WO1990014629A2 - Parallel multithreaded data processing system - Google Patents
Parallel multithreaded data processing system Download PDFInfo
- Publication number
- WO1990014629A2 WO1990014629A2 PCT/US1990/002958 US9002958W WO9014629A2 WO 1990014629 A2 WO1990014629 A2 WO 1990014629A2 US 9002958 W US9002958 W US 9002958W WO 9014629 A2 WO9014629 A2 WO 9014629A2
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- thread
- instruction
- processing element
- memory
- processing
- Prior art date
Links
- 230000015654 memory Effects 0.000 claims abstract description 120
- 230000004044 response Effects 0.000 claims abstract description 28
- 238000000034 method Methods 0.000 claims description 7
- 230000003936 working memory Effects 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 2
- 230000000717 retained effect Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 7
- 230000008520 organization Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/44—Arrangements for executing specific programs
- G06F9/448—Execution paradigms, e.g. implementations of programming paradigms
- G06F9/4494—Execution paradigms, e.g. implementations of programming paradigms data driven
Definitions
- a central feature of this model is the instruction pointer. It points to the instruction that is currently executing. Each instruction has a unique successor which is usually at the following address that immediately follows the current instruction pointer address. To execute the next instruction in a sequence, the instruction pointer is usually incremented. Execution of that next instruction is then imperative.
- the sequence of ' instructions executed in a program is often referred to as a thread of computation.
- Parallel processing machines are not limited to performing one thread at a time. They perform a number of threads simultaneously. They achieve this because they comprise multiple sequential von Neumann machines, each of which performs a single thread of computation.
- At least one data processing element is included in a data processing system. Multiple processing elements are preferred to provide added parallel processing capability. Each data processing element is itself capable of processing individual imperative control flow instructions in performing plural threads of computation. Successive instructions executed by a data processing element may be from different threads of computation. Each data processing element is preferably pipelined for high processing speed. These processing elements are preferably reduced instruction set processing elements so as to provide for easy implementation and high processing speed.
- Each processing element includes a thread descriptor storage such as a FIFO queue for storing thread descriptors. Each thread descriptor (also referred to as a token or a continuation) identifies the next instruction to be processed in a particular thread by noting an instruction pointer value. It also identifies a frame of storage locations on which the next instruction acts. This frame is indicated by a frame pointer value that points to the beginning of the frame. Multiple thread descriptors may refer to the same frame.
- the data processing system responds to load instructions to retrieve data from global memory by generating a request message.
- the load instructions are executed regardless of the current state of the global memory location.
- This request message includes a thread descriptor that need not be retained by the processing element.
- the request message is sent to a memory controller that generates a response message.
- the response message includes the data requested and a thread descriptor.
- the thread descriptor is stored in the thread descriptor storage for further processing.
- the response messages for multiple load requests may be returned to the processor in an order different from the order in which the corresponding requests are issued, and multiple requests may be issued before a response is received.
- One of the instructions of the at least one processing element is preferably the fork instruction.
- the fork instruction produces two thread descriptors from a single thread of computation.
- a joining instruction is provided to generate a single thread descriptor from two threads so as to join two threads into a single thread.
- the joining instruction specifies a memory location. The first of the two threads that reaches the join instruction marks the specified location full and stops computing. The other thread marks that location empty when it encounters a join instruction and continues computation.
- a working memory is provided.
- This working memory is preferably a cache. It is comprised of storage locations upon which the at least one processing element operates. Frames of storage held in local memory are transferred to the working memory so as to be processed. The size of the frames of storage is not fixed a priori.
- a clocking means is preferably included to clock thread descriptors out of the thread descriptor storage into the pipeline.
- the clock may also be used to clock successive stages of the pipeline such that each stage of the pipeline may operate on a separate thread of computation.
- FIG 1 illustrates the general structure of the data processing system of the present invention.
- Figure 2 illustrates the organization of a processing element.
- Figure 3 illustrates the frames held in local memory and the cache for said frames.
- Figure 4 illustrates how token descriptors are used in instruction execution.
- Figure 5 illustrates the operation of a load instruction.
- Figure 6 illustrates the operation of a load and continue instruction.
- Figure 7 illustrates the operation of a store instruction.
- Figure 8 illustrates the effect of both a fork instruction and a join instruction.
- Figure 9 illustrates a flow chart of the steps performed in a join instruction.
- Figure 10 illustrates a sample program using a fork instruction and a join instruction.
- Figure 11 illustrates a sample data flow diagram
- Figure 12 illustrates the operation of a start instruction.
- the preferred embodiment of the present invention concerns a multithreaded data processing system.
- This data processing system can operate on a plurality of threads independently.
- the basic structure of this system is shown in Figure 1.
- the system employs a plurality of processing elements 10.
- each processing element acts in parallel with the other processing elements 10.
