[mlir] [LLVMIR] add all vector reduction intrinsics to LLVM IR dialect

Summary:
This allows for lowering of VectorOps (and others) into a LLVM IR
that maps directly to efficient implementations on the target machines.

http://llvm.org/docs/LangRef.html#experimental-vector-reduction-intrinsics

Reviewers: ftynse, andydavis1, nicolasvasilache, rriddle

Reviewed By: ftynse, rriddle

Subscribers: jfb, mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, Joonsoo, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D74171

mlir/include/mlir/Dialect/LLVMIR/LLVMOpBase.td

@@ -96,4 +96,42 @@ class LLVM_EnumAttr<string name, string llvmName, string description,
   string llvmClassName = llvmName;
 }
 
+// LLVM vector reduction over a single vector.
+class LLVM_VectorReduction<string mnem> :
+  LLVM_IntrOp<"experimental.vector.reduce." # mnem, []>,
+  Arguments<(ins LLVM_Type)>, Results<(outs LLVM_Type:$res)> {
+    let llvmBuilder = [{
+      llvm::Module *module = builder.GetInsertBlock()->getModule();
+      llvm::Function *fn = llvm::Intrinsic::getDeclaration(
+        module,
+        llvm::Intrinsic::experimental_vector_reduce_}] #
+        !subst(".","_", mnem) # [{, {
+          opInst.getOperand(0).getType().cast<LLVM::LLVMType>()
+          .getUnderlyingType(),
+        });
+      auto operands = lookupValues(opInst.getOperands());
+      $res = builder.CreateCall(fn, operands);
+    }];
+}
+
+// LLVM vector reduction over a single vector, with an initial value.
+class LLVM_VectorReductionV2<string mnem> :
+  LLVM_IntrOp<"experimental.vector.reduce.v2." # mnem, []>,
+  Arguments<(ins LLVM_Type, LLVM_Type)>, Results<(outs LLVM_Type:$res)> {
+    let llvmBuilder = [{
+      llvm::Module *module = builder.GetInsertBlock()->getModule();
+      llvm::Function *fn = llvm::Intrinsic::getDeclaration(
+        module,
+        llvm::Intrinsic::experimental_vector_reduce_v2_}] #
+        !subst(".","_", mnem) # [{, {
+          opInst.getResult(0).getType().cast<LLVM::LLVMType>()
+          .getUnderlyingType(),
+          opInst.getOperand(1).getType().cast<LLVM::LLVMType>()
+          .getUnderlyingType(),
+        });
+      auto operands = lookupValues(opInst.getOperands());
+      $res = builder.CreateCall(fn, operands);
+    }];
+}
+
 #endif  // LLVMIR_OP_BASE

mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td

@@ -780,6 +780,29 @@ def LLVM_Prefetch : LLVM_ZeroResultOp<"intr.prefetch">,
   }];
 }
 
+//
+// Vector Reductions.
+//
+
+def LLVM_experimental_vector_reduce_add : LLVM_VectorReduction<"add">;
+def LLVM_experimental_vector_reduce_and : LLVM_VectorReduction<"and">;
+def LLVM_experimental_vector_reduce_mul : LLVM_VectorReduction<"mul">;
+def LLVM_experimental_vector_reduce_fmax : LLVM_VectorReduction<"fmax">;
+def LLVM_experimental_vector_reduce_fmin : LLVM_VectorReduction<"fmin">;
+def LLVM_experimental_vector_reduce_or : LLVM_VectorReduction<"or">;
+def LLVM_experimental_vector_reduce_smax : LLVM_VectorReduction<"smax">;
+def LLVM_experimental_vector_reduce_smin : LLVM_VectorReduction<"smin">;
+def LLVM_experimental_vector_reduce_umax : LLVM_VectorReduction<"umax">;
+def LLVM_experimental_vector_reduce_umin : LLVM_VectorReduction<"umin">;
+def LLVM_experimental_vector_reduce_xor : LLVM_VectorReduction<"xor">;
+
+def LLVM_experimental_vector_reduce_v2_fadd : LLVM_VectorReductionV2<"fadd">;
+def LLVM_experimental_vector_reduce_v2_fmul : LLVM_VectorReductionV2<"fmul">;
+
+//
+// Atomic operations.
+//
+
 def AtomicBinOpXchg : I64EnumAttrCase<"xchg", 0>;
 def AtomicBinOpAdd  : I64EnumAttrCase<"add", 1>;
 def AtomicBinOpSub  : I64EnumAttrCase<"sub", 2>;

mlir/test/Target/llvmir-intrinsics.mlir

@@ -99,6 +99,37 @@ llvm.func @copysign_test(%arg0: !llvm.float, %arg1: !llvm.float, %arg2: !llvm<"<
   llvm.return
 }
 
+// CHECK-LABEL: @vector_reductions
+llvm.func @vector_reductions(%arg0: !llvm.float, %arg1: !llvm<"<8 x float>">, %arg2: !llvm<"<8 x i32>">) {
+  // CHECK: call i32 @llvm.experimental.vector.reduce.add.v8i32
+  "llvm.intr.experimental.vector.reduce.add"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call i32 @llvm.experimental.vector.reduce.and.v8i32
+  "llvm.intr.experimental.vector.reduce.and"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call float @llvm.experimental.vector.reduce.fmax.v8f32
+  "llvm.intr.experimental.vector.reduce.fmax"(%arg1) : (!llvm<"<8 x float>">) -> !llvm.float
+  // CHECK: call float @llvm.experimental.vector.reduce.fmin.v8f32
+  "llvm.intr.experimental.vector.reduce.fmin"(%arg1) : (!llvm<"<8 x float>">) -> !llvm.float
+  // CHECK: call i32 @llvm.experimental.vector.reduce.mul.v8i32
+  "llvm.intr.experimental.vector.reduce.mul"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call i32 @llvm.experimental.vector.reduce.or.v8i32
+  "llvm.intr.experimental.vector.reduce.or"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call i32 @llvm.experimental.vector.reduce.smax.v8i32
+  "llvm.intr.experimental.vector.reduce.smax"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call i32 @llvm.experimental.vector.reduce.smin.v8i32
+  "llvm.intr.experimental.vector.reduce.smin"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call i32 @llvm.experimental.vector.reduce.umax.v8i32
+  "llvm.intr.experimental.vector.reduce.umax"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call i32 @llvm.experimental.vector.reduce.umin.v8i32
+  "llvm.intr.experimental.vector.reduce.umin"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  // CHECK: call float @llvm.experimental.vector.reduce.v2.fadd.f32.v8f32
+  "llvm.intr.experimental.vector.reduce.v2.fadd"(%arg0, %arg1) : (!llvm.float, !llvm<"<8 x float>">) -> !llvm.float
+  // CHECK: call float @llvm.experimental.vector.reduce.v2.fmul.f32.v8f32
+  "llvm.intr.experimental.vector.reduce.v2.fmul"(%arg0, %arg1) : (!llvm.float, !llvm<"<8 x float>">) -> !llvm.float
+  // CHECK: call i32 @llvm.experimental.vector.reduce.xor.v8i32
+  "llvm.intr.experimental.vector.reduce.xor"(%arg2) : (!llvm<"<8 x i32>">) -> !llvm.i32
+  llvm.return
+}
+
 // Check that intrinsics are declared with appropriate types.
 // CHECK-DAG: declare float @llvm.fma.f32(float, float, float)
 // CHECK-DAG: declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>) #0