SIMD-aware loop unrolling for embedded code optimization

Modern out-of-order processors have increased capacity to exploit instruction level parallelism (ILP) and memory level parallelism (MLP), e.g., by using wide superscalar pipelines and vector execution units, as well as deep buffers for inflight memory requests. These resources, however, often exhibit poor utilization rates on workloads with large working sets, e.g., in-memory databases, key-value stores, and graph analytics, as compilers and hardware struggle to expose ILP and MLP from the instruction stream automatically. In this paper, we introduce the IMLP (Instruction and Memory Level Parallelism) task programming model. IMLP tasks execute as coroutines that yield execution at annotated long-latency operations, e.g., memory accesses, divisions, or unpredictable branches. IMLP tasks are interleaved on a single thread, and integrate well with thread parallelism and vectorization. Our DSL embedded in C++, Cimple, allows exploration of task scheduling and transformations, such as buffering, vectorization, pipelining, and prefetching. We demonstrate state-of-the-art performance on core algorithms used in in-memory databases that operate on arrays, hash tables, trees, and skip lists. Cimple applications reach 2.5× throughput gains over hardware multithreading on a multi-core, and 6.4× single thread speedup.

Modern computational workloads demand exceptional performance and efficiency, necessitating the effective utilization of advanced CPU features such as SIMD (Single Instruction Multiple Data), instruction-level parallelism (ILP), and branch prediction. This paper explores optimization techniques that address inefficiencies at the algorithmic, architectural, and system levels, enabling software to align with hardware capabilities. Key techniques include resolving data dependencies, enhancing memory locality, utilizing compiler intrinsics,applying tail call optimizations, and employing strategies like loop unrolling, blocking, vectorization, and function inlining. Tail call optimization and breaking dependency chains are analyzed to improve parallelism and reduce processing overhead. Both manual and compiler-driven approaches are evaluated, providing insights into their trade-offs and synergies. Experimental results from benchmarks, such as matrix multiplication and particle simulations, demonstrate significant gains, with up to a 3x increase in instructions per cycle (IPC) and a 40% reduction in execution time. These findings highlight the critical role of optimizing software for architectural features like cache hierarchies, pipelining, and vector widths. This study provides techniques to maximize CPU efficiency, bridging the gap between hardware potential and software performance. Future directions include extending these methodologies to hybrid architectures like GPUs and integrating machine learning models for dynamic runtime optimization.

With the multitude of existing and upcoming wireless standards, it is becoming increasingly difficult for hardware-only baseband processing solutions to adapt to the rapidly changing wireless communication landscape. Software defined radio (SDR) promises to deliver a cost effective and flexible solution by implementing a wide variety of wireless protocols in software. In previous work, a fully programmable multicore architecture, SODA, was proposed that was able to meet the real-time requirements of 3G wireless protocols. SODA consists of one ARM control processor and four wide single instruction multiple data (SIMD) processing elements. Each processing element consists of a scalar and a wide 512-bit 32-lane SIMD datapath. A commercial prototype based on the SODA architecture, Ardbeg (named after a brand of Scotch whisky), has been developed. In this paper, we present the architectural evolution of going from a research design to a commercial prototype, including the goals, tradeoffs, and final design choices. Ardbegpsilas redesign process can be grouped into the following three major areas: optimizing the wide SIMD datapath, providing long instruction word (LIW) support for SIMD operations, and adding application-specific hardware accelerators. Because SODA was originally designed with 180 nm technology, the wide SIMD datapath is re-optimized in Ardbeg for 90 nm technology. This includes re-evaluating the most efficient SIMD width, designing a wider SIMD shuffle network, and implementing faster SIMD arithmetic units. Ardbeg also provides modest LIW support by allowing two SIMD operations to issue in the same cycle. This LIW execution supports SDR algorithmspsila most common parallel SIMD execution patterns with minimal hardware overhead. A viable commercial SDR solution must be competitive with existing ASIC solutions. Therefore, algorithm-specific hardware is added for performance bottleneck algorithms while still maintaining enough flexibility to support multiple wireless protocols. The combination of these architectural improvements allows Ardbeg to achieve 1.5-7x speedup over SODA across multiple wireless algorithms while consuming less power.