- Global heap memory elements 20 are provided as a global memory space.
- the numerous processing elements and heap memory elements are connected via an innerconnection network 14.
- the interconnection network 14 routes messages amongst processing elements 10 and heap memory elements 20.
- a memory controller 17 is provided for controlling access to each heap memory element.
- the processing elements 10 are preferably reduced instruction set (RISC) processors.
- RISC reduced instruction set
- the use of RISC processors allows for quick operation, as well as for a simplified hardware architecture.
- the reduced instruction set of the present invention is not that which is typically found in RISC processors, rather it is an extended set of instructions comprised of the traditional set of instructions plus additional more powerful instructions.
- the additional instructions comprise the fork, join and start instructions discussed below. Further, the load instruction is different from traditional reduced instruction sets.
- FIG. 2 shows the basic structure of a processing element 10.
- Each processing element 10 is capable of processing a plurality of threads of computation independently. It is able to achieve this independence by using a pipelining structure 36.
- the pipeline 36 is comprised of four stages. The first stage is an instruction fetch stage 32. As its name implies, the basic task of this stage is to
- the operand fetch stage 30 is the second stage of this pipeline 36. It fetches operands that are necessary to perform the fetched instruction.
- the operand fetch stage 30 access frames of memory locations 24 where the operands are found. If the fetched instruction is a load, store or start instruction, the fetched instruction is processed specially at 38.
- Frames 24 serve the role formerly served by register sets in early architectures. Specifically, operands are stored in the frames of storage 24 for
- the frames 24, however, are superior to registers because they are not limited to a fixed size.
- the size of a frame can be adapted to fit the needs of the threads that access it. Further, employing a frame pointer to specify the frame
- 25 location provides the potential for allocation of a seemingly limitless number of frames having varying sizes. Hence, the number and sizes of frames available is not limited by the number of register sets .
- the total frame memory 24 is likely too large to provide for sufficiently rapid memory access; thus, it is desirable to employ a cache 3 such as shown in Figure 3.
- the size of the cache 3 may vary but should hold at least a couple of frames. Frames can be loaded into the cache 3 when needed.
- the cache 3 should be situated as shown in Figure 3 between local frame storage 24 and the pipeline 36.
- the processing of the particular instruction continues in the third stage of the pipeline, the function units stage 28.
- This stage represents the traditional arithmetic and logical unit (ALU) of a conventional processor.
- ALU arithmetic and logical unit
- Each processing element 10 also includes a token queue 22.
- the token queue 22 stores thread descriptors (also known as tokens) for threads of computation to be performed by the processing element 10.
- thread descriptors held in the token queue 22 are placed into the processing pipeline 36, the next instruction in the thread of computation is performed.
- the processing elements 10 of the preferred embodiment of the present invention operate by feeding thread descriptors into the pipeline 36 every clock cycle. Clock cycles are generated by a processing element clock 101. Independence in processing of threads is achieved by having different threads of computation being placed into the pipeline 36 on successive clock cycles.
- Each token is comprised of two pointers: an instruction pointer (42B in Figure 4) and a frame pointer 42A.
- the instruction pointer 42B points to a particular instruction held in the code section 26 of the local memory of the processing element.
- the frame pointer 42A points to the beginning of a particular frame of memory locations in the frames section 24 of a processor element local memory.
- Figure 4 shows how the tokens are used by the processing pipeline 36. When the token 42 enters the pipeline 36, its instruction pointer 42B is used to locate a particular instruction 44 held in the code section 26 of local memory.
- the data processing system of the present invention operates on an imperative set of instructions. In other words, whether an instruction executes is determined solely by whether the instruction pointer 42B points to it. There is no state of an operand which controls execution, only a value which Is operated upon.
- Another way of characterizing this aspect of the present invention is to classify the present invention as a control flow system as opposed to a data flow system. The control mechanism, not the data, dicates whether execution of an instruction takes place. The present system, thus, may be contrasted with data flow systems wherein the state of the data and the validity of the data is typically taken into account.
- the execution of instructions in the present system is dictated by the value of the instruction pointer. Once an instruction is fetched, execution is imperative. Having already located the instruction, the processing pipeline 36 next seeks to obtain the appropriate operands for that instruction. To locate the operands, the frame pointer 42A of the token 42 is used to locate a particular frame 46 held in the frames portion 24 of local memory. As noted above, the frame pointer 42A points to the beginning of the specified frame 46. The addresses held in the operand field of the fetched instruction 44 are used as offsets to locate the memory locations where the desired operands are held.
- the a and b fields of that instruction 44 are used as offsets to locate the two operands that are to be added.
- the first operand is held at an address equal to the address at the beginning of the specified frame 46 plus a.
- the second operand to be added is, likewise, located in the same manner. It is located at the beginning address of the specified frame 46 plus an offset of b.
