The present invention generally relates to data processing, and more specifically, to secure and efficient application data processing.
In computing, asynchronous execution of an operation, such as a DEFLATE compression call (DFLTCC), occurs when a processor that does not want to wait for the execution of the DFLTCC asks another (assist) processor to execute the DFLTCC on its behalf. The assist processor then runs firmware (e.g., i390 firmware) in order to execute the DFLTCC and reports back to the requesting processor.
In accordance with embodiments of the present invention, a system architecture is provided and includes first and second processing units respectively communicative with an on-chip coherency unit and an accelerator communicative with the on-chip coherency unit. The accelerator is configured to execute an operation responsive to a call issued by one of the first and second processing units. The first processing unit is configured to set an asynchronous operation flag (AOF) to indicate that the second processing unit is to conduct an operation for the first processing unit. The second processing unit is configured to respond to the AOF by building scatter gather lists and subsequently issuing the call and feeding the scatter gather lists to the accelerator to facilitate execution of the operation by the accelerator.
In accordance with additional or alternative embodiments of the present invention, the first and second processing units are each respectively provided as ones of multiple processing units respectively communicative with the on-chip coherency unit via first interfaces.
In accordance with additional or alternative embodiments of the present invention, the on-chip coherency unit is communicative with external processors and controllers.
In accordance with additional or alternative embodiments of the present invention, the system architecture further includes page buffer units (PBUs) respectively communicative with the on-chip coherency unit via second interfaces and a memory core.
In accordance with additional or alternative embodiments of the present invention, the accelerator is communicative with the on-chip coherency unit via a second interface comprising a direct memory access (DMA)-like interface.
In accordance with additional or alternative embodiments of the present invention, the operation includes a DEFLATE operation.
In accordance with additional or alternative embodiments of the present invention, the first processing unit is further configured to determine whether to operate in one of a synchronous mode or an asynchronous mode, issue the call in an event the first processing unit determines that the synchronous mode should be in effect and set the AOF in an event the first processing unit determines that the asynchronous mode should be in effect.
In accordance with additional or alternative embodiments of the present invention, the scatter gather lists include absolute addresses and associated lengths for source data and target data of the data compression.
In accordance with additional or alternative embodiments of the present invention, the absolute addresses of the source data are non-consecutive in absolute address space and the absolute addresses of the target data are non-consecutive in absolute address space.
In accordance with additional or alternative embodiments of the present invention, the absolute addresses of the source data are non-sequential and the absolute addresses of the target data are non-sequential.
In accordance with embodiments of the present invention, a system architecture is provided and includes one or more drawers, each of which includes one or more central processors (CPs) and one or more system controllers (SCs). Each of the one or more CPs includes an accelerator configured to execute data compression faster than the first and second processing units responsive to a call issued by one of the first and second processing units. The first processing unit is configured to set an asynchronous compression flag (ACF) to indicate that the second processing unit is to conduct a compression operation for the first processing unit. The second processing unit is configured to respond to the ACF by building scatter gather lists and subsequently issuing the call and feeding the scatter gather lists to the accelerator to facilitate execution of the data compression by the accelerator.
In accordance with additional or alternative embodiments of the present invention, the one or more drawers are communitive with each other and the on-chip coherency unit of each of the one or more CPs of each of the one or more drawers is communicative with the one or more CPs and the one or more SCs of the corresponding drawer.
In accordance with additional or alternative embodiments of the present invention, the first and second processing units of each of the one or more CPs of each of the one or more drawers are provided as ones of multiple processing units respectively communicative with a corresponding on-chip coherency unit via first interfaces.
In accordance with additional or alternative embodiments of the present invention, each of the one or more CPs of each of the one or more drawers further includes page buffer units (PBUs) respectively communicative with a corresponding on-chip coherency unit via second interfaces and a memory core.
In accordance with additional or alternative embodiments of the present invention, the accelerator of each of the one or more CPs of each of the one or more drawers is communicative with a corresponding on-chip coherency unit via a second interface including a direct memory access (DMA)-like interface.
In accordance with additional or alternative embodiments of the present invention, the compression operation includes a DEFLATE operation.
In accordance with additional or alternative embodiments of the present invention, the first processing unit is further configured to determine whether to operate in one of a synchronous mode or an asynchronous mode, issue the call in an event the first processing unit determines that the synchronous mode should be in effect and set the ACF in an event the first processing unit determines that the asynchronous mode should be in effect.
In accordance with additional or alternative embodiments of the present invention, the scatter gather lists include absolute addresses and associated lengths for source data and target data of the operation.
In accordance with additional or alternative embodiments of the present invention, the absolute addresses of the source data are non-consecutive and the absolute addresses of the target data are non-consecutive.
In accordance with additional or alternative embodiments of the present invention, the absolute addresses of the source data are non-sequential in absolute address space and the absolute addresses of the target data are non-sequential in absolute address space.
In accordance with embodiments of the present invention, a computer program product is provided. The computer program product includes a processing circuit and a memory unit. The memory unit has executable instructions stored thereon, which, when executed cause the processing circuit to execute a method of securely and efficiently executing data compression or decompression. The method includes determining whether to operate in one of a synchronous mode or an asynchronous mode, setting an asynchronous operation flag (AOF) in an event the determining indicates that the asynchronous mode should be in effect and building scatter gather lists to be fed to an accelerator to facilitate execution of the data compression or decompression by the accelerator in response to the setting of the AOF.
In accordance with additional or alternative embodiments of the present invention, the scatter gather lists include absolute addresses and associated lengths for source data and target data of the data compression or decompression and the absolute addresses of the source data are non-consecutive and the absolute addresses of the target data are non-consecutive.
In accordance with embodiments of the present invention, a computer-implemented method of securely and efficiently executing data compression or decompression is provided. The computer-implemented method includes determining whether to operate in one of a synchronous mode or an asynchronous mode, setting an asynchronous operation flag (AOF) in an event the determining indicates that the asynchronous mode should be in effect and building scatter gather lists to be fed to an accelerator to facilitate execution of the data compression or decompression by the accelerator in response to the setting of the AOF.
In accordance with additional or alternative embodiments of the present invention, the scatter gather lists include absolute addresses and associated lengths for source data and target data of the data compression or decompression and the absolute addresses of the source data are non-consecutive and the absolute addresses of the target data are non-consecutive.
In accordance with embodiments of the present invention, a system architecture is provided. The system architecture includes first and second processing units respectively communicative with an on-chip coherency unit and an accelerator communicative with the on-chip coherency unit. The accelerator is configured to execute data compression or decompression faster than the first and second processing units responsive to a call issued by one of the first and second processing units. The first processing unit is configured to execute a method including determining whether to operate in one of a synchronous mode or an asynchronous mode and setting an asynchronous operation flag (AOF) in an event the determining indicates that the asynchronous mode should be in effect. The second processing unit is configured to execute a method including building scatter gather lists to be fed to the accelerator to facilitate execution of the data compression or decompression by the accelerator in response to the setting of the AOF.
Technical effects and benefits of the embodiments described herein provide for executions of certain operations in an asynchronous mode in which the ACF is set and millicode thus takes absolute addresses from scatter gather lists and feeds those addresses to an accelerator.
Additional technical features and benefits are realized through the techniques of the present invention. Embodiments and aspects of the invention are described in detail herein and are considered a part of the claimed subject matter. For a better understanding, refer to the detailed description and to the drawings.
The specifics of the exclusive rights described herein are particularly pointed out and distinctly claimed in the claims at the conclusion of the specification. The foregoing and other features and advantages of the embodiments of the invention are apparent from the following detailed description taken in conjunction with the accompanying drawings in which:
The diagrams depicted herein are illustrative. There can be many variations to the diagram or the operations described therein without departing from the spirit of the invention. For instance, the actions can be performed in a differing order or actions can be added, deleted or modified. Also, the term “coupled” and variations thereof describes having a communications path between two elements and does not imply a direct connection between the elements with no intervening elements/connections between them. All of these variations are considered a part of the specification.
In the accompanying figures and following detailed description of the disclosed embodiments, the various elements illustrated in the figures are provided with two or three digit reference numbers. With minor exceptions, the leftmost digit(s) of each reference number correspond to the figure in which its element is first illustrated.
Various embodiments of the invention are described herein with reference to the related drawings. Alternative embodiments of the invention can be devised without departing from the scope of this invention. Various connections and positional relationships (e.g., over, below, adjacent, etc.) are set forth between elements in the following description and in the drawings. These connections and/or positional relationships, unless specified otherwise, can be direct or indirect, and the present invention is not intended to be limiting in this respect. Accordingly, a coupling of entities can refer to either a direct or an indirect coupling, and a positional relationship between entities can be a direct or indirect positional relationship. Moreover, the various tasks and process steps described herein can be incorporated into a more comprehensive procedure or process having additional steps or functionality not described in detail herein.
The following definitions and abbreviations are to be used for the interpretation of the claims and the specification. As used herein, the terms “comprises,” “comprising,” “includes,” “including,” “has,” “having,” “contains” or “containing,” or any other variation thereof, are intended to cover a non-exclusive inclusion. For example, a composition, a mixture, process, method, article, or apparatus that comprises a list of elements is not necessarily limited to only those elements but can include other elements not expressly listed or inherent to such composition, mixture, process, method, article, or apparatus.
Additionally, the term “exemplary” is used herein to mean “serving as an example, instance or illustration.” Any embodiment or design described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments or designs. The terms “at least one” and “one or more” may be understood to include any integer number greater than or equal to one, i.e. one, two, three, four, etc. The terms “a plurality” may be understood to include any integer number greater than or equal to two, i.e. two, three, four, five, etc. The term “connection” may include both an indirect “connection” and a direct “connection.”
The terms “about,” “substantially,” “approximately,” and variations thereof, are intended to include the degree of error associated with measurement of the particular quantity based upon the equipment available at the time of filing the application. For example, “about” can include a range of ±8% or 5%, or 2% of a given value.
For the sake of brevity, conventional techniques related to making and using aspects of the invention may or may not be described in detail herein. In particular, various aspects of computing systems and specific computer programs to implement the various technical features described herein are well known. Accordingly, in the interest of brevity, many conventional implementation details are only mentioned briefly herein or are omitted entirely without providing the well-known system and/or process details.
Turning now to an overview of technologies that are more specifically relevant to aspects of the invention, an application running on a core processing unit (PU) operates in logical address space where each logical address that the processor sees in logical address space (which is private to the application) maps to an address in absolute address space (which is the common system wide address). If that processor executes the application itself, the execution is synchronous. Here, the source data is typically located in consecutive areas in logical address space and the target data is also typically located in consecutive areas in logical address space. On the other hand, in an event the processor does not want to wait for an execution of the application, the processor can ask another (assist) processor to execute the application on its behalf. The assist processor then runs firmware in order to execute the application in an asynchronous execution and reports back to the requesting processor upon completion. In doing so, the assist processor is unaware of the mapping between logical and absolute addresses and, therefore, the assist processor needs to operate directly on absolute addresses. Here, the source and target data are typically not located in consecutive areas in absolute address space.
In asynchronous executions, such as for asynchronous DEFLATE compression calls (DFLTCCs) or decompression calls, the assist processor has to copy data from non-consecutive source and target buffers and then place that copied data in hardware system areas (HSAs). This places a high demand on the system as additional, private areas in HSA are needed. There is also significant overhead due to copying and concatenating source data which reduces throughput and increases utilization. In addition, security concerns are presented due to customer data being placed in HSA.
