Lines Matching defs:compute

53 //   // Parallel compute function that executes the parallel body region for
55 //   // compute block index.
74 //   %block_size = ... compute parallel compute block size
75 //   %block_count = ... compute the number of compute blocks
85 //     // Call parallel compute function for a single block.
94 //   %block_size = ... compute parallel compute block size
95 //   %block_count = ... compute the number of compute blocks
138 // Helper struct to parse parallel compute function argument list.
208 // Returns a function type and implicit captures for a parallel compute
232   //   call @compute(%lb0, %lb1, ..., %ub0, %ub1, ..., %step0, %step1, ...)
248 // Create a parallel compute fuction from the parallel operation.
332   // Block first and last coordinates can be the same along the outer compute
333   // dimension when inner compute dimension contains multiple blocks.
353   // loop bounds should we use for the nested loops: bounds defined by compute
440 // Creates recursive async dispatch function for the given parallel compute
455 //     // Call parallel compute function for a single block.
471   // Compared to the parallel compute function async dispatch function takes
537     // Call parallel compute function inside the async.execute region.
559   // call the parallel compute function for the first block of the range.
575 // Launch async dispatch of the parallel compute function.
583   // Add one more level of indirection to dispatch parallel compute functions
591   // Appends operands shared by async dispatch and parallel compute functions to
610     // Call parallel compute function for the single block.
624     // execution of multiple parallel compute function. First block will be
636     // Wait for the completion of all parallel compute operations.
642   // Dispatch either single block compute function, or launch async dispatch.
646 // Dispatch parallel compute functions by submitting all async compute tasks
655   func::FuncOp compute = parallelComputeFunction.func;
661   // execution of multiple parallel compute function. First block will be
666   // Call parallel compute function for all blocks.
670   // Returns parallel compute function operands to process the given block.
688     // Call parallel compute function inside the async.execute region.
691       executeBuilder.create<func::CallOp>(executeLoc, compute.getSymName(),
692                                           compute.getResultTypes(),
704   // Iterate over all compute blocks and launch parallel compute operations.
707   // Call parallel compute function for the first block in the caller thread.
708   b.create<func::CallOp>(compute.getSymName(), compute.getResultTypes(),
711   // Wait for the completion of all async compute operations.
730   // reduce the number of parallel compute function arguments.
767     // compute function. LLVM can't always push constants across the non-trivial
779     // product is smaller than the `512`. We align the parallel compute block
812     // With large number of threads the value of creating many compute blocks
862     // Dispatch parallel compute function using async recursive work splitting,
863     // or by submitting compute task sequentially from a caller thread.
866     // Create a parallel compute function that takes a block id and computes
869     // Compute the number of parallel compute blocks.
872     // Dispatch parallel compute function without hints to unroll inner loops.
874       ParallelComputeFunction compute =
878       doDispatch(b, rewriter, compute, op, blockSize, blockCount, tripCounts);
882     // Dispatch parallel compute function with hints for unrolling inner loops.
884       ParallelComputeFunction compute = createParallelComputeFunction(
894       doDispatch(b, rewriter, compute, op, alignedBlockSize, blockCount,
899     // Dispatch to block aligned compute function only if the computed block
916   // Replace the `scf.parallel` operation with the parallel compute function.