Padding and suppressing rows and columns of data

[0001] This relates to using a streaming engine to padding and suppressing rows and columns of feature maps.

[0002] Digital signal processors (DSP) are optimized for processing streams of data that may be derived from various input signals, such as sensor data, a video stream, a voice channel, radar signals, biomedical signals, etc. Digital signal processors operating on real-time data typically receive an input data stream, perform a filter function on the data stream (such as encoding or decoding) and output a transformed data stream. The system is called real-time because the application fails if the transformed data stream is not available for output when scheduled. Typical video encoding requires a predictable but non-sequential input data pattern. A typical application requires memory access to load data registers in a data register file and then supply data from the data registers to functional units which perform the data processing.

[0003] One or more DSP processing cores can be combined with various peripheral circuits, blocks of memory, etc. on a single integrated circuit (IC) die to form a system on chip (SoC). These systems can include multiple interconnected processors that share the use of on-chip and off-chip memory. A processor can include some combination of instruction cache (ICache) and data cache (DCache) to improve processing. Furthermore, multiple processors with shared memory can be incorporated in a single embedded system. The processors can physically share the same memory without accessing data or executing code located in the same memory locations or can use some portion of the shared memory as common shared memory.

[0004] This Summary is provided to comply with 37 C.F.R. § 1.73, requiring a summary of the invention briefly indicating the nature and substance of the invention. It is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims.

[0005] According to some examples, a method is described herein. The method generally includes receiving stream parameters that defines an array, wherein the stream parameters include a first null element count and a second null element count. The method generally includes forming a stream of vectors for the multidimensional array responsive to the stream parameters. The stream of vectors generally includes a vector of null elements at a beginning of the stream of vectors based on the first null element count. The stream of vectors generally includes a null element at a beginning of each vector of the stream of vectors based on the second null element count. The stream of vectors generally includes a set of data distributed across a subset of the stream of vectors. The method generally includes providing the stream of vectors.

[0006] According to some examples, a device is described herein. The device generally includes a register configured to store a set of stream parameters, wherein the stream parameters include a first null element count and a second null element count. The device generally includes a memory configured to store a set of data. The device generally includes a circuit coupled to the register and the memory. The circuit is generally configured to form a stream of vectors for the multidimensional array responsive to the stream parameters. The stream of vectors generally includes a vector of null elements at a beginning of the stream of vectors based on the first null element count. The stream of vectors generally includes a null element at a beginning of each vector of the stream of vectors based on the second null element count. The stream of vectors generally includes the set of data distributed across a subset of the stream of vectors. The circuit is generally configured to provide the stream of vectors.

[0007] According to some examples, a method is described herein. The method generally includes receiving stream parameters into control logic of the streaming engine to define a multidimensional array, wherein the stream parameters includes a suppression count. The method generally includes receiving a first set of data by the streaming engine. The method generally includes determining a second set of data that includes a subset of the first set of data, wherein the suppression count specifies a remainder of the first set of data that is not included in the second set of data. The method generally includes writing the second set of data to memory.

[0008] These and other aspects may be understood with reference to the following detailed description.

[0060] To facilitate understanding, identical reference numerals have been used, where possible, to designate identical elements that are common to the figures. It is contemplated that elements of one example may be beneficially incorporated in other examples.

[0061] Various features are described hereinafter with reference to the figures. It should be noted that the figures may or may not be drawn to scale and that the elements of similar structures or functions are represented by like reference numerals throughout the figures. It should be noted that the figures are only intended to facilitate the description of the features. They are not intended as an exhaustive description of the description or as a limitation on the scope of the claims. In addition, an illustrated example need not have all the aspects or advantages shown. An aspect or an advantage described in conjunction with a particular example is not necessarily limited to that example and can be practiced in any other examples even if not so illustrated, or if not so explicitly described.

[0062] Digital signal processors (DSP) are optimized for processing streams of data that may be derived from various input signals, such as sensor data, a video stream, a voice channel, radar signals, biomedical signals, etc. Memory bandwidth and scheduling are concerns for digital signal processors operating on real-time data. An example DSP is described herein that includes a streaming engine to improve memory bandwidth and data scheduling.

[0063] One or more DSP can be combined with various peripheral circuits, blocks of memory, etc. on a single integrated circuit (IC) die to form a system on chip (SoC). See, for example, “66AK2Hx Multicore Keystone™ DSP+ARM® System-on-Chip,” 2013 which is incorporated by reference herein.

[0064] In the example processor described herein, an autonomous streaming engine (SE) is coupled to the DSP. In this example, the streaming engine includes two closely coupled streaming engines that can manage two data streams simultaneously. In another example, the streaming engine is capable of managing only a single stream, while in other examples the streaming engine is capable of handling more than two streams. In each case, for each stream, the streaming engine includes an address generation stage, a data formatting stage, and some storage for formatted data waiting for consumption by the processor. In the examples described herein, addresses are derived from algorithms that can involve multi-dimensional loops, each dimension maintaining an iteration count. In one example, the streaming engine supports six levels of nested iteration. In other examples, more or fewer levels of iteration are supported. In some examples, the streaming engine is a cache controller for managing a data stream to and from a cache.

[0065] In the example processor described hereinbelow, control logic and an alignment network enable the streaming engine to form null vectors or padded vectors and to insert them into a stream without accessing the null data or pad data from system memory.

[0066] Several examples of padding feature maps by inserting null elements are described in more detail with regards to FIGS. 36-44. Several examples of write suppression are described in more detail with regards to FIGS. 45-50.

[0067] An example DSP processor is described in detail herein with reference to FIGS. 1-18. An example streaming engine capable of managing two data streams using six-dimensional nested loops is described in detail herein with reference to FIGS. 19-35.

[0068] FIG. 1 illustrates an example processor 100 that includes dual scalar/vector data paths 115, 117. As used herein, the term “vector” refers to a one-dimensional array of data elements that can be accessed and operated on as a unit. Processor 100 includes a streaming engine 125 that is described in more detail herein. Processor 100 includes separate level one instruction cache (L1I) 121 and level one data cache (L1D) 123. Processor 100 includes a level 2 (L2) combined instruction/data cache 130 that holds both instructions and data. FIG. 1 illustrates connection between L1I cache and L2 combined instruction/data cache 130, 512-bit bus 142. FIG. 1 illustrates the connection between L1D cache 123 and L2 combined instruction/data cache 130, 512-bit bus 145. In the example processor 100, L2 combined instruction/data cache 130 stores both instructions to back up L1I cache 121 and data to back up L1D cache 123. In this example, L2 combined instruction/data cache 130 is further connected to higher level cache and/or main memory using known or later developed memory system techniques not illustrated in FIG. 1. As used herein, the term “higher level” memory or cache refers to a next level in a memory hierarchy that is more distant from the processor, while the term “lower level” memory or cache refers to a level in the memory hierarchy that is closer to the processor. L1I cache 121, L1D cache 123, and L2 cache 130 may be implemented in different sizes in various examples. In this example, L1I cache 121 and L1D cache 123 are each 32K bytes, and L2 cache 130 is 1024K bytes. In the example processor 100, L1I cache 121, L1D cache 123 and L2 combined instruction/data cache 130 are formed on a single integrated circuit. This single integrated circuit optionally includes other circuits.

[0069] Processing unit core 110 fetches instructions from L1I cache 121 as controlled by instruction fetch unit 111. Instruction fetch unit 111 determines the next instructions to be executed and recalls a fetch packet sized set of such instructions. The nature and size of fetch packets are further detailed below. Instructions are directly fetched from L1I cache 121 upon a cache hit if the instructions are stored in L1I cache 121. Upon a cache miss occurring when the specified instructions are not stored in L1I cache 121, the instructions are sought in L2 combined cache 130. In this example, the size of a cache line in L1I cache 121 equals the size of a fetch packet which is 512 bits. The memory locations of these instructions are either a hit in L2 combined cache 130 or a miss. A hit is serviced from L2 combined cache 130. A miss is serviced from a higher level of cache (not illustrated) or from main memory (not illustrated). In this example, the requested instruction is simultaneously supplied to both L1I cache 121 and processing unit core 110 to speed use.

[0070] In this example, processing unit core 110 includes multiple functional units to perform instruction specified data processing tasks. Instruction dispatch unit 112 determines the target functional unit of each fetched instruction. In this example, processing unit 110 operates as a very long instruction word (VLIW) processor capable of operating on multiple instructions in corresponding functional units simultaneously. A complier organizes instructions in execute packets that are executed together. Instruction dispatch unit 112 directs each instruction to its target functional unit. The functional unit assigned to an instruction is completely specified by the instruction produced by the compiler. The hardware of processing unit core 110 has no part in the functional unit assignment. In this example, instruction dispatch unit 112 operates on several instructions in parallel. The number of such parallel instructions is set by the size of the execute packet. This is further described herein.

[0071] One part of the dispatch task of instruction dispatch unit 112 is determining whether the instruction is to execute on a functional unit in scalar data path side A 115 or vector data path side B 116. An instruction bit within each instruction called the s bit determines which data path the instruction controls. This is further described herein.

[0072] Instruction decode unit 113 decodes each instruction in a current execute packet. Decoding includes identification of the functional unit performing the instruction, identification of registers used to supply data for the corresponding data processing operation from among possible register files, and identification of the register destination of the results of the corresponding data processing operation. As further explained below, instructions can include a constant field in place of one register number operand field. The result of this decoding are signals for control of the target functional unit to perform the data processing operation specified by the corresponding instruction on the specified data.

[0073] Processing unit core 110 includes control registers 114. Control registers 114 store information for control of the functional units in scalar data path side A 115 and vector data path side B 116. This information may include mode information or the like.

[0074] The decoded instructions from instruction decode 113 and information stored in control registers 114 are supplied to scalar data path side A 115 and vector data path side B 116. As a result, functional units within scalar data path side A 115 and vector data path side B 116 perform instruction specified data processing operations upon instruction specified data and store the results in an instruction specified data register or registers. Each of scalar data path side A 115 and vector data path side B 116 include multiple functional units that operate in parallel. These are further described below in conjunction with FIG. 2. There is a data path 117 between scalar data path side A 115 and vector data path side B 116 permitting data exchange.

[0075] Processing unit core 110 includes further non-instruction-based modules. Emulation unit 118 permits determination of the machine state of processing unit core 110 in response to instructions. This capability can be employed for algorithmic development. Interrupts/exceptions unit 119 enables processing unit core 110 to be responsive to external, asynchronous events (interrupts) and to respond to attempts to perform improper operations (exceptions).

[0076] Processor 100 includes streaming engine 125. Streaming engine 125 supplies two data streams from predetermined addresses cached in L2 combined cache 130 to register files of vector data path side B of processing unit core 110. This provides controlled data movement from memory (as cached in L2 combined cache 130) directly to functional unit operand inputs. This is further described herein.

[0077] FIG. 1 illustrates example data widths of busses between various parts. L1I cache 121 supplies instructions to instruction fetch unit 111 via bus 141. Bus 141 is a 512-bit bus in this example. Bus 141 is unidirectional from L1I cache 121 to processing unit 110. L2 combined cache 130 supplies instructions to L1I cache 121 via bus 142. Bus 142 is a 512-bit bus in this example. Bus 142 is unidirectional from L2 combined cache 130 to L1I cache 121.

[0078] L1D cache 123 exchanges data with register files in scalar data path side A 115 via bus 143. Bus 143 is a 64-bit bus in this example. L1D cache 123 exchanges data with register files in vector data path side B 116 via bus 144. Bus 144 is a 512-bit bus in this example. Busses 143 and 144 are illustrated as bidirectional supporting both processing unit core 110 data reads and data writes. L1D cache 123 exchanges data with L2 combined cache 130 via bus 145. Bus 145 is a 512-bit bus in this example. Bus 145 is illustrated as bidirectional supporting cache service for both processing unit core 110 data reads and data writes.

[0079] Processor data requests are directly fetched from L1D cache 123 upon a cache hit (if the requested data is stored in L1D cache 123). Upon a cache miss (the specified data is not stored in L1D cache 123), the data is sought in L2 combined cache 130. The memory locations of the requested data are either a hit in L2 combined cache 130 or a miss. A hit is serviced from L2 combined cache 130. A miss is serviced from another level of cache (not illustrated) or from main memory (not illustrated). The requested data may be simultaneously supplied to both L1D cache 123 and processing unit core 110 to speed use.

[0080] L2 combined cache 130 supplies data of a first data stream to streaming engine 125 via bus 146. Bus 146 is a 512-bit bus in this example. Streaming engine 125 supplies data of the first data stream to functional units of vector data path side B 116 via bus 147. Bus 147 is a 512-bit bus in this example. L2 combined cache 130 supplies data of a second data stream to streaming engine 125 via bus 148. Bus 148 is a 512-bit bus in this example. Streaming engine 125 supplies data of this second data stream to functional units of vector data path side B 116 via bus 149, which is a 512-bit bus in this example. Busses 146, 147, 148 and 149 are illustrated as unidirectional from L2 combined cache 130 to streaming engine 125 and to vector data path side B 116 in accordance with this example.

[0081] Streaming engine data requests are directly fetched from L2 combined cache 130 upon a cache hit (if the requested data is stored in L2 combined cache 130). Upon a cache miss (the specified data is not stored in L2 combined cache 130), the data is sought from another level of cache (not illustrated) or from main memory (not illustrated). It is technically feasible in some examples for L1D cache 123 to cache data not stored in L2 combined cache 130. If such operation is supported, then upon a streaming engine data request that is a miss in L2 combined cache 130, L2 combined cache 130 snoops L1D cache 123 for the streaming engine requested data. If L1D cache 123 stores the data, the snoop response includes the data, which is then supplied to service the streaming engine request. If L1D cache 123 does not store the data, the snoop response indicates this and L2 combined cache 130 services the streaming engine request from another level of cache (not illustrated) or from main memory (not illustrated).

[0082] In this example, both L1D cache 123 and L2 combined cache 130 can be configured as selected amounts of cache or directly addressable memory in accordance with U.S. Pat. No. 6,606,686 entitled UNIFIED MEMORY SYSTEM ARCHITECTURE INCLUDING CACHE AND DIRECTLY ADDRESSABLE STATIC RANDOM ACCESS MEMORY, which is incorporated by reference herein.

[0083] In this example, processor 100 is fabricated on an integrated chip (IC) that is mounted on a ball grid array (BGA) substrate. A BGA substrate and IC die together may be referred to as “BGA package,” “IC package,” “integrated circuit,” “IC,” “chip,” “microelectronic device,” or similar terminology. The BGA package may include encapsulation material to cover and protect the IC die from damage. In another example, other types of known or later developed packaging techniques may be used with processor 100.

[0084] FIG. 2 illustrates further details of functional units and register files within scalar data path side A 115 and vector data path side B 116. Scalar data path side A 115 includes L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226. Scalar data path side A 115 includes global scalar register file 211, L1/S1 local register file 212, M1/N1 local register file 213 and D1/D2 local register file 214. Vector data path side B 116 includes L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246. Vector data path side B 116 includes global vector register file 231, L2/S2 local register file 232, M2/N2/C local register file 233 and predicate register file 234. Which functional units can read from or write to which register files is described in more detail herein.

[0085] Scalar data path side A 115 includes L1 unit 221. L1 unit 221 generally accepts two 64-bit operands and produces one 64-bit result. The two operands are each recalled from an instruction specified register in either global scalar register file 211 or L1/S1 local register file 212. L1 unit 221 performs the following instruction selected operations: 64-bit add/subtract operations; 32-bit min/max operations; 8-bit Single Instruction Multiple Data (SIMD) instructions such as sum of absolute value, minimum and maximum determinations; circular min/max operations; and various move operations between register files. The result is written into an instruction specified register of global scalar register file 211, L1/S1 local register file 212, M1/N1 local register file 213 or D1/D2 local register file 214.

[0086] Scalar data path side A 115 includes S1 unit 222. S1 unit 222 generally accepts two 64-bit operands and produces one 64-bit result. The two operands are each recalled from an instruction specified register in either global scalar register file 211 or L1/S1 local register file 212. In this example, S1 unit 222 performs the same type operations as L1 unit 221. In another example, there may be slight variations between the data processing operations supported by L1 unit 221 and S1 unit 222. The result is written into an instruction specified register of global scalar register file 211, L1/S1 local register file 212, M1/N1 local register file 213 or D1/D2 local register file 214.

[0087] Scalar data path side A 115 includes M1 unit 223. M1 unit 223 generally accepts two 64-bit operands and produces one 64-bit result. The two operands are each recalled from an instruction specified register in either global scalar register file 211 or M1/N1 local register file 213. In this example, M1 unit 223 performs the following instruction selected operations: 8-bit multiply operations; complex dot product operations; 32-bit bit count operations; complex conjugate multiply operations; and bit-wise Logical Operations, moves, adds and subtracts. The result is written into an instruction specified register of global scalar register file 211, L1/S1 local register file 212, M1/N1 local register file 213 or D1/D2 local register file 214.

[0088] Scalar data path side A 115 includes N1 unit 224. N1 unit 224 generally accepts two 64-bit operands and produces one 64-bit result. The two operands are each recalled from an instruction specified register in either global scalar register file 211 or M1/N1 local register file 213. In this example, N1 unit 224 performs the same type operations as M1 unit 223. There are also double operations (called dual issued instructions) that employ both the M1 unit 223 and the N1 unit 224 together. The result is written into an instruction specified register of global scalar register file 211, L1/S1 local register file 212, M1/N1 local register file 213 or D1/D2 local register file 214.

[0089] Scalar data path side A 115 includes D1 unit 225 and D2 unit 226. D1 unit 225 and D2 unit 226 generally each accept two 64-bit operands and each produce one 64-bit result. D1 unit 225 and D2 unit 226 generally perform address calculations and corresponding load and store operations. D1 unit 225 is used for scalar loads and stores of 64 bits. D2 unit 226 is used for vector loads and stores of 512 bits. In this example, D1 unit 225 and D2 unit 226 also perform: swapping, pack and unpack on the load and store data; 64-bit SIMD arithmetic operations; and 64-bit bit-wise logical operations. D1/D2 local register file 214 stores base and offset addresses used in address calculations for the corresponding loads and stores. The two operands are each recalled from an instruction specified register in either global scalar register file 211 or D1/D2 local register file 214. The calculated result is written into an instruction specified register of global scalar register file 211, L1/S1 local register file 212, M1/N1 local register file 213 or D1/D2 local register file 214.

[0090] Vector data path side B 116 includes L2 unit 241. L2 unit 241 generally accepts two 512-bit operands and produces one 512-bit result. The two operands are each recalled from an instruction specified register in either global vector register file 231, L2/S2 local register file 232 or predicate register file 234. In this example, L2 unit 241 performs instruction similar to L1 unit 221 except on wider 512-bit data. The result may be written into an instruction specified register of global vector register file 231, L2/S2 local register file 232, M2/N2/C local register file 233 or predicate register file 234.

[0091] Vector data path side B 116 includes S2 unit 242. S2 unit 242 generally accepts two 512-bit operands and produces one 512-bit result. The two operands are each recalled from an instruction specified register in either global vector register file 231, L2/S2 local register file 232 or predicate register file 234. In this example, S2 unit 242 performs instructions similar to S1 unit 222. The result is written into an instruction specified register of global vector register file 231, L2/S2 local register file 232, M2/N2/C local register file 233 or predicate register file 234.

[0092] Vector data path side B 116 includes M2 unit 243. M2 unit 243 generally accepts two 512-bit operands and produces one 512-bit result. The two operands are each recalled from an instruction specified register in either global vector register file 231 or M2/N2/C local register file 233. In this example, M2 unit 243 performs instructions similar to M1 unit 223 except on wider 512-bit data. The result is written into an instruction specified register of global vector register file 231, L2/S2 local register file 232 or M2/N2/C local register file 233.

[0093] Vector data path side B 116 includes N2 unit 244. N2 unit 244 generally accepts two 512-bit operands and produces one 512-bit result. The two operands are each recalled from an instruction specified register in either global vector register file 231 or M2/N2/C local register file 233. In this example, N2 unit 244 performs the same type operations as M2 unit 243. There are also double operations (called dual issued instructions) that employ both M2 unit 243 and the N2 unit 244 together. The result is written into an instruction specified register of global vector register file 231, L2/S2 local register file 232 or M2/N2/C local register file 233.

[0094] Vector data path side B 116 includes correlation (C) unit 245. C unit 245 generally accepts two 512-bit operands and produces one 512-bit result. The two operands are each recalled from an instruction specified register in either global vector register file 231 or M2/N2/C local register file 233. In this example, C unit 245 performs “Rake” and “Search” instructions that are used for WCDMA (wideband code division multiple access) encoding/decoding. In this example, C unit 245 can perform up to 512 multiples per clock cycle of a 2-bit PN (pseudorandom number) and 8-bit I/Q (complex number), 8-bit and 16-bit Sum-of-Absolute-Difference (SAD) calculations, up to 512 SADs per clock cycle, horizontal add and horizontal min/max instructions, and vector permutes instructions. C unit 245 also contains 4 vector control registers (CUCR0 to CUCR3) used to control certain operations of C unit 245 instructions. Control registers CUCR0 to CUCR3 are used as operands in certain C unit 245 operations. In some examples, control registers CUCR0 to CUCR3 are used in control of a general permutation instruction (VPERM), and as masks for SIMD multiple DOT product operations (DOTPM) and SIMD multiple Sum-of-Absolute-Difference (SAD) operations. In further examples, control register CUCR0 is used to store the polynomials for Galois Field Multiply operations (GFMPY) and control register CUCR1 is used to store the Galois field polynomial generator function.

[0095] Vector data path side B 116 includes P unit 246. Vector predicate (P) unit 246 performs basic logic operations on registers of local predicate register file 234. P unit 246 has direct access to read from and write to predication register file 234. The logic operations include single register unary operations such as NEG (negate) which inverts each bit of the single register, BITCNT (bit count) which returns a count of the number of bits in the single register having a predetermined digital state (1 or 0), RMBD (right most bit detect) which returns a number of bit positions from the least significant bit position (right most) to a first bit position having a predetermined digital state (1 or 0), DECIMATE which selects every instruction specified Nth (1, 2, 4, etc.) bit to output, and EXPAND which replicates each bit an instruction specified N times (2, 4, etc.). The logic operations also include two register binary operations such as AND which is a bitwise AND of data of the two registers, NAND which is a bitwise AND and negate of data of the two registers, OR which is a bitwise OR of data of the two registers, NOR which is a bitwise OR and negate of data of the two registers, and XOR which is exclusive OR of data of the two registers. The logic operations include transfer of data from a predicate register of predicate register file 234 to another specified predicate register or to a specified data register in global vector register file 231. One use of P unit 246 is manipulation of the SIMD vector comparison results for use in control of a further SIMD vector operation. The BITCNT instruction can be used to count the number of 1's in a predicate register to determine the number of valid data elements from a predicate register.