Other issues with asynchronous executions are that the assist processor is often required to invoke certain instructions multiple times with each consecutive source and target data chunk being copied and concatenated. With each call having a relatively small consecutive chunk of source and target data, there can be a relatively high amount of instruction overhead and greatly reduced throughput.
Turning now to an overview of the aspects of the invention, one or more embodiments of the invention address the above-described shortcomings of the prior art by providing for an introduction of an asynchronous operation flag (AOF) or an asynchronous compression flag (ACF) for certain compression or decompression operations, such as DFLATE compression calls (DFLTCC). The AOF or ACF indicates that an asynchronous mode is in effect and is set when an assist processor executes the certain operations on behalf of a different processor. For such asynchronous execution, firmware on the assist processor builds a source scatter gather list with system absolute addresses and associated lengths for source data and target data and calls the instruction with the AOF or the ACF and the following parameters: address of source scatter gather list, length of total source data, address of target scatter gather list and length of total target data. If the AOF or the ACF is set, millicode of the certain operations works with absolute addresses provided in the scatter gather lists instead of translating consecutive source and target buffers.
That is, for a DFLTCC in synchronous mode in which the ACF is not set, the DFLTCC millicode takes base offset of source and target buffers and translates each page (4k increments) from logical to absolute and feeds those addresses to an accelerator. By contrast, for the DFLTCC in asynchronous mode in which the ACF is set, the DFTLCC millicode takes absolute addresses from the scatter gather list and feeds those addresses to the accelerator.
The above-described aspects of the invention address the shortcomings of the prior art by providing for executions of certain operations in an asynchronous mode in which the ACF is set and millicode thus takes absolute addresses from the scatter gather list and feeds those addresses to the accelerator.
Turning now to a more detailed description of aspects of the present invention, a capability is provided to facilitate processing within a computing environment. As one example, a single instruction (e.g., a single architected hardware machine instruction at the hardware/software interface) is provided to perform a function (also referred to as an operation), such as a compression or decompression function, to compress and/or decompress (also referred to as uncompress) data. The instruction is part of a general-purpose processor instruction set architecture (ISA), which is dispatched by a program (e.g., an operating system or a user program) on the general-purpose processor. By using an ISA instruction to perform compression/decompression, task switches by the operating system are not needed to perform the compression/decompression operations, thereby saving execution cycles. Further, by using a single instruction to compress and/or decompress data, execution time within a processor, such as a general-purpose processor, is reduced.
In one example, the instruction performs compressing and uncompressing operations compliant with an industry standard, referred to as the DEFLATE standard, and the instruction is referred to as a DEFLATE Conversion Call instruction. The DEFLATE standard includes descriptions for compressed data symbols which represent duplicate strings in the original form of the data (in the uncompressed form of the data). Such symbols include a pointer and a length of a duplicate string which describe the location and length of the duplicate string, which was previously processed, in relationship to the current location of data being processed. The previously processed uncompressed form of the data is referred to as history. In one example, the history is a contiguous number of bytes in memory, which can be as large as, for instance, 32 K-bytes.
One embodiment of a computing environment to incorporate and use one or more aspects of the present invention is described with reference to
In one example, processor 102 is based on the z/Architecture® hardware architecture offered by International Business Machines Corporation, Armonk, N.Y., and is part of a server, such as an IBM Z® server, which is also offered by International Business Machines Corporation and implements the z/Architecture hardware architecture. One embodiment of the z/Architecture hardware architecture is described in a publication entitled, “z/Architecture Principles of Operation,” IBM Publication No. SA22-7832-11, 12th edition, September 2017, which is hereby incorporated herein by reference in its entirety. The z/Architecture hardware architecture, however, is only one example architecture; other architectures and/or other types of computing environments may include and/or use one or more aspects of the present invention. In one example, the processor executes an operating system, such as the z/OS® operating system, also offered by International Business Machines Corporation.
Processor 102 includes a plurality of functional components used to execute instructions. As depicted in
Another example of a computing environment to incorporate and use one or more aspects of the present invention is described with reference to
Referring to
Memory 202 includes, for example, one or more logical partitions 208, a hypervisor 210 that manages the logical partitions, and processor firmware 212. One example of hypervisor 210 is the Processor Resource/System Manager (PR/SM™) hypervisor, offered by International Business Machines Corporation, Armonk, N.Y. As used herein, firmware includes, e.g., the microcode of the processor. It includes, for instance, the hardware-level instructions and/or data structures used in implementation of higher level machine code. In one embodiment, it includes, for instance, proprietary code that is typically delivered as microcode that includes trusted software or microcode specific to the underlying hardware and controls operating system access to the system hardware.
Each logical partition 208 is capable of functioning as a separate system. That is, each logical partition can be independently reset, run a guest operating system 220 such as a z/OS operating system, or another operating system, and operate with different programs 222. An operating system or application program running in a logical partition appears to have access to a full and complete system, but in reality, only a portion of it is available.
Memory 202 is coupled to processors (e.g., CPUs) 204, which are physical processor resources that may be allocated to the logical partitions. For instance, a logical partition 208 includes one or more logical processors, each of which represents all or a share of a physical processor resource 204 that may be dynamically allocated to the logical partition.
Further, memory 202 is coupled to I/O subsystem 206. I/O subsystem 206 may be a part of the central electronics complex or separate therefrom. It directs the flow of information between main storage 202 and input/output control units 230 and input/output (I/O) devices 240 coupled to the central electronics complex.
Many types of I/O devices may be used. One particular type is a data storage device 250. Data storage device 250 may store one or more programs 252, one or more computer readable program instructions 254, and/or data, etc. The computer readable program instructions may be configured to carry out functions of embodiments of aspects of the invention.
As one example, each processor 204 includes at least one cache 260 (e.g., a local cache) of a cache hierarchy that includes a plurality of levels of cache, including one or more local caches and/or one or more shared caches. Further, in one embodiment, the local caches and memory 202 are coupled to a compress/decompress component (or other component) 262 used in performing one or more of compressing and/or decompressing of data (and/or other operations of one or more aspects of the present invention). In various examples, there may be one or more components performing these tasks. Many variations are possible.
In one embodiment, a processor (e.g., processor 204) obtains the instruction (e.g., the DEFLATE Conversion Call instruction), decodes the instruction, performs set-up for the instruction, including translating addresses to be used by the instruction, and sends commands for the instruction to a component coupled to the processor, such as component 262, to perform a function specified by the instruction. Component 262 has access to the cache hierarchy and memory, such that in performing the specified function, it reads data, processes it and stores the processed data back. As an example, component 262 is a hardware component.
In a further embodiment, at least a portion of component 262 is included as part of the processor. Many variations are possible.
Central electronics complex 200 may include and/or be coupled to removable/non-removable, volatile/non-volatile computer system storage media. For example, it may include and/or be coupled to a non-removable, non-volatile magnetic media (typically called a “hard drive”), a magnetic disk drive for reading from and writing to a removable, non-volatile magnetic disk (e.g., a “floppy disk”), and/or an optical disk drive for reading from or writing to a removable, non-volatile optical disk, such as a CD-ROM, DVD-ROM or other optical media. It should be understood that other hardware and/or software components could be used in conjunction with central electronics complex 200. Examples include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.
Further, central electronics complex 200 may be operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and/or configurations that may be suitable for use with central electronics complex 200 include, but are not limited to, personal computer (PC) systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems or devices, and the like.
Although various examples of computing environments are described herein, one or more aspects of the present invention may be used with many types of environments. The computing environments provided herein are only examples.
In accordance with an aspect of the present invention, a computing environment, such as computing environment 100 or central electronics complex 200, employs a conversion facility that provides a mechanism to compress and decompress data. In one example, the conversion facility is a DEFLATE conversion facility that provides a mechanism to compress and decompress data using the DEFLATE compressed data format. In one example, the conversion facility is installed in the system when a facility indicator is set, e.g., to one. As one particular example of the z/Architecture hardware architecture, facility bit 151 is set to, e.g., one, when the conversion facility is installed in the z/Architecture architectural mode. The facility includes, for instance, the DEFLATE Conversion Call instruction, an embodiment of which is described below.
In one example, the DEFLATE Conversion Call instruction performs functions related to transforming the state of data between the original (uncompressed) form of the data, and a compressed representation of the data, as specified by a selected standard, such as the IETF (Internet Engineering Task Force) RFC (Request for Comments) 1951 specification, which is described in the DEFLATE Compressed Data Format Specification version 1.3 Internet Engineering Task Force, Request for Comments 1951, May 1996.
In one example, the uncompressed data is a sequence of bytes, and the compressed representation of the data includes symbols. Symbols represent an individual byte of uncompressed data, referred to as a literal byte, or represent a reoccurring sequence of bytes of uncompressed data, referred to as a duplicate string. A Huffman table, as an example, specifies the encoding and decoding between compressed data symbols and uncompressed data. There are two types of Huffman tables: a fixed-Huffman table (FHT), which is a predetermined specification which includes, for instance, all possible codings; and a dynamic-Huffman table (DHT), which is a set of codings created specifically for the data to be compressed, which may be a subset of all possible codings. A compressed representation of data generated with a DHT is typically smaller than a compressed representation of the same data generated with an FHT. A portion of the most recently processed uncompressed data, referred to as history, is maintained for encoding and decoding compressed data symbols representing duplicate strings. The history is the reference source for duplicate strings. The history is updated as data is processed during an operation.
As indicated, in one example, the DEFLATE Conversion Call instruction uses the DEFLATE compressed data format, which is described in RCF 1951, DEFLATE Compressed Data Format Specification version 1.3. Attributes of the DEFLATE standard which apply to the DEFLATE Conversion Call instruction include, for instance:
When a compressed data element occupies part of, and not all of, a byte in storage, the entire byte in storage is accessed. Storage operand lengths specify the number of addressable bytes, which may specify more bits than the compressed data occupies.
Additional details regarding compressed data blocks are described further below.
One embodiment of the DEFLATE Conversion Call (DFLTCC) instruction is described with reference to
In one embodiment, a program (e.g., an operating system or user program) may execute the DEFLATE Conversion Call instruction multiple times to compress or uncompress a single data stream. For instance, when an application compresses or decompresses a large data stream (e.g., greater than 1 M-bytes), the operation may include multiple calls to compress or decompress buffered portions of the data stream. In accordance with one aspect of the present invention, a program declares a buffer (e.g., a 32 K-byte buffer), which is used to accumulate the history of uncompressed data processed during an operation which spans multiple executions of the DEFLATE Conversion Call instruction. The buffer is referred to as a circular history buffer, which is defined using the DEFLATE Conversion Call instruction, as described herein.
Referring to
In one embodiment, execution of the instruction includes the use of one or more implied general registers (i.e., registers not explicitly designated by the instruction). For instance, general registers 0 and 1 are used in execution of the DEFLATE Conversion Call instruction, as described herein. General register 0 is used, in one example, to specify a function to be performed (as well as a history buffer type, described below), and general register 1 is used to provide a location of a parameter block used by the instruction.