The emerging applications have imposed strong requirements, such as high processing capacity, low‐power consumption, reduced size and many others, on the sensor‐based systems. Owing to their balanced combination of flexibility and hardware performance, digital signal processors (DSPs) have become more and more popular used in sensor‐based systems. Many DSPs have adopted very long instruction words (VLIW) style architecture, for its ability to greatly enhance instruction level parallelism. However, as VLIW codes are statically scheduled, the behaviour of VLIW architecture is dominated by the efficiency of its compiler. Single instruction multiple data (SIMD) instructions, which perform multiple operations in parallel on multiple data packed in registers, have been widely used in DSPs to meet the requirements of sensor‐based systems. Although hand programming still yields the best performing SIMD codes, it is both time consuming and error prone. Advanced compiler techniques to automatically generate SIMD instructions, are under urgent demand. In this study, the authors proposed an SIMD code auto generation approach for VLIW architecture. It recognises candidates of operations in the intermediate representation, evaluates the possibility of grouping them into SIMD code, reconstructs the verified ones according to the cost model and finally generates the SIMD code. The authors have implemented this approach in the compiler of a VLIW DSP named Magnolia, which is designed for sensor‐based systems. The results show that the authors’ approach is very efficient, and can largely enhance the performance.

Array indirection causes several challenges for compilers to utilize single instruction, multiple data (SIMD) instructions. Disjoint memory references, arbitrarily misaligned memory references, and dependence cycles in loops are main challenges to handle for SIMD compilers. Due to those challenges, existing SIMD compilers have excluded loops with array indirection from their candidate loops for SIMD vectorization. However, addressing those challenges is inevitable, since many important compute-intensive applications extensively use array indirection to reduce memory and computation requirements. In this work, we propose a method to generate efficient SIMD code for loops containing indirected memory references. We extract both inter- and intra-iteration parallelism, taking data reorganization overhead into consideration. We also optimally place data reorganization code in order to amortize the reorganization overhead through the performance gain of SIMD vectorization. Experiments on four array indirection kernels, which are extracted from real-world scientific applications, show that our proposed method effectively generates SIMD code for irregular kernels with array indirection. Compared to the existing SIMD vectorization methods, our proposed method significantly improves the performance of irregular kernels by 91%, on average.

Array indirection causes several challenges for compilers to utilize single instruction, multiple data (SIMD) instructions. Disjoint memory references, arbitrarily misaligned memory references, and dependence cycles in loops are main challenges to handle for SIMD compilers. Due to those challenges, existing SIMD compilers have excluded loops with array indirection from their candidate loops for SIMD vectorization. However, addressing those challenges is inevitable, since many important compute-intensive applications extensively use array indirection to reduce memory and computation requirements. In this work, we propose a method to generate efficient SIMD code for loops containing indirected memory references. We extract both inter- and intra-iteration parallelism, taking data reorganization overhead into consideration. We also optimally place data reorganization code in order to amortize the reorganization overhead through the performance gain of SIMD vectorization. Experiments on four array indirection kernels, which are extracted from real-world scientific applications, show that our proposed method effectively generates SIMD code for irregular kernels with array indirection. Compared to the existing SIMD vectorization methods, our proposed method significantly improves the performance of irregular kernels by 91%, on average.

SIMD (Single Instruction Multiple Data) extensions are acceleration components integrated in general processor, aiming at extracting instruction and data level parallelism of multimedia and scientific calculation programs. Currently, most of the automatic vectorization methods for SIMD architectures are based on innermost loops. Inner loop vectorization is the common approach for auto-vectorization. This method has been used for many years and its efficiency is widely accepted by people. In this paper, we put forward a better method than inner loop vectorization for some loop nests, which is outer loop vectorization. Outer loop vectorization method means vectorizing the outer loop directly. It can extract more data level parallelism and make the most use of the spatial locality to improve the program efficiency than the inner one which is more suitable for some loop nests. This paper presents the realization. In this paper, we first revisit the preliminary analysis of outer loop vectorization based on Open64 complier. And then, data type conversion and code generation is presented in detail. Finally, we propose two optimization methods, capable of boosting the performance of outer loop vectorization to achieve the acceleration of 20% on average, 50% at most.

Exploitation of instruction-level parallelism is an effective mechanism for improving the performance of modern super-scalar/VLIW processors. Various software techniques can be applied to increase instruction-level parallelism. This paper describes and evaluates a software technique, dynamic memory disambiguation, that permits loops containing loads and stores to be scheduled more aggressively, thereby exposing more instruction-level parallelism. The results of our evaluation show that when dynamic memory disambiguation is applied in conjunction with loop unrolling, register renaming, and static memory disambiguation, the ILP of memory-intensive benchmarks can be increased by as much as 300 percent over loops where dynamic memory disambiguation is not performed. Our measurements also indicate that for the programs that benefit the most from these optimizations, the register usage does not exceed the number of registers on mast high-performance processors.