[0096] FIG. 3 illustrates global scalar register file 211. There are 16 independent 64-bit wide scalar registers designated A0 to A15. Each register of global scalar register file 211 can be read from or written to as 64-bits of scalar data. All scalar data path side A 115 functional units (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) can read or write to global scalar register file 211. Global scalar register file 211 can be read from as 32-bits or as 64-bits and written to as 64-bits. The instruction executing determines the read data size. Vector data path side B 116 functional units (L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246) can read from global scalar register file 211 via cross path 117 under restrictions that are described below.

[0097] FIG. 4 illustrates D1/D2 local register file 214. There are sixteen independent 64-bit wide scalar registers designated D0 to D16. Each register of D1/D2 local register file 214 is read from or written to as 64-bits of scalar data. All scalar data path side A 115 functional units (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) can write to global scalar register file 211. Only D1 unit 225 and D2 unit 226 can read from D1/D2 local scalar register file 214. Data stored in D1/D2 local scalar register file 214 can include base addresses and offset addresses used in address calculation.

[0098] FIG. 5 illustrates L1/S1 local register file 212. In this example, L1/S1 local register file 212 includes eight independent 64-bit wide scalar registers designated AL0 to AL7. In this example, the instruction coding permits L1/S1 local register file 212 to include up to 16 registers. In this example, eight registers are implemented to reduce circuit size and complexity. Each register of L1/S1 local register file 212 can be read from or written to as 64-bits of scalar data. All scalar data path side A 115 functional units (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) can write to L1/S1 local scalar register file 212. L1 unit 221 and S1 unit 222 can read from L1/S1 local scalar register file 212.

[0099] FIG. 6 illustrates M1/N1 local register file 213. In this example, eight independent 64-bit wide scalar registers designated AM0 to AM7 are implemented. In this example, the instruction coding permits M1/N1 local register file 213 to include up to 16 registers. In this example, eight registers are implemented to reduce circuit size and complexity. Each register of M1/N1 local register file 213 can be read from or written to as 64-bits of scalar data. All scalar data path side A 115 functional units (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) can write to M1/N1 local scalar register file 213. M1 unit 223 and N1 unit 224 can read from M1/N1 local scalar register file 213.

[0100] FIG. 7 illustrates global vector register file 231. There are sixteen independent 512-bit wide vector registers. Each register of global vector register file 231 can be read from or written to as 64-bits of scalar data designated B0 to B15. Each register of global vector register file 231 can be read from or written to as 512-bits of vector data designated VB0 to VB15. The instruction type determines the data size. All vector data path side B 116 functional units (L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246) can read or write to global vector register file 231. Scalar data path side A 115 functional units (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) can read from global vector register file 231 via cross path 117 under restrictions that are described below.

[0101] FIG. 8 illustrates predicate (P) local register file 234. There are eight independent 64-bit wide registers designated P0 to P7. Each register of P local register file 234 can be read from or written to as 64-bits of scalar data. Vector data path side B 116 functional units L2 unit 241, S2 unit 242, C unit 244 and P unit 246 can write to P local register file 234. L2 unit 241, S2 unit 242 and P unit 246 can read from P local scalar register file 234. One use of P local register file 234 is writing one-bit SIMD vector comparison results from L2 unit 241, S2 unit 242 or C unit 244, manipulation of the SIMD vector comparison results by P unit 246, and use of the manipulated results in control of a further SIMD vector operation.

[0102] FIG. 9 illustrates L2/S2 local register file 232. In this example, eight independent 512-bit wide vector registers are implemented. In this example, the instruction coding permits L2/S2 local register file 232 to include up to sixteen registers. In this example, eight registers are implemented to reduce circuit size and complexity. Each register of L2/S2 local vector register file 232 can be read from or written to as 64-bits of scalar data designated BL0 to BL7. Each register of L2/S2 local vector register file 232 can be read from or written to as 512-bits of vector data designated VBL0 to VBL7. The instruction type determines the data size. All vector data path side B 116 functional units (L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246) can write to L2/S2 local vector register file 232. L2 unit 241 and S2 unit 242 can read from L2/S2 local vector register file 232.

[0103] FIG. 10 illustrates M2/N2/C local register file 233. In this example, eight independent 512-bit wide vector registers are implemented. In this example, the instruction coding permits M2/N2/C local register file 233 to include up to sixteen registers. In this example, eight registers are implemented to reduce circuit size and complexity. Each register of M2/N2/C local vector register file 233 can be read from or written to as 64-bits of scalar data designated BMO to BM7. Each register of M2/N2/C local vector register file 233 can be read from or written to as 512-bits of vector data designated VBM0 to VBM7. All vector data path side B 116 functional units (L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246) can write to M2/N2/C local vector register file 233. M2 unit 243, N2 unit 244 and C unit 245 can read from M2/N2/C local vector register file 233.

[0104] The provision of global register files accessible by all functional units of a side and local register files accessible by some of the functional units of a side is a design choice. In another example, a different accessibility provision could be made, such as employing one type of register file corresponding to the global register files described herein.

[0105] Cross path 117 permits limited exchange of data between scalar data path side A 115 and vector data path side B 116. During each operational cycle one 64-bit data word can be recalled from global scalar register file A 211 for use as an operand by one or more functional units of vector data path side B 116 and one 64-bit data word can be recalled from global vector register file 231 for use as an operand by one or more functional units of scalar data path side A 115. Any scalar data path side A 115 functional unit (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) can read a 64-bit operand from global vector register file 231. This 64-bit operand is the least significant bits of the 512-bit data in the accessed register of global vector register file 231. Multiple scalar data path side A 115 functional units can employ the same 64-bit cross path data as an operand during the same operational cycle. However, a single 64-bit operand is transferred from vector data path side B 116 to scalar data path side A 115 in a single operational cycle. Any vector data path side B 116 functional unit (L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246) can read a 64-bit operand from global scalar register file 211. If the corresponding instruction is a scalar instruction, the cross-path operand data is treated as a 64-bit operand. If the corresponding instruction is a vector instruction, the upper 448 bits of the operand are zero filled. Multiple vector data path side B 116 functional units can employ the same 64-bit cross path data as an operand during the same operational cycle. In one example, a single 64-bit operand is transferred from scalar data path side A 115 to vector data path side B 116 in a single operational cycle.

[0106] Streaming engine 125 (FIG. 1) transfers data in certain restricted circumstances. Streaming engine 125 controls two data streams. A stream includes of a sequence of elements of a particular type. Programs that operate on streams read the data sequentially, operating on each element in turn. Every stream has the following basic properties: the stream data have a well-defined beginning and ending in time; the stream data have fixed element size and type throughout the stream; and, the stream data have a fixed sequence of elements. Once a stream is opened, streaming engine 125 performs the following operations: calculates the address; fetches the defined data type from L2 unified cache 130 (which may require cache service from a higher level memory, e.g., in the event of a cache miss in L2); performs data type manipulation such as zero extension, sign extension, data element sorting/swapping such as matrix transposition; and delivers the data directly to the programmed data register file within processor core 110. Streaming engine 125 is thus useful for real-time digital filtering operations on well-behaved data. Streaming engine 125 frees the corresponding processor from these memory fetch tasks, thus enabling other processing functions.

[0107] Streaming engine 125 provides several benefits. For example, streaming engine 125 permits multi-dimensional memory accesses, increases the available bandwidth to the functional units minimizes the number of cache miss stalls since the stream buffer bypasses L1D cache 123, and reduces the number of scalar operations required to maintain a loop. Streaming engine 125 also manages address pointers and handles address generation which frees up the address generation instruction slots and D1 unit 225 and D2 unit 226 for other computations.

[0108] Processor core 110 (FIG. 1) operates on an instruction pipeline. Instructions are fetched in instruction packets of fixed length as further described below. All instructions require the same number of pipeline phases for fetch and decode but require a varying number of execute phases.

[0109] FIG. 11 illustrates the following pipeline phases: program fetch phase 1110, dispatch and decode phases 1120, and execution phases 1130. Program fetch phase 1110 includes three stages for all instructions. Dispatch and decode phases 1120 include three stages for all instructions. Execution phase 1130 includes one to four stages depending on the instruction.

[0110] Fetch phase 1110 includes program address generation (PG) stage 1111, program access (PA) stage 1112 and program receive (PR) stage 1113. During program address generation stage 1111, the program address is generated in the processor and the read request is sent to the memory controller for the L1I cache. During the program access stage 1112, the L1I cache processes the request, accesses the data in its memory and sends a fetch packet to the processor boundary. During the program receive stage 1113, the processor registers the fetch packet.

[0111] Instructions are fetched in a fetch packet that includes sixteen 32-bit wide words. FIG. 12 illustrates sixteen instructions 1201 to 1216 of a single fetch packet. Fetch packets are aligned on 512-bit (16-word) boundaries. This example employs a fixed 32-bit instruction length which enables decoder alignment. A properly aligned instruction fetch can load multiple instructions into parallel instruction decoders. Such a properly aligned instruction fetch can be achieved by predetermined instruction alignment when stored in memory by having fetch packets aligned on 512-bit boundaries coupled with a fixed instruction packet fetch. Conversely, variable length instructions require an initial step of locating each instruction boundary before decoding. A fixed length instruction set generally permits more regular layout of instruction fields which simplifies the construction of each decoder which is an advantage for a wide issue VLIW processor.

[0112] The execution of the individual instructions is partially controlled by a p bit in each instruction. In this example, the p bit is bit 0 of the 32-bit wide slot. The p bit determines whether an instruction executes in parallel with the next instruction. In this example, instructions are scanned from lower to higher address. If the p bit of an instruction is 1, then the next following instruction (higher memory address) is executed in parallel with (in the same cycle as) that instruction. If the p bit of an instruction is 0, then the next following instruction is executed in the cycle after the instruction.

[0113] Processor core 110 (FIG. 1) and L1I cache 121 pipelines (FIG. 1) are de-coupled from each other. Fetch packet returns from L1I cache can take a different number of clock cycles, depending on external circumstances such as whether there is a hit in Lil cache 121 or a hit in L2 combined cache 130. Therefore, program access stage 1112 can take several clock cycles instead of one clock cycle as in the other stages.

[0114] The instructions executing in parallel constitute an execute packet. In this example, an execute packet can contain up to sixteen 32-bit wide slots for sixteen instructions. No two instructions in an execute packet can use the same functional unit. A slot is one of five types: 1) a self-contained instruction executed on one of the functional units of processor core 110 (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225, D2 unit 226, L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246); 2) a unitless instruction such as a NOP (no operation) instruction or multiple NOP instructions; 3) a branch instruction; 4) a constant field extension; and 5) a conditional code extension. Some of these slot types are further explained herein.

[0115] Dispatch and decode phases 1120 (FIG. 11) include instruction dispatch to appropriate execution unit (DS) stage 1121, instruction pre-decode (DC1) stage 1122, and instruction decode, operand read (DC2) stage 1123. During instruction dispatch to appropriate execution unit stage 1121, the fetch packets are split into execute packets and assigned to the appropriate functional units. During the instruction pre-decode stage 1122, the source registers, destination registers, and associated paths are decoded for the execution of the instructions in the functional units. During the instruction decode, operand read stage 1123, more detailed unit decodes are performed and operands are read from the register files.

[0116] Execution phase 1130 includes execution (E1 to E5) stages 1131 to 1135. Different types of instructions require different numbers of such stages to complete execution. The execution stages of the pipeline play an important role in understanding the device state at processor cycle boundaries.

[0117] During E1 stage 1131, the conditions for the instructions are evaluated and operands are operated on. As illustrated in FIG. 11, E1 stage 1131 can receive operands from a stream buffer 1141 and one of the register files shown schematically as 1142. For load and store instructions, address generation is performed, and address modifications are written to a register file. For branch instructions, the branch fetch packet in PG phase is affected. As illustrated in FIG. 11, load and store instructions access memory here shown schematically as memory 1151. For single-cycle instructions, results are written to a destination register file when any conditions for the instructions are evaluated as true. If a condition is evaluated as false, the instruction does not write any results or have any pipeline operation after E1 stage 1131.

[0118] During E2 stage 1132, load instructions send the address to memory. Store instructions send the address and data to memory. Single-cycle instructions that saturate results set the SAT bit in the control status register (CSR) if saturation occurs. For 2-cycle instructions, results are written to a destination register file.

[0119] During E3 stage 1133, data memory accesses are performed. Any multiply instructions that saturate results set the SAT bit in the control status register (CSR) if saturation occurs. For 3-cycle instructions, results are written to a destination register file.

[0120] During E4 stage 1134, load instructions bring data to the processor boundary. For 4-cycle instructions, results are written to a destination register file.

[0121] During E5 stage 1135, load instructions write data into a register as illustrated schematically in FIG. 11 with input from memory 1151 to E5 stage 1135.

[0122] FIG. 13 illustrates an example of the instruction coding 1300 of functional unit instructions used by this example. Each instruction includes 32 bits and controls the operation of one of the individually controllable functional units (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225, D2 unit 226, L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, C unit 245 and P unit 246).

[0123] The creg field 1301 (bits 29 to 31) and the z bit 1302 (bit 28) are optional fields used in conditional instructions. The bits are used for conditional instructions to identify the predicate register and the condition. The z bit 1302 (bit 28) indicates whether the predication is based upon zero or not zero in the predicate register. If z=1, the test is for equality with zero. If z=0, the test is for nonzero. The case of creg=0 and z=0 is treated as true to allow unconditional instruction execution. The creg field 1301 and the z field 1302 are encoded in the instruction as shown in Table 1.

[0124]

TABLE 1
Conditional	creg	z
Register	31	30	29	28
Unconditional	0	0	0	0
Reserved	0	0	0	1
A0	0	0	1	z
A1	0	1	0	z
A2	0	1	1	z
A3	1	0	0	z
A4	1	0	1	z
A5	1	1	0	z
Reserved	1	1	x	x

[0125] Execution of a conditional instruction is conditional upon the value stored in the specified data register. The data register is in the global scalar register file 211 for all functional units. Note that “z” in the z bit column refers to the zero/not zero comparison selection noted above and “x” is a don't care state. This coding specifies a subset of the sixteen global registers as predicate registers which preserves bits in the instruction coding. Note that unconditional instructions do not have the optional bits. For unconditional instructions, the bits in fields 1301 and 1302 (28 to 31) are used as additional opcode bits.

[0126] The dst field 1303 (bits 23 to 27) specifies a register in a corresponding register file as the destination of the instruction results.

[0127] The src2/cst field 1304 (bits 18 to 22) has several meanings depending on the instruction opcode field (bits 3 to 12 for all instructions and additionally bits 28 to 31 for unconditional instructions). One meaning specifies a register of a corresponding register file as the second operand. Another meaning is an immediate constant. Depending on the instruction type, the field 1304 is treated as an unsigned integer and zero extended to a specified data length or is treated as a signed integer and sign extended to the specified data length.

[0128] The src1 field 1305 (bits 13 to 17) specifies a register in a corresponding register file as the first source operand.

[0129] The opcode field 1306 (bits 3 to 12) for all instructions (and additionally bits 28 to 31 for unconditional instructions) specifies the type of instruction and designates appropriate instruction options including unambiguous designation of the functional unit used and operation performed. A detailed explanation of the opcode is beyond the scope of this description except for the instruction options described below.

[0130] The e bit 1307 (bit 2) is used for immediate constant instructions where the constant can be extended. If e=1, then the immediate constant is extended in a manner described below. If e=0, then the immediate constant is not extended and the immediate constant is specified by the src2/cst field 1304 (bits 18 to 22). Note that the e bit 1307 is used for some instructions. Accordingly, with proper coding, the e bit 1307 can be omitted from some instructions and the bit can be used as an additional opcode bit.

[0131] The s bit 1308 (bit 1) designates scalar data path side A 115 or vector data path side B 116. If s=0, then scalar data path side A 115 is selected which limits the functional unit to L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226 and the corresponding register files illustrated in FIG. 2. Similarly, s=1 selects vector data path side B 116 which limits the functional unit to L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244, P unit 246 and the corresponding register file illustrated in FIG. 2.

[0132] The p bit 1309 (bit 0) marks the execute packets. The p-bit determines whether the instruction executes in parallel with the following instruction. The p-bits are scanned from lower to higher address. If p=1 for the current instruction, then the next instruction executes in parallel with the current instruction. If p=0 for the current instruction, then the next instruction executes in the cycle after the current instruction. All instructions executing in parallel constitute an execute packet. An execute packet can contain up to sixteen instructions. Each instruction in an execute packet uses a different functional unit.

[0133] There are two different condition code extension slots. Each execute packet can contain one each of these unique 32-bit condition code extension slots which contains the 4-bit creg/z fields for the instructions in the same execute packet. FIG. 14 illustrates the coding for condition code extension slot 0 and FIG. 15 illustrates the coding for condition code extension slot 1.

[0134] FIG. 14 illustrates the coding for condition code extension slot 0 having 32 bits. Field 1401 (bits 28 to 31) specifies 4 creg/z bits assigned to the L1 unit 221 instruction in the same execute packet. Field 1402 (bits 27 to 24) specifies four creg/z bits assigned to the L2 unit 241 instruction in the same execute packet. Field 1403 (bits 20 to 23) specifies four creg/z bits assigned to the S1 unit 222 instruction in the same execute packet. Field 1404 (bits 16 to 19) specifies four creg/z bits assigned to the S2 unit 242 instruction in the same execute packet. Field 1405 (bits 12 to 15) specifies four creg/z bits assigned to the D1 unit 225 instruction in the same execute packet. Field 1406 (bits 8 to 11) specifies four creg/z bits assigned to the D2 unit 226 instruction in the same execute packet. Field 1407 (bits 6 and 7) is unused/reserved. Field 1408 (bits 0 to 5) is coded as a set of unique bits (CCEX0) to identify the condition code extension slot 0. Once the unique ID of condition code extension slot 0 is detected, the corresponding creg/z bits are employed to control conditional execution of any L1 unit 221, L2 unit 241, S1 unit 222, S2 unit 242, D1 unit 225 and D2 unit 226 instruction in the same execution packet. The creg/z bits are interpreted as shown in Table 1. If the corresponding instruction is conditional (includes creg/z bits), the corresponding bits in the condition code extension slot 0 override the condition code bits in the instruction. Setting the creg/z bits equal to “0000” makes the instruction unconditional. Thus, a properly coded condition code extension slot 0 can make some corresponding instructions conditional and some unconditional.

[0135] FIG. 15 illustrates the coding for condition code extension slot 1 having 32 bits. Field 1501 (bits 28 to 31) specifies four creg/z bits assigned to the M1 unit 223 instruction in the same execute packet. Field 1502 (bits 27 to 24) specifies four creg/z bits assigned to the M2 unit 243 instruction in the same execute packet. Field 1503 (bits 19 to 23) specifies four creg/z bits assigned to the C unit 245 instruction in the same execute packet. Field 1504 (bits 16 to 19) specifies four creg/z bits assigned to the N1 unit 224 instruction in the same execute packet. Field 1505 (bits 12 to 15) specifies four creg/z bits assigned to the N2 unit 244 instruction in the same execute packet. Field 1506 (bits 6 to 11) is unused/reserved. Field 1507 (bits 0 to 5) is coded as a set of unique bits (CCEX1) to identify the condition code extension slot 1. Once the unique ID of condition code extension slot 1 is detected, the corresponding creg/z bits are employed to control conditional execution of any M1 unit 223, M2 unit 243, C unit 245, N1 unit 224 and N2 unit 244 instruction in the same execution packet. These creg/z bits are interpreted as shown in Table 1. If the corresponding instruction is conditional (includes creg/z bits), the corresponding bits in the condition code extension slot 1 override the condition code bits in the instruction. Setting the creg/z bits equal to “0000” makes the instruction unconditional. Thus, a properly coded condition code extension slot 1 can make some instructions conditional and some unconditional.

[0136] Both condition code extension slot 0 and condition code extension slot 1 can include a p bit to define an execute packet as described above in conjunction with FIG. 13. In this example, as illustrated in FIGS. 14 and 15, code extension slot 0 and condition code extension slot 1 have bit 0 (p bit) encoded as 1. Thus, neither condition code extension slot 0 nor condition code extension slot 1 can be in the last instruction slot of an execute packet.

[0137] There are two different 32-bit constant extension slots. Each execute packet can contain one each of the unique constant extension slots which contains 27 bits to be concatenated as high order bits with the 5-bit constant field 1305 to form a 32-bit constant. As noted in the instruction coding description above, some instructions define the src2/cst field 1304 as a constant rather than a source register identifier. At least some of such instructions can employ a constant extension slot to extend the constant to 32 bits.

[0138] FIG. 16 illustrates the fields of constant extension slot 0. Each execute packet can include one instance of constant extension slot 0 and one instance of constant extension slot 1. FIG. 16 illustrates that constant extension slot 0 1600 includes two fields. Field 1601 (bits 5 to 31) constitutes the most significant 27 bits of an extended 32-bit constant including the target instruction scr2/cst field 1304 as the five least significant bits. Field 1602 (bits 0 to 4) is coded as a set of unique bits (CSTX0) to identify the constant extension slot 0. In this example, constant extension slot 0 1600 can be used to extend the constant of one of an L1 unit 221 instruction, data in a D1 unit 225 instruction, an S2 unit 242 instruction, an offset in a D2 unit 226 instruction, an M2 unit 243 instruction, an N2 unit 244 instruction, a branch instruction, or a C unit 245 instruction in the same execute packet. Constant extension slot 1 is similar to constant extension slot 0 except that bits 0 to 4 are coded as a set of unique bits (CSTX1) to identify the constant extension slot 1. In this example, constant extension slot 1 can be used to extend the constant of one of an L2 unit 241 instruction, data in a D2 unit 226 instruction, an S1 unit 222 instruction, an offset in a D1 unit 225 instruction, an M1 unit 223 instruction or an N1 unit 224 instruction in the same execute packet.