As an example, with reference to
Example assigned function codes for the DEFLATE Conversion Call instruction are shown in
When the specified function is DFLTCC-CMPR or DFLTCC-XPND, bit 56 of general register 0 specifies the history buffer type (HBT) used during the operation. When HBT is zero, the history buffer is called an in-line history buffer. When using an in-line history buffer, the history is, e.g., immediately to the left of the second operand when DFLTCC-CMPR is specified and is, e.g., immediately to the left of the first operand when DFLTCC-XPND is specified. When HBT is one, the history buffer is called a circular history buffer. When using a circular history buffer, the history is a portion of, or all of, the third operand when either DFLTCC-CMPR or DFLTCC-XPND is specified. When the DFLTCC-QAF or DFLTCC-GDHT function is specified, bit 56 of general register 0 is ignored. In one example, bit positions 0-31 of general register 0 are ignored. Further, in one example, bit positions 32-55 of general register 0 are reserved and should contain zeros; otherwise, the program may not operate compatibly in the future.
Further details regarding another implied register, general register 1, used by the DEFLATE Conversion Call instruction are described with reference to
For the specified functions (e.g., DFLTCC-QAF, DFLTCC-GDHT, DFLTCC-CMPR, DFLTCC-XPND), the contents of general registers 0, 1 and R3 are not modified. Further, in one example, R1 field 304 designates an even-odd pair of general registers. It is to designate an even-numbered register and is not to designate general register 0; otherwise, a specification exception is recognized.
As depicted in
When the specified function is DFLTCC-XPND, the contents of general register R1+1 specify the length of the first operand. When the specified function is DFLTCC-CMPR or DFLTCC-XPND, the results of compressing or decompressing data are stored at the first operand location. When the DFLTCC-QAF or DFLTCC-GDHT function is specified, the contents of general registers R1 and R1+1 are ignored.
Moreover, for the specified functions (e.g., DFLTCC-QAF, DFLTCC-GDHT, DFLTCC-CMPR and DFLTCC-XPND), in one example, R2 field 306 designates an even-odd pair of general registers. It is to designate an even-numbered register and is not to designate general register 0; otherwise, a specification exception is recognized.
As depicted in
When the DFLTCC-QAF function is specified, the contents of general registers R2 and R2+1 are ignored. When the DFLTCC-GDHT function is specified and the contents of general register R2+1 specify a length equal to zero, a specification exception is recognized, and the second operand is not accessed. When the DFLTCC-CMPR or DFLTCC-XPND function is specified, the continuation flag (CF) field of the parameter block is zero at the beginning of the execution of the instruction, and the contents of general register R2+1 specify a length equal to zero, a specification exception is recognized, and the second operand is not accessed.
As shown in
As part of the operation, when the specified function is DFLTCC-CMPR, the address in general register R1 is incremented by the number of bytes processed of the first operand that included processing bit position 0, and the length in general register R1+1 is decremented by the same number; the address in general register R2 is incremented by the number of bytes processed of the second operand, and the length in general register R2+1 is decremented by the same number. The number of bytes processed of the first operand that included processing bit position 0 is, for instance, the integer quotient resulting from an integer division with the dividend being the sum of the number of output bits processed and the original value of the SBB, and the divisor being a value of eight. The formation and updating of the addresses and lengths are dependent on the addressing mode, as described below.
As part of the operation, when the specified function is DFLTCC-XPND, the address in general register R1 is incremented by the number of bytes processed of the first operand, and the length in general register R1+1 is decremented by the same number; the address in general register R2 is incremented by the number of bytes processed of the second operand that included processing bit position 0, and the length in general register R2+1 is decremented by the same number. The number of bytes processed of the second operand that included processing bit position 0 is the integer quotient resulting from an integer division with the dividend being the sum of the number of input bits processed and the original value of the SBB, and the divisor being a value of eight. The formation and updating of the addresses and lengths are dependent on the addressing mode, as described below.
In the 24-bit addressing mode, the following apply, in one embodiment:
In the 31-bit addressing mode, the following apply, in one embodiment:
In the 64-bit addressing mode, the following apply, in one embodiment:
In the access-register mode, access registers 1, R1, R2, and R3 specify the address spaces containing the parameter block, first operand, second operand, and circular history buffer, respectively. When DFTCC-CMPR with an in-line history buffer is specified in the access-register mode, access register R2 specifies the address space containing the in-line history. When DFTCC-XPND with an in-line history buffer is specified in the access-register mode, access register R1 specifies the address space containing the in-line history.
Further details regarding the various functions are described below:
Function Code 0: DFLTCC-QAF (Query Available Functions)
The DFLTCC-QAF (query available functions) function provides a mechanism to indicate the availability of the installed functions and installed parameter block formats. One example format of a parameter block for the DFLTCC-QAF function is described with reference to
As an example, bits 0-127 of installed functions vector 342 correspond to function codes 0-127, respectively, of the DEFLATE Conversion Call instruction. When a bit is, e.g., one, the corresponding function is installed; otherwise, the function is not installed.
Further, in one example, bits 0-15 of installed parameter block formats vector 346 correspond to parameter block formats 0-15, respectively for the DFLTCC-GDHT, DFLTCC-CMPR and DFLTCC-XPND functions. When a bit is, e.g., one, the corresponding parameter block format is installed; otherwise, the parameter block format is not installed. In one example, zeros are stored to reserved bytes 16-23 and 26-31 of the parameter block.
Although certain fields are described with respect to parameter block 340, additional, fewer and/or other fields may be included in other embodiments.
The contents of general registers R1, R2, R3, R1+1, and R2+1 are ignored by the DFLTCC-QAF function, in one embodiment.
A PER (program event recording) storage alteration event is recognized, when applicable, for the parameter block. A PER zero address detection event is recognized, when applicable, for the parameter block.
In one example, condition code 0 is set when execution of the DFLTCC-QAF function completes; condition codes 1, 2, and 3 are not applicable to the query function, in one example.
Function Code 1: DFLTCC-GDHT (Generate Dynamic-Huffman Table)
When the DFLTCC-GDHT function is specified, the second operand is used, for instance, as a source to generate a compressed representation of a dynamic-Huffman table (DHT), as specified by the DEFLATE standard.
In one example, the DFLTCC-GDHT function uses a parameter block, an example of which is described with reference to
Additionally, in one example, the parameter block includes one or more preserved fields and one or more reserved fields. Preserved fields are not modified by the DFLTCC-GDHT function. Preserved fields are distinguished from reserved fields to enable a program to initialize a single storage location, use that storage location for the parameter block of a DFLTCC-GDHT function, and subsequently use the same storage location for the parameter block of a DFLTCC-CMPR function. Reserved fields are to contain zeros; otherwise, the program may not operate compatibly in the future. When an operation ends, reserved fields may be stored as zeros or may remain unchanged.
Yet further, some of the fields are used by other functions (e.g., DFLTCC-CMPR or DFLTCC-XPND), and thus, aspects relating to those functions may also be described with the description of those fields.
In one example, a parameter block 360 for the DFLTCC-GDHT function includes the following fields:
Parameter Block Version Number (PBVN) 362: Bytes 0-1 of the parameter block specify the version and size of the parameter block. Bits 0-11 of the PBVN are reserved and should contain zeros; otherwise, the program may not operate compatibly in the future. Bits 12-15 of the PBVN contain an unsigned binary integer specifying the format of the parameter block. The DFLTCC-QAF function provides a mechanism to indicate the parameter block formats available. When the format of the parameter block specified is not supported by the model, a general operand data exception is recognized. The PBVN is specified by the program and is not modified during execution of the instruction.
Model Version Number (MVN) 363: Byte 2 of the parameter block is an unsigned binary integer identifying the model which executed the instruction. The program is not required to initialize the MVN. The MVN is updated during execution of the instruction. The value stored in the MVN is model-dependent.
Dynamic-Huffman Table (DHT) Generation Control (DHTGC) 364: Bit 2 of byte 17 of the parameter block applies to generating a dynamic-Huffman table (DHT). The DHT specifies Huffman codes for symbols representing literal bytes, duplicate string lengths, end-of-block (EOB) symbol, and duplicate string pointer distances. The value of a Huffman code for a particular symbol is a function of the count of occurrences for the entity, which the symbol represents, in the uncompressed form of the data. When the count for a symbol is zero, there is no Huffman code in the DHT for the symbol. The DHTGC specifies counts equal to zero will be treated as follows, in one example:
A DHT which specifies a Huffman code for every possible value of literal bytes, an EOB symbol, duplicate string lengths, and duplicate string pointer distances is called a universal DHT. A DHT which does not specify Huffman codes for values of literal bytes, duplicate string lengths, or duplicate string pointer distances which do not occur in the uncompressed form of the data is called a non-universal DHT.
For all values of the DHTGC, the resulting DHT specifies Huffman codes for all possible duplicate string lengths and pointer distances, as defined by the DEFLATE standard. Therefore, the HLIT (Huffman literal) and HDIST (Huffman distances) sub-elements of the resulting compressed form of the DHT, described further below, each contain the value of, for instance, 29.
The DHTGC is an input to the operation when the DFLTCC-GDHT function is specified. The DHTGC does not apply to the operation when the DFLTCC-CMPR or DFLTCC-XPND function is specified. The DHTGC is not modified, in one embodiment, during execution of the instruction.
Operation Ending Supplemental Code (OESC) 365: Byte 19 of the parameter block is an unsigned binary integer providing additional information on the condition being reported to the program. Since this field is used by multiple functions, some of the conditions refer to fields of a parameter block used by other functions (e.g., the parameter block of
When the operation ends without reporting a general operand data exception, zeros are stored to the OESC field.
Support for supplemental codes other than zero is model-dependent. When multiple conditions exist, it is model-dependent which code, if any, is reported in the OESC field.
Compressed Dynamic-Huffman Table Length (CDHTL) 366: Twelve bits, starting with bit 4 of byte 56 through bit 7 of byte 57, of the parameter block contain an unsigned binary integer which specifies the length, as a bit count, of the compressed format of the DHT in the CDHT field of the parameter block (e.g., CDHT 367).
The CDHTL is an output from the operation when the DFLTCC-GDHT function is specified.
The CDHTL is an input to the operation when the DFLTCC-CMPR function is specified and the Huffman table type (e.g., HTT 376 of
When the DFLTCC-XPND function is specified and the operation ends after decoding only a portion of a block with BTYPE 10 binary, the length of the compressed representation of the DHT in the block is stored to this field. When the DFLTCC-XPND function is specified and the operation ends at a block boundary or after decoding only a portion of a block with BTYPE 00 or 01 binary, zeros are stored to this field. When an uncompressing operation is resumed within a block with BTYPE 10 binary (that is when CF (continuation flag 373 of
Compressed Dynamic-Huffman Table (CDHT) 367: Bytes 64-351 of the parameter block contain a compressed format of a dynamic-Huffman table (DHT).
The DHT specifies Huffman codes (bit sequences) to represent two sets of elements. The elements for one set include literal bytes, an EOB symbol, and duplicate string lengths. The elements for the other set include duplicate string pointer distances. The compressed representation of the DHT defines a set of code lengths and specifies a code length (CL) for each element of each set. The Huffman code for an element expected to be referenced during an operation is derived from the CL specified for that element and the number of elements in the same set with the same specified CL. Specifically, the compressed representation of the DHT includes the following, as an example:
Further details of a compressed representation of a DHT are described below with reference to the description of a compressed data block with block type 10 binary.
In one example, the compressed representation of the DHT is left justified in the CDHT field. That is, the rightmost bit of byte 64 contains the least-significant bit of the HLIT sub-element of the compressed representation of the DHT.