- the processing pipeline 36 next performs the desired function specified by the instruction 44.
- the two operands are added.
- the addition specified by the instruction of the example case takes place in the third stage of the pipeline 36, the function units stage 28.
- This reduced instruction set includes typical arithmetic and logic operations, such as, addition, subtraction, multiplication, etc.
- all arithmetic logical operations are three address operations.
- the operands specify three addresses.
- the plus instruction used as an example in Figure 4, falls within this class of instructions. The three
- addresses specified by that instruction are the addresses of the two operands to be added and the l l -
- control instructions control the order and flow of program execution.
- the continuation emerging from the processing pipeline 36 is the instruction pointer value specified as an operand in the jump instruction and the current frame pointer.
- a conditional jump instruction a location specified as an operand in the instruction is tested for some condition, and, depending on the outcome, one of two possible continuations emerges out of the processing pipeline 36. It contains either the current instruction pointer plus 1 or a new instruction pointer specified as an operand in the instruction. In either case, it contains the current frame pointer.
- the frames are typically stored, as previously noted, in a frames portion 24 of local processing element memory.
- a difficulty arises, however, when data values held in a frame need to be shared by more than one processing element.
- the preferred embodiment of the present invention utilizes the global memory available in the heap memory elements 20 ( Figure 1).
- shared data values are stored in the global heap memory elements 20.
- the processing elements 10 must first bring the data values into local frame storage 24. This can be done using a load instruction. Results of these operations may be returned to the global heap memory elements 20 using a store instruction.
- the load and store instructions are the only instructions for moving data in and out of local memory of processing elements 10. They do not perform any arithmetic operations. They move data between frames 24 of local memory and heap memory elements 20.
- FIG. 5 shows the load instruction in more detail.
- the processing element 10 is executing a given thread of computation 48. Included within this thread 48 is a local instruction to which the instruction pointer (IP) points.
- IP instruction pointer
- the processing element 10 attempts to execute it.
- the instruction specified in Figure 5 tells the processing element to load a value held in a heap memory element 20 location to a memory location held within a local frame 24.
- the processing element 10 looks for the heap memory location address by looking at the memory location specified by the frame pointer plus a. That location contains a pointer to a specific heap memory element 20 location. Processing then continues outside the pipeline 36 at the load or store component 38 of the processing element 10. This component generates a message 50 that is sent through the network 14 to the heap memory elements 20.
- the message 50 is comprised of several fields. Specifically, a first field tells the memory controller 17 that a read is desired of the heap memory element 20 location specified in the second field of the message 50. This second field includes the address taken from the frame as previously discussed. In the example case of Figure 5, the pointer was held at frame pointer plus a. The third field of the message specifies a continuation which is a thread to be initiated after the read value has been loaded. The fourth and final field denoted as "d" specifies the offset of the memory location within the frame 24 where the read value is to be loaded.
- the memory controller 17 of the heap memory element responds to this message by sending a re- sponse message to the processing element 10.
- the response message 52 is also comprised of several fields.
- the first field specifies that this is a start message.
- the second field is the value taken from address [a] of the heap memory to be loaded into the memory location at frame pointer plus d in the processing element 10.
- the third field is the continuation that was passed in the request message 50.
- the fourth field is the offset field which also was sent in the request message 50.
- the processing element 10 When the processing element 10 receives this message 52, it does two things. First, it loads the value read from heap memory element 20 to the memory location specified by the frame pointer plus the offset. In Figure 5, the value V is loaded into the memory location pointed to by the frame pointer plus d. Second, it forwards the thread descriptor specified in the third field of the response message 52 to the token queue 22.
- loadc terminates the current thread in which it is contained (i.e. no continuation is issued).
- a special load and continue Instruction is included to allow the thread to continue operating after the load instruction is performed.
- the operation of the load and continue instruction (denoted as loadc) is Illustrated in Figure 6. It generates the same request message 50 as the load instruction, and the memory controller 17 responds with the same response message 52.
- the loadc instruction differs from the load instruction in two ways. First, the instruction pointer in the request that is sent to memory Is not the current instruction pointer plus 1 but a different instruction pointer (designated as IP in Figure 6) that is an operand of the loadc instruction.
- the loadc instruction allows a continuation of the thread such as 48A shown in Figure 6.
- the load and loadc instructions defer the task of retrieving the heap memory element data to the memory controller 17. By doing this, they overcome the delay that usually occurs waiting for access to memory. In typical data processing systems (both sequential and parallel) , the processor must wait during the time period in which memory is accessed. The waiting time period is usually quite large relative to the time required to perform other processor operations. This memory latency is overcome with the load and loadc instructions by not requiring a processor to idle while memory is being accessed.
- the processor After deferring the task of accessing memory, the processor takes another continuation from the token queue 22 to process so that it does not idle.