Modern compilers are still built using technology that existed decades ago. These include basic algorithms and techniques for lexing, parsing, data-flow analysis, data dependence analysis, vectorization, register allocation, instruction selection, and instruction scheduling. It is high time that we modernize our compiler toolchain. In this talk, I will show the path to the modernization of one important compiler technique -- vectorization. Vectorization was first introduced in the era of Cray vector processors during the 1980's. In modernizing vectorization, I will first show how to use new techniques that better target modern hardware. While vector supercomputers need large vectors, which are only available by parallelizing loops, modern SIMD instructions efficiently work on short vectors. Thus, in 2000, we introduced Superword Level Parallelism (SLP) based vectorization. SLP finds short vector instructions within basic blocks, and by loop unrolling we can convert vector parallelism to SLP. Next, I will show how we can take advantage of the power of modern computers for compilation, by using more accurate but expensive techniques to improve SLP vectorization. Due to the hardware resource constraints of the era, like many other compiler optimizations, SLP implementation was a greedy algorithm. In 2018, we introduced goSLP, which uses integer linear programming to find an optimal instruction packing strategy and achieves 7.58% geomean performance improvement over the LLVM's SLP implementation on SPEC2017fp C/C++ programs. Finally, I will show how to truly modernize a compiler by automatically learning the necessary components of the compiler with Ithemal and Vemal. The optimality of goSLP is under LLVM's simple per instruction additive cost model that fits within the Integer programming framework. However, the actual cost of execution in a modern out-of-order, pipelined, superscalar processor is much more complex. Manually building such cost models as well as manually developing compiler optimizations is costly, tedious, error-prone and is hard to keep up with the architectural changes. Ithemal is the first learnt cost model for predicting the throughput of x86 basic blocks. It not only significantly outperforms (more than halves the error) state-of-the-art analytical hand-written tools like llvm-mca, but also is learnt from data requiring minimal human effort. Vemal is a learnt policy for end-to-end vectorization as opposed to tuning heuristics, which outperforms LLVM's SLP vectorizer. These data-driven techniques can help achieve state-of-the-art results while also reducing the development and maintenance burden of the compiler developer.

SIMD (single-instruction multiple-data) instruction set extensions are quite common today in both high performance and embedded microprocessors, and enable the exploitation of a specific type of data parallelism called SLP (Superword Level Parallelism). While prior research shows that significant performance savings are possible when SLP is exploited, placing SIMD instructions in an application code manually can be very difficult and error prone. In this paper, we propose a novel automated compiler framework for improving superword level parallelism exploitation. The key part of our framework consists of two stages: superword statement generation and data layout optimization. The first stage is our main contribution and has two phases, statement grouping and statement scheduling, of which the primary goals are to increase SIMD parallelism and, more importantly, capture more superword reuses among the superword statements through global data access and reuse pattern analysis. Further, as a complementary optimization, our data layout optimization organizes data in memory space such that the price of memory operations for SLP is minimized. The results from our compiler implementation and tests on two systems indicate performance improvements as high as 15.2% over a state-of-the-art SLP optimization algorithm.

Nearly all personal computer and workstation processors, and virtually all high-performance embedded processor cores, now embody instruction level parallel (ILP) processing in the form of superscalar or very long instruction word (VLIW) architectures. ILP processors put much more of a burden on compilers; without heroic compiling techniques, most such processors fall far short of their performance goals. Those techniques are largely found in the high-level optimization phase and in the code generation phase; they are also collectively called instruction scheduling. This paper reviews the state of the art in code generation for ILP parallel processors. Modern ILP code generation methods move code across basic block boundaries. These methods grew out of techniques for generating horizontal microcode, so we introduce the problem by describing its history. Most modem approaches can be categorized by the shape of the scheduling region. Some of these regions are loops, and for those techniques known broadly as are used. Software Pipelining techniques are only considered here when there are issues relevant to the region-based techniques presented. The selection of a type of region to use in this process is one of the most controversial questions in code generation; the paper surveys the best known alternatives. The paper then considers two questions: First, given a type of region, how does one pick specific regions of that type in the intermediate code. In conjunction with region selection, we consider region enlargement techniques such as unrolling and branch target expansion. The second question, how does one construct a schedule once regions have been selected, occupies the next section of the paper. Finally, schedule construction using recent, innovative resource modeling based on finite-state automata is then reexamined. The paper includes an extensive bibliography.

