[llvm] effacc1 - [AArch64] Constant fold sve_convert_from_svbool(zero) to zero

[Joe Ellis via llvm-commits]

Author: Joe Ellis
Date: 2021-04-20T10:02:49Z
New Revision: effacc15999dde16de3e991f861769bc899aa910

URL: https://github.com/llvm/llvm-project/commit/effacc15999dde16de3e991f861769bc899aa910
DIFF: https://github.com/llvm/llvm-project/commit/effacc15999dde16de3e991f861769bc899aa910.diff

LOG: [AArch64] Constant fold sve_convert_from_svbool(zero) to zero

Co-authored-by: Paul Walker <paul.walker at arm.com>

Differential Revision: https://reviews.llvm.org/D100463

Added: 
    llvm/test/Transforms/InstSimplify/ConstProp/AArch64/aarch64-sve-convert-from-svbool.ll
    llvm/test/Transforms/InstSimplify/ConstProp/AArch64/lit.local.cfg

Modified: 
    llvm/lib/Analysis/ConstantFolding.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Analysis/ConstantFolding.cpp b/llvm/lib/Analysis/ConstantFolding.cpp
index d326af554e5a4..cd65e63695677 100644
--- a/llvm/lib/Analysis/ConstantFolding.cpp
+++ b/llvm/lib/Analysis/ConstantFolding.cpp
@@ -41,6 +41,7 @@
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsAArch64.h"
 #include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/IR/IntrinsicsARM.h"
 #include "llvm/IR/IntrinsicsWebAssembly.h"
@@ -1486,6 +1487,7 @@ bool llvm::canConstantFoldCallTo(const CallBase *Call, const Function *F) {
   case Intrinsic::arm_mve_vctp16:
   case Intrinsic::arm_mve_vctp32:
   case Intrinsic::arm_mve_vctp64:
+  case Intrinsic::aarch64_sve_convert_from_svbool:
   // WebAssembly float semantics are always known
   case Intrinsic::wasm_trunc_signed:
   case Intrinsic::wasm_trunc_unsigned:
@@ -2847,20 +2849,10 @@ static Constant *ConstantFoldScalarCall(StringRef Name,
   return nullptr;
 }
 
-static Constant *ConstantFoldVectorCall(StringRef Name,
-                                        Intrinsic::ID IntrinsicID,
-                                        VectorType *VTy,
-                                        ArrayRef<Constant *> Operands,
-                                        const DataLayout &DL,
-                                        const TargetLibraryInfo *TLI,
-                                        const CallBase *Call) {
-  // Do not iterate on scalable vector. The number of elements is unknown at
-  // compile-time.
-  if (isa<ScalableVectorType>(VTy))
-    return nullptr;
-
-  auto *FVTy = cast<FixedVectorType>(VTy);
-
+static Constant *ConstantFoldFixedVectorCall(
+    StringRef Name, Intrinsic::ID IntrinsicID, FixedVectorType *FVTy,
+    ArrayRef<Constant *> Operands, const DataLayout &DL,
+    const TargetLibraryInfo *TLI, const CallBase *Call) {
   SmallVector<Constant *, 4> Result(FVTy->getNumElements());
   SmallVector<Constant *, 4> Lane(Operands.size());
   Type *Ty = FVTy->getElementType();
@@ -2977,6 +2969,24 @@ static Constant *ConstantFoldVectorCall(StringRef Name,
   return ConstantVector::get(Result);
 }
 
+static Constant *ConstantFoldScalableVectorCall(
+    StringRef Name, Intrinsic::ID IntrinsicID, ScalableVectorType *SVTy,
+    ArrayRef<Constant *> Operands, const DataLayout &DL,
+    const TargetLibraryInfo *TLI, const CallBase *Call) {
+  switch (IntrinsicID) {
+  case Intrinsic::aarch64_sve_convert_from_svbool: {
+    auto *Src = dyn_cast<Constant>(Operands[0]);
+    if (!Src || !Src->isNullValue())
+      break;
+
+    return ConstantInt::getFalse(SVTy);
+  }
+  default:
+    break;
+  }
+  return nullptr;
+}
+
 } // end anonymous namespace
 
 Constant *llvm::ConstantFoldCall(const CallBase *Call, Function *F,
@@ -2990,9 +3000,15 @@ Constant *llvm::ConstantFoldCall(const CallBase *Call, Function *F,
 
   Type *Ty = F->getReturnType();
 
-  if (auto *VTy = dyn_cast<VectorType>(Ty))
-    return ConstantFoldVectorCall(Name, F->getIntrinsicID(), VTy, Operands,
-                                  F->getParent()->getDataLayout(), TLI, Call);
+  if (auto *FVTy = dyn_cast<FixedVectorType>(Ty))
+    return ConstantFoldFixedVectorCall(
+        Name, F->getIntrinsicID(), FVTy, Operands,
+        F->getParent()->getDataLayout(), TLI, Call);
+
+  if (auto *SVTy = dyn_cast<ScalableVectorType>(Ty))
+    return ConstantFoldScalableVectorCall(
+        Name, F->getIntrinsicID(), SVTy, Operands,
+        F->getParent()->getDataLayout(), TLI, Call);
 
   return ConstantFoldScalarCall(Name, F->getIntrinsicID(), Ty, Operands, TLI,
                                 Call);

diff  --git a/llvm/test/Transforms/InstSimplify/ConstProp/AArch64/aarch64-sve-convert-from-svbool.ll b/llvm/test/Transforms/InstSimplify/ConstProp/AArch64/aarch64-sve-convert-from-svbool.ll
new file mode 100644
index 0000000000000..c5115e46175ae
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/AArch64/aarch64-sve-convert-from-svbool.ll
@@ -0,0 +1,10 @@
+; RUN: opt -instsimplify -S -o - < %s | FileCheck %s
+
+define <vscale x 2 x i1> @reinterpret_zero() {
+; CHECK-LABEL:  @reinterpret_zero(
+; CHECK: ret <vscale x 2 x i1> zeroinitializer
+  %pg = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> zeroinitializer)
+  ret <vscale x 2 x i1> %pg
+}
+
+declare <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1>)

diff  --git a/llvm/test/Transforms/InstSimplify/ConstProp/AArch64/lit.local.cfg b/llvm/test/Transforms/InstSimplify/ConstProp/AArch64/lit.local.cfg
new file mode 100644
index 0000000000000..7184443994b69
--- /dev/null
+++ b/llvm/test/Transforms/InstSimplify/ConstProp/AArch64/lit.local.cfg
@@ -0,0 +1,2 @@
+if not 'AArch64' in config.root.targets:
+    config.unsupported = True