[0139] Constant extension slot 0 and constant extension slot 1 are used as follows. The target instruction is of the type permitting constant specification. In this example, the extension is implemented by replacing one input operand register specification field with the least significant bits of the constant as described above with respect to scr2/cst field 1304. Instruction decoder 113 determines this case, known as an immediate field, from the instruction opcode bits. The target instruction also includes one constant extension bit (e bit 1307) dedicated to signaling whether the specified constant is not extended (constant extension bit=0) or extended (constant extension bit=1). If instruction decoder 113 detects a constant extension slot 0 or a constant extension slot 1, instruction decoder 113 further checks the other instructions within the execute packet for an instruction corresponding to the detected constant extension slot. A constant extension is made if one corresponding instruction has a constant extension bit (e bit 1307) equal to 1.

[0140] FIG. 17 is a partial block diagram 1700 illustrating constant extension. FIG. 17 assumes that instruction decoder 113 (FIG. 1) detects a constant extension slot and a corresponding instruction in the same execute packet. Instruction decoder 113 supplies the twenty-seven extension bits from the constant extension slot (bit field 1601) and the five constant bits (bit field 1305) from the corresponding instruction to concatenator 1701. Concatenator 1701 forms a single 32-bit word from these two parts. In this example, the twenty-seven extension bits from the constant extension slot (bit field 1601) are the most significant bits and the five constant bits (bit field 1305) are the least significant bits. The combined 32-bit word is supplied to one input of multiplexer 1702. The five constant bits from the corresponding instruction field 1305 supply a second input to multiplexer 1702. Selection of multiplexer 1702 is controlled by the status of the constant extension bit. If the constant extension bit (e bit 1307) is 1 (extended), multiplexer 1702 selects the concatenated 32-bit input. If the constant extension bit is 0 (not extended), multiplexer 1702 selects the five constant bits from the corresponding instruction field 1305. The output of multiplexer 1702 supplies an input of sign extension unit 1703.

[0141] Sign extension unit 1703 forms the final operand value from the input from multiplexer 1702. Sign extension unit 1703 receives control inputs Scalar/Vector and Data Size. The Scalar/Vector input indicates whether the corresponding instruction is a scalar instruction or a vector instruction. The functional units of data path side A 115 (L1 unit 221, S1 unit 222, M1 unit 223, N1 unit 224, D1 unit 225 and D2 unit 226) perform scalar instructions. Any instruction directed to one of these functional units is a scalar instruction. Data path side B functional units L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244 and C unit 245 can perform scalar instructions or vector instructions. Instruction decoder 113 determines whether the instruction is a scalar instruction or a vector instruction from the opcode bits. P unit 246 may perform scalar instructions. The Data Size can be eight bits (byte B), sixteen bits (half-word H), 32 bits (word W), or 64 bits (double word D).

[0142] Table 2 lists the operation of sign extension unit 1703 for the various options.

[0143]

TABLE 2
Instruction	Operand	Constant
Type	Size	Length	Action
Scalar	B/H/W/D	5 bits	Sign extend to 64 bits
Scalar	B/H/W/D	32 bits	Sign extend to 64 bits
Vector	B/H/W/D	5 bits	Sign extend to operand size and
			replicate across whole vector
Vector	B/H/W	32 bits	Replicate 32-bit constant across
			each 32-bit (W) lane
Vector	D	32 bits	Sign extend to 64 bits
			and replicate
			across each 64-bit (D) lane

[0144] Both constant extension slot 0 and constant extension slot 1 can include a p bit to define an execute packet as described above in conjunction with FIG. 13. In this example, as in the case of the condition code extension slots, constant extension slot 0 and constant extension slot 1 have bit 0 (p bit) encoded as 1. Thus, neither constant extension slot 0 nor constant extension slot 1 can be in the last instruction slot of an execute packet.

[0145] An execute packet can include a constant extension slot 0 or 1 and more than one corresponding instruction marked constant extended (e bit=1). For such an occurrence, for constant extension slot 0, more than one of an L1 unit 221 instruction, data in a D1 unit 225 instruction, an S2 unit 242 instruction, an offset in a D2 unit 226 instruction, an M2 unit 243 instruction or an N2 unit 244 instruction in an execute packet can have an e bit of 1. For such an occurrence, for constant extension slot 1, more than one of an L2 unit 241 instruction, data in a D2 unit 226 instruction, an S1 unit 222 instruction, an offset in a D1 unit 225 instruction, an M1 unit 223 instruction or an N1 unit 224 instruction in an execute packet can have an e bit of 1. In one example, instruction decoder 113 determines that such an occurrence is an invalid operation and not supported. Alternately, the combination can be supported with extension bits of the constant extension slot applied to each corresponding functional unit instruction marked constant extended.

[0146] L1 unit 221, S1 unit 222, L2 unit 241, S2 unit 242 and C unit 245 often operate in a single instruction multiple data (SIMD) mode. In this SIMD mode, the same instruction is applied to packed data from the two operands. Each operand holds multiple data elements disposed in predetermined slots. SIMD operation is enabled by carry control at the data boundaries. Such carry control enables operations on varying data widths.

[0147] FIG. 18 illustrates the carry control logic. AND gate 1801 receives the carry output of bit N within the operand wide arithmetic logic unit (64 bits for scalar data path side A 115 functional units and 512 bits for vector data path side B 116 functional units). AND gate 1801 also receives a carry control signal which is further explained below. The output of AND gate 1801 is supplied to the carry input of bit N+1 of the operand wide arithmetic logic unit. AND gates such as AND gate 1801 are disposed between every pair of bits at a possible data boundary. For example, for 8-bit data such an AND gate will be between bits 7 and 8, bits 15 and 16, bits 23 and 24, etc. Each such AND gate receives a corresponding carry control signal. If the data size is the minimum size, each carry control signal is 0, effectively blocking carry transmission between the adjacent bits. The corresponding carry control signal is 1 if the selected data size requires both arithmetic logic unit sections. Table 3 below shows example carry control signals for the case of a 512-bit wide operand as used by vector data path side B 116 functional units which can be divided into sections of 8 bits, 16 bits, 32 bits, 64 bits, 128 bits or 256 bits. In Table 3, the upper 32 bits control the upper bits (bits 128 to 511) carries and the lower 32 bits control the lower bits (bits 0 to 127) carries. No control of the carry output of the most significant bit is needed, thus only 63 carry control signals are required.

[0148]

	TABLE 3
	Data Size	Carry Control Signals
	8 bits (B)	−000 0000 0000 0000 0000 0000 0000 0000
		0000 0000 0000 0000 0000 0000 0000 0000
	16 bits (H)	−101 0101 0101 0101 0101 0101 0101 0101
		0101 0101 0101 0101 0101 0101 0101 0101
	32 bits (W)	−111 0111 0111 0111 0111 0111 0111 0111
		0111 0111 0111 0111 0111 0111 0111 0111
	64 bits (D)	−111 1111 0111 1111 0111 1111 0111 1111
		0111 1111 0111 1111 0111 1111 0111 1111
	128 bits	−111 1111 1111 1111 0111 1111 1111 1111
		0111 1111 1111 1111 0111 1111 1111 1111
	256 bits	−111 1111 1111 1111 1111 1111 1111 1111
		0111 1111 1111 1111 1111 1111 1111 1111

[0149] Operation on data sizes that are integral powers of 2 (2^N) is common. However, the carry control technique is not limited to integral powers of 2 and can be applied to other data sizes and operand widths.

[0150] In this example, at least L2 unit 241 and S2 unit 242 employ two types of SIMD instructions using registers in predicate register file 234. In this example, the SIMD vector predicate instructions operate on an instruction specified data size. The data sizes include byte (8 bit) data, half word (16 bit) data, word (32 bit) data, double word (64 bit) data, quad word (128 bit) data and half vector (256 bit) data. In the first of these instruction types, the functional unit (L unit 241 or S unit 242) performs a SIMD comparison on packed data in two general data registers and supplies results to a predicate data register. The instruction specifies a data size, the two general data register operands, and the destination predicate register. In this example, each predicate data register includes one bit corresponding to each minimal data size portion of the general data registers. In the current example, the general data registers are 512 bits (64 bytes) and the predicate data registers are 64 bits (8 bytes). Each bit of a predicate data register corresponds to eight bits of a general data register. The comparison is performed on a specified data size (8, 16, 32, 64, 128 or 256 bits). If the comparison is true, then the functional unit supplies 1's to all predicate register bits corresponding the that data size portion. If the comparison is false, the functional unit supplies zeroes to the predicate register bits corresponding to that data size portion. In this example, the enabled comparison operations include: less than, greater than, and equal to.

[0151] In the second of the instruction types, the functional unit (L2 unit 241 or S2 unit 242) separately performs a first SIMD operation or a second SIMD operation on packed data in general data registers based upon the state of data in a predicate data register. The instruction specifies a data size, one or two general data register operands, a controlling predicate register, and a general data register destination. For example, a functional unit can select, for each data sized portion of two vector operands, a first data element of a first operand or a second data element of a second operand dependent upon the I/O state of corresponding bits in the predicate data register to store in the destination register. In another example, the data elements of a single vector operand can be saved to memory or not saved dependent upon the data of the corresponding bits of the predicate register.

[0152] The operations of P unit 245 permit a variety of compound vector SIMD operations based upon more than one vector comparison. For example, a range determination can be made using two comparisons. In a SIMD operation, a candidate vector is compared with a vector reference having the minimum of the range packed within a data register. The greater than result is scalar data with bits corresponding to the SIMD data width set to 0 or 1 depending upon the SIMD comparison and is stored in a predicate data register. Another SIMD comparison of the candidate vector is performed with another reference vector having the maximum of the range packed within a different data register produces another scalar with less than results stored in another predicate register. The P unit then ANDs the two predicate registers. The AND result indicates whether each SIMD data part of the candidate vector is within range or out of range. A P unit BITCNT instruction of the AND result can produce a count of the data elements within the comparison range. The P unit NEG function can be used to convert various expressions, such as: a less than comparison result to a greater than or equal comparison result; a greater than comparison result to a less than or equal to comparison result; or, an equal to comparison result to a not equal to comparison result.

Streaming Engine

[0153] FIG. 19 is a conceptual view of the streaming engine 125 of the example processor 100 of FIG. 1. FIG. 19 illustrates the processing of a single stream representative of the two streams controlled by streaming engine 125. Streaming engine 1900 includes stream address generator 1901. Stream address generator 1901 sequentially generates addresses of the elements of the stream and supplies these element addresses to system memory 1910. Memory 1910 recalls data stored at the element addresses (data elements) and supplies these data elements to data first-in-first-out (FIFO) buffer 1902. Data FIFO buffer 1902 provides buffering between memory 1910 and processor 1920. Data formatter 1903 receives the data elements from data FIFO memory 1902 and provides data formatting according to the stream definition. This process is described in more detail herein. Streaming engine 1900 supplies the formatted data elements from data formatter 1903 to the processor 1920. A program executing on processor 1920 consumes the data and generates an output.

[0154] Stream elements typically reside in system memory. The memory imposes no particular structure upon the stream. Programs define streams and thereby impose structure by specifying the stream attributes such as address of the first element of the stream, size and type of the elements in the stream, formatting for data in the stream, and the address sequence associated with the stream.

[0155] The streaming engine defines an address sequence for elements of the stream in terms of a pointer walking through memory. A multiple-level nested loop controls the path the pointer takes. An iteration count for a loop level indicates the number of times the level repeats. A dimension gives the distance between pointer positions of the loop level.

[0156] In a basic forward stream, the innermost loop consumes physically contiguous elements from memory as the implicit dimension of the innermost loop is one element. The pointer moves from element to element in consecutive, increasing order. In each level outside the inner loop, that loop moves the pointer to a new location based on the size of the dimension of the loop level.

[0157] This form of addressing allows programs to specify regular paths through memory using a small number of parameters. Table 4 lists the addressing parameters of a basic stream.

[0158]

	TABLE 4
	Parameter	Definition
	ELEM_	Size of each element in bytes
	BYTES
	ICNT0	Number of iterations for the innermost loop
		level 0. At loop level 0 all
		elements are physically contiguous.
		Implied DIM0 = ELEM_BYTES
	ICNT1	Number of iterations for loop level 1
	DIM1	Number of bytes between the starting points for
		consecutive iterations of loop level 1
	ICNT2	Number of iterations for loop level 2
	DIM2	Number of bytes between the starting points for
		consecutive iterations of loop level 2
	ICNT3	Number of iterations for loop level 3
	DIM3	Number of bytes between the starting points for
		consecutive iterations of loop level 3
	ICNT4	Number of iterations for loop level 4
	DIM4	Number of bytes between the starting points for
		consecutive iterations of loop level 4
	ICNT5	Number of iterations for loop level 5
	DIM5	Number of bytes between the starting points for
		consecutive iterations of loop level 5

[0159] In this example, ELEM_BYTES ranges from 1 to 64 bytes as shown in Table 5.

[0160]

TABLE 5
ELEM_BYTES	Stream Element Length
000	1 byte
001	2 bytes
010	4 bytes
011	8 bytes
100	16 bytes
101	32 bytes
110	64 bytes
111	Reserved

[0161] The definition above maps consecutive elements of the stream to increasing addresses in memory which is appropriate for many algorithms. Some algorithms are better served by reading elements in decreasing memory address order or reverse stream addressing. For example, a discrete convolution computes vector dot-products, as illustrated by expression (1).
(f*g)[t]=Σ_x=−∞^∞f[x]g[t−x]  (1)

[0162] In expression (1), f[ ] and g[ ] represent arrays in memory. For each output, the algorithm reads f[ ] in the forward direction and reads g[ ] in the reverse direction. Practical filters limit the range of indices for [x] and [t−x] to a finite number of elements. To support this pattern, the streaming engine supports reading elements in decreasing address order.

[0163] Matrix multiplication presents a unique problem to the streaming engine. Each element in the matrix product is a vector dot product between a row from the first matrix and a column from the second. Programs typically store matrices in row-major or column-major order. Row-major order stores all the elements of a single row contiguously in memory. Column-major order stores all elements of a single column contiguously in memory. Matrices are typically stored in the same order as the default array order for the language. As a result, only one of the two matrices in a matrix multiplication map on to the 2-dimensional stream definition of the streaming engine. In a typical example, an index steps through columns on one array and rows of the other array. The streaming engine supports implicit matrix transposition with transposed streams. Transposed streams avoid the cost of explicitly transforming the data in memory. Instead of accessing data in strictly consecutive-element order, the streaming engine effectively interchanges the inner two loop dimensions of the traversal order, fetching elements along the second dimension into contiguous vector lanes.

[0164] This algorithm works but is impractical to implement for small element sizes. Some algorithms work on matrix tiles which are multiple columns and rows together. Therefore, the streaming engine defines a separate transposition granularity. The hardware imposes a minimum granularity. The transpose granularity needs to be at least as large as the element size. Transposition granularity causes the streaming engine to fetch one or more consecutive elements from dimension 0 before moving along dimension 1. When the granularity equals the element size, a single column from a row-major array is fetched. Otherwise, the granularity specifies fetching two, four or more columns at a time from a row-major array. This is also applicable for column-major layout by exchanging row and column in the description. A parameter GRANULE indicates the transposition granularity in bytes.

[0165] Another common matrix multiplication technique exchanges the innermost two loops of the matrix multiply. The resulting inner loop no longer reads down the column of one matrix while reading across the row of another. For example, the algorithm may hoist one term outside the inner loop, replacing it with the scalar value. The innermost loop can be implemented with a single scalar by vector multiply followed by a vector add. Or, the scalar value can be duplicated across the length of the vector and a vector by vector multiply used. The streaming engine of this example directly supports the latter case and related use models with an element duplication mode. In this mode, the streaming engine reads a granule smaller than the full vector size and replicates that granule to fill the next vector output.

[0166] The streaming engine treats each complex number as a single element with two sub-elements that give the real and imaginary (rectangular) or magnitude and angle (polar) portions of the complex number. Not all programs or peripherals agree what order these sub-elements should appear in memory. Therefore, the streaming engine offers the ability to swap the two sub-elements of a complex number with no cost. The feature swaps the halves of an element without interpreting the contents of the element and can be used to swap pairs of sub-elements of any type, not just complex numbers.

[0167] Algorithms generally prefer to work at high precision, but high precision values require more storage and bandwidth than lower precision values. Commonly, programs store data in memory at low precision, promote those values to a higher precision for calculation, and then demote the values to lower precision for storage. The streaming engine supports such operations directly by allowing algorithms to specify one level of type promotion. In this example, every sub-element can be promoted to a larger type size with either sign or zero extension for integer types. In some examples, the streaming engine supports floating point promotion, promoting 16-bit and 32-bit floating point values to 32-bit and 64-bit formats, respectively.

[0168] While the streaming engine defines a stream as a discrete sequence of data elements, the processing unit core 110 consumes data elements packed contiguously in vectors. The vectors resemble streams as the vectors contain multiple homogeneous elements with some implicit sequence. Because the streaming engine reads streams, but the processing unit core 110 consumes vectors, the streaming engine maps streams onto vectors in a consistent way.

[0169] Vectors include equal-sized lanes, each lane containing a sub-element. The processing unit core 110 designates the rightmost lane of the vector as lane 0, regardless of current endian mode. Lane numbers increase right-to-left. The actual number of lanes within a vector varies depending on the length of the vector and the data size of the sub-element.

[0170] FIG. 20 illustrates the sequence of the formatting operations of formatter 1903. Formatter 1903 includes three sections: input section 2010, formatting section 2020, and output section 2030. Input section 2010 receives the data recalled from system memory 1910 as accessed by stream address generator 1901. The data can be via linear fetch stream 2011 or transposed fetch stream 2012.

[0171] Formatting section 2020 includes various formatting blocks. The formatting performed within formatter 1903 by the blocks is further described below. Complex swap block 2021 optionally swaps two sub-elements forming a complex number element. Type promotion block 2022 optionally promotes each data element into a larger data size. Promotion includes zero extension for unsigned integers and sign extension for signed integers. Decimation block 2023 optionally decimates the data elements. In this example, decimation can be 2:1 retaining every other data element or 4:1 retaining every fourth data element. Element duplication block 2024 optionally duplicates individual data elements. In this example, the data element duplication is an integer power of 2 (2N, where N is an integer) including 2×, 4×, 8×, 16×, 32× and 64×. In this example, data duplication can extend over multiple destination vectors. Vector length masking/group duplication block 2025 has two primary functions. An independently specified vector length VECLEN controls the data elements supplied to each output data vector. When group duplication is off, excess lanes in the output data vector are zero filled and these lanes are marked invalid. When group duplication is on, input data elements of the specified vector length are duplicated to fill the output data vector.

[0172] Output section 2030 holds the data for output to the corresponding functional units. Register and buffer for processor 2031 stores a formatted vector of data to be used as an operand by the functional units of processing unit core 110 (FIG. 1).

[0173] FIG. 21 illustrates an example of lane allocation in a vector. Vector 2100 is divided into eight 64-bit lanes (8×64 bits=512 bits, the vector length). Lane 0 includes bits 0 to 63, line 1 includes bits 64 to 127, lane 2 includes bits 128 to 191, lane 3 includes bits 192 to 255, lane 4 includes bits 256 to 319, lane 5 includes bits 320 to 383, lane 6 includes bits 384 to 447. and lane 7 includes bits 448 to 511.

[0174] FIG. 22 illustrates another example of lane allocation in a vector. Vector 2200 is divided into sixteen 32-bit lanes (16×32 bits=512 bits, the vector length). Lane 0 includes bits 0 to 31, line 1 includes bits 32 to 63, lane 2 includes bits 64 to 95, lane 3 includes bits 96 to 127, lane 4 includes bits 128 to 159, lane 5 includes bits 160 to 191, lane 6 includes bits 192 to 223, lane 7 includes bits 224 to 255, lane 8 includes bits 256 to 287, lane 9 includes bits 288 to 319, lane 10 includes bits 320 to 351, lane 11 includes bits 352 to 383, lane 12 includes bits 384 to 415, lane 13 includes bits 416 to 447, lane 14 includes bits 448 to 479, and lane 15 includes bits 480 to 511.

[0175] The streaming engine maps the innermost stream dimension directly to vector lanes. The streaming engine maps earlier elements within the innermost stream dimension to lower lane numbers and later elements to higher lane numbers, regardless of whether the stream advances in increasing or decreasing address order. Whatever order the stream defines, the streaming engine deposits elements in vectors in increasing-lane order. For non-complex data, the streaming engine places the first element in lane 0 of the vector processing unit core 110 (FIG. 1) fetches, the second in lane 1, and so on. For complex data, the streaming engine places the first element in lanes 0 and 1, the second element in lanes 2 and 3, and so on. Sub-elements within an element retain the same relative ordering regardless of the stream direction. For non-swapped complex elements, the sub-elements with the lower address of each pair are placed in the even numbered lanes, and the sub-elements with the higher address of each pair are placed in the odd numbered lanes. For swapped complex elements, the placement is reversed.

[0176] The streaming engine fills each vector processing unit core 110 fetches with as many elements as possible from the innermost stream dimension. If the innermost dimension is not a multiple of the vector length, the streaming engine zero pads the dimension to a multiple of the vector length. As noted below, the streaming engine also marks the lanes invalid. Thus, for higher-dimension streams, the first element from each iteration of an outer dimension arrives in lane 0 of a vector. The streaming engine maps the innermost dimension to consecutive lanes in a vector. For transposed streams, the innermost dimension includes groups of sub-elements along dimension 1, not dimension 0, as transposition exchanges these two dimensions.

[0177] Two-dimensional (2D) streams exhibit greater variety as compared to one-dimensional streams. A basic 2D stream extracts a smaller rectangle from a larger rectangle. A transposed 2D stream reads a rectangle column-wise instead of row-wise. A looping stream, where the second dimension overlaps first, executes a finite impulse response (FIR) filter taps which loops repeatedly over FIR filter samples providing a sliding window of input samples.