- the load instruction thereby allows the processor to switch to another thread of computation if it would idle otherwise.
- the token queue 22 assures that another thread will generally be available and thus, facilitates maximum use of the processor despite memory latency.
- loads can be considered as split-phase transactions comprised of a request phase and a response phase. Performing the loads in this manner enables the preferred embodiment of the present invention to issue multiple load requests before receiving any response to a particular one of these requests. Moreover, responses can be generated in an order different from the order in which requests are made. A continuation for the thread to which the response is targeted is specified in the response message 52.
- a store instruction is similar to a load instruction except that the data held in the local frame of memory 24 is written into the heap memory element location 20.
- Figure 7 illustrates a sample operation of a store instruction. The store instruction has two operands.
- the first operand "a" specifies an address in local memory where a value to be stored is held
- the second operand "x" specifies an address in local memory that contains the heap memory element address where the value is to be stored.
- the value is stored at the local memory address specified by the frame pointer plus a.
- the heap memory location address is stored at a local memory address specified by the frame pointer plus x.
- the request message 54 specifies that a write is requested. It notes the heap memory element address where write will take place and the value to be stored there. In response to this message, the memory controller 17 writes the value into the heap memory element address.
- the processing element 10 after issuing this message 54, issues a continuation designated by the instruction pointer plus one and the frame pointer. No response message is generated.
- New control instructions in the present invention are the fork, join and start instructions which control operation of the processing system.
- a fork instruction produces two continuations from a current operating thread.
- One continuation is the next instruction in the thread in which the fork instruction is executed.
- the other continuation represents an entirely different thread that operates on the same frame.
- two concurrent thread descriptors are passed to the token queue 22.
- the first thread descriptor is associated with A[i] 62 which contains the next instruction to be executed in the current thread.
- the other thread descriptor is associated with B[i] 64 which contains an instruction of an entirely different thread of computation.
- the fork instruction thus provides an ability to initiate a new thread that can execute concurrently with the current thread. Any synchronization that needs to be achieved between the threads can be instituted using a join instruction.
- the join instruction (66 in Figure 8) performs the opposite operation of the fork instruction. It causes two independently running threads of computation to produce only a single continuation. In other words, it takes two threads and produces a single thread from it.
- Figure 9 shows a flow chart of how the join operation joins together two threads.
- the syntax of the join operation is "join w" , where w specifies a memory location.
- the single join instruction must be executed by both threads in order for a join to occur.
- the join instruction acts on the specified memory location w which Is marked as either all ones or all zeroes.
- the value at location w is toggled to the other state of ones or zeroes that is opposite the current state (Step 68). Thus, for instance, If w were all ones, it would be switched to all zeroes and vice versa.
- a thread is the first thread to execute the join instruction for memory location w, the memory location w is marked as all ones .
- Step 70 The system then looks to see if the memory location is marked all ones (Step 70) . If the memory location is already all ones, the thread that caused it to be marked all ones issues no continuation (Step 74). On the other hand, if the thread is not the first thread to execute the join instruction but is instead the second thread to execute the join instruction, then, the memory location w is marked all zeroes, and a continuation issues so that the processing of a new thread is initiated beginning at the next instruction (Step 76). Hence, the rule of the join operation is that both threads terminate and a new thread is initiated after the arrival of the second one .
- Figure 10 shows an example of a program where fork instructions and join instructions are used.
- the instruction pointer starts off with a value of IP.
- a fork to IP is performed.
- expressions el and e2 will subsequently be performed independently in two separate threads that share the same frame pointer.
- the thread having an instruction pointer value of IP next executes a jump to the instruction specified by IP. . .
- the second thread accordingly next executes join the instruction specified by IP. . .
- This join instruction 98 is a join instruction specifying memory location r. It does not matter which thread
- J 3 executes the join first.
- the one that executes it first will issue no continuation.
- the other will, in contrast, issue a continuation.
- the thread that issues a continuation specifies e3 as the next instruction 100 to be executed.
- the join assures that both el and e2 have been performed by the two threads respectively prior to e3 being performed.
- the start instruction is used to terminate a thread of computation in the current frame and to initiate a new thread of computation in a different frame.
- a start instruction has three operands v, c and d where v, c and d specify memory locations in the current frame.
- the location designated by c contains a new continuation, that is, a new frame pointer and a new instruction pointer.
- the location designated by d specifies an offset in the frame designated by the new frame pointer.
- the new continuation may be on a different data processing element or it may be on the same one.
- Figure 12 shows the execution of the start instruction in more detail.
- start v c d the start instruction "start v c d" is placed in the processing pipeline 36 of data processing element A 200.
- the three operands [v] , [c] and [d] are read from the current frame.
- Processing then continues in the load/store component 38.
- a message 202 is generated and sent to data processing element B 201.
