[VP,Integer,#1] Vector-predicated integer intrinsics

Summary: This patch adds IR intrinsics for vector-predicated integer arithmetic. It is subpatch #1 of the [integer slice](https://reviews.llvm.org/D57504#1732277) of [LLVM-VP](https://reviews.llvm.org/D57504). LLVM-VP is a larger effort to bring native vector predication to LLVM. Reviewed By: andrew.w.kaylor Differential Revision: https://reviews.llvm.org/D69891
llvm · Mar 19, 2020 · 733b319 · 733b319
1 parent bb7d2b1
commit 733b319
Show file tree

Hide file tree

Showing 13 changed files with 1,258 additions and 4 deletions.
diff --git a/llvm/docs/LangRef.rst b/llvm/docs/LangRef.rst
diff --git a/llvm/include/llvm/Analysis/TargetTransformInfo.h b/llvm/include/llvm/Analysis/TargetTransformInfo.h
@@ -1167,6 +1167,15 @@ class TargetTransformInfo {
   /// to a stack reload.
   unsigned getGISelRematGlobalCost() const;
 
+  /// \name Vector Predication Information
+  /// @{
+  /// Whether the target supports the %evl parameter of VP intrinsic efficiently in hardware.
+  /// (see LLVM Language Reference - "Vector Predication Intrinsics")
+  /// Use of %evl is discouraged when that is not the case.
+  bool hasActiveVectorLength() const;
+
+  /// @}
+
   /// @}
 
 private:
@@ -1420,6 +1429,7 @@ class TargetTransformInfo::Concept {
                                      ReductionFlags) const = 0;
   virtual bool shouldExpandReduction(const IntrinsicInst *II) const = 0;
   virtual unsigned getGISelRematGlobalCost() const = 0;
+  virtual bool hasActiveVectorLength() const = 0;
   virtual int getInstructionLatency(const Instruction *I) = 0;
 };
 
@@ -1913,6 +1923,10 @@ class TargetTransformInfo::Model final : public TargetTransformInfo::Concept {
     return Impl.getGISelRematGlobalCost();
   }
 
+  bool hasActiveVectorLength() const override {
+    return Impl.hasActiveVectorLength();
+  }
+
   int getInstructionLatency(const Instruction *I) override {
     return Impl.getInstructionLatency(I);
   }

diff --git a/llvm/include/llvm/Analysis/TargetTransformInfoImpl.h b/llvm/include/llvm/Analysis/TargetTransformInfoImpl.h
@@ -628,6 +628,10 @@ class TargetTransformInfoImplBase {
     return 1;
   }
 
+  bool hasActiveVectorLength() const {
+    return false;
+  }
+
 protected:
   // Obtain the minimum required size to hold the value (without the sign)
   // In case of a vector it returns the min required size for one element.

diff --git a/llvm/include/llvm/IR/IntrinsicInst.h b/llvm/include/llvm/IR/IntrinsicInst.h
@@ -206,6 +206,48 @@ namespace llvm {
     /// @}
   };
 
+  /// This is the common base class for vector predication intrinsics.
+  class VPIntrinsic : public IntrinsicInst {
+  public:
+    static Optional<int> GetMaskParamPos(Intrinsic::ID IntrinsicID);
+    static Optional<int> GetVectorLengthParamPos(Intrinsic::ID IntrinsicID);
+
+    /// The llvm.vp.* intrinsics for this instruction Opcode
+    static Intrinsic::ID GetForOpcode(unsigned OC);
+
+    // Whether \p ID is a VP intrinsic ID.
+    static bool IsVPIntrinsic(Intrinsic::ID);
+
+    /// \return the mask parameter or nullptr.
+    Value *getMaskParam() const;
+
+    /// \return the vector length parameter or nullptr.
+    Value *getVectorLengthParam() const;
+
+    /// \return whether the vector length param can be ignored.
+    bool canIgnoreVectorLengthParam() const;
+
+    /// \return the static element count (vector number of elements) the vector
+    /// length parameter applies to.
+    ElementCount getStaticVectorLength() const;
+
+    // Methods for support type inquiry through isa, cast, and dyn_cast:
+    static bool classof(const IntrinsicInst *I) {
+      return IsVPIntrinsic(I->getIntrinsicID());
+    }
+    static bool classof(const Value *V) {
+      return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
+    }
+
+    // Equivalent non-predicated opcode
+    unsigned getFunctionalOpcode() const {
+      return GetFunctionalOpcodeForVP(getIntrinsicID());
+    }
+
+    // Equivalent non-predicated opcode
+    static unsigned GetFunctionalOpcodeForVP(Intrinsic::ID ID);
+  };
+
   /// This is the common base class for constrained floating point intrinsics.
   class ConstrainedFPIntrinsic : public IntrinsicInst {
   public:

diff --git a/llvm/include/llvm/IR/Intrinsics.td b/llvm/include/llvm/IR/Intrinsics.td
@@ -27,6 +27,10 @@ class IntrinsicProperty;
 // effects.  It may be CSE'd deleted if dead, etc.
 def IntrNoMem : IntrinsicProperty;
 