Java bytecodes can be executed with the following three methods: a Java interpreter running on a particular machine interprets bytecodes; a Just-in-Time (JIT) compiler translates bytecodes to the native primitives of the particular machine and the machine executes the translated codes; and a Java processor executes bytecodes directly. The first two methods require no special hardware support for the execution of Java bytecodes and are widely used currently. The last method requires an embedded Java processor, picoJavaI or picoJavaII for instance. The picoJavaI and picoJavaII are simple pipelined processors with no ILP (instruction level parallelism) and TLP (thread level parallelism) supports. A so-called MAJC (microprocessor architecture for Java computing) design can exploit ILP and TLP by using a modified VLIW (very long instruction word) architecture and vertical multithreading technique, but it has its own instruction set and cannot execute Java bytecodes directly. In this paper, we investigate a processor architecture which can directly execute Java bytecodes meanwhile can exploit Java ILP and TLP simultaneously. The proposed processor consists of multiple slots implementing horizontal multithreading and multiple functional units shared by all threads executed in parallel. Our architectural simulation results show that the Java processor could achieve an average 20 IPC (instructions per cycle), or 7.33 EIPC (effective IPC), with 8 slots and a 4-instruction scheduling window for each slot. We also check other configurations and give the utilization of functional units as well as the performance improvement with various kinds of working loads.

As wireless video products evolve, they demand more sophisticated processing at higher resolutions and frame rates. Computational performance and energy efficiency have become critical design issues. This paper presents the Quantized Color Pack eXtension (QCPX) combined with a loop unrolling (LU) technique to improve execution performance and energy efficiency of color image and video processing applications. QCPX applied to a 32-bit datapath processor supports parallel operations on two packed 16-bit YCbCr (Y: luminance, Cr and Cb: chrominance) color pixels, providing greater subword-level parallelism by increasing the number of smaller color pixels packed into a word. Instruction-level parallelism can be further enhanced through loop unrolling. These techniques provide greater performance and efficiency for multimedia workloads on mobile systems. Experimental results on a set of media benchmark applications indicate that the LU plus QCPX-optimized version achieves a speedup ranging from 3.8 to 7.9 while reducing the energy consumption from 76% to 87% over the baseline version on identically configured, dynamically scheduled ILP superscalar processors. The LU plus QCPX-optimized version also outperforms the LU plus MDMX-like (MIPS’s multimedia extension) version.

Use of single instruction multiple data (SIMD) functional units enables multimedia systems to exploit parallelism to a higher degree resulting in significant system performance improvements. While implementation of whole SIMD system functionality for an application results in wastage of area resources, we have observed that for a specific multimedia application, we only need to implement a customized SIMD unit that is a subset of whole SIMD standard implementation. Based on this study, we have proposed an extension to the traditional system design and synthesis flow by integrating a methodology of SIMD unit synthesis. Our system synthesizes a customized SIMD unit along with an extended instruction set and generates an equivalent version of assembly code for the application using the extended instruction set. The results of area and performance obtained by experimenting over our implementation of AltiVec compatible customized SIMD units show the effectiveness of our approach.

Transistor budgets have been increasing at a very fast pace in the last years. This increasing transistor density will lead next generation processors to have a billion transistors available. It is the task of the computer architect to find the best way to use them.Out of order superscalar processors exploit parallelism at the finer grain, exploiting Instruction Level Parallelism (ILP). They issue multiple instructions per cycle, often in an order other than the specified by the programmer, using branch prediction and other speculative execution techniques in order to increase the available parallelism.Very Long Instruction Word (VLIW) processors also exploit parallelism at the instruction level, but they mostly rely on the compiler to detect the available parallelism. This increased compiler role allows a simpler design, and can be run at a faster clock rate, compensating for the loss of ILP.Chip Multiprocessors (CMP) join several narrow superscalar/VLIW components into a single processor, and mostly rely on the Thread Level Parallelism (TLP) for performance. The small and simple components can also run at a faster clock rate compensating for the loss of ILP, and significantly increasing throughput.Simultaneous Multithreaded (SMT) processors are based on wide superscalars, and exploit both ILP and TLP by issuing instructions from several different threads to the same pipeline, obtaining the benefits of TLP without sacrificing the ILP on single-threaded applications.For each of these options we will need to find the best balance between performance, design complexity, and power consumption among other factors. Also, the frontiers between them are not clear, and many intermediate design points can be found which leads to better/simpler/cheaper processors for the next generation of high performance computers.