201     coords[i] = b.create<arith::RemSIOp>(index, tripCounts[i]);  in delinearize()
202     index = b.create<arith::DivSIOp>(index, tripCounts[i]);  in delinearize()
261   func::FuncOp func = func::FuncOp::create(  in createParallelComputeFunction()
286   Value c0 = b.create<arith::ConstantIndexOp>(0);  in createParallelComputeFunction()
287   Value c1 = b.create<arith::ConstantIndexOp>(1);  in createParallelComputeFunction()
294             return b.create<arith::ConstantOp>(attr);  in createParallelComputeFunction()
313     tripCount = b.create<arith::MulIOp>(tripCount, tripCounts[i]);  in createParallelComputeFunction()
317   Value blockFirstIndex = b.create<arith::MulIOp>(blockIndex, blockSize);  in createParallelComputeFunction()
321   Value blockEnd0 = b.create<arith::AddIOp>(blockFirstIndex, blockSize);  in createParallelComputeFunction()
322   Value blockEnd1 = b.create<arith::MinSIOp>(blockEnd0, tripCount);  in createParallelComputeFunction()
323   Value blockLastIndex = b.create<arith::SubIOp>(blockEnd1, c1);  in createParallelComputeFunction()
336     blockEndCoord[i] = b.create<arith::AddIOp>(blockLastCoord[i], c1);  in createParallelComputeFunction()
379       computeBlockInductionVars[loopIdx] = b.create<arith::AddIOp>(  in createParallelComputeFunction()
380           lowerBounds[loopIdx], b.create<arith::MulIOp>(iv, steps[loopIdx]));  in createParallelComputeFunction()
383       isBlockFirstCoord[loopIdx] = b.create<arith::CmpIOp>(  in createParallelComputeFunction()
385       isBlockLastCoord[loopIdx] = b.create<arith::CmpIOp>(  in createParallelComputeFunction()
390         isBlockFirstCoord[loopIdx] = b.create<arith::AndIOp>(  in createParallelComputeFunction()
392         isBlockLastCoord[loopIdx] = b.create<arith::AndIOp>(  in createParallelComputeFunction()
401           b.create<scf::ForOp>(c0, tripCounts[loopIdx + 1], c1, ValueRange(),  in createParallelComputeFunction()
407           auto lb = b.create<arith::SelectOp>(isBlockFirstCoord[loopIdx],  in createParallelComputeFunction()
410           auto ub = b.create<arith::SelectOp>(isBlockLastCoord[loopIdx],  in createParallelComputeFunction()
414           b.create<scf::ForOp>(lb, ub, c1, ValueRange(),  in createParallelComputeFunction()
418         b.create<scf::YieldOp>(loc);  in createParallelComputeFunction()
432   b.create<scf::ForOp>(blockFirstCoord[0], blockEndCoord[0], c1, ValueRange(),  in createParallelComputeFunction()
434   b.create<func::ReturnOp>(ValueRange());  in createParallelComputeFunction()
480   func::FuncOp func = func::FuncOp::create(loc, "async_dispatch_fn", type);  in createAsyncDispatchFunction()
494   Value c1 = b.create<arith::ConstantIndexOp>(1);  in createAsyncDispatchFunction()
495   Value c2 = b.create<arith::ConstantIndexOp>(2);  in createAsyncDispatchFunction()
510   scf::WhileOp whileOp = b.create<scf::WhileOp>(types, operands);  in createAsyncDispatchFunction()
520     Value distance = b.create<arith::SubIOp>(end, start);  in createAsyncDispatchFunction()
522         b.create<arith::CmpIOp>(arith::CmpIPredicate::sgt, distance, c1);  in createAsyncDispatchFunction()
523     b.create<scf::ConditionOp>(dispatch, before->getArguments());  in createAsyncDispatchFunction()
532     Value distance = b.create<arith::SubIOp>(end, start);  in createAsyncDispatchFunction()
533     Value halfDistance = b.create<arith::DivSIOp>(distance, c2);  in createAsyncDispatchFunction()
534     Value midIndex = b.create<arith::AddIOp>(start, halfDistance);  in createAsyncDispatchFunction()
545       executeBuilder.create<func::CallOp>(executeLoc, func.getSymName(),  in createAsyncDispatchFunction()
547       executeBuilder.create<async::YieldOp>(executeLoc, ValueRange());  in createAsyncDispatchFunction()
551     auto execute = b.create<ExecuteOp>(TypeRange(), ValueRange(), ValueRange(),  in createAsyncDispatchFunction()
553     b.create<AddToGroupOp>(indexTy, execute.token(), group);  in createAsyncDispatchFunction()
554     b.create<scf::YieldOp>(ValueRange({start, midIndex}));  in createAsyncDispatchFunction()
566   b.create<func::CallOp>(computeFunc.func.getSymName(),  in createAsyncDispatchFunction()
569   b.create<func::ReturnOp>(ValueRange());  in createAsyncDispatchFunction()
587   Value c0 = b.create<arith::ConstantIndexOp>(0);  in doAsyncDispatch()
588   Value c1 = b.create<arith::ConstantIndexOp>(1);  in doAsyncDispatch()
604       b.create<arith::CmpIOp>(arith::CmpIPredicate::eq, blockCount, c1);  in doAsyncDispatch()
613     b.create<func::CallOp>(parallelComputeFunction.func.getSymName(),  in doAsyncDispatch()