The compressed representation of a DHT is an output from the operation when the DFLTCC-GDHT function is specified.
The compressed representation of a DHT is an input to the operation when the DFLTCC-CMPR function is specified and HTT, described below, is one. The CDHT field is not modified by the DFLTCC-CMPR function.
When the DFLTCC-XPND function is specified and the operation ends after decoding only a portion of a block with BTYPE 10 binary, the compressed representation of the DHT in the block is stored to this field. When the DFLTCC-XPND function is specified and the operation ends at a block boundary or after decoding only a portion of a block with BTYPE 00 or 01 binary, zeros are stored to this field. When an uncompressing operation is resumed within a block with BTYPE 10 binary (that is when CF equals one and IFS equals C or D hex), this field is an input to the operation.
When the CDHT is modified, bits of the field not used to represent the compressed representation of the DHT are stored as zeros.
Although various fields are described above with respect to parameter block 360, additional, fewer and/or other fields may be included in other embodiments.
Aspects of DHT generation are specified by the program to the machine using dynamic-Huffman table generation control (DHTGC) field 364 of the parameter block. It is intended that the source contains uncompressed data and subsequent to completing the operation, the generated result is specified with the DFLTCC-CMPR function to compress the same source.
In one embodiment, there is no history to reference from prior operations while processing the current operation.
When the contents of general register R2+1 specify a length greater than, e.g., 32 K-bytes, the following applies, in one example:
When the contents of general register R2+1 specify a length equal to zero, a specification exception is recognized, and the second operand is not accessed.
The resulting compressed DHT includes a Huffman code representing an end-of-block (EOB) symbol.
A compressed format of the generated DHT is stored to compressed dynamic-Huffman Table (CDHT) field 367 of the parameter block. The length of the compressed format of the generated DHT is stored to CDHTL field 366 of the parameter block.
The operation includes storing a model identification to a model version number field 363 of the parameter block.
When the operation ends without recognizing a general operand data exception, zeros are stored to operation ending supplemental code (OESC) field 365 of the parameter block.
Condition code 0 is set when execution of the DFLTCC-GDHT function completes; condition codes 1, 2 and 3 are not applicable to the DFLTCC-GDHT function.
General registers R2 and R2+1 are not modified by the operation.
The contents of general registers R1, R1+1 and R3 are ignored when the DFLTCC-GDHT function is specified.
A PER zero address detection event is recognized, when applicable, for the second operand location and for the parameter block.
Function Code 2: DFLTCC-CMPR (Compress)
When the DFLTCC-CMPR function is specified, a compressing operation is performed. The operation includes encoding data from the second operand location into compressed data symbols, which are stored to the first operand location.
In one example, the DFLTCC-CMPR function uses a parameter block, an example of which is described with reference to
In one example, parameter block 370 includes:
Parameter Block Version Number (PBVN) 362.
Model Version Number (MVN) 363.
Continuation Flag (CF) 373: Bit 63 of the parameter block, when one, indicates the operation is partially complete and the contents of the continuation state buffer (e.g., in continuation state buffer field 392) may be used to resume the operation. The program is to initialize the continuation flag (CF) to zero and not modify CF in the event the instruction is to be re-executed for the purpose of resuming the operation; otherwise results are unpredictable.
New Task (NT) 374: Bit 0 of byte 16 of the parameter block, when one, indicates the operation applies to the beginning of a compressed data set. Therefore, no history and no check value from a prior operation applies to the current operation. When NT is one at the beginning of the operation, and the operation ends after partial completion, zero is stored to the NT field. When NT is zero, history and a check value from a prior operation apply to the current operation.
Check Value Type (CVT) 375: Bit 2 of byte 16 of the parameter block specifies the type of check value contained in the check value field of the parameter block (e.g., field 387). When CVT is zero, the check value type is, e.g., a 32-bit cyclic-redundancy-check (CRC-32). When CVT is one, the check value type is, e.g., a 32-bit Adler checksum (Adler-32). The CVT bit is not modified during execution of the instruction.
Huffman Table Type (HTT) 376: Bit 4 of byte 16 of the parameter block, when zero, specifies a table containing fixed-Huffman codes (FHT), as defined by the DEFLATE standard, is used during a compression operation. When the HTT is one, a table containing dynamic-Huffman codes (DHT), as specified in the CDHT field of the parameter block, is used during a compression operation. The HTT does not apply to decompressing operations. The HTT bit is not modified during execution of the instruction.
Block Continuation Flag (BCF) 377: Bit 5 of byte 16 of the parameter block applies when the DFLTCC-CMPR function is specified. When zero, a 3-bit block header, and when applicable, the compressed format of a dynamic-Huffman table, as specified in the CDHT field of the parameter block (e.g., field 367), is stored to the first operand location prior to storing any compressed data elements. When one, neither a block header nor a compressed format of a DHT is stored to the first operand location. When NT is one, BCF is treated as equal to zero. The BCF bit is not modified during execution of the instruction.
Block Closing Control (BCC) 378: Bit 6 of byte 16 of the parameter block applies when the DFLTCC-CMPR function is specified. When one, subsequent to storing all compressed data symbols, an end-of block (EOB) symbol is stored to the first operand location. When the HTT specifies using an FHT, Huffman code 0000000 binary (which corresponds to the intermediate integer representation of 256 in the table specifying codes for literal bytes, an EOB symbol, and duplicate string lengths), as an example, is used for the EOB symbol. When the HTT specifies using a DHT, the Huffman code for the EOB symbol is specified in the DHT. When the BCC bit is zero, an EOB symbol is not stored to the first operand location. The BCC bit is not modified during execution of the instruction.
Block Header Final (BHF) 379: Bit 7 of byte 16 of the parameter block applies when the DFLTCC-CMPR function is specified and either BCF 377 is zero or NT 374 is one; otherwise the BHF does not apply. When applicable and one, the first bit of the block header (BFINAL) is set to one before storing the block header to the first operand location. When applicable and zero, the first bit of the block header (BFINAL) is set to zero before storing the block header to the first operand location. The BHF bit is not modified during execution of the instruction.
DHT Generation Control (DHTGC) 364: The DHTGC does not apply to the operation when the DFLTCC-CMPR function is specified. The DHTGC is not modified during execution of the instruction.
Sub-Byte Boundary (SBB) 381: Bits 5-7 of byte 18 of the parameter block contain an unsigned binary integer specifying the boundary between processed and unprocessed bits within a byte of the compressed data stream. The byte of the stream referenced is the last byte referenced, meaning the rightmost byte, when an operation ends, and is the first byte to be referenced, meaning the leftmost byte, when an operation begins or resumes. When the DFLTCC-CMPR function is specified, the SBB applies to the byte designated by the first operand address. When the DFLTCC-XPND function is specified, the SBB applies to the byte designated by the second operand address. The SBB specifies the number of rightmost bits that have been processed. The SBB is an input to the operation and an output of the operation.
One example of a compressed data stream when SBB has a value of 011 binary is depicted in
Further,
Returning to
Operation Ending Supplemental Code (OESC) 365.
Incomplete Function Status (IFS) 383: Bits 4-7 of byte 21 of the parameter block contain status information when certain operations end. When a decompressing operation ends, the IFS conveys information about the second operand as follows, in on example:
In one embodiment, an uncompressing operation may end with IFS equal 0000 binary and not satisfy normal completion. In such cases, the operation ends with condition code 1 or 3 set.
When a compressing operation ends, the IFS field is undefined, but may be modified.
The IFS is not an input to the operation.
Incomplete Function Length (IFL) 384: Bytes 22-23 of the parameter block contain length information when certain operations end. For a decompressing operation, the IFL applies to the second operand. When a decompressing operation ends after decoding some, but not all of a block with BTYPE equal 00 binary, the IFL contains an unsigned binary integer specifying the number of bytes of the block in the second operand, which have not yet been processed. Bytes 22-23 contain the IFL in, for instance, big-endian byte order unlike the LEN field of a block with BTYPE equal 00 binary, which is in, for instance, little-endian byte order.
When a decompressing operation ends after decoding a complete block with BTYPE equal 00 binary and BFINAL equal to one, zeros are stored to the IFL field. When a decompressing operation ends after decoding some, but not all of a block with a non-zero BTYPE, or ends at a block boundary, the IFL field is undefined, but may be modified.
When a compressing operation ends, the IFL field is undefined, but may be modified.
The IFL is not an input to the operation.
History Length (HL) 385: Bytes 44-45 of the parameter block contain an unsigned binary integer specifying the number of bytes of history in the history buffer which can be referenced during an operation. The HL applies to in-line and circular history buffers. When new task (NT) equals one, no history applies to the beginning of the operation and the history length is treated as zero as an input to the operation.
A general operand data exception is recognized when the history length is greater than, e.g., 32,768 and NT equals zero.
The history length is modified during compressing and decompressing operations. When the sum of the original HL and the number of uncompressed data bytes processed during the operation is less than, or equal to, e.g., 32,768, the updated HL is equal to the sum of the original HL and the number of uncompressed data bytes processed during the operation; otherwise the updated HL is equal to the value of 32,768.
History Offset (HO) 386: Fifteen bits, starting with bit 1 of byte 46, through bit 7 of byte 47, of the parameter block, contain an unsigned binary integer specifying an offset in the third operand when the history buffer type is circular. The sum of the contents of R3 and the history offset designates the location of the first byte of history within the circular history buffer, which is the least recently processed byte of uncompressed data in the buffer. When the history buffer type is circular, history offset is an input to the operation and is updated at the end of the operation. When the sum of the original HL and the number of uncompressed data bytes processed during the operation is less than, or equal to, e.g., 32,768, the updated HO is equal to the original HO; otherwise, the updated HO is equal to the sum of the original HO, the original HL, and the number of uncompressed data bytes processed during the operation, modulo 32,768.
When the history buffer type is in-line, the HO field of the parameter block is undefined, but may be modified.
Check Value 387: Bytes 48-51 of the parameter block contain a check value. As part of the operation, a check value is generated. The check value applies to the uncompressed data operand. That is, the check value applies to the second operand for the DFLTCC-CMPR function and applies to the first operand for the DFLTCC-XPND function. When CVT bit 375 is zero, a, e.g., 32-bit cyclic-redundancy-check check value (CRC-32) is generated. When the CVT bit is one, a, e.g., 32-bit Adler checksum check value (Adler-32) is generated.
The inputs to generating a check value are, for instance, a 4-byte base and the uncompressed data processed during the operation. The base input provides the means to compute a single and consistent check value for a set of compressed data blocks, regardless of the number of times the DFLTCC instruction is executed to process the complete set of compressed data blocks. When the NT bit is zero, the original value in the check value field is used for the base input in generating a check value.
When an Adler-32 check value is generated, the following apply, in one example:
When a CRC-32 check value is generated, the following apply, in one embodiment:
In one example, the check value is only meaningful to the program when the operation ends with condition code 0 set; otherwise, the check value is only an intermediate result and only meaningful to resume the operation. When the DFLTCC-CMPR function is specified and the operation ends with condition code 1, 2, or 3 set some bytes to the left of the byte designated by the second operand address may not be included in the computation of the resulting check value. When the DFLTCC-XPND function is specified and the operation ends with condition code 1, 2, or 3 set, some result bytes not yet stored to the right of the byte designated by the first operand address may already be included in the computation of the resulting check value.
