75     Value tmp1 = create<arith::MulIOp>(int32Type, tidZ, dimY);  in rewrite()
76     Value tmp2 = create<arith::AddIOp>(int32Type, tmp1, tidY);  in rewrite()
77     Value tmp3 = create<arith::MulIOp>(int32Type, tmp2, dimX);  in rewrite()
78     Value tmp4 = create<arith::MulIOp>(int32Type, dimX, dimY);  in rewrite()
79     Value invocationIdx = create<arith::AddIOp>(int32Type, tmp3, tidX);  in rewrite()
80     Value workgroupSize = create<arith::MulIOp>(int32Type, tmp4, dimZ);  in rewrite()
84         create<arith::ConstantIntOp>(kSubgroupSize - 1, int32Type);  in rewrite()
85     Value laneId = create<arith::AndIOp>(invocationIdx, subgroupMask);  in rewrite()
87         create<arith::CmpIOp>(arith::CmpIPredicate::eq, laneId,  in rewrite()
88                               create<arith::ConstantIntOp>(0, int32Type));  in rewrite()
91         create<arith::SubIOp>(invocationIdx, laneId);  in rewrite()
96         create<arith::SubIOp>(workgroupSize, numThreadsWithSmallerSubgroupId);  in rewrite()
113       Value index = create<arith::IndexCastOp>(indexType, subgroupId);  in rewrite()
114       create<memref::StoreOp>(subgroupReduce, buffer, index);  in rewrite()
116     create<gpu::BarrierOp>();  in rewrite()
120         create<arith::AddIOp>(int32Type, workgroupSize, subgroupMask);  in rewrite()
122     Value isValidSubgroup = create<arith::CmpIOp>(arith::CmpIPredicate::slt,  in rewrite()
128     Value zero = create<arith::ConstantIndexOp>(0);  in rewrite()
130       Value index = create<arith::IndexCastOp>(indexType, invocationIdx);  in rewrite()
131       Value value = create<memref::LoadOp>(valueType, buffer, index);  in rewrite()
134       create<memref::StoreOp>(result, buffer, zero);  in rewrite()
138     create<gpu::BarrierOp>();  in rewrite()
139     Value result = create<memref::LoadOp>(valueType, buffer, zero);  in rewrite()
147   T create(Args... args) {  in create()  function
148     return rewriter.create<T>(loc, std::forward<Args>(args)...);  in create()
154     Value dim = create<T>(indexType, dimension);  in getDimOp()
155     return create<arith::IndexCastOp>(int32Type, dim);  in getDimOp()
197       create<cf::BranchOp>(block, ValueRange());  in getFactory()
251       return create<T>(lhs.getType(), lhs, rhs);  in getFactory()
260       Value cmp = rewriter.create<T>(loc, predicate, lhs, rhs);  in getCmpFactory()
261       return rewriter.create<arith::SelectOp>(loc, cmp, lhs, rhs);  in getCmpFactory()
288     create<cf::CondBranchOp>(condition, thenBlock,  in createIf()
294     create<cf::BranchOp>(continueBlock, thenOperands);  in createIf()
298     create<cf::BranchOp>(continueBlock, elseOperands);  in createIf()
325     Value subgroupSize = create<arith::ConstantIntOp>(kSubgroupSize, int32Type);  in createSubgroupReduce()
326     Value isPartialSubgroup = create<arith::CmpIOp>(arith::CmpIPredicate::slt,  in createSubgroupReduce()
339             Value offset = create<arith::ConstantIntOp>(i, int32Type);  in createSubgroupReduce()
340             auto shuffleOp = create<gpu::ShuffleOp>(  in createSubgroupReduce()
361             Value offset = create<arith::ConstantIntOp>(i, int32Type);  in createSubgroupReduce()
363                 create<gpu::ShuffleOp>(shuffleType, value, offset, subgroupSize,  in createSubgroupReduce()
374     Value subgroupSize = create<arith::ConstantIntOp>(kSubgroupSize, int32Type);  in getDivideBySubgroupSize()
375     return create<arith::DivSIOp>(int32Type, value, subgroupSize);  in getDivideBySubgroupSize()