- the format of the message 202 is " ⁇ start, [v] , [c] , [d]>" .
- the format of the message is identical to the format of the previously described response messages 52 from a memory controller to a data processing element.
- Data processing element A 200 may continue to perform instructions from other threads taken from its thread descriptor storage 22.
- the message arrives at the store component 40 of data processing element B 201, it is treated in exactly the same way as the previously described response messages 52 from a memory controller.
- the value [v] is stored at offset [d] in the frame specified In the continuation [c] , and the continuation [c] is placed in the token queue 22.
- data processing element B 201 may then execute instructions from the thread designated by the continuation [c] .
- the start instruction provides a way to simultaneously do two things: deliver a data value from one data processing element (A) to another (B), and initiate a thread of computation In B.
- the present invention can also be used in a producer/consumer environment. In that case it may be desirable to implement the heap memory elements as I-structures .
- the I-structure organization of the heap memory elements 20 prevents potential conflicts for memory locations that could occur otherwise.
- two new types of messages are desirable for the load and store instructions, respectively. Instead of merely sending a read request when loading, an I-read request is sent.
- the I-read behaves like an ordinary read in most respects. However, if the value sought is not available, the
- memory controller 17 stores the request message in the heap memory element 20 location where the value is supposed to be. If other read requests arrive while the value is still not available, messages indicating such requests are placed on a deferred
- the memory controller 17 stores the value in the heap memory element 20 location, removes the message and generates a response message containing the value and l_ continuation to the appropriate processing element 10.
- an I-write request is sent.
- the data value is
- FIG. 11 shows a sample data flow graph which the present invention can execute.
- the data flow graph shown in Figure 11 corresponds to the 0 equation (x + 1) x 3 + (x + l)/4.
- the instruction performed at node 110 is implemented using an add instruction.
- the syntax of that instruction is add x,l,y. It adds x + 1 and store it at location y. Once the add is performed, a fork instruction Is performed.
- the computations performed at node 112 and 114 are performed concurrently.
- the computation performed at node 114 is performed as a continuation of the same thread in which the fork instruction was executed.
- An instruction to multiply y and 3 and store it in a location performs the task by node 114.
- an instruction to divide the value held in location y by 4 and to store the resulting produce in a register r performs the operation specified by node 112.
- an instruction to join at location j is executed after the multiplication instruction.
- the instruction to join location j is performed.
- the second thread Presuming that the second thread completes its calculation after the first thread, it initiates the thread that performs the adding operation specified by node 116 of adding the values held at location a and location r and storing them in a location queue.
- the first thread In contrast, halts operation after executing the join, without initiating any thread.
- the preferred embodiment of the present invention can adequately execute data flow graphs. It is, thus, a sort of hybrid which executes data flow graphs on a Von Neumann architecture. Moreover it can execute these data flow graphs independently and in parallel.
- the reduced Instruction set nature of the processing element further heightens performance by allowing computations to be formed quickly with non-complex architectural hardware. Further, the unique nature of the loading instructions prevents wasted processor idle time attributable to memory latency.
- a particularly attractive aspect of the present invention is that, since its instruction set is a proper extension of the instruction set of a conventional reduced instruction set processor, compatibility with existing software can be maintained.
Abstract
Description
Claims
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US35893789A | 1989-05-26 | 1989-05-26 | |
US358,937 | 1989-05-26 | ||
US527,122 | 1990-05-21 | ||
US07/527,122 US5353418A (en) | 1989-05-26 | 1990-05-21 | System storing thread descriptor identifying one of plural threads of computation in storage only when all data for operating on thread is ready and independently of resultant imperative processing of thread |
Publications (2)
Publication Number | Publication Date |
---|---|
WO1990014629A2 true WO1990014629A2 (en) | 1990-11-29 |
WO1990014629A3 WO1990014629A3 (en) | 1991-02-07 |