[0178] FIG. 23 illustrates a region of memory that can be accessed using a basic two-dimensional stream. The inner two dimensions, represented by ELEM_BYTES, ICNT0, DIM1 and ICNT1 (refer to Table 4), give sufficient flexibility to describe extracting a smaller rectangle 2320 having dimensions 2321 and 2322 from a larger rectangle 2310 having dimensions 2311 and 2312. In this example, rectangle 2320 is a 9 by 13 rectangle of 64-bit values and rectangle 2310 is a larger 11 by 19 rectangle. The following stream parameters define this stream: ICNT0=9, ELEM_BYTES=8, ICNT1=13, and DIM1=88 (11 times 8).

[0179] Thus, the iteration count in the 0-dimension 2321 is nine and the iteration count in the 1-dimension 2322 is thirteen. Note that the ELEM_BYTES scales the innermost dimension. The first dimension has ICNT0 elements of size ELEM_BYTES. The stream address generator does not scale the outer dimensions. Therefore, DIM1=88, which is eleven elements scaled by eight bytes per element.

[0180] FIG. 24 illustrates the order of elements within the example stream of FIG. 23. The streaming engine fetches elements for the stream in the order illustrated in order 2400. The first nine elements come from the first row of rectangle 2320, left-to-right in hops 1 to 8. The 10th through 24th elements comes from the second row, and so on. When the stream moves from the 9th element to the 10th element (hop 9 in FIG. 24), the streaming engine computes the new location based on the position of the pointer at the start of the inner loop, not the position of the pointer at the end of the first dimension. Thus, DIM1 is independent of ELEM_BYTES and ICNT0. DIM1 represents the distance between the first bytes of each consecutive row.

[0181] Transposed streams are accessed along dimension 1 before dimension 0. The following examples illustrate transposed streams with varying transposition granularity. FIG. 25 illustrates extracting a smaller rectangle 2520 (12×8) having dimensions 2521 and 2522 from a larger rectangle 2510 (14×13) having dimensions 2511 and 2512. In FIG. 25, ELEM_BYTES equal 2.

[0182] FIG. 26 illustrates how the streaming engine fetches the stream of the example stream of FIG. 25 with a transposition granularity of four bytes. Fetch pattern 2600 fetches pairs of elements from each row (because the granularity of four is twice the ELEM_BYTES of two), but otherwise moves down the columns. Once the streaming engine reaches the bottom of a pair of columns, the streaming engine repeats the pattern with the next pair of columns.

[0183] FIG. 27 illustrates how the streaming engine fetches the stream of the example stream of FIG. 25 with a transposition granularity of eight bytes. The overall structure remains the same. The streaming engine fetches four elements from each row (because the granularity of eight is four times the ELEM_BYTES of two) before moving to the next row in the column as shown in fetch pattern 2700.

[0184] The streams examined so far read each element from memory exactly once. A stream can read a given element from memory multiple times, in effect looping over a portion of memory. FIR filters exhibit two common looping patterns: re-reading the same filter taps for each output and reading input samples from a sliding window. Two consecutive outputs need inputs from two overlapping windows.

[0185] FIG. 28 illustrates the details of streaming engine 125 of FIG. 1. Streaming engine 125 contains three major sections: Stream 0 engine 2810; Stream 1 engine 2820; and Shared L2 Interfaces 2830. Stream 0 engine 2810 and Stream 1 2820 both contain identical hardware that operates in parallel. Stream 0 engine 2810 and Stream 1 engine 2820 both share L2 interfaces 2830. Each stream 0 engine 2810 and stream 1 engine 2820 provides processing unit core 110 (FIG. 1) data at a rate of up to 512 bits/cycle, every cycle, which is enabled by the dedicated stream paths and shared dual L2 interfaces.

[0186] Each streaming engine 125 includes a respective dedicated 6-dimensional (6D) stream address generator 2811/2821 that can each generate one new non-aligned request per cycle. As is further described herein, address generators 2811/2821 output 512-bit aligned addresses that overlap the elements in the sequence defined by the stream parameters.

[0187] Each address generator 2811/2821 connects to a respective dedicated micro table look-aside buffer (μTLB) 2812/2822. The μTLB 2812/2822 converts a single 48-bit virtual address to a 44-bit physical address each cycle. Each μTLB 2812/2822 has 8 entries, covering a minimum of 32 kB with 4 kB pages or a maximum of 16 MB with 2 MB pages. Each address generator 2811/2821 generates 2 addresses per cycle. The μTLB 2812/2822 only translates one address per cycle. To maintain throughput, streaming engine 125 operates under the assumption that most stream references are within the same 4 kB page. Thus, the address translation does not modify bits 0 to 11 of the address. If aout0 and aout1 line in the same 4 kB page (aout0[47:12] are the same aout1 [47:12]), then the μTLB 2812/2822 only translates aout0 and reuses the translation for the upper bits of both addresses.

[0188] Translated addresses are queued in respective command queue 2813/2823. These addresses are aligned with information from the respective corresponding Storage Allocation and Tracking block 2814/2824. Streaming engine 125 does not explicitly manage μTLB 2812/2822. The system memory management unit (MMU) invalidates μTLBs as necessary during context switches.

[0189] Storage Allocation and Tracking 2814/2824 manages the internal storage of the stream, discovering data reuse and tracking the lifetime of each piece of data. The block accepts two virtual addresses per cycle and binds those addresses to slots in the internal storage. The data store is organized as an array of slots. The streaming engine maintains following metadata to track the contents and lifetime of the data in each slot: 49-bit virtual address associated with the slot, valid bit indicating valid address, ready bit indicating data has arrived for the address, active bit indicating if there are any references outstanding to this data, and a last reference value indicating the most recent reference to this slot in the reference queue. The storage allocation and tracking are further described herein.

[0190] Respective reference queue 2815/2825 stores the sequence of references generated by the respective corresponding address generator 2811/2821. The reference sequence enables the data formatting network to present data to processing unit core 110 in the correct order. Each entry in respective reference queue 2815/2825 contains the information necessary to read data out of the data store and align the data for processing unit core 110. Respective reference queue 2815/2825 maintains the information listed in Table 6 in each slot.

[0191]

	TABLE 6
	Data Slot Low	Slot number for the lower half
		of data associated with aout0
	Data Slot High	Slot number for the upper half
		of data associated with aout1
	Rotation	Number of bytes to rotate data
		to align next element with lane 0
	Length	Number of valid bytes in this reference

[0192] Storage allocation and tracking 2814/2824 inserts references in reference queue 2815/2825 as address generator 2811/2821 generates new addresses. Storage allocation and tracking 2814/2824 removes references from reference queue 2815/2825 when the data becomes available and there is room in the stream head registers. As storage allocation and tracking 2814/2824 removes slot references from reference queue 2815/2825 and formats data, the references are checked for the last reference to the corresponding slots. Storage allocation and tracking 2814/2824 compares reference queue 2815/2825 removal pointer against the recorded last reference of the slot. If the pointer and the recorded last reference match, then storage allocation and tracking 2814/2824 marks the slot inactive once the data is no longer needed.

[0193] Streaming engine 125 has respective data storage 2816/2826 for a selected number of elements. Deep buffering allows the streaming engine to fetch far ahead in the stream, hiding memory system latency. Each data storage 2816/2826 accommodates two simultaneous read operations and two simultaneous write operations per cycle and each is therefore referred to a two-read, two-write (2r2w) data storage. In other examples, the amount of buffering can be different. In the current example, streaming engine 125 dedicates 32 slots to each stream with each slot tagged by a virtual address. Each slot holds 64 bytes of data in eight banks of eight bytes.

[0194] Data storage 2816/2826 and the respective storage allocation/tracking logic 2814/2824 and reference queues 2815/2825 implement the data FIFO 1902 discussed with reference to FIG. 19.

[0195] Respective butterfly network 2817/2827 includes a seven-stage butterfly network that implements the formatter 1903 (FIG. 19, FIG. 20). Butterfly network 2817/2827 receives 128 bytes of input and generates 64 bytes of output. The first stage of the butterfly is actually a half-stage that collects bytes from both slots that match a non-aligned fetch and merges the collected bytes into a single, rotated 64-byte array. The remaining six stages form a standard butterfly network. Respective butterfly network 2817/2827 performs the following operations: rotates the next element down to byte lane 0; promotes data types by a power of two, if requested; swaps real and imaginary components of complex numbers, if requested; and converts big endian to little endian if processing unit core 110 is presently in big endian mode. The user specifies element size, type promotion, and real/imaginary swap as part of the parameters of the stream.

[0196] Streaming engine 125 attempts to fetch and format data ahead of processing unit core 110's demand in order to maintain full throughput. Respective stream head registers 2818/2828 provide a small amount of buffering so that the process remains fully pipelined. Respective stream head registers 2818/2828 are not directly architecturally visible. Each stream also has a respective stream valid register 2819/2829. Valid registers 2819/2829 indicate which elements in the corresponding stream head registers 2818/2828 are valid. The outputs of stream head registers 2818/2828 and valid registers 2819/2829 are provided to processing unit core 110 via buses 2840/2841.

[0197] The two streams 2810/2820 share a pair of independent L2 interfaces 2830: L2 Interface A (IFA) 2833 and L2 Interface B (IFB) 2834. Each L2 interface provides 512 bits/cycle throughput direct to the L2 130 (FIG. 1) via respective buses 147/149 for an aggregate bandwidth of 1024 bits/cycle. The L2 interfaces use the credit-based multicore bus architecture (MBA) protocol. The MBA protocol is described in more detail in U.S. Pat. No. 9,904,645, “Multicore Bus Architecture with Non-Blocking High Performance Transaction Credit System,” which is incorporated by reference herein. The L2 controller assigns a pool of command credits to each interface. The pool has sufficient credits so that each interface can send sufficient requests to achieve full read-return bandwidth when reading L2 RAM, L2 cache and multicore shared memory controller (MSMC) memory, as described in more detail herein.

[0198] To maximize performance, in this example both streams can use both L2 interfaces, allowing a single stream to send a peak command rate of two requests per cycle. Each interface prefers one stream over the other, but this preference changes dynamically from request to request. IFA 2833 and IFB 2834 prefer opposite streams, when IFA 2833 prefers Stream 0, IFB 2834 prefers Stream 1 and vice versa.

[0199] Respective arbiter 2831/2832 ahead of each respective interface 2833/2834 applies the following basic protocol on every cycle having credits available. Arbiter 2831/2832 checks if the preferred stream has a command ready to send. If so, arbiter 2831/2832 chooses that command. Arbiter 2831/2832 next checks if an alternate stream has at least two requests ready to send, or one command and no credits. If so, arbiter 2831/2832 pulls a command from the alternate stream. If either interface issues a command, the notion of preferred and alternate streams swap for the next request. Using this algorithm, the two interfaces dispatch requests as quickly as possible while retaining fairness between the two streams. The first rule ensures that each stream can send a request on every cycle that has available credits. The second rule provides a mechanism for one stream to borrow the interface of the other when the second interface is idle. The third rule spreads the bandwidth demand for each stream across both interfaces, ensuring neither interface becomes a bottleneck.

[0200] Respective coarse grain rotator 2835/2836 enables streaming engine 125 to support a transposed matrix addressing mode. In this mode, streaming engine 125 interchanges the two innermost dimensions of the multidimensional loop to access an array column-wise rather than row-wise. Respective rotators 2835/2836 are not architecturally visible.

[0201] FIG. 29 illustrates an example stream template register 2900. The stream definition template provides the full structure of a stream that contains data. The iteration counts and dimensions provide most of the structure, while the various flags provide the rest of the details. In this example, a single stream template 2900 is defined for all data-containing streams. All stream types supported by the streaming engine are covered by the template 2900. The streaming engine supports a six-level loop nest for addressing elements within the stream. Most of the fields in the stream template 2900 map directly to the parameters in that algorithm. The numbers above the fields are bit numbers within a 256-bit vector. Table 7 shows the stream field definitions of a stream template.

[0202]

	TABLE 7
		FIG. 29
	Field	Reference		Size
	Name	Number	Description	Bits
	ICNT0	2901	Iteration count for loop 0	32
	ICNT1	2902	Iteration count for loop 1	16
	ICNT2	2903	Iteration count for loop 2	16
	ICNT3	2904	Iteration count for loop 3	32
	ICNT4	2905	Iteration count for loop 4	32
	ICNT5	2906	Iteration count for loop 5	32
	DIM1	2911	Signed dimension for loop 1	32
	DIM2	2912	Signed dimension for loop 2	32
	DIM3	2913	Signed dimension for loop 3	32
	DIM4	2914	Signed dimension for loop 4	32
	DIM5	2915	Signed dimension for loop 5	32
	FLAGS	2921	Stream modifier flags	64

[0203] Loop 0 is the innermost loop and loop 5 is the outermost loop. In the current example, DIM0 is equal to ELEM_BYTES defining physically contiguous data. Thus, the stream template register 2900 does not define DIM0. Streaming engine 125 interprets iteration counts as unsigned integers and dimensions as unscaled signed integers. An iteration count of zero at any level (ICNT0, ICNT1, ICNT2, ICNT3, ICNT4 or ICNT5) indicates an empty stream. Each iteration count must be at least one to define a valid stream. The template above specifies the type of elements, length and dimensions of the stream. The stream instructions separately specify a start address, e.g., by specification of a scalar register in scalar register file 211 which stores the start address. Thus, a program can open multiple streams using the same template but different registers storing the start address.

[0204] As described further below, the stream template register 2900 includes two width parameters (INVERSE_DECDIMx_WIDTH), where “x” is “1” or “2” in this example. Further details about these width parameters is described below.

[0205] FIG. 30 illustrates an example of sub-field definitions of the flags field 2921 shown in FIG. 29. As shown in FIG. 30, the flags field 2911 is 6 bytes or 48 bits. FIG. 30 shows bit numbers of the fields. Table 8 shows the definition of these fields.

[0206]

TABLE 8
	FIG. 30
	Reference		Size
Field Name	Number	Description	Bits
ELTYPE	3001	Type of data element	4
TRANSPOSE	3002	Two-dimensional transpose mode	3
PROMOTE	3003	Promotion mode	3
VECLEN	3004	Stream vector length	3
ELDUP	3005	Element duplication	3
GRDUP	3006	Group duplication	1
DECIM	3007	Element decimation	2
DIMFMT	3009	Stream dimensions format	3
DIR	3010	Stream direction	1
		0 forward direction
		1 reverse direction
CBK0	3011	First circular block size number	4
CBK1	3012	Second circular block size number	4
AM0	3013	Addressing mode for loop 0	2
AM1	3014	Addressing mode for loop 1	2
AM2	3015	Addressing mode for loop 2	2
AM3	3016	Addressing mode for loop 3	2
AM4	3017	Addressing mode for loop 4	2
AM5	3018	Addressing mode for loop 5	2
TEMPLATE_	3024	Template format selection	3
FMT

[0207] The Element Type (ELTYPE) field 3001 defines the data type of the elements in the stream. The coding of the four bits of the ELTYPE field 3001 is defined as shown in Table 9.

[0208]

TABLE 9
		Sub-element	Total Element
ELTYPE	Real/Complex	Size Bits	Size Bits
0000	real	8	8
0001	real	16	16
0010	real	32	32
0011	real	64	64
0100	reserved
0101	reserved
0110	reserved
0111	reserved
1000	complex	8	16
	no swap
1001	complex	16	32
	no swap
1010	complex	32	64
	no swap
1011	complex	64	128
	no swap
1100	complex	8	16
	swapped
1101	complex	16	32
	swapped
1110	complex	32	64
	swapped
1111	complex	64	128
	swapped

[0209] Real/Complex Type determines whether the streaming engine treats each element as a real number or two parts (real/imaginary or magnitude/angle) of a complex number and also specifies whether to swap the two parts of complex numbers. Complex types have a total element size twice the sub-element size. Otherwise, the sub-element size equals the total element size.

[0210] Sub-Element Size determines the type for purposes of type promotion and vector lane width. For example, 16-bit sub-elements get promoted to 32-bit sub-elements or 64-bit sub-elements when a stream requests type promotion. The vector lane width matters when processing unit core 110 (FIG. 1) operates in big endian mode, as the core 110 lays out vectors in little endian order.

[0211] Total Element Size specifies the minimal granularity of the stream which determines the number of bytes the stream fetches for each iteration of the innermost loop. Streams read whole elements, either in increasing or decreasing order. Therefore, the innermost dimension of a stream spans ICNT0×total-element-size bytes.

[0212] The TRANSPOSE field 3002 determines whether the streaming engine accesses the stream in a transposed order. The transposed order exchanges the inner two addressing levels. The TRANSPOSE field 3002 also indicated the granularity for transposing the stream. The coding of the three bits of the TRANSPOSE field 3002 is defined as shown in Table 10 for normal 2D operations.

[0213]

TABLE 10
Transpose	Meaning
000	Transpose disabled
001	Transpose on 8-bit boundaries
010	Transpose on 16-bit boundaries
011	Transpose on 32-bit boundaries
100	Transpose on 64-bit boundaries
101	Transpose on 128-bit boundaries
110	Transpose on 256-bit boundaries
111	Reserved

[0214] Streaming engine 125 can transpose data elements at a different granularity than the element size thus allowing programs to fetch multiple columns of elements from each row. The transpose granularity cannot be smaller than the element size. The TRANSPOSE field 3002 interacts with the DIMFMT field 3009 in a manner further described below.

[0215] The PROMOTE field 3003 controls whether the streaming engine promotes sub-elements in the stream and the type of promotion. When enabled, streaming engine 125 promotes types by powers-of-2 sizes. The coding of the three bits of the PROMOTE field 3003 is defined as shown in Table 11.

[0216]

TABLE 11
PRO-	Promotion	Promotion	Resulting Sub-element Size
MOTE	Factor	Type	8-bit	16-bit	32-bit	64-bit
000	1x	N/A	8-bit	16-bit	32-bit	64-bit
001	2x	zero extend	16-bit	32-bit	64-bit	Invalid
010	4x	zero extend	32-bit	64-bit	Invalid	Invalid
011	8x	zero extend	64-bit	Invalid	Invalid	Invalid
100	reserved
101	2x	sign extend	16-bit	32-bit	64-bit	Invalid
110	4x	sign extend	32-bit	64-bit	Invalid	Invalid
111	8x	sign extend	64-bit	Invalid	Invalid	Invalid

[0217] When PROMOTE is 000, corresponding to a 1×promotion, each sub-element is unchanged and occupies a vector lane equal in width to the size specified by ELTYPE. When PROMOTE is 001, corresponding to a 2× promotion and zero extend, each sub-element is treated as an unsigned integer and zero extended to a vector lane twice the width specified by ELTYPE. A 2× promotion is invalid for an initial sub-element size of 64 bits. When PROMOTE is 010, corresponding to a 4× promotion and zero extend, each sub-element is treated as an unsigned integer and zero extended to a vector lane four times the width specified by ELTYPE. A 4× promotion is invalid for an initial sub-element size of 32 or 64 bits. When PROMOTE is 011, corresponding to an 8× promotion and zero extend, each sub-element is treated as an unsigned integer and zero extended to a vector lane eight times the width specified by ELTYPE. An 8× promotion is invalid for an initial sub-element size of 16, 32 or 64 bits. When PROMOTE is 101, corresponding to a 2× promotion and sign extend, each sub-element is treated as a signed integer and sign extended to a vector lane twice the width specified by ELTYPE. A 2× promotion is invalid for an initial sub-element size of 64 bits. When PROMOTE is 110, corresponding to a 4× promotion and sign extend, each sub-element is treated as a signed integer and sign extended to a vector lane four times the width specified by ELTYPE. A 4× promotion is invalid for an initial sub-element size of 32 or 64 bits. When PROMOTE is 111, corresponding to an 8× promotion and zero extend, each sub-element is treated as a signed integer and sign extended to a vector lane eight times the width specified by ELTYPE. An 8× promotion is invalid for an initial sub-element size of 16, 32 or 64 bits.

[0218] The VECLEN field 3004 defines the stream vector length for the stream in bytes. Streaming engine 125 breaks the stream into groups of elements that are VECLEN bytes long. The coding of the three bits of the VECLEN field 3004 is defined as shown in Table 12.

[0219]

TABLE 12
VECLEN	Stream Vector Length
000	1 byte
001	2 bytes
010	4 bytes
011	8 bytes
100	16 bytes
101	32 bytes
110	64 bytes
111	Reserved

[0220] VECLEN cannot be less than the product of the element size in bytes and the duplication factor. As shown in Table 11, the maximum VECLEN of 64 bytes equals the preferred vector size of vector data path side B 116. When VECLEN is shorter than the native vector width of processing unit core 110, streaming engine 125 pads the extra lanes in the vector provided to processing unit core 110. The GRDUP field 3006 determines the type of padding. The VECLEN field 3004 interacts with ELDUP field 3005 and GRDUP field 3006 in a manner detailed below.

[0221] The ELDUP field 3005 specifies the number of times to duplicate each element. The element size multiplied with the element duplication amount cannot exceed the 64 bytes. The coding of the three bits of the ELDUP field 3005 is defined as shown in Table 13.

[0222]

TABLE 13
ELDUP	Duplication Factor
000	No Duplication
001	2 times
010	4 times
011	8 times
100	16 times
101	32 times
110	64 times
111	Reserved

[0223] The ELDUP field 3005 interacts with VECLEN field 3004 and GRDUP field 3006 in a manner detailed below. The nature of the relationship between the permitted element size, the element duplication factor, and the destination vector length requires that a duplicated element that overflows the first destination register fills an integer number of destination registers upon completion of duplication. The data of the additional destination registers eventually supplies the respective stream head register 2818/2828. Upon completion of duplication of a first data element, the next data element is rotated down to the least significant bits of source register 3100 discarding the first data element. The process then repeats for the new data element.

[0224] The GRDUP bit 3006 determines whether group duplication is enabled. If GRDUP bit 3006 is 0, then group duplication is disabled. If the GRDUP bit 3006 is 1, then group duplication is enabled. When enabled by GRDUP bit 3006, streaming engine 125 duplicates a group of elements to fill the vector width. VECLEN field 3004 defines the length of the group to replicate. When VECLEN field 3004 is less than the vector length of processing unit core 110 and GRDUP bit 3006 enables group duplication, streaming engine 125 fills the extra lanes (see FIGS. 21 and 22) with additional copies of the stream vector. Because stream vector length and vector length of processing unit core 110 are integral powers of two, group duplication produces an integral number of duplicate copies. Note GRDUP and VECLEN do not specify the number of duplications. The number of duplications performed is based upon the ratio of VECLEN to the native vector length, which is 64 bytes/512 bits in this example.