End-Of-Block Symbol (EOBS) 388: Fifteen bits, starting with bit 0 of byte 52, through bit 6 of byte 53, of the parameter block, contain an end-of-block (EOB) symbol. The end-of-block length (EOBL) field 389 of the parameter block specifies the length of the EOB symbol in the EOBS field. The EOB symbol is left justified in the EOBS field. Bits of the EOBS field not occupied by the EOB symbol are stored as zeros. The EOBS field is an output of the operation when compressing data, regardless of which type of Huffman table applies. The EOBS field is not used as an input to the operation.
Bit 0 of byte 52 contains the most significant bit of the EOB symbol. When the length of the EOB symbol is 7 bits, bit 6 of byte 52 contains the least significant bit of the EOB symbol. When the length of the EOB symbol is 15 bits, bit 6 of byte 53 contains the least significant bit of the EOB symbol.
For blocks using a FHT, the EOB symbol is 0000000 binary, as defined by the DEFLATE standard. For blocks using a DHT, the EOB symbol is defined by the DHT. The EOB symbol is conveyed in order to provide the capability for the program to close a block.
The EOBS field is undefined when the DFLTCC-XPND function is specified but may be modified.
End-Of-Block Length (EOBL) 389: Bits 0-3 of byte 54 of the parameter block contain an unsigned binary integer specifying the length of the end-of-block (EOB) symbol in EOBS field 388 of the parameter block. The length specifies the number of bits which the EOB symbol occupies in the EOBS field. The EOBL field is an output of the operation when compressing data, regardless of which type of Huffman table applies. The EOBL field is not used as an input to the operation.
The EOBL field is undefined when the DFLTCC-XPND function is specified, but may be modified.
Compressed Dynamic-Huffman Table Length (CDHTL) 366.
Compressed Dynamic-Huffman Table (CDHT) 367: The compressed representation of a DHT is an input to the operation when the DFLTCC-CMPR function is specified and HTT is one. The CDHT field is not modified by the DFLTCC-CMPR function.
Continuation State Buffer (CSB) 392: When conditions cause a value of one to be stored in CF field 373, internal state data is stored to bytes 384-1535 of the parameter block; otherwise, bytes 384-1535 of the parameter block are undefined and may be modified. The internal state data stored is model-dependent and may be used subsequently to resume the operation. It is expected, but not required, for the program to initialize the continuation state buffer to contain, e.g., all zeros. Subsequent to the instruction ending with a non-zero condition code set, and prior to re-executing the instruction for the purpose of resuming the operation, the program should not modify the continuation state buffer; otherwise results are unpredictable.
Although various fields are described above with respect to parameter block 370, additional, fewer and/or other fields may be included in other embodiments.
One example of the compress operation is described below with respect to Compressing Data.
Normal completion for the DFLTCC-CCMPR function occurs when the entire second operand is compressed and stored to the first operand location. When the operation ends due to normal completion, the following occurs, in one example:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
When normal completion occurs, CSB field 392 of the parameter block is undefined after the operation ends.
When a CPU-determined number of bytes have been processed, the operation ends and the following occurs, in one example:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
The CPU-determined number of bytes depends on the model, and may be a different number each time the instruction is executed.
Subsequent to the instruction ending with condition code 3 set, it is expected the program does not modify any input or output specification for the instruction and branches back to re-execute the instruction to resume the operation.
In certain situations, despite ending the instruction with condition code 3 set, the parameter block and general registers are not updated. These situations may occur when the CPU performs a quiescing operation or CPU retry while executing the DEFLATE Conversion Call instruction. In these cases, the CPU-determined number of bytes processed is zero, data may have been stored to the first operand location, data may have been stored to the third operand location, when applicable, and corresponding change bits have been set.
The first operand length is insufficient to complete the operation when any of the following conditions apply, in one example:
In one example, the first operand length is zero when the content of general register R1+1 is zero, regardless of the values in the NT and SBB fields of the parameter block.
When the first operand length becomes equal to zero during execution of the instruction, the operation ends and the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
When the first operand length is zero at the beginning of the execution of the instruction, the operation ends and the following occurs, in one embodiment:
Subsequent to the instruction ending with condition code 1 set, it is expected that the program modifies the first operand length, first operand address, or both and re-executes the instruction to resume the operation.
A PER storage alteration event is recognized, when applicable, for the following:
When the entire parameter block overlaps the PER storage area designation, a PER storage alteration event is recognized, when applicable, for the parameter block. When only a portion of the parameter block overlaps the PER storage area designation, it is model-dependent which of the following occurs:
A PER zero-address-detection event is recognized when applicable, for the parameter block, first operand location, second operand location, and third operand location when the HBT is one (circular).
Condition code 2 is not applicable to the DFLTC-CCMPR function.
When the instruction ends with condition code 1 or 3 set, input data referenced from the second operand location may be completely, or only partially, processed. When input data is only partially processed, results in the first operand location, first operand address, first operand length, and SBB field of the parameter block do not represent a state consistent with the updated second operand address and length. In these cases, partially processed data and internal-state information may be placed in the CSB field of the parameter block. The amount of partially processed data depends on conditions existing at the time the operation ends and the model. Although some data may only be partially processed, results stored to the left of the location designated by the updated first operand address are complete and will not be modified when the operation resumes. Furthermore, it is expected that the program subsequently re-executes the instruction to resume the operation, at which time the contents of the CSB field are referenced prior to resuming the operation. When the instruction ends with condition code 0 set, all data is completely processed and all results associated with input and output data represent a consistent state.
Subsequent to the instruction ending with a non-zero condition code set, and prior to re-executing the instruction for the purpose of resuming the operation, the program should not modify any fields of the parameter block; otherwise results are unpredictable.
Function Code 4: DFLTCC-XPND (Expand)
When the DFLTCC-XPND function is specified, an uncompressing operation is performed. The operation includes decoding compressed data symbols from the second operand location into uncompressed data, which is stored to the first operand location.
In one example, the DFLTCC-XPND function uses a parameter block, an example of which is described above with respect to
One example of the DFLTCC-XPND operation is described below with respect to Uncompressing Data.
Normal completion occurs when all elements of the final block of the data set in the second operand are decoded and all uncompressed data is stored to the first operand location. The last block of the data set is identified when the BFINAL bit of the block header is one. When the operation ends due to normal completion, the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
When normal completion occurs, CSB field 392 of the parameter block is undefined after the operation ends.
When a CPU-determined number of bytes have been processed, the operation ends and the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
The CPU-determined number of bytes depends on the model, and may be a different number each time the instruction is executed.
Subsequent to the instruction ending with condition code 3 set, it is expected the program does not modify any input or output specification for the instruction and branches back to re-execute the instruction to resume the operation.
In certain situations, despite ending the instruction with condition code 3 set, the parameter block and general registers are not updated. These situations may occur when the CPU performs a quiescing operation or CPU retry while executing the DEFLATE Conversion Call instruction. In these cases, the CPU-determined number of bytes processed is zero, data may have been stored to the first operand location, data may have been stored to the third operand location, when applicable, and corresponding change bits have been set.
The second operand length is insufficient to complete the operation when the following applies, for instance:
When the second operand length is insufficient to complete the operation, the operation has been partially completed, the operation ends, and the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
Subsequent to the instruction ending with condition code 2 set, it is expected that the program modifies the second operand length, second operand address, or both and re-executes the instruction to resume the operation.
The first operand length is insufficient to complete the operation when the following applies, for instance:
When the first operand length is insufficient to complete the operation, the operation has been partially completed, the operation ends, and the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
Subsequent to the instruction ending with condition code 1 set, it is expected that the program modifies the first operand length, first operand address, or both and re-executes the instruction to resume the operation.
A PER storage-alteration event is recognized, when applicable, for the following:
In one example, when the entire parameter block overlaps the PER storage area designation, a PER storage alteration event is recognized, when applicable, for the parameter block. When only a portion of the parameter block overlaps the PER storage area designation, it is model-dependent which of the following occurs, in one embodiment:
A PER zero-address-detection event is recognized when applicable, for the parameter block, first operand location, second operand location, and third operand location when the HBT is one (circular).
When the instruction ends with condition code 1, 2, or 3 set, input data referenced from the second operand location may be completely, or only partially, processed. When input data is only partially processed, results in the first operand location, first operand address, first operand length, SBB field of the parameter block, check value field of the parameter block, HL field of the parameter block, IFS field of the parameter block, and when applicable, the third operand location and HO field of the parameter block, do not represent a state consistent with the updated second operand address and length. In these cases, partially processed data and internal state information may be placed in the CSB field of the parameter block. The amount of partially processed data depends on conditions existing at the time the operation ends and the model. Although some data may only be partially processed, results stored to the left of the location designated by the updated first operand address are complete and will not be modified when the operation resumes. Furthermore, it is expected that the program subsequently re-executes the instruction to resume the operation, at which time the contents of the CSB field are referenced prior to resuming the operation. When the operation ends with condition code 0 set, all data is completely processed and all results associated with input and output data represent a consistent state.
Subsequent to the instruction ending with a non-zero condition code set, and prior to re-executing the instruction for the purpose of resuming the operation, the program should not modify any fields of the parameter block; otherwise results are unpredictable.
Compressed Data Blocks
In one example, the bytes of a compressed data block in storage are processed from, e.g., left to right. Compressed data blocks may, or may not, start or end on byte boundaries. A compressed data block is, for instance, a bit stream. Elements of the block are loaded into storage one bit at a time. The bit stream is loaded, e.g., from right to left within each byte of storage and in byte order from, e.g., left to right. When the element is a Huffman code, the bits are stored in order from, e.g., most significant bit to least significant bit of the element. When the element is not a Huffman code, the bits are stored in order from, e.g., least significant bit to most significant bit of the element.
1. A Huffman code of variable length. The most significant bits of the code designate the length of the code. The code is encountered in the bit stream starting with the most significant bit of the code and ending with the least significant bit of the code. When the code represents a literal value or the end-of-block symbol, the code is the only sub-element of the compressed data symbol. When the code represents a length of a pointer to the history buffer, the code is followed by subsequent sub-elements of the compressed data symbol.
2. When applicable, as specified by the DEFLATE standard, extra length bits may follow the Huffman code representing a pointer length. Extra length bits are encountered in the bit stream starting with the least significant bit and ending with the most significant bit of the extra length bits.
3. The next sub-element encountered in the bit stream is a 5-bit distance code of a pointer to the history buffer. The distance code is encountered in the bit stream starting with, e.g., the most significant bit of the code and ending with the least significant bit of the distance code.
4. When applicable, as specified by the DEFLATE standard, extra distance bits may follow the distance code. Extra distance bits are encountered in the bit stream starting with the least significant bit and ending with the most significant bit of the extra distance bits.
1. HLIT: The sum of the 5-bit HLIT sub-element and 257 specifies the number of Huffman codes representing literal bytes, an EOB symbol, and duplicate string lengths. Valid values of HLIT range from, e.g., 0 to 29. HLIT bits are encountered in the bit stream starting with the least significant bit and ending with the most significant bit of the HLIT sub-element. In this example, bit 1 of byte 0, identified as b3 is the least significant bit of the HLIT sub-element.
2. HDIST: The sum of the 5-bit HDIST sub-element and 1 specifies the number of Huffman codes representing duplicate string pointer distances. Valid values of HDIST range from, e.g., 0 to 29. HDIST bits are encountered in the bit stream starting with the least significant bit and ending with the most significant bit of the HDIST sub-element.