Family
ID=27000260
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US1990/002958 WO1990014629A2 (en) | 1989-05-26 | 1990-05-25 | Parallel multithreaded data processing system |
Country Status (3)
Country | Link |
---|---|
US (1) | US5499349A (en) |
EP (1) | EP0473714A1 (en) |
WO (1) | WO1990014629A2 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001065367A2 (en) * | 2000-02-28 | 2001-09-07 | Sun Microsystems, Inc. | Supporting inter-process communication through a conditional trap instruction |
Families Citing this family (65)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5812811A (en) * | 1995-02-03 | 1998-09-22 | International Business Machines Corporation | Executing speculative parallel instructions threads with forking and inter-thread communication |
US5710923A (en) * | 1995-04-25 | 1998-01-20 | Unisys Corporation | Methods and apparatus for exchanging active messages in a parallel processing computer system |
US6343309B1 (en) * | 1996-09-30 | 2002-01-29 | International Business Machines Corporaton | Method and apparatus for parallelizing a graphics pipeline |
US5842003A (en) * | 1997-03-26 | 1998-11-24 | Unisys Corporation | Auxiliary message arbitrator for digital message transfer system in network of hardware modules |
US6064818A (en) * | 1997-04-10 | 2000-05-16 | International Business Machines Corporation | Straight path optimization for compilers |
US6182177B1 (en) * | 1997-06-13 | 2001-01-30 | Intel Corporation | Method and apparatus for maintaining one or more queues of elements such as commands using one or more token queues |
US6092155A (en) | 1997-07-10 | 2000-07-18 | International Business Machines Corporation | Cache coherent network adapter for scalable shared memory processing systems |
US6044438A (en) | 1997-07-10 | 2000-03-28 | International Business Machiness Corporation | Memory controller for controlling memory accesses across networks in distributed shared memory processing systems |
US6105051A (en) * | 1997-10-23 | 2000-08-15 | International Business Machines Corporation | Apparatus and method to guarantee forward progress in execution of threads in a multithreaded processor |
US6697935B1 (en) * | 1997-10-23 | 2004-02-24 | International Business Machines Corporation | Method and apparatus for selecting thread switch events in a multithreaded processor |
US6212544B1 (en) | 1997-10-23 | 2001-04-03 | International Business Machines Corporation | Altering thread priorities in a multithreaded processor |
US6567839B1 (en) | 1997-10-23 | 2003-05-20 | International Business Machines Corporation | Thread switch control in a multithreaded processor system |
US6076157A (en) * | 1997-10-23 | 2000-06-13 | International Business Machines Corporation | Method and apparatus to force a thread switch in a multithreaded processor |
US6061710A (en) * | 1997-10-29 | 2000-05-09 | International Business Machines Corporation | Multithreaded processor incorporating a thread latch register for interrupt service new pending threads |
US6256775B1 (en) | 1997-12-11 | 2001-07-03 | International Business Machines Corporation | Facilities for detailed software performance analysis in a multithreaded processor |
US6018759A (en) * | 1997-12-22 | 2000-01-25 | International Business Machines Corporation | Thread switch tuning tool for optimal performance in a computer processor |
US6341338B1 (en) * | 1999-02-04 | 2002-01-22 | Sun Microsystems, Inc. | Protocol for coordinating the distribution of shared memory |
US6378066B1 (en) | 1999-02-04 | 2002-04-23 | Sun Microsystems, Inc. | Method, apparatus, and article of manufacture for developing and executing data flow programs, and optimizing user input specifications |
US6434714B1 (en) | 1999-02-04 | 2002-08-13 | Sun Microsystems, Inc. | Methods, systems, and articles of manufacture for analyzing performance of application programs |
US6535905B1 (en) * | 1999-04-29 | 2003-03-18 | Intel Corporation | Method and apparatus for thread switching within a multithreaded processor |
US6542921B1 (en) | 1999-07-08 | 2003-04-01 | Intel Corporation | Method and apparatus for controlling the processing priority between multiple threads in a multithreaded processor |
US6629236B1 (en) * | 1999-11-12 | 2003-09-30 | International Business Machines Corporation | Master-slave latch circuit for multithreaded processing |
US6357016B1 (en) | 1999-12-09 | 2002-03-12 | Intel Corporation | Method and apparatus for disabling a clock signal within a multithreaded processor |
US6889319B1 (en) * | 1999-12-09 | 2005-05-03 | Intel Corporation | Method and apparatus for entering and exiting multiple threads within a multithreaded processor |
US6496925B1 (en) | 1999-12-09 | 2002-12-17 | Intel Corporation | Method and apparatus for processing an event occurrence within a multithreaded processor |
US7051329B1 (en) | 1999-12-28 | 2006-05-23 | Intel Corporation | Method and apparatus for managing resources in a multithreaded processor |