[0225] The GRDUP field 3006 specifies how streaming engine 125 pads stream vectors for bits following the VECLEN length to the vector length of processing unit core 110. When GRDUP bit 3006 is 0, streaming engine 125 fills the extra lanes with zeros and marks the extra vector lanes invalid. When GRDUP bit 3006 is 1, streaming engine 125 fills extra lanes with copies of the group of elements in each stream vector. Setting GRDUP bit 3006 to 1 has no effect when VECLEN is set to the native vector width of processing unit core 110. VECLEN must be at least as large as the product of ELEM_BYTES and the element duplication factor ELDUP. That is, an element or the duplication factor number of elements cannot be separated using VECLEN.

[0226] Group duplication operates to the destination vector size. Group duplication does not change the data supplied when the product of the element size ELEM_BYTES and element duplication factor ELDUP equals or exceeds the destination vector width. Under such conditions, the states of the GRDUP bit 3006 and the VECLEN field 3004 have no effect on the supplied data.

[0227] The set of examples below illustrate the interaction between VECLEN and GRDUP. Each of the following examples show how the streaming engine maps a stream onto vectors across different stream vector lengths and the vector size of vector data path side B 116. The stream of this example includes twenty-nine elements (E0 to E28) of 64 bits/8 bytes. The stream can be a linear stream of twenty-nine elements or an inner loop of 29 elements. The tables illustrate eight byte lanes such as shown in FIG. 21. Each illustrated vector is stored in the respective stream head register 2818/2828 in turn.

[0228] Table 14 illustrates how the example stream maps onto bits within the 64-byte processor vectors when VECLEN is 64 bytes.

[0229]

TABLE 14
Processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	E7	E6	E5	E4	E3	E2	E1	E0
2	E15	E14	E13	E12	E11	E10	E9	E8
3	E23	E22	E21	E20	E19	E18	E17	E16
4	0	0	0	E28	E27	E26	E25	E24

[0230] As shown in Table 14, the stream extends over four vectors. As previously described, the lanes within vector 4 that extend beyond the stream are zero filled. When VECLEN has a size equal to the native vector length, the value of GRDUP does not matter as no duplication can take place with such a VECLEN.

[0231] Table 15 shows the same parameters as shown in Table 14, except with VECLEN of 32 bytes. Group duplicate is disabled (GRDUP=0).

[0232]

TABLE 15
Processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	0	0	0	0	E3	E2	E1	E0
2	0	0	0	0	E7	E6	E5	E4
3	0	0	0	0	E11	E10	E9	E8
4	0	0	0	0	E15	E14	E13	E12
5	0	0	0	0	E19	E18	E17	E16
6	0	0	0	0	E23	E22	E21	E20
7	0	0	0	0	E27	E26	E25	E24
8	0	0	0	0	0	0	0	E28

[0233] The twenty-nine elements of the stream are distributed over lanes 0 to 3 in eight vectors. Extra lanes 4 to 7 in vectors 1-7 are zero filled. In vector 8, lane 1 has a stream element (E28) and the other lanes are zero filled.

[0234] Table 16 shows the same parameters as shown in Table 14, except with VECLEN of sixteen bytes. Group duplicate is disabled (GRDUP=0).

[0235]

TABLE 16
Processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	0	0	0	0	0	0	E1	E0
2	0	0	0	0	0	0	E3	E2
3	0	0	0	0	0	0	E5	E4
4	0	0	0	0	0	0	E7	E6
5	0	0	0	0	0	0	E9	E8
6	0	0	0	0	0	0	E11	E10
7	0	0	0	0	0	0	E13	E12
8	0	0	0	0	0	0	E15	E14
9	0	0	0	0	0	0	E17	E16
10	0	0	0	0	0	0	E19	E18
11	0	0	0	0	0	0	E21	E20
12	0	0	0	0	0	0	E23	E22
13	0	0	0	0	0	0	E25	E24
14	0	0	0	0	0	0	E27	E26
15	0	0	0	0	0	0	0	E28

[0236] The twenty-nine elements of the stream are distributed over lane 0 and lane 1 in fifteen vectors. Extra lanes 2 to 7 in vectors 1-14 are zero filled. In vector 15, lane 1 has a stream element (E28) and the other lanes are zero filled.

[0237] Table 17 shows the same parameters as shown in Table 14, except with VECLEN of eight bytes. Group duplicate is disabled (GRDUP=0).

[0238]

TABLE 17
Processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	0	0	0	0	0	0	0	E0
2	0	0	0	0	0	0	0	E1
3	0	0	0	0	0	0	0	E2
4	0	0	0	0	0	0	0	E3
5	0	0	0	0	0	0	0	E4
6	0	0	0	0	0	0	0	E5
7	0	0	0	0	0	0	0	E6
8	0	0	0	0	0	0	0	E7
9	0	0	0	0	0	0	0	E8
10	0	0	0	0	0	0	0	E9
11	0	0	0	0	0	0	0	E10
12	0	0	0	0	0	0	0	E11
13	0	0	0	0	0	0	0	E12
14	0	0	0	0	0	0	0	E13
15	0	0	0	0	0	0	0	E14
16	0	0	0	0	0	0	0	E15
17	0	0	0	0	0	0	0	E16
18	0	0	0	0	0	0	0	E17
19	0	0	0	0	0	0	0	E18
20	0	0	0	0	0	0	0	E19
21	0	0	0	0	0	0	0	E20
22	0	0	0	0	0	0	0	E21
23	0	0	0	0	0	0	0	E22
24	0	0	0	0	0	0	0	E23
25	0	0	0	0	0	0	0	E24
26	0	0	0	0	0	0	0	E25
27	0	0	0	0	0	0	0	E26
28	0	0	0	0	0	0	0	E27
29	0	0	0	0	0	0	0	E28

[0239] The twenty-nine elements of the stream appear in lane 0 in twenty-nine vectors. Extra lanes 1-7 in vectors 1-29 are zero filled.

[0240] Table 18 shows the same parameters as shown in Table 15, except with VECLEN of thirty-two bytes and group duplicate is enabled (GRDUP=1).

[0241]

TABLE 18
Processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	E3	E2	E1	E0	E3	E2	E1	E0
2	E7	E6	E5	E4	E7	E6	E5	E4
3	E11	E10	E9	E8	E11	E10	E9	E8
4	E15	E14	E13	E12	E15	E14	E13	E12
5	E19	E18	E17	E16	E19	E18	E17	E16
6	E23	E22	E21	E20	E23	E22	E21	E20
7	E27	E26	E25	E24	E27	E26	E25	E24
8	0	0	0	E28	0	0	0	E28

[0242] The twenty-nine elements of the stream are distributed over lanes 0-7 in eight vectors. Each vector 1-7 includes four elements duplicated. The duplication factor (2) results because VECLEN (32 bytes) is half the native vector length of 64 bytes. In vector 8, lane 0 has a stream element (E28) and lanes 1-3 are zero filled. Lanes 4-7 of vector 9 duplicate this pattern.

[0243] Table 19 shows the same parameters as shown in Table 16, except with VECLEN of sixteen bytes. Group duplicate is enabled (GRDUP=1).

[0244]

TABLE 19
processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	E1	E0	E1	E0	E1	E0	E1	E0
2	E3	E2	E3	E2	E3	E2	E3	E2
3	E5	E4	E5	E4	E5	E4	E5	E4
4	E7	E6	E7	E6	E7	E6	E7	E6
5	E9	E8	E9	E8	ES	E8	ES	E8
6	E11	E10	E11	E10	E11	E10	E11	E10
7	E13	E12	E13	E12	E13	E12	E13	E12
8	E15	E14	E15	E14	E15	E14	E15	E14
9	E17	E16	E17	E16	E17	E16	E17	E16
10	E19	E18	E19	E18	E19	E18	E19	E18
11	E21	E20	E21	E20	E21	E20	E21	E20
12	E23	E22	E23	E22	E23	E22	E23	E22
13	E25	E24	E25	E24	E25	E24	E25	E24
14	E27	E26	E27	E26	E27	E26	E27	E26
15	0	E28	0	E28	0	E28	0	E28

[0245] The twenty-nine elements of the stream are distributed over lanes 0-7 in fifteen vectors. Each vector 1-7 includes two elements duplicated four times. The duplication factor (4) results because VECLEN (16 bytes) is one quarter the native vector length of 64 bytes. In vector 15, lane 0 has a stream element (E28) and lane 1 is zero filled. This pattern is duplicated in lanes 2 and 3, lanes 4 and 5, and lanes 6 and 7 of vector 15.

[0246] Table 20 shows the same parameters as shown in Table 17, except with VECLEN of eight bytes. Group duplicate is enabled (GRDUP=1).

[0247]

TABLE 20
Processor	Lane	Lane	Lane	Lane	Lane	Lane	Lane	Lane
Vectors	7	6	5	4	3	2	1	0
1	E0	E0	E0	E0	E0	E0	E0	E0
2	E1	E1	E1	E1	E1	E1	E1	E1
3	E2	E2	E2	E2	E2	E2	E2	E2
4	E3	E3	E3	E3	E3	E3	E3	E3
5	E4	E4	E4	E4	E4	E4	E4	E4
6	E5	E5	E5	E5	E5	E5	E5	E5
7	E6	E6	E6	E6	E6	E6	E6	E6
8	E7	E7	E7	E7	E7	E7	E7	E7
9	E8	E8	E8	E8	E8	E8	E8	E8
10	E9	E9	E9	E9	E9	E9	E9	E9
11	E10	E10	E10	E10	E10	E10	E10	E10
12	E11	E11	E11	E11	E11	E11	E11	E11
13	E12	E12	E12	E12	E12	E12	E12	E12
14	E13	E13	E13	E13	E13	E13	E13	E13
15	E14	E14	E14	E14	E14	E14	E14	E14
16	E15	E15	E15	E15	E15	E15	E15	E15
17	E16	E16	E16	E16	E16	E16	E16	E16
18	E17	E17	E17	E17	E17	E17	E17	E17
19	E18	E18	E18	E18	E18	E18	E18	E18
20	E19	E19	E19	E19	E19	E19	E19	E19
21	E20	E20	E20	E20	E20	E20	E20	E20
22	E21	E21	E21	E21	E21	E21	E21	E21
23	E22	E22	E22	E22	E22	E22	E22	E22
24	E23	E23	E23	E23	E23	E23	E23	E23
25	E24	E24	E24	E24	E24	E24	E24	E24
26	E25	E25	E25	E25	E25	E25	E25	E25
27	E26	E26	E26	E26	E26	E26	E26	E26
28	E27	E27	E27	E27	E27	E27	E27	E27
29	E28	E28	E28	E28	E28	E28	E28	E28

[0248] The twenty-nine elements of the stream all appear on lanes 0 to 7 in twenty-nine vectors. Each vector includes one element duplicated eight times. The duplication factor (8) results because VECLEN (8 bytes) is one eighth the native vector length of 64 bytes. Thus, each lane is the same in vectors 1-29.

[0249] FIG. 31 illustrates an example of vector length masking/group duplication block 2025 (see FIG. 20) that is included within formatter block 1903 of FIG. 19. Input register 3100 receives a vector input from element duplication block 2024 shown in FIG. 20. Input register 3100 includes 64 bytes arranged in 64 1-byte blocks byte0 to byte63. Note that bytes byte0 to byte63 are each equal in length to the minimum of ELEM_BYTES. A set of multiplexers 3101 to 3163 couple input bytes from source register 3100 to output register 3170. Each respective multiplexer 3101 to 3163 supplies an input to a respective byte1 to byte63 of output register 3170. Not all input bytes byte0 to byte63 of input register 3100 are coupled to every multiplexer 3101 to 3163. Note there is no multiplexer supplying byte0 of output register 3170. In this example, byte0 of output register 3170 is supplied by byte0 of input register 3100.

[0250] Multiplexers 3101 to 3163 are controlled by multiplexer control encoder 3180. Multiplexer control encoder 3180 receives ELEM_BYTES, VECLEN and GRDUP input signals and generates respective control signals for multiplexers 3101 to 3163. ELEM_BYTES and ELDUP are supplied to multiplexer control encoder 3180 to check to see that VECLEN is at least as great as the product of ELEM_BYTES and ELDUP. In operation, multiplexer control encoder 3180 controls multiplexers 3101 to 3163 to transfer least significant bits equal in number to VECLEN from input register 3100 to output register 3170. If GRDUP=0 indicating group duplication disabled, then multiplexer control encoder 3180 controls the remaining multiplexers 3101 to 3163 to transfer zeros to all bits in the remaining most significant lanes of output register 3170. If GRDUP=1 indicating group duplication enabled, then multiplexer control encoder 3180 controls the remaining multiplexers 3101 to 3163 to duplicate the VECLEN number of least significant bits of input register 3100 into the most significant lanes of output register 3170. This control is similar to the element duplication control described above and fills the output register 3170 with the first vector. For the next vector, data within input register 3100 is rotated down by VECLEN, discarding the previous VECLEN least significant bits. The rate of data movement in formatter 1903 (FIG. 19) is set by the rate of consumption of data by processing unit core 110 (FIG. 1) via stream read and advance instructions described below. The group duplication formatting repeats as long as the stream includes additional data elements.

[0251] Element duplication (ELDUP) and group duplication (GRUDP) are independent. Note these features include independent specification and parameter setting. Thus, element duplication and group duplication can be used together or separately. Because of how these are specified, element duplication permits overflow to the next vector while group duplication does not.

[0252] Referring again to FIG. 30, the DECIM field 3007 controls data element decimation of the corresponding stream. Streaming engine 125 deletes data elements from the stream upon storage in respective stream head registers 2818/2828 for presentation to the requesting functional unit. Decimation removes whole data elements, not sub-elements. The DECIM field 3007 is defined as listed in Table 21.

[0253]

TABLE 21
DECIM	Decimation Factor
00	No Decimation
01	2 times
10	4 times
11	Reserved

[0254] If DECIM field 3007 equals 00, then no decimation occurs. The data elements are passed to the corresponding stream head registers 2818/2828 without change. If DECIM field 3007 equals 01, then 2:1 decimation occurs. Streaming engine 125 removes odd number elements from the data stream upon storage in the stream head registers 2818/2828. Limitations in the formatting network require 2:1 decimation to be employed with data promotion by at least 2×(PROMOTE cannot be 000), ICNT0 must be multiple of 2, and the total vector length (VECLEN) must be large enough to hold a single promoted, duplicated element. For transposed streams (TRANSPOSE 0), the transpose granule must be at least twice the element size in bytes before promotion. If DECIM field 3007 equals 10, then 4:1 decimation occurs. Streaming engine 125 retains every fourth data element removing three elements from the data stream upon storage in the stream head registers 2818/2828. Limitations in the formatting network require 4:1 decimation to be employed with data promotion by at least 4×(PROMOTE cannot be 000, 001 or 101), ICNT0 must be a multiple of 4 and the total vector length (VECLEN) must be large enough to hold a single promoted, duplicated element. For transposed streams (TRANSPOSE 0), in one example, decimation removes columns, and does not remove rows. Thus, in such cases, the transpose granule must be at least twice the element size in bytes before promotion for 2:1 decimation (GRANULE≥2×ELEM_BYTES) and at least four times the element size in bytes before promotion for 4:1 decimation (GRANULE≥4×ELEM_BYTES).

[0255] The DIMFMT field 3009 defines which of the loop count fields ICNT0 2901, ICNT1 2902, ICNT2 2903, ICNT3 2904, ICNT4 2905 and ICNT5 2906, of the loop dimension fields DIM1 2911, DIM2 2912, DIM3 2913, DIM4 2914 and DIM5 2915 and of the addressing mode fields AM0 3013, AM1 3014, AM2 3015, AM3 3016, AM4 3017 and AM5 3018 (part of FLAGS field 2921) of the stream template register 2900 are active for the particular stream. Table 23 lists the active loops for various values of the DIMFMT field 3009. Each active loop count must be at least 1 and the outer active loop count must be greater than 1.

[0256]

TABLE 23
DIMFMT	Loop5	Loop4	Loop3	Loop2	Loop1	Loop0
000	Inactive	Inactive	Inactive	Inactive	Inactive	Active
001	Inactive	Inactive	Inactive	Inactive	Active	Active
010	Inactive	Inactive	Inactive	Active	Active	Active
011	Inactive	Inactive	Active	Active	Active	Active
100	Inactive	Active	Active	Active	Active	Active
101	Active	Active	Active	Active	Active	Active
110-111	Reserved

[0257] The DIR bit 3010 determines the direction of fetch of the inner loop (Loop0). If the DIR bit 3010 is 0, Loop0 fetches are in the forward direction toward increasing addresses. If the DIR bit 3010 is 1, Loop0 fetches are in the backward direction toward decreasing addresses. The fetch direction of other loops is determined by the sign of the corresponding loop dimension DIM1, DIM2, DIM3, DIM4 and DIM5.

[0258] The CBK0 field 3011 and the CBK1 field 3012 control the circular block size upon selection of circular addressing. The manner of determining the circular block size is described herein.

[0259] The AM0 field 3013, AM1 field 3014, AM2 field 3015, AM3 field 3016, AM4 field 3017 and AM5 field 3018 control the addressing mode of a corresponding loop, thus permitting the addressing mode to be independently specified for each loop. Each of AM0 field 3013, AM1 field 3014, AM2 field 3015, AM3 field 3016, AM4 field 3017 and AM5 field 3018 are three bits and are decoded as listed in Table 24.

[0260]

TABLE 24
AMx field	Meaning
00	Linear addressing
01	Circular addressing block size set by CBK0
10	Circular addressing block size set by CBK0 + CBK1 + 1
11	reserved

[0261] In linear addressing, the address advances according to the address arithmetic whether forward or reverse. In circular addressing, the address remains within a defined address block. Upon reaching the end of the circular address block the address wraps around to the beginning limit of the block. Circular addressing blocks are limited to 2N addresses where N is an integer. Circular address arithmetic can operate by cutting the carry chain between bits and not allowing a selected number of most significant bits to change. Thus, arithmetic beyond the end of the circular block changes only the least significant bits. The block size is set as listed in Table 25.

[0262]

TABLE 25
Encoded Block Size
CBK0 or
CBK0 + CBK1 + 1	Block Size (bytes)
0	512
1	1K
2	2K
3	4K
4	8K
5	16K
6	32K
7	64K
8	128K
9	256K
10	512K
11	1M
12	2M
13	4M
14	8M
15	16M
16	32M
17	64M
18	128M
19	256M
20	512M
21	1G
22	2G
23	4G
24	8G
25	16G
26	32G
27	64G
28	Reserved
29	Reserved
30	Reserved
31	Reserved

[0263] In this example, the circular block size is set by the number encoded by CBK0 (first circular address mode 01) or the number encoded by CBK0+CBK1+1 (second circular address mode 10). For example, in the first circular address mode, the circular address block size can range from 512 bytes to 16 M bytes. For the second circular address mode, the circular address block size can range from 1 K bytes to 64 G bytes. Thus, the encoded block size is 2(B+9) bytes, where B is the encoded block number which is CBK0 for the first block size (AMx of 01) and CBK0+CBK1+1 for the second block size (AMx of 10).

[0264] The processing unit 110 (FIG. 1) exposes the streaming engine 125 (FIG. 28) to programs through a small number of instructions and specialized registers. Programs start and end streams with SEOPEN and SECLOSE. SEOPEN opens a new stream and the stream remains open until terminated explicitly by SECLOSE or replaced by a new stream with SEOPEN. The SEOPEN instruction specifies a stream number indicating opening stream 0 or stream 1. The SEOPEN instruction specifies a data register storing the start address of the stream. The SEOPEN instruction also specifies a stream template register that stores the stream template as described above. The arguments of the SEOPEN instruction are listed in Table 26.

[0265]

TABLE 26
Argument	Description
Stream Start	Scalar register storing stream start address
Address Register
Stream Number	Stream 0 or Stream 1
Stream Template	Vector register storing stream template data
Register

[0266] The stream start address register is a register in general scalar register file 211 (FIG. 2) in this example. The SEOPEN instruction can specify the stream start address register via scr1 field 1305 (FIG. 13) of example instruction coding 1300 (FIG. 13). The SEOPEN instruction specifies stream 0 or stream 1 in the opcode. The stream template register is a vector register in general vector register file 231 in this example. The SEOPEN instruction can specify the stream template register via scr2/cst field 1304 (FIG. 13). If the specified stream is active, the SEOPEN instruction closes the prior stream and replaces the stream with the specified stream.

[0267] SECLOSE explicitly marks a stream inactive, flushing any outstanding activity. Any further references to the stream trigger exceptions. SECLOSE also allows a program to prematurely terminate one or both streams.

[0268] An SESAVE instruction saves the state of a stream by capturing sufficient state information of a specified stream to restart that stream in the future. An SERSTR instruction restores a previously saved stream. An SESAVE instruction saves the stream metadata and does not save any of the stream data. The stream re-fetches stream data in response to an SERSTR instruction.

[0269] Each stream can be in one of three states: inactive, active, or frozen after reset. Both streams begin in the inactive state. Opening a stream moves the stream to the active state. Closing the stream returns the stream to the inactive state. In the absence of interrupts and exceptions, streams ordinarily do not make other state transitions. To account for interrupts, the streaming engine adds a third state: frozen. The frozen state represents an interrupted active stream.

[0270] In this example, four bits, two bits per stream, define the state of both streams. One bit per stream resides within the streaming engine, and the other bit resides within the processor core 110. The streaming engine internally tracks whether each stream holds a parameter set associated with an active stream. This bit distinguishes an inactive stream from a not-inactive stream. The processor core 110 separately tracks the state of each stream with a dedicated bit per stream in the Task State Register (TSR): TSR.SE0 for stream 0, and TSR.SE1 for stream 1. These bits distinguish between active and inactive streams.

[0271] Opening a stream moves the stream to the active state. Closing a stream moves the stream to the inactive state. If a program opens a new stream over a frozen stream, the new stream replaces the old stream and the streaming engine discards the contents of the previous stream. The streaming engine supports opening a new stream on a currently active stream. The streaming engine discards the contents of the previous stream, flushes the pipeline, and starts fetching data for the new opened stream. Data to processor is asserted once the data has returned. If a program closes an already closed stream, nothing happens. If a program closes an open or frozen stream, the streaming engine discards all state related to the stream, clears the internal stream-active bit, and clears the counter, tag and address registers. Closing a stream serves two purposes. Closing an active stream allows a program to specifically state the stream and the resources associated with the stream are no longer needed. Closing a frozen stream also allows context switching code to clear the state of the frozen stream, so that other tasks do not see it.