3. HCLEN: The sum of the 4-bit HCLEN sub-element and 4 specifies the number of Huffman codes representing code lengths. Valid values of HCLEN range from, e.g., 0 to 15. HCLEN bits are encountered in the bit stream starting with the least significant bit and ending with the most significant bit of the HCLEN sub-element.
4. A sequence of codes specifying a bit length for each of the code lengths defined for the compressed DHT. The number of codes is equal to the sum of HCLEN and 4. Each code is 3 bits.
5. A sequence of codes specifying a code length for each of the elements of the set consisting of literal bytes, an EOB symbol, and duplicate string lengths. The number of code lengths specified is equal to the sum of HLIT and 257.
When the last code length (CL) for the set of literal bytes, an EOB symbol, and duplicate string lengths is 16, 17, or 18, and the extra bits following the CL specify repeating the CL for more elements than are defined for the set, the code length also applies to the set of duplicate string pointer distances. The sequence of codes specifying code lengths for the set of literal bytes, an EOB symbol and duplicate string lengths, followed by the sequence of codes specifying code lengths for duplicate string pointer distances is a contiguous sequence for both sets.
6. A sequence of codes specifying a code length for each of the elements of the set consisting of duplicate string pointer distances. The number of code lengths specified is equal to the sum of HDIST and 1.
1. A Huffman code of variable length. The most significant bits of the code designate the length of the code. The code is encountered in the bit stream starting with the most significant bit of the code and ending with the least significant bit of the code. When the code represents a literal value or the end-of-block symbol, the code is the only sub-element of the compressed data symbol. When the code represents a length of a pointer to the history buffer, the code is followed by subsequent sub-elements of the compressed data symbol.
2. When applicable, as specified by the DEFLATE standard, extra length bits may follow the Huffman code representing a pointer length. Extra length bits are encountered in the bit stream starting with, e.g., the least significant bit and ending with the most significant bit of the extra length bits.
3. The next sub-element encountered in the bit stream is a 5-bit distance code of a pointer to the history buffer. The distance code is encountered in the bit stream starting with, e.g., the most significant bit of the code and ending with the least significant bit of the distance code.
4. When applicable, as specified by the DEFLATE standard, extra distance bits may follow the distance code. Extra distance bits are encountered in the bit stream starting with, e.g., the least significant bit and ending with the most significant bit of the extra distance bits.
In the above descriptions of the various block types, certain constant values are specified, as well as specific bits, bytes, directions, etc. These are only examples. Other constant values, bits, bytes, directions, etc. may be specified in other embodiments.
Processing a Compressed Data Set
Examples of processing a compressed data set to illustrate example uses of the DEFLATE Conversion Call instruction and augment the descriptions of various fields of the parameter block are provided. The examples do not describe all possible scenarios, requirements, and capabilities, but illustrate various of the scenarios, requirements and/or capabilities. The examples and descriptions apply, for instance, to a compressed data set in storage, an example of which is illustrated in
For the examples described herein, it is intended for a program processing the compressed data set to consider the following, in one embodiment:
Similarly,
Moreover, referring to
Compressing Data
The process of compressing data includes generating one or more compressed data blocks. The compress function of the DEFLATE Conversion Call instruction is used to construct a portion of an individual block. The portion may be the entire block. This function generates portions of a block with block type (BTYPE) 01 or 10 binary, and not 00 binary. When the new task bit (NT) of the parameter block is one, the first block of compressed data is generated and there is no history to reference from previously performed compressing operations.
In one example, an individual block contains the following elements in the order which they are listed:
1. Final block indication (BFINAL).
2. Block type (BTYPE).
3. Compressed format of a dynamic-Huffman table, when applicable.
4. Compressed data symbols.
5. End-of-block (EOB) symbol.
The compression operation generates the elements specified in the order defined for a block. The elements may begin or end between byte boundaries in storage. The sub-byte boundary (SBB) applies to storing of the first element to the first operand location. A compressed data block is a bit stream. Components of the block are loaded into storage one bit at a time. As an example, the bit stream is loaded from right to left within each byte of storage and in byte order from left to right.
When the SBB is non-zero, the reference to the first byte at the first operand location is an update reference.
Uncompressed data from the second operand location is compressed and stored as compressed data symbols to the first operand location.
When the first operand length is zero at the beginning of the execution of the instruction, the first operand is not accessed, and the first operand address and first operand length in general registers R1 and R1+1, respectively, are not changed. This applies when the value of CF field 373 (
When the second operand length is zero at the beginning of execution of the instruction, the second operand is not accessed, and the second operand address and second operand length in general registers R2 and R2+1, respectively, are not changed. The second operand length is zero at the beginning of execution of the instruction for the following case, as an example:
In one embodiment, the program is not to use the DEFLATE Conversion Call instruction to perform the following operations:
The compression algorithm includes searching an updated history of recently compressed data for a string of bytes which matches data currently being compressed from the second operand location. Before the compression operation begins or resumes, the following applies, in one embodiment:
During the compression operation, fetch-type references to the entire history may be made, regardless of which bytes of history are used to perform the operation. Furthermore, when the history buffer type is circular, fetch-type references to the entire 32 K-byte history buffer may be made, regardless of which bytes of history are used to perform the operation.
During the compression operation, history is updated. Subsequent to encoding one or more bytes of source data into a compressed data symbol without encountering a general operand data exception condition, the source bytes are concatenated to the end of the history. The most recently processed bytes of source data, up to a maximum of 32 K-bytes, constitute the updated history available to reference while processing subsequent bytes of source data.
When the compression operation ends, the following applies, in one example, to the resulting history available to subsequently resume the operation, or begin another operation:
As examples,
When the HBT (history buffer type) specified by bit 56 of general register 0 is circular (e.g., bit 310=1), the history is maintained in, for instance, a 32 K-byte buffer located at the third operand location. The location of the first byte of history within the buffer (HB) is designated by, e.g., the sum of the contents of general register R3 and history offset (HO) 386 (
HE=R3+modulo32K(HO+HL−1)
The last byte of history is the most recently processed byte of uncompressed data in the buffer. When the sum of history offset (HO) 386 (
As examples,
When the HBT is circular and the number of bytes processed from the second operand location is less than, e.g., 32,768, the following applies, in one example:
The range of bytes includes and starts with the location designated by, for instance:
R3+modulo32K(HOO+HLO), where
HOO: The history offset before the instruction executes.
HLO: The history length before the instruction executes.
The range of bytes includes and ends with the location designated by, e.g.:
R3+modulo32K(HOO+HLO+BP−1), where
BP: The number of bytes processed from the second operand location during execution of the instruction.
Stores made to the range of bytes just described are subject to store-type access exceptions, PER storage alteration events, and setting change bits, as an example.
When the HBT is circular and the number of bytes processed from the second operand location is more than, or equal to, e.g., 32,768, stores are made to all bytes of the third operand location and subject to store type access exceptions, PER storage alteration events, and setting change bits.
When the block continuation flag (BCF) 377 is zero, a 3 bit block header, including BFINAL followed by BTYPE, is stored to the first operand location. The BFINAL bit of the block header is set equal to the block header final bit (BHF) 379 of the parameter block. When the Huffman table type (HTT) 376 is zero, the BTYPE field of the block header is set to, e.g., 01 binary and when the HTT is one, the BTYPE field of the block header is set to, e.g., 10 binary. When a block header is stored, the BFINAL bit is stored to the bit specified by the SBB in the first byte of the first operand. Subsequently, the BTYPE is stored to the first operand location. When the BCF is one, a block header is not stored.
When the Huffman table type (HTT) is one, the compressed format of the dynamic-Huffman table (DHT) 367 specified in the parameter block is examined for general operand data exception conditions. When a general operand data exception condition exists for the specified compressed format of the DHT, the compressed DHT is referred to as invalid and is not to be used to compress data. Example definitions of general operand data exception conditions are described further below. When the compressed format of the DHT specifies a bit length for a code length, or a code length for a literal byte, the EOB symbol, a duplicate string length, or a duplicate string pointer distance, which is greater than the length required by the Huffman algorithm to specify a proper and functional Huffman tree, the compressed DHT is still used to derive a functional DHT and compress data. When the block continuation flag (BCF) is zero and the HTT is one, the compressed format of the DHT, as specified in CDHT field 367 of the parameter block is stored to the first operand location.
During the compression operation, source data from the second operand location is encoded into compressed data symbols. As part of the encoding, source data is compared to the history. When no match is found, the intermediate representation of the source data is literal bytes, which is the same as the source data. When a match is found, the intermediate representation of the source data is a pointer to a location within the history which contains a duplicate copy of the source data. A pointer consists of a length and a distance. The length is the number of source data bytes which match a string in the history. The distance is the number of bytes from the end of the history to the beginning of the string which matches the source data. Two Huffman code trees from the Huffman table are used, in one example, to encode the intermediate representation of the source data into compressed data symbols. When the Huffman table type (HTT) is zero, a fixed-Huffman table (FHT), as described by the DEFLATE standard, specifies the two Huffman code trees used for encoding intermediate results. When HTT 376 is one, the dynamic-Huffman table (DHT), which is derived from the compressed representation of the DHT, specified in CDHT field 367 of the parameter block, specifies the two Huffman code trees used for encoding intermediate results. The encoding is performed as described by the DEFLATE standard. When a non-universal DHT is used which does not specify a Huffman code to be used to encode the intermediate representation of the source data, a general operand data exception is recognized. The bits of the resulting compressed data symbol are arranged in the order specified by the DEFLATE standard before storing the result to the first operand location.
In one example, duplicate-string lengths range from 3 to 258 bytes.
Prior to processing further source data, the history is updated, as described herein.
The process is repeated, in one example, until all source bytes have been processed.
After the source bytes (e.g., all source bytes) have been processed and the block closing control (BCC) 378 is one, an end-of-block (EOB) symbol is stored to the first operand location. When a fixed-Huffman table is used, Huffman code 0000000 binary is used for the EOB symbol. When a dynamic-Huffman table (DHT) is used, the Huffman code used for the EOB symbol is specified by the DHT. The bits of the EOB symbol are arranged in the order specified by the DEFLATE standard before storing the EOB symbol to the first operand location.
When the last compressed data symbol of the operation (including the EOB symbol), only occupies a portion of the last byte to store, the bits that do not contain a portion of the last symbol are stored as zeros, in one example.
Subsequent to processing the last compressed data symbol, the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
Coincident with compressing the source data, the source data is an input to generating a 32-bit check value, described above. The resulting check value is stored to check value field 387 of the parameter block.
Uncompressing Data
In one embodiment, the expand function of the DEFLATE Conversion Call instruction is used to decode a compressed data set into uncompressed data. The compressed data set in the second operand location includes one or more consecutive compressed data blocks. The blocks of the data set are processed from left to right, in one example, and the bytes of a block are processed, e.g., from left to right. The blocks may or may not start or end on byte boundaries. Each block is decoded independent of other blocks in the data set. General register R2 specifies the logical address of the leftmost byte of the first block in the data set. The last block in the data set is the block encountered during processing with the BFINAL bit equal to one. In one example, there are three types of blocks to process. The technique of decoding the contents of a block is a function of the block type (BTYPE).
When the operation begins (e.g., when continuation flag field 373 of the parameter block is zero), the bit designated by general register R2, new task (NT) field 374, and sub-byte boundary (SBB) field 381 is interpreted as the first bit of a compressed data block (the BFINAL bit of a block header).
