small-vector.ll
-Transforms/SLPVectorizer/RISCV/reordered-interleaved-loads.ll
-Transforms/SLPVectorizer/RISCV/revec.ll
-Transforms/SLPVectorizer/RISCV/select-profitability.ll
-Transforms/SLPVectorizer/RISCV/shuffled-gather-casted.ll
-Transforms/SLPVectorizer/RISCV/unsigned-node-trunc-with-signed-users.ll
-Transforms/SLPVectorizer/slp-deleted-inst.ll
-Transforms/SLPVectorizer/SystemZ/cmp-ptr-minmax.ll
-Transforms/SLPVectorizer/SystemZ/ext-not-resized-op-resized.ll
-Transforms/SLPVectorizer/SystemZ/minbitwidth-trunc.ll
-Transforms/SLPVectorizer/X86/bool-mask.ll
-Transforms/SLPVectorizer/X86/bv-root-part-of-graph.ll
-Transforms/SLPVectorizer/X86/cmp-after-intrinsic-call-minbitwidth.ll
-Transforms/SLPVectorizer/X86/cmp-as-alternate-ops.ll
-Transforms/SLPVectorizer/X86/cmp_sel.ll
-Transforms/SLPVectorizer/X86/crash_7zip.ll
-Transforms/SLPVectorizer/X86/crash_clear_undefs.ll
-Transforms/SLPVectorizer/X86/crash_cmpop.ll
-Transforms/SLPVectorizer/X86/debug-counter.ll
-Transforms/SLPVectorizer/X86/debug-info-salvage.ll
-Transforms/SLPVectorizer/X86/extractelement-single-use-many-nodes.ll
-Transforms/SLPVectorizer/X86/extracts-non-extendable.ll
-Transforms/SLPVectorizer/X86/ext-used-scalar-different-bitwidth.ll
-Transforms/SLPVectorizer/X86/gather-node-same-as-vect-but-order.ll
-Transforms/SLPVectorizer/X86/horizontal-minmax.ll
-Transforms/SLPVectorizer/X86/insert-after-bundle.ll
-Transforms/SLPVectorizer/X86/jumbled-load-multiuse.ll
-Transforms/SLPVectorizer/X86/minbitwidth-icmp-to-trunc.ll
-Transforms/SLPVectorizer/X86/minbw-user-non-sizable.ll
-Transforms/SLPVectorizer/X86/non-load-reduced-as-part-of-bv.ll
-Transforms/SLPVectorizer/X86/ordering-bug.ll
-Transforms/SLPVectorizer/X86/phi-node-bitwidt-op-not.ll
-Transforms/SLPVectorizer/X86/phi-node-reshuffled-part.ll
-Transforms/SLPVectorizer/X86/pr46983.ll
-Transforms/SLPVectorizer/X86/pr49933.ll
-Transforms/SLPVectorizer/X86/propagate_ir_flags.ll
-Transforms/SLPVectorizer/X86/reduction-bool-logic-op-inside.ll
-Transforms/SLPVectorizer/X86/reduction-logical.ll
-Transforms/SLPVectorizer/X86/resized-bv-values-non-power-of2-node.ll
-Transforms/SLPVectorizer/X86/reused-reductions-with-minbitwidth.ll
-Transforms/SLPVectorizer/X86/select-reduction-op.ll
-Transforms/SLPVectorizer/X86/shrink_after_reorder.ll
-Transforms/SLPVectorizer/X86/subvector-minbitwidth-unsigned-value.ll
-Transforms/SLPVectorizer/X86/undef_vect.ll
-Transforms/SLPVectorizer/X86/used-reduced-op.ll
-Transforms/SLPVectorizer/X86/vec3-reorder-reshuffle.ll
-Transforms/SLPVectorizer/X86/vectorize-widest-phis.ll
-Transforms/SLPVectorizer/X86/whole-registers-compare.ll
 Transforms/SROA/addrspacecast.ll
 Transforms/SROA/phi-and-select.ll
 Transforms/SROA/phi-gep.ll