[0272] As noted above, there are circumstances when some data within a stream holding register 2818 or 2828 is not valid. As described above, such a state can occur at the end of an inner loop when the number of stream elements is less than the respective stream holding register 2818/2828 size or at the end of an inner loop when the number of stream elements remaining is less than the lanes defined by VECLEN. For times not at the end of an inner loop, if VECLEN is less than the width of stream holding register 2818/2828 and GRDUP is disabled, then lanes in stream holding register 2818/2828 in excess of VECLEN are invalid.

[0273] Referring again to FIG. 28, in this example streaming engine 125 further includes valid registers 2819 and 2829. Valid register 2819 indicates the valid lanes in stream head register 2818. Valid register 2829 indicates the valid lanes in stream head register 2828. Respective valid registers 2819/2829 include one bit for each minimum ELEM_BYTES lane within the corresponding stream head register 2818/2828. In this example, the minimum ELEM_BYTES 15 1 byte. The preferred data path width of processor 100 and the data length of stream head registers 2818/2828 is 64 bytes (512 bits). Valid registers 2819/2829 accordingly have a data width of 64 bits. Each bit in valid registers 2819/2829 indicates whether a corresponding byte in stream head registers 2818/2828 is valid. In this example, a 0 indicates the corresponding byte within the stream head register is invalid, and a 1 indicates the corresponding byte is valid.

[0274] In this example, upon reading a respective one of the stream head registers 2818/2828 and transferring of data to the requesting functional unit, the invalid/valid data in the respective valid register 2819/2829 is automatically transferred to a data register within predicate register file 234 (FIG. 2) corresponding to the particular stream. In this example the valid data for stream 0 is stored in predicate register P0 and the valid data for stream 1 is stored in predicate register P1.

[0275] The valid data stored in the predicate register file 234 can be used in a variety of ways. The functional unit can combine the vector stream data with another set of vectors and then store the combined data to memory using the valid data indications as a mask, thus enabling the same process to be used for the end of loop data as is used for cases where all the lanes are valid which avoids storing invalid data. The valid indication stored in predicate register file 234 can be used as a mask or an operand in other processes. P unit 246 (FIG. 2) can have an instruction to count the number of 1's in a predicate register (BITCNT, which can be used to determine the count of valid data elements from a predicate register.

[0276] FIG. 32 illustrates example hardware 3200 to produce the valid/invalid indications stored in the valid register 2819 (FIG. 28). FIG. 32 illustrates hardware for stream 0; stream 1 includes corresponding hardware. Hardware 3200 operates to generate one valid word each time data is updated in stream head register 2818 (FIG. 28). A first input ELTYPE is supplied to decoder 3201. Decoder 3201 produces an output TOTAL ELEMENT SIZE corresponding to the minimum data size based upon the element size ELEM_BYTES and whether the elements are real numbers or complex numbers. The meanings of various codings of ELTYPE are shown in Table 9. Table 27 shows an example output of decoder 3201 in bytes for the various ELTYPE codings. Note Table 9 lists bits and Table 27 lists bytes. As shown in Table 27, TOTAL ELEMENT SIZE is 1, 2, 4 or 8 bytes if the element is real and 2, 4, 8 or 16 bytes if the element is complex.

[0277]

TABLE 27
		Total Element
ELTYPE	Real/Complex	Size Bytes
0000	Real	1
0001	Real	2
0010	Real	4
0011	Real	8
0100	Reserved	Reserved
0101	Reserved	Reserved
0110	Reserved	Reserved
0110	Reserved	Reserved
1000	Complex, Not Swapped	2
1001	Complex, Not Swapped	4
1010	Complex, Not Swapped	8
1011	Complex, Not Swapped	16
1100	Complex, Swapped	2
1101	Complex, Swapped	4
1110	Complex, Swapped	8
1111	Complex, Swapped	16

[0278] A second input PROMOTE is supplied to decoder 3202. Decoder 3202 produces an output promotion factor corresponding to the PROMOTE input. The meaning of various codings of PROMOTE are shown in Table 28, which shows an example output of decoder 3202 in bytes for the various PROMOTE codings. The difference in extension type (zero extension or sign extension) is not relevant to decoder 3202.

[0279]

TABLE 28
PROMOTE	Promotion Factor
000	1
001	2
010	4
011	8
100	Reserved
101	2
110	4
111	8

[0280] The outputs of decoders 3201 and 3202 are supplied to multiplier 3203. The product produced by multiplier 3203 is the lane size corresponding to the TOTAL ELEMENT SIZE and the promotion factor. Because the promotion factor is an integral power of 2 (2^N), the multiplication can be achieved by corresponding shifting of the TOTAL ELEMENT SIZE, e.g., no shift for a promotion factor of 1, a one-bit shift for a promotion factor of 2, a two-bit shift for a promotion factor of 4, and a three-bit shift for a promotion factor of 8.

[0281] NUMBER OF LANES unit 3204 receives the vector length VECLEN and the LANE SIZE and generates the NUMBER OF LANES. Table 29 shows an example decoding of the number of lanes for lane size in bytes and the vector length VECLEN.

[0282]

TABLE 29
LANE	VECLEN
SIZE	000	001	010	011	100	101	110
1	1	2	4	8	16	32	64
2	—	1	2	4	8	16	32
4	—	—	1	2	4	8	16
8	—	—	—	1	2	4	8
16	—	—	—	—	1	2	4
32	—	—	—	—	—	1	2
64	—	—	—	—	—	—	1

[0283] As previously stated, VECLEN must be greater than or equal to the product of the element size and the duplication factor. As shown in Table 29, VECLEN must also be greater than or equal to the product of the element size and the promotion factor. This means that VECLEN must be large enough to guarantee that an element cannot be separated from its extension produced by type promotion block 2022 (FIG. 20). The cells below the diagonal in Table 29 marked “-” indicate an unpermitted combination of parameters.

[0284] The NUMBER OF LANES output of unit 3204 serves as one input to LANE/REMAINING ELEMENTS CONTROL WORD unit 3211. A second input comes from multiplexer 3212. Multiplexer 3212 receives a Loop0 input and a Loop1 input. The Loop0 input and the Loop1 input represent the number of remaining elements in the current iteration of the corresponding loop.

[0285] FIG. 33 illustrates a partial schematic view of address generator 2811 shown in FIG. 28. Address generator 2811 forms an address for fetching the next element in the defined stream of the corresponding streaming engine based on a start address register 3301 and loop address values for each of the various loops (e.g., Loop0, Loop1, Loop2, Loop3, Loop4, Loop5, etc.) using a final sum adder 3303. Start address register 3301 stores a start address of the data stream. As previously described above, in this example, start address register 3301 is a scalar register in global scalar register file 211 designated by the SEOPEN instruction that opened the corresponding stream. The start address can be copied from the specified scalar register and stored locally at the respective address generator 2811/2821 by control logic included with address generator 2811. The first loop of the stream employs Loop0 count register 3311, adder 3312, multiplier 3313 and comparator 3314. Loop0 count register 3311 stores the working copy of the iteration count of the first loop (Loop0). For each iteration of Loop0, adder 3312, as triggered by the Next Address signal, adds 1 to the loop count, which is stored back in Loop0 count register 3311. Multiplier 3313 multiplies the current loop count and the quantity ELEM_BYTES. ELEM_BYTES is the size of each data element in loop0 in bytes. Loop0 traverses data elements physically contiguous in memory with an iteration step size of ELEM_BYTES.

[0286] Comparator 3314 compares the count stored in Loop0 count register 3311 (after incrementing by adder 3312) with the value of ICNT0 2901 (FIG. 29) from the corresponding stream template register 2900 (FIG. 29). When the output of adder 3312 equals the value of ICNT0 2901 of the stream template register 2900, an iteration of Loop0 is complete. Comparator 3314 generates an active Loop0 End signal. Loop0 count register 3311 is reset to 0 and an iteration of the next higher loop, in this case Loop1, is triggered.

[0287] Circuits for the higher loops (Loop1, Loop2, Loop3, Loop4 and Loop5) are similar to that illustrated in FIG. 33. Each loop includes a respective working loop count register, adder, multiplier and comparator. The adder of each loop is triggered by the loop end signal of the prior loop. The second input to each multiplier is the corresponding dimension DIME DIM2, DIM3, DIM4 and DIM5 from the corresponding stream template. The comparator of each loop compares the working loop register count with the corresponding iteration value ICNT1, ICNT2, ICNT3, ICNT4 and ICNT5 of the corresponding stream template register 2900. A loop end signal generates an iteration of the next higher loop. A loop end signal from Loop5 ends the stream.

[0288] FIG. 33 also illustrates the generation of Loop0 count. Loop0 count equals the updated data stored in the corresponding working count register 3311. Loop0 count is updated on each change of working Loop0 count register 3311. The loop counts for the higher loops (Loop1, Loop2, Loop3, Loop4 and Loop5) are similarly generated.

[0289] FIG. 33 also illustrates the generation of Loop0 address. Loop0 address equals the data output from multiplier 3313. Loop0 address is updated on each change of working Loop0 count register 3311. Similar circuits for Loop1, Loop2, Loop3, Loop4 and Loop5 produce corresponding loop addresses. In this example, Loop0 count register 3311 and the other loop count registers are implemented as count up registers. In another example, initialization and comparisons operate as count down circuits.

[0290] Referring again to FIG. 32, the value of the loop down count, such as Loop0/, is given by expression (2).
Loopx/=ICNTx−Loopx  (2)
That is, the loop down count is the difference between the initial iteration count specified in the stream template register and the loop up count produced as illustrated in FIG. 33.

[0291] LANE/REMAINING ELEMENTS CONTROL WORD unit 3211 (FIG. 32) generates a control word 3213 based upon the number of lanes from NUMBER OF LANES unit 3204 and the loop down count selected by multiplexer 3212. The control input to multiplexer 3212 is the TRANSPOSE signal from field 3002 of FIG. 30. If TRANSPOSE is disabled (“000”), multiplexer 3212 selects the Loop0 down count Loop0/. For all other legal values of TRANSPOSE (“001”, “010”, “011”, “100”, “101” and “110”) multiplexer 3212 selects the Loop1 down count Loop1/. The streaming engine maps the innermost dimension to consecutive lanes in a vector. For normal streams this is Loop0. For transposed streams, this is Loop1, because transposition exchanges the two dimensions.

[0292] LANE/REMAINING ELEMENTS CONTROL WORD unit 3211 generates control word 3213 as follows. Control word 3213 has a number of bits equal to the number of lanes from unit 3204. If the remaining count of elements of the selected loop is greater than or equal to the number of lanes, then all lanes are valid. For this case, control word 3213 is all ones, indicating that all lanes within the vector length VECLEN are valid. If the remaining count of elements of the selected loop is nonzero and less than the number of lanes, then some lanes are valid and some are invalid. According to the lane allocation described above in conjunction with FIGS. 21 and 22, stream elements are allocated lanes starting with the least significant lanes. Under these circumstances, control word 3213 includes a number of least significant bits set to one equal to the number of the selected loop down count. All other bits of control word 3213 are set to zero. In the example illustrated in FIG. 32, the number of lanes equals eight and there are five valid (1) least significant bits followed by three invalid (0) most significant bits which corresponds to a loop having five elements remaining in the final iteration.

[0293] Control word expansion unit 3214 expands the control word 3213 based upon the magnitude of LANE SIZE. The expanded control word includes one bit for each minimally sized lane. In this example, the minimum stream element size, and thus the minimum lane size, is one byte (8 bits). In this example, the size of holding registers 2818/2828 equals the vector size of 64 bytes (512 bits). Thus, the expanded control word has 64 bits, one bit for each byte of stream holding registers 2818/2828. This expanded control word fills the least significant bits of the corresponding valid register 2819 and 2829 (FIG. 28).

[0294] For the case when VECLEN equals the vector length, the description is complete. The expanded control word includes bits for all places within respective valid register 2819/2829. There are some additional considerations when VECLEN does not equal the vector length. When VECLEN does not equal the vector length, the expanded control word does not have enough bits to fill the corresponding valid register 2819/2829. As illustrated in FIG. 32, the expanded control word fills the least significant bits of the corresponding valid register 2819/2829, thus providing the valid/invalid bits for lanes within the VECLEN width. Another mechanism is provided for lanes beyond the VECLEN width up to the data width of stream head register 2818.

[0295] Referring still to FIG. 32, multiplexer 3215 and group duplicate unit 3216 are illustrated to provide the needed additional valid/invalid bits. Referring to the description of VECLEN, if group duplication is not enabled (GRDUP=0), then the excess lanes are not valid. A first input of multiplexer 3215 is an INVALID 0 signal that includes multiple bits equal in number to VECLEN. When GRDUP=0, multiplexer 3215 selects this input. Group duplicate unit 3216 duplicates this input to all excess lanes of stream head register 2818. Thus, the most significant bits of valid register 2819 are set to zero indicating the corresponding bytes of stream head register 2818 are invalid. This occurs for vectors 1-8 of the example shown in Table 15, vectors 1-15 of the example shown in Table 16, and vectors 1-29 of the example shown in Table 17.

[0296] In another example, multiplexer 3215 and group duplicate block 3216 are replaced with group duplicate logic that is similar to the group duplicate logic 2025 illustrated in FIG. 31.

[0297] As previously described, if group duplication is enabled (GRDUP=1), then the excess lanes of stream head register 2818 (FIG. 28) are filled with copies of the least significant bits. A second input of multiplexer 3215 is the expanded control word from control word expansion unit 3214. When GRDUP=1, multiplexer 3215 selects this input. Group duplicate unit 3216 duplicates this input to all excess lanes of stream head register 2818.

[0298] There are two possible outcomes. In one outcome, in most cases, all the lanes within VECLEN are valid and the bits from control word expansion unit 3214 are all ones. This occurs for vectors 1-7 of the group duplication example shown in Table 18 and vectors 1-14 of the group duplication example shown in Table 19. Under these conditions, all bits of the expanded control word from control word expansion unit 3214 are one and all lanes of stream head register 2818 are valid. Group duplicate unit 3216 thus fills all the excess lanes with ones. In the other outcome, the number of remaining stream data elements is less than the number of lanes within VECLEN. This occurs for vector 8 in the group duplication example shown in Table 18 and vector 15 in the group duplication example shown in Table 19. Under these conditions, some lanes within VECLEN are valid and some are invalid. Group duplicate unit 3216 fills the excess lanes with bits having the same pattern as the expanded control word bits. In either case, the excess lanes are filled corresponding to the expanded control bits.

[0299] Referring still to FIG. 32, a boundary 3217 is illustrated between the least significant bits and the most significant bits. The location of this boundary is set by the size of VECLEN relative to the size of stream head register 2818.

[0300] FIG. 34 is a partial schematic diagram 3400 illustrating the stream input operand coding described above. FIG. 34 illustrates a portion of instruction decoder 113 (see FIG. 1) decoding src1 field 1305 of one instruction to control corresponding src1 input of functional unit 3420. These same or similar circuits are duplicated for src2/cst field 1304 of an instruction controlling functional unit 3420. In addition, these circuits are duplicated for each instruction within an execute packet capable of employing stream data as an operand that are dispatched simultaneously.

[0301] Instruction decoder 113 receives bits 13-17 of src1 field 1305 of an instruction. The opcode field (bits 3-12 for all instructions and additionally bits 28-31 for unconditional instructions) unambiguously specifies a corresponding functional unit 3420 and the function to be performed. In this example, functional unit 3420 can be L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244 or C unit 245. The relevant part of instruction decoder 113 illustrated in FIG. 34 decodes src1 bit field 1305. Sub-decoder 3411 determines whether src1 bit field 1305 is in the range from 00000 to 01111. If this is the case, sub-decoder 3411 supplies a corresponding register number to global vector register file 231. In this example, the register number is the four least significant bits of src1 bit field 1305. Global vector register file 231 recalls data stored in the register corresponding to the register number and supplies the data to the src1 input of functional unit 3420.

[0302] Sub-decoder 3412 determines whether src1 bit field 1305 is in the range from 10000 to 10111. If this is the case, sub-decoder 3412 supplies a corresponding register number to the corresponding local vector register file. If the instruction is directed to L2 unit 241 or S2 unit 242, the corresponding local vector register file is local vector register file 232. If the instruction is directed to M2 unit 243, N2 unit 244 or C unit 245, the corresponding local vector register file is local vector register file 233. In this example, the register number is the three least significant bits of src1 bit field 1305. The corresponding local vector register file 232/233 recalls data stored in the register corresponding to the register number and supplies the data to the src1 input of functional unit 3420.

[0303] Sub-decoder 3413 determines whether src1 bit field 1305 is 11100. If this is the case, sub-decoder 3413 supplies a stream 0 read signal to streaming engine 125. Streaming engine 125 then supplies stream 0 data stored in holding register 2818 to the src1 input of functional unit 3420.

[0304] Sub-decoder 3414 determines whether src1 bit field 1305 is 11101. If this is the case, sub-decoder 3414 supplies a stream 0 read signal to streaming engine 125. Streaming engine 125 then supplies stream 0 data stored in holding register 2818 to the src1 input of functional unit 3420. Sub-decoder 3414 also supplies an advance signal to stream 0. As previously described, streaming engine 125 advances to store the next sequential vector of data elements of stream 0 in holding register 2818.

[0305] Supply of a stream 0 read signal to streaming engine 125 by either sub-decoder 3413 or sub-decoder 3414 triggers another data movement. Upon such a stream 0 read signal, streaming engine 125 supplies the data stored in valid register 2819 to predicate register file 234 for storage. In accordance with this example, this is a predetermined data register within predicate register file 234. In this example, data register P0 corresponds to stream 0.

[0306] Sub-decoder 3415 determines whether src1 bit field 1305 is 11110. If this is the case, sub-decoder 3415 supplies a stream 1 read signal to streaming engine 125. Streaming engine 125 then supplies stream 1 data stored in holding register 2828 to the src1 input of functional unit 3420.

[0307] Sub-decoder 3416 determines whether src1 bit field 1305 is 11111. If this is the case, sub-decoder 3416 supplies a stream 1 read signal to streaming engine 125. Streaming engine 125 then supplies stream 1 data stored in holding register 2828 to the src1 input of functional unit 3420. Sub-decoder 3414 also supplies an advance signal to stream 1. As previously described, streaming engine 125 advances to store the next sequential vector of data elements of stream 1 in holding register 2828.

[0308] Supply of a stream 1 read signal to streaming engine 125 by either sub-decoder 3415 or sub-decoder 3416 triggers another data movement. Upon such a stream 1 read signal, streaming engine 125 supplies the data stored in valid register 2829 to predicate register file 234 for storage. In accordance with this example, this is a predetermined data register within predicate register file 234. In this example, data register P1 corresponds to stream 1.

[0309] Similar circuits are used to select data supplied to scr2 input of functional unit 3420 in response to the bit coding of src2/cst field 1304. The src2 input of functional unit 3420 can be supplied with a constant input in a manner described above. If instruction decoder 113 generates a read signal for stream 0 from either scr1 field 1305 or scr2/cst field 1304, streaming engine 125 supplies the data stored in valid register 2819 to predicate register P0 of predicate register file 234 for storage. If instruction decode 113 generates a read signal for stream 1 from either scr1 field 1305 or scr2/cst field 1304, streaming engine 125 supplies the data stored in valid register 2829 to predicate register P1 of predicate register file 234 for storage.

[0310] The exact number of instruction bits devoted to operand specification and the number of data registers and streams are design choices. In particular, the specification of a single global vector register file and omission of local vector register files is feasible. This example employs a bit coding of an input operand selection field to designate a stream read and another bit coding to designate a stream read and advancing the stream.

[0311] The process illustrated in FIG. 34 automatically transfers valid data into predicate register file 234 each time stream data is read. The transferred valid data can then be used by P unit 246 for further calculation of meta data. The transferred valid data can also be used as a mask or as an operand for other operations by one or more of vector data path side B 116 functional units including L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244 and C unit 245. There are numerous feasible compound logic operations employing this stream valid data.

[0312] FIG. 35 is a partial schematic diagram 3500 illustrating another example configuration for selecting operand sources. In this example, the respective stream valid register 2819/2829 need not be automatically loaded to a predetermined register in predicate register file 234. Instead, an explicit instruction to P unit 246 is used to move the data. FIG. 35 illustrates a portion of instruction decoder 113 (see FIG. 1) decoding src1 field 1305 of one instruction to control a corresponding src1 input of P unit 246. These same or similar circuits can be duplicated for src2/cst field 1304 (FIG. 13) of an instruction controlling P unit 246.

[0313] Instruction decoder 113 receives bits 13-17 of src1 field 1305 of an instruction. The opcode field opcode field (bits 3-12 for all instructions and additionally bits 28-31 for unconditional instructions) unambiguously specifies P unit 246 and the function to be performed. The relevant part of instruction decoder 113 illustrated in FIG. 35 decodes src1 bit field 1305. Sub-decoder 3511 determines whether src1 bit field 1305 is in the range 00000 to 01111. If this is the case, sub-decoder 3511 supplies a corresponding register number to global vector register file 231. In this example, the register number is the four least significant bits of src1 bit field 1305. Global vector register file 231 recalls data stored in the register corresponding to the register number and supplies the data to the src1 input of P unit 246.

[0314] Sub-decoder 3512 determines whether src1 bit field 1305 is in the range 10000 to 10111. If this is the case, sub-decoder 3512 supplies a decoded register number to the predicate register file 234. In this example, the register number is the three least significant bits of src1 bit field 1305. The predicate register file 234 recalls data stored in the register corresponding to the register number and supplies the data to the src1 input of predicate unit 246.

[0315] Sub-decoder 3513 determines whether src1 bit field 1305 is 11100. If this is the case, sub-decoder 3513 supplies a stream 0 valid read signal to streaming engine 125. Streaming engine 125 then supplies valid data stored in valid register 2819 to the src1 input of P unit 246.

[0316] Sub-decoder 3514 determines whether src1 bit field 1305 is 11101. If this is the case, sub-decoder 3514 supplies a stream 1 valid read signal to streaming engine 125. Streaming engine 125 then supplies stream 1 valid data stored in valid register 2829 to the src1 input of P unit 246.