The expand function includes referencing an updated history of recently decoded uncompressed data. Before the uncompressing operation begins or resumes, the following applies, in one embodiment:
During the operation, fetch-type references to the entire history may be made, regardless of which bytes of history are used to perform the operation. Furthermore, when the history buffer type is circular, fetch-type references to the entire history buffer (e.g., 32 K-byte) may be made, regardless of which bytes of history are used to perform the operation.
During the uncompressing operation, history is updated. Subsequent to decoding source data without encountering a general operand data exception condition, the resulting bytes of uncompressed data are concatenated to the end of the history. The most recently decoded bytes of uncompressed data, up to a maximum of, e.g., 32 K-bytes, constitute the updated history available to reference while processing subsequent source data.
When the uncompressing operation ends, the following applies to the resulting history available to subsequently resume the operation, or begin another operation, in one example:
As examples,
When the HBT specified by bit 56 of general register 0 is circular, the history is maintained in, for instance, a 32 K-byte buffer located at the third operand location. The location of the first byte of history within the buffer (HB) is designated by the sum of the contents of general register R3 and history offset (HO) 386. The first byte of history is the least recently processed byte of uncompressed data in the buffer. The location of the last byte of history within the buffer (HE) is designated by, e.g., the following equation:
HE=R3+modulo32K(HO+HL−1).
The last byte of history is the most recently processed byte of uncompressed data in the buffer. When the sum of the history offset (HO) and history length (HL) exceeds the size of the third operand (e.g., 32 K-bytes), the history wraps from the end of the third operand to the beginning of the third operand.
When the HBT is circular and the number of bytes stored to the first operand location is less than, e.g., 32,768, the following applies, in one example:
HOO: The history offset before the instruction executes.
HLO: The history length before the instruction executes.
The range of bytes includes and ends with the location designated by, e.g.,
R3+modulo32K(HOO+HLO+BP−1), where
BP: The number of bytes stored to the first operand location during execution of the instruction.
Stores made to the range of bytes just described are subject to store-type access exceptions, PER storage-alteration events, and setting change bits.
When the HBT is circular and the number of bytes stored to the first operand location is more than, or equal to, e.g., 32,768, stores are made to, e.g., all bytes of the third operand location and subject to store-type access exceptions, PER storage-alteration events, and setting change bits.
When the BTYPE is 00 binary, the block does not contain compressed data.
When the BTYPE is 01 binary, the block contains compressed data symbols that were generated using a fixed-Huffman table (FHT). The FHT is defined by the DEFLATE standard and is not part of the block.
Prior to processing further source data, the history is updated as previously described.
The updated history applies to decoding the next symbol of the block. When the EOB symbol is encountered, processing of the block is complete.
When the BTYPE is 10 binary, the block contains compressed data symbols that were generated using a dynamic-Huffman table (DHT). A compressed format of the DHT used is an element of the compressed data block.
Coincident with uncompressing the second operand, the uncompressed data is an input to generating a check value (e.g., a 32-bit check value). The resulting check value is stored to check value field 387 of the parameter block.
Subsequent to processing the last block of the data set, the following occurs, in one embodiment:
The formation and updating of the addresses and lengths are dependent on the addressing mode.
When the first operand length is zero at the beginning of the execution of the instruction, the first operand is not accessed, and the first operand address and first operand length in general registers R1 and R1+1, respectively, are not changed. This applies when the value of CF field 373 is zero or one at the beginning of the execution of the instruction.
When the second operand length is zero at the beginning of the execution of the instruction, the second operand is not accessed, and the second operand address and second operand length in general registers R2 and R2+1, respectively, are not changed. The second operand length is zero at the beginning of the execution of the instruction for the following case, in one embodiment:
The uncompressing operation may end without storing any results to the first operand location, even though data was processed from the second operand location. This occurs when the data processed from the second operand location only contains any of the following compressed data block elements, in an example:
The following conditions apply to executing the DEFLATE Conversion Call instruction, in one or more embodiments:
A general operand data exception is recognized when the DFLTCC-GDHT function is specified and the following condition occurs, in one example:
A general operand data exception is recognized when the DFLTCC-CMPR function is specified and any of the following conditions occur, in one example:
A general operand data exception is recognized when, for instance, the DFLTCC-XPND function is specified and any of the following conditions occur, as examples:
When a general operand data exception is recognized, the operation is considered suppressed, even though operation ending supplemental code (OESC) 365 and model version number (MVN) fields 363 of the parameter block are updated to provide additional information associated with the exception.
When a DFLTCC-CMPR or DFLTCC-XPND function is being executed and a general operand data exception is due to be recognized for the second operand, the result is that either the exception is recognized, or the operation ends with partial completion and condition code, e.g., 3 is set. If condition code 3 is set, the exception will be recognized when the instruction is executed again to continue processing the same operands and the exception condition still exists.
Other Conditions include, for instance:
The execution of the instruction is interruptible. When an interruption occurs, the addresses in general registers R1 and R2, the lengths in general registers R1+1 and R2+1, and specific fields of the parameter block are updated, so that the instruction, when reexecuted, resumes at the point of interruption.
When a DFLTCC-CMPR or DFLTCC-XPND function is being executed and an access exception is due to be recognized for the first or second operand, the result is that either the exception is recognized, or the operation ends with partial completion and condition code, e.g., 3 is set. If condition code 3 is set, the exception will be recognized when the instruction is executed again to continue processing the same operands and the exception condition still exists.
As observed by this CPU, other CPUs, and channel programs, references to the parameter block, first, second, and third operands may be multiple-access references, accesses to these storage locations are not necessarily block concurrent, and the sequence of these accesses or references is undefined.
Results are unpredictable if the DFLTCC-CMPR or DFLTCC-XPND function is specified and any of the following apply, in one embodiment:
In certain situations, despite ending the execution of the DEFLATE Conversion Call instruction with a CPU-determined number of bytes processed being zero, data may have been stored to the first operand location, data may have been stored to the third operand location, when applicable, and corresponding change bits have been set, when applicable. In these cases, the contents of the parameter block and general registers have not been modified from original values. These situations may occur when the CPU performs a quiescing operation or a CPU retry while executing the DEFLATE Conversion Call instruction.
The following are example Resulting Condition Codes from executing the DEFLATE Conversion Call instruction:
0 Normal completion
1 The first operand length is insufficient to complete the operation
2 The second operand length is insufficient to complete the operation (DFLTCC-XPND)
3 CPU-determined amount of data processed
Program Exceptions:
Example priorities of execution for the DEFLATE CONVERSION CALL instruction are shown below:
1.-6. Exceptions with the same priority as the priority of program-interruption conditions for the general case.
7.A Access exceptions for second instruction halfword.
7.B Operation exception.
7.0 Transaction constraint.
8.A Specification exception due to invalid function code or invalid register number.
8.B Specification exception due to parameter block not designated on a 4 K-byte boundary.
8.0 Specification exception due to circular history buffer not designated on a 4 K-byte boundary.
9. Access exceptions for an access to the parameter block.
10. General operand data exception when specified format of the parameter block is not supported by the mode.
11. Specification exception due to second operand length equal to zero and CF equal to zero at the beginning of the execution of the instruction.
12. Condition code 1 due to first operand length equal to zero at the beginning of the execution of the instruction and DFLTCC-CMPR is specified.
13.A General operand data exception due to the history length field greater than 32,768 and the new task field is zero when DFLTCC-CMPR or DFLTCC-XPND is specified.
13.B Access exceptions for an access to the first operand and the first operand length is non-zero.
13.0 Access exceptions for an access to the second operand and the second operand length is non-zero.
13.D Access exceptions for an access to in-line history specified at the beginning of the execution of the instruction.
13.E Access exceptions for an access to the third operand.
14.A General operand data exception due to conditions other than those included in items 10 and 13.A above.
14.B Condition codes 1, 2 or 3 due to conditions other than those included in item 12 above.
15. Condition code 0.
Prior to usage, the compressed format of a DHT is examined for the existence of general operand data exception conditions. When the length of the compressed format of a DHT is not precisely defined due to a general operand data exception condition, the interpreted length may depend on the condition, be model-dependent, and does not exceed, e.g., 286 bytes. As a result, when the DFLTCC-XPND function is specified and a compressed format of a DHT with a general operand data exception condition is encountered in the, e.g., rightmost 286 bytes of the second operand, it is model-dependent whether the exception condition (priority 14.A) or condition code 2 (priority 14.B) is recognized.
Example programming notes are provided below:
1. When compressing or uncompressing data, it may be more efficient overall when the operation is performed with a minimum number of times the DEFLATE Conversion Call instruction is executed. In other words, executing DFLTCC with a large operand may be more efficient than executing DFLTCC with small operands multiple times.
2. For the compressing and uncompressing operations, when condition code 3 is set, the general registers used by the instruction and the parameter block have been updated such that the program can branch back to the instruction to continue the operation.
3. In one embodiment, the DEFLATE Conversion Call instruction may be completed after performing a CPU-determined sub-portion of the processing specified by the parameters of the instruction. When the instruction is completed after performing only a CPU-determined amount of processing instead of all specified processing, the instruction sets condition code 3. On such completion, the instruction address in the PSW (program status word) designates the next sequential instruction, and the operand parameters of the instruction have been adjusted so that the processing of the instruction can be resumed by branching back to the instruction to execute it again. When the instruction has performed all specified processing, it sets a condition code other than 3.
4. When the DFLTCC-CMPR function is specified and the operation ends with a non-zero value in the sub-byte boundary (SBB) field of the parameter block, the operation included storing to the byte designated by the resulting first operand address. When the DFLTCC-XPND function is specified and the operation ends with a non-zero value in the SBB, the operation included fetching the byte designated by the resulting second operand address.
5. When the operation ends with a non-zero condition code set, CSB field 392 of the parameter block may contain partially processed data, and it is expected that the program re-executes the instruction to resume the operation.
6. Subsequent to an operation ending with a non-zero condition code set, and prior to re-executing the instruction for the purpose of resuming the operation, the program is not to modify any fields of the parameter block; otherwise results are unpredictable.
7. When the DFLTCC-GDHT function is specified, the compressed representation of a DHT generated describes three proper-full Huffman code trees, according to the Huffman algorithm. That is, no under-full Huffman code trees are described. An under-full Huffman code tree is derived from a compressed representation of a DHT which specifies a code length for an element which is greater than the length required by the Huffman algorithm to specify a proper and functional Huffman tree.
When the DFLTCC-CMPR function is specified, HTT is one, and the compressed representation of the DHT includes a description of an under-full Huffman code tree, the compressed data results can be transformed to the original uncompressed data by using the DFLTCC-XPND function, but not all decoders, which comply to the DEFLATE standard may be able to transform the results to the original uncompressed data. This may occur, for instance, when the compressed representation of a DHT specified by the program, for the DFLTCC-CMPR function was not generated as a result of performing the DFLTCC-GDHT function.
8. When the DFLTCC-CMPR function ends with condition code 1 set, the result stored to sub-byte boundary (SBB) field 381 of the parameter block is 000 binary. Recognizing this scenario may be relevant to a program allocating output buffers for use with the DEFLATE Conversion Call instruction.
As described herein, in one aspect, a single instruction (e.g., a single architected machine instruction at the hardware/software interface, e.g., DEFLATE Conversion Call instruction) is provided to perform compress and/or decompress operations using a general-purpose processor. This instruction is, for instance, a hardware instruction defined in an Instruction Set Architecture (ISA). As a result, the complexity of the program related to the compressing and/or decompressing operations is reduced. Further, performance of the operations, and thus, the processor, is improved.