US7035989B1 (en) | 2000-02-16 | 2006-04-25 | Sun Microsystems, Inc. | Adaptive memory allocation |
US7856633B1 (en) | 2000-03-24 | 2010-12-21 | Intel Corporation | LRU cache replacement for a partitioned set associative cache |
US7093109B1 (en) * | 2000-04-04 | 2006-08-15 | International Business Machines Corporation | Network processor which makes thread execution control decisions based on latency event lengths |
US6546359B1 (en) | 2000-04-24 | 2003-04-08 | Sun Microsystems, Inc. | Method and apparatus for multiplexing hardware performance indicators |
US6647546B1 (en) | 2000-05-03 | 2003-11-11 | Sun Microsystems, Inc. | Avoiding gather and scatter when calling Fortran 77 code from Fortran 90 code |
US6802057B1 (en) | 2000-05-03 | 2004-10-05 | Sun Microsystems, Inc. | Automatic generation of fortran 90 interfaces to fortran 77 code |
US6986130B1 (en) | 2000-07-28 | 2006-01-10 | Sun Microsystems, Inc. | Methods and apparatus for compiling computer programs using partial function inlining |
US6910107B1 (en) | 2000-08-23 | 2005-06-21 | Sun Microsystems, Inc. | Method and apparatus for invalidation of data in computer systems |
US6957208B1 (en) | 2000-10-31 | 2005-10-18 | Sun Microsystems, Inc. | Method, apparatus, and article of manufacture for performance analysis using semantic knowledge |
US7320065B2 (en) | 2001-04-26 | 2008-01-15 | Eleven Engineering Incorporated | Multithread embedded processor with input/output capability |
US7047395B2 (en) * | 2001-11-13 | 2006-05-16 | Intel Corporation | Reordering serial data in a system with parallel processing flows |
CN1212567C (en) * | 2002-03-28 | 2005-07-27 | 徐肇昌 | Structure and method for software simulation of N+1 parallel program in sequence network |
US7200738B2 (en) | 2002-04-18 | 2007-04-03 | Micron Technology, Inc. | Reducing data hazards in pipelined processors to provide high processor utilization |
US8024735B2 (en) * | 2002-06-14 | 2011-09-20 | Intel Corporation | Method and apparatus for ensuring fairness and forward progress when executing multiple threads of execution |
US7849297B2 (en) | 2003-08-28 | 2010-12-07 | Mips Technologies, Inc. | Software emulation of directed exceptions in a multithreading processor |
WO2005022385A1 (en) * | 2003-08-28 | 2005-03-10 | Mips Technologies, Inc. | Mechanisms for dynamic configuration of virtual processor resources |
DE602004017879D1 (en) | 2003-08-28 | 2009-01-02 | Mips Tech Inc | INTEGRATED MECHANISM FOR SUSPENDING AND FINAL PROCESSOR |
US7711931B2 (en) * | 2003-08-28 | 2010-05-04 | Mips Technologies, Inc. | Synchronized storage providing multiple synchronization semantics |
US7418585B2 (en) | 2003-08-28 | 2008-08-26 | Mips Technologies, Inc. | Symmetric multiprocessor operating system for execution on non-independent lightweight thread contexts |
US20050050305A1 (en) * | 2003-08-28 | 2005-03-03 | Kissell Kevin D. | Integrated mechanism for suspension and deallocation of computational threads of execution in a processor |
US9032404B2 (en) * | 2003-08-28 | 2015-05-12 | Mips Technologies, Inc. | Preemptive multitasking employing software emulation of directed exceptions in a multithreading processor |
US7836450B2 (en) | 2003-08-28 | 2010-11-16 | Mips Technologies, Inc. | Symmetric multiprocessor operating system for execution on non-independent lightweight thread contexts |
US7376954B2 (en) * | 2003-08-28 | 2008-05-20 | Mips Technologies, Inc. | Mechanisms for assuring quality of service for programs executing on a multithreaded processor |
US7870553B2 (en) * | 2003-08-28 | 2011-01-11 | Mips Technologies, Inc. | Symmetric multiprocessor operating system for execution on non-independent lightweight thread contexts |
US7594089B2 (en) * | 2003-08-28 | 2009-09-22 | Mips Technologies, Inc. | Smart memory based synchronization controller for a multi-threaded multiprocessor SoC |
US20080052687A1 (en) * | 2003-11-03 | 2008-02-28 | Agustin Gonzales-Tuchmann | Development environment for data transformation applications |
US8230423B2 (en) | 2005-04-07 | 2012-07-24 | International Business Machines Corporation | Multithreaded processor architecture with operational latency hiding |
US20060230409A1 (en) * | 2005-04-07 | 2006-10-12 | Matteo Frigo | Multithreaded processor architecture with implicit granularity adaptation |
EP1783604A3 (en) * | 2005-11-07 | 2007-10-03 | Slawomir Adam Janczewski | Object-oriented, parallel language, method of programming and multi-processor computer |
GB2474521B (en) * | 2009-10-19 | 2014-10-15 | Ublox Ag | Program flow control |
US8230410B2 (en) | 2009-10-26 | 2012-07-24 | International Business Machines Corporation | Utilizing a bidding model in a microparallel processor architecture to allocate additional registers and execution units for short to intermediate stretches of code identified as opportunities for microparallelization |
US9672132B2 (en) * | 2009-11-19 | 2017-06-06 | Qualcomm Incorporated | Methods and apparatus for measuring performance of a multi-thread processor |