[0317] The P unit 246 instruction employing the stream valid register 2819/2829 as an operand can be any P unit instruction previously described such as NEG, BITCNT, RMBD, DECIMATE, EXPAND, AND, NAND, OR, NOR, and XOR.

[0318] The special instructions noted above can be limited to P unit 242. Thus, the operations outlined in FIGS. 34 and 35 can be used together. If the functional unit specified by the instruction is L2 unit 241, S2 unit 242, M2 unit 243, N2 unit 244 or C unit 245, then src1 field 1305 is interpreted as outlined with respect to FIG. 34. If the functional unit specified by the instruction is P unit 246, then src1 field 1305 is interpreted as outlined with respect to FIG. 35. Alternatively, the automatic saving of the stream valid register to a predetermined predicate register illustrated in FIG. 34 can be implemented in one example and not implemented in another example.

Matrix Multiplication Accelerator

[0319] FIG. 36 is a block diagram of a system 3600 that includes processing unit core 110, streaming engine 125, system memory 130, and matrix multiplication accelerator (MMA) 3640. MMA 3640 is a tightly coupled matrix multiplication acceleration unit that is a third type of function unit for processing unit core 110. The other two are the traditional scalar data path 115 (FIG. 1) and the vector data path 116 (FIG. 1). MMA 3640 enables system 3600 to fulfill a large group of image and data processing application requirements. MMA 3640 supports high computational performance requirements in matrix multiplications. With support from streaming engine 125, processing unit core 110, and L2 memory 130, MMA 3640 efficiently computes large numbers of MACs (multiple accumulation) required by various vision algorithms, dense linear algebra, FFT operations and high level apps including convolutional neural networks (CNNs), structure from motion (SFM), Radar, etc. without increasing the memory bandwidth into the processing unit core 110.

[0320] Generally speaking, MMA 3640 supports matrix multiplication of two matrices. As shown in expression (3), where matrix A is an [n×m] matrix and matrix B is an [m×p] matrix, the matrix multiplication outputs matrix C.
C_ij=Σ_k=1^mA_ik*B_kj  (3)
where each i,j entry is given by multiplying the entry A_ik (across row i of A) by the entries B_kj (down column j of B), for k=1, 2, . . . , m.

[0321] FIG. 37 illustrates an example matrix A, matrix B and a resulting matrix C in more detail. Each result element C_ij is a summation of products of elements from a row of matrix A and a column of matrix B as defined by expression (3).

[0322] Referring back to FIG. 36, MMA 3640 includes an A[.] buffer 3641 to hold a matrix A, a B[.] buffer 3642 to hold a matrix B, and a C[.] buffer 3643 to collect the result elements for matrix C. MMA 3640 includes an array of individual multipliers and a set of accumulators as indicated at 3644 to allow an entire row of the C matrix to be calculated in one execution cycle of MMA 3640. In an example, MMA 3640 is equipped to handle a 32×32 16-bit fixed/floating point matrix multiply and to produce a 32×32 16-bit product matrix in 32 cycles.

[0323] In an example, MMA 3640 is also equipped to multiply two 64×64 8-bit matrices by breaking each into four 32×32 sub-matrices, multiplying various combinations of the sub-matrices and then combining the results to produce a final 64×64 8-bit matrix result. Other examples may be implemented to support larger or smaller matrices having a larger or smaller precision than 8 or 16-bits.

[0324] In order for MMA 3640 to operate correctly on a matrix that is smaller than the native size of the MMA 3640, such as the 32×32 example mentioned above, the input matrices A and B can be expanded such that all of the elements of each input matrix A and B should contain valid data. Unused elements can be set to zero, or some other agreed upon null value. Streaming engine 125 includes support for nulling unused elements of an array during a stream access without performing additional memory accesses, as described in more detail hereinbelow. In an example, streaming engine 125 includes support for inserting zeros or a selected value, such as a max value or a min value, during a stream access without performing memory accesses to provide the selected values.

[0325] FIG. 38 shows an example datapath 3800 block diagram of the first streaming engine 2810 of FIG. 28. In some examples, the datapath 3800 also applies to the second streaming engine 2820 of FIG. 28. The datapath 3800 includes data storage 2816 (an internal cache of the streaming engine), a butterfly network 2817, a mask generation block 3806, an AND logic gate 3808, holding registers 3810, and a reference queue 2815. The reference queue 2815 stores a sequence of references generated by the address generator 2811 (FIG. 28). This information drives the butterfly network 2817 (the data formatting network) to enable the first streaming engine 2810 (FIG. 28) to present data to the processing unit 110 in the correct order, to enable processing as described above.

[0326] Entries in the reference queue 2815 contain information used to read data out of the L2 cache 130 and align the data for the processing unit 110. For example, a reference includes all tag identifiers (IDs) associated with a memory access, alignment and rotation to be performed on retrieved data, a number of valid bytes in the retrieved memory vector prior to applying masking, a number of invalid bytes in the retrieved memory vector to mask prior to a first valid byte, and masking control information. Relevant portions of this information are passed by the reference queue 2815 to the data storage 2816, the butterfly network 2817, and the mask generation block 3806. In some examples, the streaming engine's data storage 2816 (internal cache) is fully associative, so that any memory system bus can be placed in any of the entries within the data storage 2816. A tag ID indicates which cache entry index a memory system bus is to be placed in. In some examples, a reference can include multiple tag IDs.

[0327] In some examples, an allocation and tracking logic 2814 (not shown in FIG. 38; see FIG. 28) inserts references in the reference queue 2815 as the address generator 2811 (FIG. 27) generates new addresses. The allocation and tracking logic 2814 removes references from the reference queue 2815 when the data corresponding to the references becomes available and there is room in the formatting pipeline to process the corresponding data. Data corresponding to a reference becomes available when the data retrieved from the L2 cache 130 is stored in the data storage 2816 and can be read out via read ports of the data storage 2816. Also, as the first streaming engine 2810 removes a reference from the reference queue 2815, the first streaming engine 2810 checks whether the reference represents a last reference in the reference queue 2815 containing a corresponding tag ID. If so, the allocation and tracking logic 2814 marks the corresponding tag ID inactive once the first streaming engine 2810 reads corresponding data from the data storage 2816 for a final time (to complete formatting of the data to enable passing formatted data to the processing unit 110).

[0328] The butterfly network 2817 receives data from the L2 cache 130 via the data storage 2816, and performs alignment and rotation on the data. For example, the butterfly network 2817 can be used to format the data so that it can be treated as contiguous memory vectors, arranged as portions of input feature maps in preparation for matrix operations as described above. The formatted data is then passed to a first input of the AND logic gate 3808. The mask generation block 3806 generates a stream of ones and zeroes, which are output to a second input of the AND logic gate 3808. The AND logic gate 3808 accordingly accomplishes padding with null elements, as further described below, by performing an AND logical operation on the butterfly network 2817 and mask generation block 3806 outputs. The padded, formatted data is then passed to the holding registers 3810 for buffering and outputting to the processing unit 110.

Inserting Padding Elements into a Stream

[0329] Complex signal processing applications such as various vision algorithms, dense linear algebra, FFT operations and high-level apps including convolutional neural networks (CNNs), structure from motion (SFM), and radar may require numeric manipulation of a complex multidimensional data structure. Boundaries of the data structure need to be clean so that anomalous data is not included in computations done at the boundary. Generally, clean boundaries for the data structure requires a programmer to allocate memory space at the boundary of a data structure that can be preset to a known value, such as zero or null, or a maximum or minimum data value. Presetting the boundary data then takes additional processing cycles to access and set the boundary data values.

[0330] In the examples described herein, several ways of padding stream vectors with constant values and several ways of forming null stream vectors without accessing system memory were described that are based on a specified value for an INVERSE_DECDIMx_WIDTH count. In another example, it may be useful to specify a number of null vectors to be inserted in a stream by the streaming engine.

[0331] Creation of a Convolution neural network (CNN) Toeplitz style matrix on the fly requires zero or constant values fed into the Matrix Multiplication Accelerator after the last feature map in the CNN layer. In linear algebra, a Toeplitz matrix or diagonal-constant matrix in which each descending diagonal from left to right is constant. As described with regard to FIG. 36, MMA 3640 has a fixed number of rows which need to be filled prior to starting matrix multiplication. The rows remaining during creation of Toeplitz matrix cannot have a junk value, otherwise the results will be wrong.

[0332] Assuring that all unused data elements are set to zero or to a selected null value using software requires allocation of memory and execution of instructions to write out the null values.

[0333] An example steaming engine 125 (FIG. 1) includes a feature to allow a programmer to specify an inverse decrement dimension (INVERSE_DECDIM_WIDTH) parameter for one or more of the six-dimensional nested loops supported by steaming engine 125 that is different from a dimension defined by a respective loop dimension parameter. In this manner, streaming engine 125 can be programmed to insert the appropriate known value into the data stream as the data stream is being fetched from system memory by the streaming engine. Furthermore, streaming engine 125 can be programmed to insert null or known value stream vectors that correspond to boundary area of the data structure without accessing the system memory for these boundary area stream vectors, thereby reducing processing cycles and time.

[0334] Referring to FIG. 38, control logic (not illustrated) of the data storage 2816 can control each of the multiplexers on alignment networks (not illustrated) in the data storage 2816 to force a null value onto selected elements of a stream vector based on the metadata provided by reference queue 2815. In this manner, an entire stream vector can be set to a null value without fetching any data from system memory. In an example, the null value is “0.” In another example, a predefined value or pattern may be used to represent a null value.

[0335] In another example, control logic of the data storage 2816 can control each of the multiplexers on alignment networks in the data storage 2816 to force a selected value onto selected elements of a stream vector based on the metadata provided by reference queue 2815. In this manner, an entire stream vector can be set to a selected value, such as a min value or a max value without fetching any data from system memory.

[0336] In this example, control logic in the address generator 2811/2821 (see FIG. 28) performs count tracking for the six levels of nested iterations. As will be described in more detail hereinbelow, various conditions may be detected that are associated with one or more of the six levels of nested iterations and used to signal that a null value or a selected value is to be inserted into the vector stream without fetching data from memory for the null value or selected value vector.

[0337] Metadata produced by the address generator is pushed into the stream reference queue 2815/2825 (see FIG. 28). On the backend, when processing unit core 110 (FIG. 1) performs a read, this metadata is popped out from the stream reference queue 2815/2825 and sent to control logic in the data storage 2816 in alignment network in the data storage 2816 with the lanes to be nulled or set to a selected value without reading data from system memory. The metadata keeps track of decrement dimensions (DECDIM, INVERSE_DECDIM) and dimension widths (DECDIMx_WIDTH, INVERSE_DECDIMx_WIDTH) and is fed to backend logic, as described hereinbelow in more detail.

[0338] FIG. 39 shows an example diagram 3900 of padded data structures such as input feature maps. Each input feature map represents a set of valid data elements loaded from memory having an N-dimensional arrangement. Any shaded padding area around an input feature map represents a set of data elements (e.g., zero value data elements, null data elements, etc.) that are not germane to the operation being performed other than possibly for alignment of the input feature map. The diagram 3900 illustrates that filter kernels of different dimensions correspond to different amounts of zero padding of input feature maps to enable convolution to generate output feature maps with the same dimensions as the input feature maps. For example, a first input feature map 3902 is windowed by filter kernels (not shown) of dimensions 1×1. The first input feature map 3902 receives no zero padding because generated output feature maps (not shown) will have the same dimensions as the input feature map 3902 without zero padding. This is because Fc−1=Fr−1=0, where Fc (number of filter kernel columns) and Fr (number of filter kernel rows). A second input feature map 3906 is windowed by filter kernels of dimensions 2×2. The second input feature map 3906 receives one row and one column of zero padding 3904 because Fc−1=Fr−1=1. A third input feature map 3908 is windowed by filter kernels of dimensions 3×3. The third input feature map 3908 receives two rows and two columns of zero padding 3904 because Fc−1=Fr−1=2.

[0339] FIG. 40 shows an example diagram of the input feature map 3902 of FIG. 39, with the addition of padding zeroes to form zero padded input feature maps 3908 of FIG. 39, where such padding can be implemented to compensate for certain of the above-described architectural and processing attributes. (Padding zeroes are shown herein as “zero” instead of as decimal numbers.) The input feature map 3902 has dimensions (Lr, Lc) of 10×10, corresponding to DECDIM1_WIDTH×DECDIM2_WIDTH, and each feature kernel (not shown) has dimensions (Fr, Fc) of 3×3. The zero padded input feature map 3908 includes a row of zeroes 4004 above a first row (row zero) and below a last row (row ten, the Lr^th row) of the input feature map 3902 (referred to herein as vertical pad zeroes 4004), each with Lc+Fc−1 padding zeroes; and a column of zeroes 4006 to the left of the first column (column zero) and to the right of the last column (column ten, the Lc^th column) of the input feature map 3902 (referred to herein as side pad zeroes 4006), each with Lr+Fr−1 padding zeroes. The number of rows of padding corresponds to INVERSE_DECDIM2_WIDTH, and the number of columns of padding corresponds to INVERSE_DECDIM1_WIDTH. Each row of the input feature map 3902 is padded with Fc−1 zeroes, and each column of the input feature map 3902 is padded with Fr−1 zeroes. Accordingly, with the inclusion of the zero padding, convolution of the zero padded input feature maps 3908 and the feature kernels returns output feature maps (not shown) with the same dimensions as the input feature map 3902. This avoids the earlier-described progressive data loss that occurs, in some examples, due to the limitations of convolution of input feature maps with filter kernels of the CNN.

[0340] FIG. 41 shows a diagram 4100 illustrating an example of contents 4102 of an MMA accelerator memory after reading feature map data from an L2 memory and writing it into the MMA accelerator memory. A cycle column 4104 provides a cycle on which each respective contiguous memory vector 4109 is read from the L2 memory to be written into the MMA accelerator memory as a row of the contents 4102. The feature map data from which the contents 4102 are generated corresponds to input feature maps written in L2 memory with zero padding in a manner of the zero padded input feature map 4000 of FIG. 40, but that does not include in the L2 memory the column of zero padding to the right of the last column (right-hand zero padding column) of the input feature map 3902 (tenth column, the Lc^th column). Instead, the zero padding column preceding the first column (left-hand zero padding column) of the input feature map 3902 is treated as fulfilling the functions of both the left- and right-hand zero padding columns. This enables construction of the contents 4102 with one skip column 4106 per group of columns corresponding to a single row of an input feature map 3902. Accordingly, there are ten (Lc) non-skip entries (values corresponding to windowed elements in a generating input feature map 3902, and in some rows, a padding zero) between each skip column 4106. Note that each contiguous memory vector of the contents 4102 includes a continuous sequence of feature map 3902 data values (some of which are located in skip columns 4106), with padding zeroes included additionally to (rather than instead of) the feature map 3902 data values and interspersed in or adjacent to skip columns 4106.

[0341] Referring to FIG. 29, two INVERSE_DECDIMx_WIDTH fields 2922, 2924 are defined within stream template register 2900. In some examples, the INVERSE_DECDIM1 feature and its field INVERSE_DECDIM1_WIDTH 2922 is used in conjunction with the DECDIM1 feature and its respective fields (e.g., DECDIM1 WIDTH 2931 (FIG. 29), DECDIM1 flag 3019 (FIG. 30), and DECDIM1SD flag 3020 (FIG. 30)) that are included in the flag field 2921 of stream template register 2900. Similarly, in some examples, the INVERSE_DECDIM2 feature and its field INVERSE_DECDIM2_WIDTH 2924 is used in conjunction with the DECDIM2 feature and its respective fields (e.g., DECDIM2_WIDTH 2932 (FIG. 29), DECDIM2 flag 3019 (FIG. 30), and DECDIM2SD flag 3020 (FIG. 30). As illustrated in FIG. 29, the INVERSE_DECDIMx_WIDTH fields 2922, 2924 are disposed between the ICNT3 field 2904, ICNT2 field 2903, and ICNT1 field 2902.

[0342] In some examples, the value in the INVERSE_DECDIM1_WIDTH field 2922 is not the same as the INVERSE_DECDIM2_WIDTH field 2924, and accordingly, the resulting feature map can have asymmetric padding. The values in the INVERSE_DECDIMx_WIDTH fields 2922, 2924 can depend on the kernel filter size. For example, for a kernel filter size of 3×3, the INVERSE_DECDIMx_WIDTH fields 2922, 2924 can have a value of 1 and thus a row of padding can be inserted on top of and to the left of the feature map. Similarly, for a kernel filter size of 5×5, the INVERSE_DECDIMx_WIDTH fields 2922, 2924 can have a value of 2 and thus two rows of padding can be inserted on top of and to the left of the feature map. In some examples, one of the INVERSE_DECDIMx_WIDTH fields 2922, 2924 can be based on the size of a dimension of the feature map. For example, the INVERSE_DECDIM2_WIDTH field 2924 can be based on the width of the feature map.

[0343] In some examples, the DECDIMx flag field 3019, 3021 (FIG. 30) and INVERSE_DECDIMx_WIDTH fields 2922, 2924 (FIG. 29) supports a padding feature that allows a programmer to define the amount of padding to add to the top and to the left of an image using the INVERSE_DECDIMx_WIDTH count to provide the padding amount. In this mode, the streaming engine 125 inserts padding for a selected dimension based on the value in the INVERSE_DECDIMx_WIDTH field 2922, 2924. In some examples, when the streaming engine 125 inserts padding, the streaming engine 125 decrements the remaining INVERSE_DECDIMx_WIDTH based on the selected dimension (DIM1-DIM5) when the address generator 2811/2821 (FIG. 28) enters that loop dimension. For example, a null element is inserted to pad the left and top of the image and the current value of the remaining INVERSE_DECDIMx_WIDTH is decremented by the DIMx value. When the value in the remaining INVERSE_DECDIMx_WIDTH equals zero, the streaming engine 125 fetches and includes the data of the input feature map for the selected dimension.

[0344] The remaining INVERSE_DECDIMx_WIDTH count value reloads to the INVERSE_DECDIMx_WIDTH programmed value again when the selected DECDIMx dimension loop count (ICNT) expires. In the above example, the remaining INVERSE_DECDIM2_WIDTH count value would reload when dimension 2 (i.e., ICNT2) expires, and dimension 3 is entered. In other words, the remaining INVERSE_DECDIMx_WIDTH reloads when entering any dimension higher than the selected INVERSE_DECDIMx dimension. Thus, it is possible to program the INVERSE_DECDIMx_WIDTH smaller than the loop iterations count for the selected dimension, which could cause the width count to underflow. In some examples, the remaining INVERSE_DECDIMx_WIDTH is decremented using the settings of the respective DECDIMx and DECDIMxSD when null elements are inserted into the stream of vectors for the respective selected dimension.

[0345] In some examples, the streaming engine 125 defines a pointer virtualSrcPtr based on the padding amount to the left and to the top of the input feature map, based on the following equation:
virtualSrcPtr=actualSrcPtr−INVERSE_DECDIM2WIDTH−INVERSE_DECDIM1_WIDTH
The actualSrcPtr is a pointer that indicates the position of the first pixel of the input feature map. In such examples, the virtualSrcPtr assists with tracking the number of inserted null elements into the stream of vectors that form the input feature map.

[0346] Referring back to FIG. 38, the INVERSE_DECDIM feature in the streaming engine 125 can have two effects on data flow. First, in some examples, the data masked out by a combination of the DECDIM feature and INVERSE_DECDIM feature is not fetched, lowering memory system bandwidth utilization at the tops and bottoms of feature maps. In some examples, the address generator 2811/2821 (FIG. 28) transmits a memory address, a number of valid bytes (i.e., the number of bytes to be fetched in the current clock cycle), the DECDIMx_WIDTH, and the INVERSE_DECDIMx_WIDTH to storage control of data storage 2816 (FIG. 38). The storage control of data storage 2816 receives the memory address, the number of valid bytes, the DECDIMx_WIDTH, and the INVERSE_DECDIMx_WIDTH from the address generator 2811/2821 (FIG. 28), and the storage control of data storage 2816 fetches data from memory based on the received information. In some examples, the storage control of data storage 2816 fetches data from memory if the number of valid bytes is greater than the INVERSE_DECDIMx and the DECDIMx is greater than INVERSE_DECDIMx. The INVERSE_DECDIMx field provides details on which dimension of the streaming engine or the address generator for which the feature is enabled.

[0347] Second, elements falling into the INVERSE_DECDIM region can be masked out by a byte enable mask before being sent to the processing unit 110. The byte enable mask can be defined by the following equation:
Byte enable mask=(2^{vector read bytes}−1)&(2^{remaining ICNT0}−1)&(2^DECDIMxWIDTH−1)&˜(2^{INVERSE_DECDIMx_WIDTH}−1)

[0348] In some examples, the storage control of data storage 2816 (FIG. 38) sends the number of valid bytes, DECDIMx_WIDTH, and INVERSE_DECDIMx_WIDTH to the mask generator 3806 (as illustrated in FIG. 38). Based on this information from storage control of data storage 2816 (FIG. 38), the mask generator 3806 generates the Byte enable mask applied to the AND gate 3808 (as illustrated in FIG. 38). Accordingly, the Byte enable mask is able to mask out elements in the region defined by INVERSE_DECDIM.

[0349] FIG. 42 is an exemplary circuit for generating fetches from memory for the input feature map, according to some examples. As mentioned previously, the streaming engine 125 generates all fetches that are required to produce the next vector of data of the input feature map for the processing unit 110. Some vectors of the input feature map could require data from as many as 3 distinct, consecutive cache lines (e.g., Address0, Address1, Address2). However, with the INVERSE_DECDIM feature, each of these fetches could be suppressed by the inverse DECDIM feature.

[0350] The fetch suppression circuit 4200 includes various inputs: the number of valid bytes, the DECDIMx_WIDTH, the INVERSE_DECDIMx_WIDTH, and an address offset. The fetch suppression circuit 4200 includes a MIN circuit 4202, adders 4204 and 4206, and comparators 4208, 4210, 4212, 4214, and 4216. The fetch suppression circuit 4200 also includes AND logic gates 4218, 4220, and 4222. While the fetch suppression circuit 4200 includes such components, other fetch suppression circuits can include any number of adders, comparators, and/or logic gates.