Advantageously, the DEFLATE Conversion Call instruction is dispatched, by, e.g., a programmer, on a general-purpose processor (e.g., a central processing unit, referred to herein as a processor), rather than a special-purpose processor, such as an I/O device, an application specific device connected through an I/O interface, or other types of special-purpose processors. Compared to a software implementation, executing the disclosed instruction requires significantly fewer execution cycles to perform the same operation. Further, compared to dispatching an operation to an I/O device, executing the disclosed instruction does not require I/O operations by an operating system and does not trigger the operating system to perform a task switch while waiting for the operation to complete.
Although various fields and registers are described, one or more aspects of the present invention may use other, additional or fewer fields or registers, or other sizes of fields and registers, etc. Many variations are possible. For instance, implied registers may be used instead of explicitly specified registers or fields of the instruction and/or explicitly specified registers or fields may be used instead of implied registers or fields. Other variations are also possible.
One embodiment of using the DEFLATE Conversion Call instruction is described with reference to
Thereafter, the program executes the DEFLATE Conversion Call (DFLTCC) instruction, STEP 1708. In one example, the instruction is dispatched on a general-purpose processor. As examples, it is processed on the general-purpose processor, or, at least in part, is processed by hardware coupled to the general-purpose processor and accessible absent using an I/O interface.
Based on the instruction terminating, a determination is made as to whether the condition code resulting from execution is equal to a first defined value, e.g., 0, INQUIRY 1710. If the condition code is equal to the first defined value, then processing of the instruction is complete, STEP 1712. However, if the condition code is not equal to the first defined value, then a further determination is made as to whether the condition code is equal to a second defined value, e.g., 3, INQUIRY 1714. If the condition code is equal to the second defined value indicating there is additional data to be processed, then the instruction is re-executed, STEP 1708. However, if the condition code is not equal to the second defined value, then another determination is made as to whether the condition code is set to a third defined value, e.g., 1, INQUIRY 1716. If the condition code is set to the third defined value indicating the first operand length is insufficient, then processing continues with STEP 1706; otherwise, the second operand length is insufficient for the function and processing continues with STEP 1704.
As indicated, the DEFLATE Conversion Call instruction may be executed multiple times to compress or decompress a single data stream. Therefore, in one aspect, the DEFLATE Conversion Call instruction includes an attribute which provides a mechanism for a program to declare a buffer (e.g., a 32 K-byte buffer), which is used to accumulate the history of uncompressed data processed during an operation which spans multiple executions of the DEFLATE Conversion Call instruction. The buffer is, for instance, a circular history buffer.
In one aspect, the DEFLATE Conversion Call instruction uses an indicator (e.g., a bit) in an implied register (e.g., GR0.56) to indicate use of a circular history buffer. When the circular history buffer is indicated and the specified function to be performed by the DEFLATE Conversion Call instruction is compressing or uncompressing data, a field of the instruction (e.g., R3) specifies the location in memory of, e.g., a 32 K-byte buffer, which the processor uses to fetch history from at the beginning of an operation and store history to at the end of an operation. The length of the history within the circular history buffer is specified by a field of a parameter block associated with the DEFLATE Conversion Call instruction (e.g., HL field 385), and the beginning of the history within the buffer is specified by an offset included in another field of the parameter block (e.g., HO field 386).
Further details of using a circular history buffer are described with reference to
Further, in one example, the program allocates and specifies a location in memory of a pre-defined sized (e.g., 32 K-byte) circular buffer, STEP 1802. Additionally, the program places a portion of an uncompressed data stream into a buffer and specifies the location and the size of the buffer as an input to the DEFLATE Conversion Call instruction, STEP 1804, and specifies or updates the location and size of a result buffer in storage, STEP 1806.
The DEFLATE Conversion Call instruction is then executed, STEP 1808. Based on executing the instruction, the processor fetches history from, e.g., a circular history buffer, as an input to the operation, STEP 1820, and performs the specified operation, STEP 1822, as described herein. Further, the processor modifies the history in the circular history buffer as an output of the operation, STEP 1824. A determination is made as to whether the entire data stream has been processed, INQUIRY 1826. If not, then processing continues with STEP 1804. Otherwise, processing is complete.
Use of a circular history buffer provides the following, as examples:
When the size of the input or output buffer, specified for use with an individual execution of the DEFLATE Conversion Call instruction, is relatively small (e.g., 512 bytes), a history spanning multiple segments of buffered data, up to, e.g., 32 K-bytes, may be used as an input to the DEFLATE Conversion Call instruction, which processes a small number of bytes.
When the size of the input or output buffer, specified for use with an individual execution of the DEFLATE Conversion Call instruction, is relatively large (e.g., 128 K-bytes), a history of the prior segment of buffered data, up to, e.g., 32 K-bytes, may be used as a input to the DEFLATE Conversion Call instruction which is processing the first 32 K-bytes of data.
In both cases, more history is available to process data than would otherwise be available. As a result, the effectiveness of detecting duplicate strings is improved, resulting in improved overall compression ratios. This facilitates processing within the computing environment and improves performance.
One or more aspects of the present invention are inextricably tied to computer technology and facilitate processing within a computer, improving performance thereof. The use of a single architected machine instruction to perform compressing and/or decompressing improves performance within the computing environment. The compressed/decompressed data may be used in many technical fields that manage and/or use data, such as in computer processing, medical processing, security, inventory control, etc. By providing optimizations in compressing/decompressing, these technical fields are improved by reducing execution time.
The accelerator 15000 is uniquely configured to execute certain functions and compression or decompression operations including, but not limited to DEFLATE operations and other lossless data compression or decompression algorithms. As noted above, DEFLATE, in particular, is an industry standard algorithm for compression of data that can be several gigabytes (GBs) in size where an application may only have small buffers at a time and the compression has to be completed in relatively small blocks that can be 1 megabyte (MB) or smaller.
While the DEFLATE operations and the other lossless data compression or decompression algorithms could be executed by any one of the PUs 12000, the accelerator 15000 is capable of completion the execution faster than any of the PUs 12000 even when taking into account the time required to communicate between the accelerator 15000 and other components of the CP 11000.
To the extent that the accelerator 15000 can be provided as an NXU (DEFLATE) accelerator, the accelerator 15000 is effectively a DMA-attached hardware accelerator that implements actual DEFLATE algorithm standards. When an application executes DEFLATE instructions, millicode operates the hardware of the accelerator 15000 on behalf of the application to perform the actual DEFLATE operations. As used herein, millicode is low level code that is transparent from an application perspective and is used to implement instructions or parts of instructions. For DEFLATE, millicode acquires and operates the accelerator 15000 on behalf of the application.
With reference to
With reference to
With reference to
In an event the determining indicates that the asynchronous mode should be in effect, the first processing unit 120001 sets an ACF (40200). In this case, the data compression operation is conducted in asynchronous mode. Here, the second processing unit 120002 issues a call to the accelerator 15000. In doing so, the second processing unit 120002 builds scatter gather lists (40300) to be fed to the accelerator 15000 to facilitate execution of the data compression by the accelerator 15000 in response to the setting of the ACF (40400). That is, DFLTCC millicode in the second processing unit 120002 takes absolute address from the scatter gather lists and feeds those absolute addresses to the accelerator 15000.
In an event the determining indicates that the synchronous mode should be in effect, the data compression operation is conducted in synchronous mode (40500). Here, DFLTCC millicode in the first processing unit 120001 translates logical addresses of each page (in, e.g., 4k increments) of source and target buffers into absolute addresses and feeds those absolute addresses to the accelerator 15000.
With reference to
With reference to
The present invention may be a system, a method, and/or a computer program product at any possible technical detail level of integration. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present invention.
The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.
Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. A network adapter card or network interface in each computing/processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing/processing device.
Computer readable program instructions for carrying out operations of the present invention may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instruction by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present invention.
Aspects of the present invention are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer readable program instructions.
These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function/act specified in the flowchart and/or block diagram block or blocks.
The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.
The descriptions of the various embodiments of the present invention have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments described herein.
Number | Name | Date | Kind |
---|---|---|---|
4477881 | Kobayashi et al. | Oct 1984 | A |
5561809 | Elko et al. | Oct 1996 | A |
5636366 | Robinson | Jun 1997 | A |
5948060 | Gregg et al. | Sep 1999 | A |
6105075 | Ghaffari | Aug 2000 | A |
6678754 | Soulier | Jan 2004 | B1 |
6820264 | Bashkansky | Nov 2004 | B1 |
6822589 | Dye | Nov 2004 | B1 |
7249202 | Simon et al. | Jul 2007 | B2 |
7356001 | Jones | Apr 2008 | B1 |
7971030 | O'Sullivan | Jun 2011 | B2 |
8346929 | Lai | Jan 2013 | B1 |
8578490 | Moran | Nov 2013 | B2 |
9411535 | Shemer | Aug 2016 | B1 |
9852076 | Garg | Dec 2017 | B1 |
9886691 | Desai | Feb 2018 | B2 |
9916622 | Parsons | Mar 2018 | B2 |
20010037419 | Hagersten | Nov 2001 | A1 |
20050044197 | Lai | Feb 2005 | A1 |
20070109153 | Ma | May 2007 | A1 |
20090041129 | Suh | Feb 2009 | A1 |
20090172364 | Sprangle et al. | Jul 2009 | A1 |
20090210437 | Kuila | Aug 2009 | A1 |
20110258049 | Ramer | Oct 2011 | A1 |
20130204886 | Faith | Aug 2013 | A1 |
20130290234 | Harris | Oct 2013 | A1 |
20130325998 | Hormuth | Dec 2013 | A1 |
20150127565 | Chevalier | May 2015 | A1 |
20160233882 | Agarwal | Aug 2016 | A1 |
20170187388 | Satpathy | Jun 2017 | A1 |
20180203743 | Myerscough | Jul 2018 | A1 |
Entry |
---|
“Method for Improving the Accuracy and Efficiency of Exception Based Instruction Tracing”; IP.com; IP.com No. IPCOM000238423D; IP.com Electronic Publication Date: Aug. 26, 2014; 4 Pages. |
“Method for Verifying Intermediate Results of a Randomly Generated Testcase of Processor Functional Validation Exerciser”; IP.com; IP.com No. IPCOM000243758D; IP.com Electronic Publication Date: Oct. 16, 2015; 10 Pages. |
“System and Method for Accounting Flush Cycles in a Cycles Per Instruction Stack”; IP.com; IP.com No. IPCOM000202948D; IP.com Electronic Publication Date: Jan. 11, 2011; 3 Pages. |
List of IBM Patents or Patent Applications Treated as Related; Appendix P filed Nov. 14, 2019; 2 Pages. |
Tune et al.; “Dynamic Prediction of Critical Path Instructions”; In Proceedings of the 7th International Symposium on High Performance Computer Architecture; Jan. 2001; pp. 1-11. |
Matthias Klein et al.; “Functional Completion When Retrying a Non-Interruptible Instruction in a Bi-Modal Execution Environment”, U.S. Appl. No. 16/286,987, filed Feb. 27, 2019. |
International Search Report; International Application No. PCT/EP2020/053383; International Filing Date: Feb. 11, 2020; dated May 19, 2020; 11 pages. |