616     b.create<scf::YieldOp>();  in doAsyncDispatch()
625     Value groupSize = b.create<arith::SubIOp>(blockCount, c1);  in doAsyncDispatch()
626     Value group = b.create<CreateGroupOp>(GroupType::get(ctx), groupSize);  in doAsyncDispatch()
632     b.create<func::CallOp>(asyncDispatchFunction.getSymName(),  in doAsyncDispatch()
637     b.create<AwaitAllOp>(group);  in doAsyncDispatch()
639     b.create<scf::YieldOp>();  in doAsyncDispatch()
643   b.create<scf::IfOp>(TypeRange(), isSingleBlock, syncDispatch, asyncDispatch);  in doAsyncDispatch()
657   Value c0 = b.create<arith::ConstantIndexOp>(0);  in doSequentialDispatch()
658   Value c1 = b.create<arith::ConstantIndexOp>(1);  in doSequentialDispatch()
663   Value groupSize = b.create<arith::SubIOp>(blockCount, c1);  in doSequentialDispatch()
664   Value group = b.create<CreateGroupOp>(GroupType::get(ctx), groupSize);  in doSequentialDispatch()
691       executeBuilder.create<func::CallOp>(executeLoc, compute.getSymName(),  in doSequentialDispatch()
694       executeBuilder.create<async::YieldOp>(executeLoc, ValueRange());  in doSequentialDispatch()
698     auto execute = b.create<ExecuteOp>(TypeRange(), ValueRange(), ValueRange(),  in doSequentialDispatch()
700     b.create<AddToGroupOp>(rewriter.getIndexType(), execute.token(), group);  in doSequentialDispatch()
701     b.create<scf::YieldOp>();  in doSequentialDispatch()
705   b.create<scf::ForOp>(c1, blockCount, c1, ValueRange(), loopBuilder);  in doSequentialDispatch()
708   b.create<func::CallOp>(compute.getSymName(), compute.getCallableResults(),  in doSequentialDispatch()
712   b.create<AwaitAllOp>(group);  in doSequentialDispatch()
748     tripCount = b.create<arith::MulIOp>(tripCount, tripCounts[i]);  in matchAndRewrite()
752   Value c0 = b.create<arith::ConstantIndexOp>(0);  in matchAndRewrite()
754       b.create<arith::CmpIOp>(arith::CmpIPredicate::eq, tripCount, c0);  in matchAndRewrite()
758     nestedBuilder.create<scf::YieldOp>(loc);  in matchAndRewrite()
808       numWorkerThreadsVal = b.create<arith::ConstantIndexOp>(numWorkerThreads);  in matchAndRewrite()
810       numWorkerThreadsVal = b.create<async::RuntimeNumWorkerThreadsOp>();  in matchAndRewrite()
829     Value scalingFactor = b.create<arith::ConstantFloatOp>(  in matchAndRewrite()
832       Value bracketBegin = b.create<arith::ConstantIndexOp>(p.first);  in matchAndRewrite()
833       Value inBracket = b.create<arith::CmpIOp>(  in matchAndRewrite()
835       Value bracketScalingFactor = b.create<arith::ConstantFloatOp>(  in matchAndRewrite()
837       scalingFactor = b.create<arith::SelectOp>(inBracket, bracketScalingFactor,  in matchAndRewrite()
841         b.create<arith::IndexCastOp>(b.getI32Type(), numWorkerThreadsVal);  in matchAndRewrite()
843         b.create<arith::SIToFPOp>(b.getF32Type(), numWorkersIndex);  in matchAndRewrite()
845         b.create<arith::MulFOp>(scalingFactor, numWorkersFloat);  in matchAndRewrite()
847         b.create<arith::FPToSIOp>(b.getI32Type(), scaledNumWorkers);  in matchAndRewrite()
849         b.create<arith::IndexCastOp>(b.getIndexType(), scaledNumInt);  in matchAndRewrite()
851     Value maxComputeBlocks = b.create<arith::MaxSIOp>(  in matchAndRewrite()
852         b.create<arith::ConstantIndexOp>(1), scaledWorkers);  in matchAndRewrite()
858     Value bs0 = b.create<arith::CeilDivSIOp>(tripCount, maxComputeBlocks);  in matchAndRewrite()
859     Value bs1 = b.create<arith::MaxSIOp>(bs0, minTaskSize);  in matchAndRewrite()
860     Value blockSize = b.create<arith::MinSIOp>(tripCount, bs1);  in matchAndRewrite()
870     Value blockCount = b.create<arith::CeilDivSIOp>(tripCount, blockSize);  in matchAndRewrite()
879       b.create<scf::YieldOp>();  in matchAndRewrite()
890       Value numIters = b.create<arith::ConstantIndexOp>(  in matchAndRewrite()
892       Value alignedBlockSize = b.create<arith::MulIOp>(  in matchAndRewrite()
893           b.create<arith::CeilDivSIOp>(blockSize, numIters), numIters);  in matchAndRewrite()
896       b.create<scf::YieldOp>();  in matchAndRewrite()
903       Value numIters = b.create<arith::ConstantIndexOp>(  in matchAndRewrite()
905       Value useBlockAlignedComputeFn = b.create<arith::CmpIOp>(  in matchAndRewrite()
908       b.create<scf::IfOp>(TypeRange(), useBlockAlignedComputeFn,  in matchAndRewrite()
910       b.create<scf::YieldOp>();  in matchAndRewrite()
917   b.create<scf::IfOp>(TypeRange(), isZeroIterations, noOp, dispatch);  in matchAndRewrite()
932         return builder.create<arith::ConstantIndexOp>(minTaskSize);  in runOnOperation()