[0351] The MIN circuit 4202 of the fetch suppression circuit 4200 receives the number of valid bytes and the DECDIMx_WIDTH, and the resulting output of the MIN circuit 4202 is added, via the adder 4206, to an address offset. The result from the adder 4206 is then provided to both comparators 4214 and 4216.

[0352] The address offset input is also added to the INVERSE_DECDIMx_WIDTH, via the adder 4204, and the resulting output is provided to comparator 4210 and comparator 4212.

[0353] The comparator 4208 receives the output of the MIN circuit 4202 and the INVERSE_DECDIM_WIDTH. Based on the values from the MIN circuit 4202 and the INVERSE_DECDIM_WIDTH, the output of comparator 4208 indicates that a vector contains unmasked bytes. In some examples, the comparator 4208 performs a comparison to determine if the values from the MIN circuit 4202 is smaller than or equal to the INVERSE_DECDIM_WIDTH. The comparator 4208 outputs a 1-bit value where 1 indicates that the comparison was true and 0 indicates the comparison was false.

[0354] The comparator 4210 receives the resulting output of the adder 4204, which adds the INVERSE_DECDIM_WIDTH and an address offset. Based on this output from the adder 4204, and the memory system bus width, the output of comparator 4210 indicates for the INVERSE DECDIM feature to stop masking during address0. In some examples, the comparator 4210 performs a comparison to determine if the resulting output of the adder 4204 is smaller than the memory system bus width. The comparator 4210 outputs a 1-bit value where 1 indicates that the comparison was true and 0 indicates the comparison was false.

[0355] Like comparator 4210, comparator 4212 receives the resulting output of the adder 4204. Based on the output of the adder 4204 and two times the memory system bus width, the output of comparator 4212 indicates that the INVERSE_DECDIM feature stops masking before address 2. In some examples, the comparator 4212 performs a comparison to determine if the resulting output of the adder 4204 is smaller than twice the memory system bus width. The comparator 4212 outputs a 1-bit value where 1 indicates that the comparison was true and 0 indicates the comparison was false.

[0356] The comparator 4214 receives the resulting output of the adder 4206, which adds the output of the MIN circuit 4202 and the address offset. Based on this output from the adder 4206 and the memory system bus width, the output of comparator 4214 indicates that valid bytes and/or the DECDIM feature extend beyond address0. In some examples, the comparator 4214 performs a comparison to determine if the resulting output of the adder 4206 is smaller than the memory system bus width. The comparator 4214 outputs a 1-bit value where 1 indicates that the comparison was true and 0 indicates the comparison was false.

[0357] Like comparator 4214, comparator 4216 receives the resulting output of the adder 4206. Based on the output of the adder 4204 and two times the cache line width, the output of comparator 4216 indicates that valid bytes and/or the DECDIM feature extend beyond address 1. In some examples, the comparator 4216 performs a comparison to determine if the resulting output of the adder 4206 is smaller than twice the memory system bus width. The comparator 4216 outputs a 1-bit value where 1 indicates that the comparison was true and 0 indicates the comparison was false.

[0358] The outputs of the comparators 4208, 4210, 4212, 4214, and 4216 are provided to the AND logic gates 4218, 4220, and 4222. Specifically, the AND gate 4218 is coupled to the outputs of comparator 4208 and comparator 4210; the AND gate 4220 is coupled to the outputs of comparator 4208, comparator 4212, and comparator 4214; and the AND gate 4222 is coupled to the outputs of comparator 4208 and comparator 4216. The output of AND gate 4218 indicates whether Address0 is valid, the output of AND gate 4220 indicates whether Address1 is valid, and the output of AND gate 4222 indicates whether Address2 is valid.

[0359] FIG. 43 is a flowchart illustrating operations 4300 for formation of a stream by inserting null or predefined data vectors by a streaming engine, such as streaming engine 125 of FIG. 28.

[0360] In this example, at 4302, a streaming engine receives stream parameters and a stream is opened on the streaming engine by storing the stream parameters in a stream template register (2900, FIG. 29) within the streaming engine. The stream parameters include a number of vectors to include in the stream for each dimension of the array, and a width indicator for a selected dimension of the array, such as a DECDIMx flag and a DECDIMx_CNT. The stream parameters also include the INVERSE_DECDIMx_WIDTH, which indicates the number of null elements to include with a particular dimension.

[0361] At 4304, the streaming engine checks whether a current matrix line includes any null elements. That is, the streaming engine checks whether the current matrix line needs padding. In some examples, the streaming engine uses a pointer to check whether the current matrix line is a vector of padding or is vector of matrix data with padding at the ends of the vector.

[0362] If the current matrix line does not include any null elements, then at 4308A, an address stream is generated according to the stream parameters stored in the stream template. Metadata is saved that indicates loop counts, end of loop, remaining width count, etc.

[0363] At 4312A, a line of matrix data is fetched by the streaming engine from system memory using the sequence of addresses generated by the address generator 2811/2821 (FIG. 28).

[0364] At 4314A, a stream vector is formed representing either a column or a row of the array.

[0365] At 4316A, the line of matrix data is inserted into the stream vector.

[0366] If the current matrix line does include null elements, then, at 4306, the streaming engine checks whether the current matrix line includes only null elements. That is, the streaming engine checks whether the current matrix is a vector of padding or if the current matrix line is to be a vector of matrix data with padding at the ends.

[0367] If the current matrix line does not only include null elements, then at 4307, null elements are inserted without accessing system memory. In some examples, the INVERSE_DECDIMx_WIDTH is decremented if it is associated with the current dimension. In an example, the null elements may be set to all zeros. In another example, the null elements may be set to a predetermined pad value, such as a min value, a max value, etc. A predetermined pad value may be designated by a flag field, such a FILLVAL flag 3008 (FIG. 30) in the stream template. The number of null elements inserted depends on the INVERSE_DECDIMx_WIDTH.

[0368] At 4308B, an address stream is generated according to the stream parameters stored in the stream template. Metadata is saved that indicates loop counts, end of loop, remaining width count, etc.

[0369] At 4312B, a line of matrix data is fetched by the streaming engine from system memory using the sequence of addresses generated by the address generator 2811/2821 (FIG. 28).

[0370] At 4314B, a stream vector is formed representing either a column or a row of the array.

[0371] At 4316B, the line of matrix data is inserted into the stream vector.

[0372] In some examples, at 4318, the streaming engine inserts more null elements. The additional null elements inserted can correspond to right padding, e.g., padding on the right side of the matrix.

[0373] If the current matrix line includes only null elements, then at 4320, the streaming engine inserts a vector of null elements without accessing system memory. That is, in some examples, the streaming engine adds the top padding or the bottom padding to the matrix. In some examples, depending on the dimension of the vector and the INVERSE_DECDIMx_WIDTH, the streaming engine inserts a null vector corresponding to the padding of the matrix. In such examples, the null vector has a width equal to a dimension of the array.

[0374] At 4322, a check is made to determine whether the matrix or the stream is complete. That is, the check involves determining whether the data of each row and/or column of the matrix has been fetched from system memory. In some examples, if the current dimension is complete, then the width count is decremented if it is associated with the current dimension. Otherwise, the width count is not decremented. If the matrix or the stream is incomplete, the process repeats to insert more null elements, to access more matrix data from the system, and/or to form stream vectors.

[0375] At 4324, once the entire matrix has been accessed from memory, the data stream is closed.

Write Suppression of Rows and Columns

[0376] While performing computation on the MMA, multiple K blocks (e.g., elements of a matrix) are processed and the results are accumulated before being written into memory. Once all K blocks for one N block (e.g., all elements in a column of a matrix) are processed, the results are written to memory. However, such requirement causes software complications due to corner cases. Accordingly, to satisfy the write requirements, the streaming address generator (e.g., address generator 2811/2821 (FIG. 28)) can be programmed to suppress invalid output writes using the INVERSE_DECDIM feature.

[0377] FIG. 44 is an illustration of write suppression of rows and/or columns, according to some examples. Specifically, FIG. 44 is an illustration of a matrix 4400 with write suppressed elements, written elements, and a zero data element. FIG. 44 also shows a portion of a linear version of the matrix 4400 as processed by the streaming address generator 2811/2821 (FIG. 28).

[0378] The matrix 4400 of FIG. 44 includes multiple columns (corresponding to N blocks 4442) and multiple rows (corresponding to K blocks 4452). Accordingly, as illustrated, the streaming address generator 2811/2821 (FIG. 28) suppresses the writes of each element of the first column 4440 and allows for the write of the first row 4450 of the next column. For example, elements 4404, 4406, 4408, 4414, 4416, 4418, 4424, 4426, 4428, 4434, 4436, and 4438 are write suppressed, while elements 4410, 4420, and 4430 are written to memory. In such example, when elements of the matrix 4400 is displayed linearly, the streaming address generator 2811/2821 (FIG. 28) suppresses the writes of elements 4404, 4406, and 4408 before allowing the write of element 4410.

[0379] The streaming address generator 2811/2821 (FIG. 28) is configured with the number of elements of the N block to be suppressed and the number of elements of the N blocks to be written to memory. For example, as illustrated, every fourth element is written to memory, and so all other elements are suppressed. Accordingly, elements 4404, 4406, 4408, 4414, 4416, 4418, 4424, 4426, 4428, 4434, 4436, and 4438 are write suppressed, while elements 4410, 4420, and 4430 are written to memory.

[0380] In some examples, the streaming address generator 2811/2821 (FIG. 28) includes a null element 4402 at the beginning of the matrix 4400. The inclusion of the null element 4402 addresses corner cases that involve correctly counting elements of the matrix 4400 when changing between different rows and/or columns. As illustrated in the linear version of the matrix 4400 of FIG. 4, every fourth element is written to memory; however, in the matrix version of the matrix 4400, element 4410 is not an element of the first column (i.e., the first N block 4440) of the matrix 4400, but instead is an element of the second column (i.e., the next N block) of the matrix 4400. Accordingly, in order to ensure correct positioning of elements of the matrix 4400 when written to memory, the streaming address generator 2811/2821 (FIG. 28) inserts null elements (via the INVERSE DECDIM feature) to adjust the positioning of elements of the matrix 4400.

[0381] In some examples, the streaming address generator is configured with the dimension of the matrix 4400 to suppress. For example, DECDIM1 can be configured with SA_DECDIM_DIM2. In some examples, the streaming address generator 2811/2821 (FIG. 28) is configured with the DECDIM1_WIDTH, which can be based on the number of rows of the matrix 4400 and the number columns of the matrix 4400. For example, DECDIM1_WIDTH can be Rows*(K blocks−1).

[0382] Like FIG. 29, FIG. 45 illustrates an example streaming address configuration register 4500. The streaming address configuration register 4500 of FIG. 45 provides the full structure of a stream that contains data and can be used with write suppression by the streaming address generator 2811/2821 (FIG. 28). The iteration counts and dimensions provide most of the structure, while the various flags provide the rest of the details. Table 30 shows the stream field definitions of a stream template.

[0383]

TABLE 30
	FIG. 45
	Reference		Size
Field Name	Number	Description	Bits
ICNT0	4502	Iteration count for loop 0	32
ICNT1	4504	Iteration count for loop 1	16
ICNT2	4506	Iteration count for loop 2	16
ICNT3	4508	Iteration count for loop 3	32
ICNT4	4510	Iteration count for loop 4	32
ICNT5	4512	Iteration count for loop 5	32
DECDIM1_WIDTH	4514	First dimension size	32
DECDIM2_WIDTH	4516	Second dimension size	32
DIM1	4518	Signed dimension for loop 1	32
DIM2	4520	Signed dimension for loop 2	32
DIM3	4522	Signed dimension for loop 3	32
DIM4	4524	Signed dimension for loop 4	32
DIM5	4526	Signed dimension for loop 5	32
FLAGS	4530	Stream modifier flags	64

[0384] The iteration count ICNT0, ICNT1, ICNT2, ICNT3, ICNT4, ICNT5 for a loop level indicates the total number of iterations in a level. Though, as described below, the number of iterations of loop 0 does not depend only on the value of ICNT0. The dimensions DIM0, DIM1, DIM2, DIM3, DIM4, and DIM5 indicate the distance between pointer positions for consecutive iterations of the respective loop level. DECDIM1_WIDTH and DECDIM2_WIDTH define, in conjunction with other parameters in the FLAGS field, any vertical strip mining—i.e., any portions of the memory pattern that will not be written.

[0385] Like FIG. 30, FIG. 46 illustrates an example of sub-field definitions of the flags field 4530 shown in FIG. 45. As shown in FIG. 45, the flags field 4530 is 8 bytes or 64 bits. FIG. 46 shows bit numbers of the fields. Table 31 shows the definition of these fields.

[0386]

TABLE 31
	FIG. 46
	Reference		Size
Field Name	Number	Description	Bits
DECDIM2SD	4622	Decrement Dimension 2 Flag	2
DECDIM2	4620	Decrement Dimension 2	2
DECDIM1SD	4618	Decrement Dimension 1 Flag	2
DECDIM1	4616	Decrement Dimension 1	2
DECDIM3SD	4614	Decrement Dimension 3 Flag	2
DECDIM3	4612	Decrement Dimension 3	2
DIMFMT	4610	Stream dimensions format	3
INV_DD3	4608	Enable Inverse Decrement	1
		Dimension 3
INV_DD2	4606	Enable Inverse Decrement	1
		Dimension 2
INV_DD1	4604	Enable Inverse Decrement	1
		Dimension 1
VECLEN	4602	Stream vector length	3

[0387] Like the VECLEN field 3004 of FIG. 30, the VECLEN field 4602 defines the stream vector length for the stream in bytes. The use of the VECLEN field 4602 of the streaming address configuration register 4600 is similar the use of the VECLEN field 3004 of FIG. 30.

[0388] Like the DIMFMT field 3009 of FIG. 30, the DIMFMT field 4610 defines which of the loop count of the loop dimensions and of the addressing modes stream template register 4600 are active for the particular stream. The values of the DIMFMT field 4610 can be the same as the DIMFMT field 3009 of FIG. 30.

[0389] FIGS. 47 and 48 illustrate exemplary streaming address count registers. CNT5, CNT4, CNT3, CNT2, CNT1 and CNT0 represent the intermediate iteration counts for each respective loop level. When the iteration count CNTX of loop X becomes zero, assuming that the loop counts are decremented and not incremented, the starting address of the iteration of the next loop is computed using the next loop dimension. Table 32 show the definition of these fields:

[0390]

TABLE 32
	FIG. 47
	Reference		Size
Field Name	Number	Description	Bits
CNT0	4720	Count for loop 0	32
CNT1	4722	Count for loop 1	16
CNT2	4723	Count for loop 2	16
CNT3	4724	Count for loop 3	32
CNT4	4726	Count for loop 4	32
CNT5	4728	Count for loop 5	32
CURRENT	4702	Pointer offset for loop 0	32
OFFSET
I1 OFFSET	4704	Pointer offset for loop 1	32
I2 OFFSET	4708	Pointer offset for loop 2	32
I3 OFFSET	4710	Pointer offset for loop 3	32
I4 OFFSET	4712	Pointer offset for loop 4	32
I5 OFFSET	4714	Pointer offset for loop 5	32

[0391] Table 33 and 34 illustrate exemplary streaming address count registers. In such count registers, DEC_DIMSD_CNT, DEC_DIM2SD_CNT, and DEC_DIM3SD_CNT define a counter for additional dimensions or loops to which each of DECDIM1_WIDTH 4514, DECDIM2_WIDTH 4516, and DECDIM3_WIDTH 4528 may apply, thereby allowing for the definition of multi-dimensional data exclusion.

[0392]

TABLE 33
	FIG. 47
	Reference		Size
Field Name	Number	Description	Bits
DECDIM1SD_CNT	4716	Decrement Dimension 1 Count	32
DECDIM1_CNT	4730	Decrement Dimension 1 Count	32
DECDIM2SD_CNT	4718	Decrement Dimension 2 Count	32
DECDIM2_CNT	4732	Decrement Dimension 2 Count	32

[0393]

TABLE 34
	FIG. 48
	Reference		Size
Field Name	Number	Description	Bits
DECDIM3SD_CNT	4804	Decrement Dimension 3 Count	32
DECDIM3_CNT	4802	Decrement Dimension 3 Count	32

[0394] Suppressing writes to memory using the INVERSE DECDIM feature reduces software complexity (e.g., uses a single loop structure) and requires a smaller memory footprint, resulting in fewer data transfers in and out of on-chip memory and thus improving performance.

[0395] FIG. 49 illustrates operations 4900 for suppressing writes of elements of a matrix using a streaming address generator of a processing unit, such as processing unit 110. In this example, at 4902 a stream is opened on a processing unit by storing stream parameters in a stream template register (2900, FIG. 29) within the processing unit. The stream parameters include an element size of the array, a number of elements to include in each vector of a stream, a number of iterations for each dimension of the array, a loop end count, such as loop end zero count 2933 (FIG. 29), and a flag to associate the loop end zero count with a selected dimension of the multidimensional loop.

[0396] At 4906, a vector of the matrix is processed by a processing unit (e.g., processing unit 110). In some examples, the matrix comprises K×N elements, and a single N block includes K blocks. Accordingly, in such examples, at 4906, the processing unit processes all K blocks in an N block. In some examples, once the vector of the matrix is processed, the processing unit is prepared to write the vector to memory using the streaming address generator of the processing unit.

[0397] At 4908, the streaming address generator checks whether all K block of the N block has either been written to memory or has been write suppressed. According, the streaming address generator can ensure that all K blocks of the N block are either written or suppressed, and that it has looped through all K blocks in the N block.

[0398] If the streaming address generator has not looped through all K blocks in the N block, then at 4910, a check is made to see if the current K block is to be written to memory or if the write of the current K block is suppressed. The check can be based on a predefined interval between the last K block written to memory and the current K block. For example, every fourth K block is to be written to memory and the other K blocks are to be suppressed. Accordingly, the streaming address generator checks whether the current K block is the fourth K block after the last K block written to memory. Other parameters can be used to determine whether the current K block is to be written to memory.

[0399] If the current K block is to be written to memory, then at 4914, the K block is written to memory. Once the K block is written to memory, the streaming address generator loops back to check whether all K blocks of the N block have been either written or suppressed (at 4908).

[0400] If the current K block is to be write suppressed, then at 4912, the streaming address generator suppresses the write of the K block. Once the K block is suppressed, the streaming address generator loops back to check whether all K blocks of the N block have been either written or suppressed (at 4908).

[0401] If all K blocks of the N block have been either written or suppressed, then another check 4916 is made to determine whether all N blocks of the matrix have been processed.

[0402] If at least one N block of the matrix still needs to be processed, then the streaming address generator continues with processing each N block of the matrix (at 4906).

[0403] At 4918, once the entire matrix has been processed (i.e., all K blocks of all N blocks have been either written to memory or suppressed), the data stream is closed.

[0404] FIG. 50 illustrates an example multiprocessor system. In this example, SoC 5000 includes processor 100 (FIG. 1) (referred to as “processor A”) and it is combined with a second processor 5011 (referred to as “processor B”). Each processor is coupled to a block of shared level three (L3) memory 5050 via bus 5051. Processor B includes a block of unshared level two memory 5012. A direct memory access (DMA) engine 5060 may be programmed to transfer blocks of data/instructions from L3 memory to L2 memory 130 or L2 memory 5012 using known or later developed DMA techniques. Various types of peripherals 5062 are also coupled to memory bus 5051, such as wireless and/or wired communication controllers, etc.

[0405] In this example, processor A, processor B, L3 memory 5050 are all included in a SoC 5000 that may be encapsulated to form a package that may be mounted on a substrate such as a printed circuit board (PCB) using known or later developed packaging techniques. For example, SoC 5000 may be encapsulated in a ball grid array (BGA) package. In this example, external memory interface (EMI) 5052 allows additional external bulk memory 5054 to be accessed by processor A and/or processor B.

[0406] In this example, processor B is an ARM® processor that may be used for scalar processing and control functions. In other examples, various types of known or later developed processors may be combined with processor 100. While two processors are illustrated in this example, in another example, multiple copies of processor 100 and/or multiple copies of processor B may be included within an SoC and make use of the techniques for forming masked and null vectors without accessing system memory provided by streaming engine 125 that are described herein in more detail.

[0407] In a described example, a streaming engine includes two closely coupled streaming engines that can manage two data streams simultaneously. In another example, the streaming engine maybe capable of managing only a single stream, while in other examples the streaming engine is capable of handling more than two streams. In each case, for each stream, the streaming engine includes an address generation stage, a data formatting stage, and some storage for formatted data waiting for consumption by the processor.

[0408] In a described example, addresses are derived from algorithms that can involve multi-dimensional loops, each dimension maintaining an iteration count. In one example, the streaming engine supports six levels of nested iteration. In other examples, more or fewer levels of iteration are supported.

[0409] In a described example, one-dimensional zero padding of stream vectors is provided. In another example, two-dimensional zero padding of stream vectors is provided. In yet another example, more than two dimensions of zero padding may be provided.

[0410] In described examples, a complex DSP processor with multiple function units and dual data paths is described. In another example, a simpler DSP that is coupled to a stream processor may be used. In another example, other types of known or later developed processors may be coupled to a stream processor, such as a reduced instruction set computer (RISC), a traditional microprocessor, etc.

[0411] In a described example, the MMA supports 32×32 16-bit matrix multiplication, and 64×64 8-bit matrix multiplication, and the streaming engine is configured to provide 64-byte stream vectors. In another example, the MMA may be configured to support large or smaller matrix sizes. An associated streaming engine may be configured to provide stream vectors that have a size that is larger or smaller than 64 bytes.

[0412] In described examples, a processor that consumes a stream of data and a streaming engine that retrieves the stream of data from system memory are all included within a single integrated circuit (IC) as a system on a chip. In another example, the processor that consumes the stream of data may be packaged in a first IC and the streaming engine may be packaged in a second separate IC that is coupled to the first IC by a known or later developed communication channel or bus.

[0413] In this description, the term “couple” and derivatives thereof mean an indirect, direct, optical, and/or wireless electrical connection. Thus, if a first device couples to a second device, that connection may be through a direct electrical connection, through an indirect electrical connection via other devices and connections, through an optical electrical connection, and/or through a wireless electrical connection.

[0414] Modifications are possible in the described examples, and other examples are possible, within the scope of the claims.