+// IntrNoSync - Threads executing the intrinsic will not synchronize using
+// memory or other means.
+def IntrNoSync : IntrinsicProperty;
+
 // IntrReadMem - This intrinsic only reads from memory. It does not write to
 // memory and has no other side effects. Therefore, it cannot be moved across
 // potentially aliasing stores. However, it can be reordered otherwise and can
@@ -1153,6 +1157,79 @@ def int_is_constant : Intrinsic<[llvm_i1_ty], [llvm_any_ty], [IntrNoMem, IntrWil
 def int_ptrmask: Intrinsic<[llvm_anyptr_ty], [llvm_anyptr_ty, llvm_anyint_ty],
                            [IntrNoMem, IntrSpeculatable, IntrWillReturn]>;
 
+//===---------------- Vector Predication Intrinsics --------------===//
+
+// Binary operators
+let IntrProperties = [IntrNoMem, IntrNoSync, IntrWillReturn] in {
+  def int_vp_add : Intrinsic<[ llvm_anyvector_ty ],
+                             [ LLVMMatchType<0>,
+                               LLVMMatchType<0>,
+                               LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                               llvm_i32_ty]>;
+  def int_vp_sub : Intrinsic<[ llvm_anyvector_ty ],
+                             [ LLVMMatchType<0>,
+                               LLVMMatchType<0>,
+                               LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                               llvm_i32_ty]>;
+  def int_vp_mul  : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_sdiv : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_udiv : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_srem : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_urem : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_ashr : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_lshr : Intrinsic<[ llvm_anyvector_ty ],
+                              [ LLVMMatchType<0>,
+                                LLVMMatchType<0>,
+                                LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                                llvm_i32_ty]>;
+  def int_vp_shl : Intrinsic<[ llvm_anyvector_ty ],
+                             [ LLVMMatchType<0>,
+                               LLVMMatchType<0>,
+                               LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                               llvm_i32_ty]>;
+  def int_vp_or : Intrinsic<[ llvm_anyvector_ty ],
+                            [ LLVMMatchType<0>,
+                              LLVMMatchType<0>,
+                              LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                              llvm_i32_ty]>;
+  def int_vp_and : Intrinsic<[ llvm_anyvector_ty ],
+                             [ LLVMMatchType<0>,
+                               LLVMMatchType<0>,
+                               LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                               llvm_i32_ty]>;
+  def int_vp_xor : Intrinsic<[ llvm_anyvector_ty ],
+                             [ LLVMMatchType<0>,
+                               LLVMMatchType<0>,
+                               LLVMScalarOrSameVectorWidth<0, llvm_i1_ty>,
+                               llvm_i32_ty]>;
+
+}
+
+
 //===-------------------------- Masked Intrinsics -------------------------===//
 //
 def int_masked_store : Intrinsic<[], [llvm_anyvector_ty,

diff --git a/llvm/include/llvm/IR/VPIntrinsics.def b/llvm/include/llvm/IR/VPIntrinsics.def
@@ -0,0 +1,84 @@
+//===-- IR/VPIntrinsics.def - Describes llvm.vp.* Intrinsics -*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains descriptions of the various Vector Predication intrinsics.
+// This is used as a central place for enumerating the different instructions
+// and should eventually be the place to put comments about the instructions.
+//
+//===----------------------------------------------------------------------===//
+
+// NOTE: NO INCLUDE GUARD DESIRED!
+
+// Provide definitions of macros so that users of this file do not have to
+// define everything to use it...
+//
+#ifndef REGISTER_VP_INTRINSIC
+#define REGISTER_VP_INTRINSIC(VPID, MASKPOS, VLENPOS)
+#endif
+
+// Map this VP intrinsic to its functional Opcode
+#ifndef HANDLE_VP_TO_OC
+#define HANDLE_VP_TO_OC(VPID, OC)
+#endif
+
+///// Integer Arithmetic /////
+
+// llvm.vp.add(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_add, 2, 3)
+HANDLE_VP_TO_OC(vp_add, Add)
+
+// llvm.vp.and(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_and, 2, 3)
+HANDLE_VP_TO_OC(vp_and, And)
+
+// llvm.vp.ashr(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_ashr, 2, 3)
+HANDLE_VP_TO_OC(vp_ashr, AShr)
+
+// llvm.vp.lshr(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_lshr, 2, 3)
+HANDLE_VP_TO_OC(vp_lshr, LShr)
+
+// llvm.vp.mul(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_mul, 2, 3)
+HANDLE_VP_TO_OC(vp_mul, Mul)
+
+// llvm.vp.or(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_or, 2, 3)
+HANDLE_VP_TO_OC(vp_or, Or)
+
+// llvm.vp.sdiv(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_sdiv, 2, 3)
+HANDLE_VP_TO_OC(vp_sdiv, SDiv)
+
+// llvm.vp.shl(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_shl, 2, 3)
+HANDLE_VP_TO_OC(vp_shl, Shl)
+
+// llvm.vp.srem(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_srem, 2, 3)
+HANDLE_VP_TO_OC(vp_srem, SRem)
+
+// llvm.vp.sub(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_sub, 2, 3)
+HANDLE_VP_TO_OC(vp_sub, Sub)
+
+// llvm.vp.udiv(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_udiv, 2, 3)
+HANDLE_VP_TO_OC(vp_udiv, UDiv)
+
+// llvm.vp.urem(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_urem, 2, 3)
+HANDLE_VP_TO_OC(vp_urem, URem)
+
+// llvm.vp.xor(x,y,mask,vlen)
+REGISTER_VP_INTRINSIC(vp_xor, 2, 3)
+HANDLE_VP_TO_OC(vp_xor, Xor)
+
+#undef REGISTER_VP_INTRINSIC
+#undef HANDLE_VP_TO_OC