US9996403B2 (en) * | 2011-09-30 | 2018-06-12 | Oracle International Corporation | System and method for providing message queues for multinode applications in a middleware machine environment |
US9195516B2 (en) | 2011-12-01 | 2015-11-24 | International Business Machines Corporation | Determining collective barrier operation skew in a parallel computer |
US8924763B2 (en) * | 2011-12-15 | 2014-12-30 | International Business Machines Corporation | Synchronizing compute node time bases in a parallel computer |
US9747108B2 (en) * | 2015-03-27 | 2017-08-29 | Intel Corporation | User-level fork and join processors, methods, systems, and instructions |
US20160381050A1 (en) | 2015-06-26 | 2016-12-29 | Intel Corporation | Processors, methods, systems, and instructions to protect shadow stacks |
US10394556B2 (en) * | 2015-12-20 | 2019-08-27 | Intel Corporation | Hardware apparatuses and methods to switch shadow stack pointers |
US10430580B2 (en) | 2016-02-04 | 2019-10-01 | Intel Corporation | Processor extensions to protect stacks during ring transitions |
Family Cites Families (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3614745A (en) * | 1969-09-15 | 1971-10-19 | Ibm | Apparatus and method in a multiple operand stream computing system for identifying the specification of multitasks situations and controlling the execution thereof |
US4229790A (en) * | 1978-10-16 | 1980-10-21 | Denelcor, Inc. | Concurrent task and instruction processor and method |
US4481573A (en) * | 1980-11-17 | 1984-11-06 | Hitachi, Ltd. | Shared virtual address translation unit for a multiprocessor system |
US4530051A (en) * | 1982-09-10 | 1985-07-16 | At&T Bell Laboratories | Program process execution in a distributed multiprocessor system |
DE3751503T2 (en) * | 1986-03-26 | 1996-05-09 | Hitachi Ltd | Data processor in pipeline structure with the ability to decode and execute multiple instructions in parallel. |
US4819155A (en) * | 1987-06-01 | 1989-04-04 | Wulf William A | Apparatus for reading to and writing from memory streams of data while concurrently executing a plurality of data processing operations |
US4943908A (en) * | 1987-12-02 | 1990-07-24 | International Business Machines Corporation | Multiple branch analyzer for prefetching cache lines |
US5050070A (en) * | 1988-02-29 | 1991-09-17 | Convex Computer Corporation | Multi-processor computer system having self-allocating processors |
US5050068A (en) * | 1988-10-03 | 1991-09-17 | Duke University | Method and apparatus for using extracted program flow information to prepare for execution multiple instruction streams |
US5241635A (en) * | 1988-11-18 | 1993-08-31 | Massachusetts Institute Of Technology | Tagged token data processing system with operand matching in activation frames |
US5226131A (en) * | 1989-12-27 | 1993-07-06 | The United States Of America As Represented By The United States Department Of Energy | Sequencing and fan-out mechanism for causing a set of at least two sequential instructions to be performed in a dataflow processing computer |
-
1990
- 1990-05-25 WO PCT/US1990/002958 patent/WO1990014629A2/en not_active Application Discontinuation
- 1990-05-25 EP EP90909259A patent/EP0473714A1/en not_active Ceased
-
1994
- 1994-10-03 US US08/317,708 patent/US5499349A/en not_active Expired - Fee Related
Non-Patent Citations (5)
Title |
---|
Computer Architecture Conference Proceedings, Volume 17, No. 3, June 1989, ACM, (Washington, US), R.S. NIKHIL et al.: "Can Dataflow Subsume von Neumann Computing?", pages 262-272 see the whole article * |
Conference Proceedings of the 15th Annual International Symposium on Computer Architecture, 30 May - 2 June 1988, Honolulu, Hawaii, IEEE, (US), R.A. IANNUCCI: "Toward a Dataflow/ von Neumann Hybrid Architecture", pages 131-140 see section 2.1, first paragraph; section 3.3; section 3.3.1; figures 8,9 * |
Conference Proceedings of the 15th Annual International Symposium on Computer Architecture, 30 May - 2 June 1988, Honolulu, Hawaii, IEEE, (US), R.H. HALSTEAD, Jr. et al.: "MASA: A Multithreaded Processor Architecture for Parallel Symbolic Computing", pages 443-451 see figure 6 * |
IEEE Transactions on Computers, Volume C-36, No. 12, December 1987, IEEE, (New York, US), R. BUEHRER et al.: "Incorporating Data Flow Ideas Into von Neumann Processors for Parallel Execution", pages 1515-1522 see figures 2,3; page 1517, right-hand column, line 32 - page 1518, right-hand column, line 41 * |
Proceedings of the 1988 International Conference on Parallel Processing, 15-19 August 1988, Volume II Software, The Pennsylvania State University Press, (PE, US), B.R. PREISS et al.: "Semi-Static Dataflow", pages 127-134 see section 4.1 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2001065367A2 (en) * | 2000-02-28 | 2001-09-07 | Sun Microsystems, Inc. | Supporting inter-process communication through a conditional trap instruction |
WO2001065367A3 (en) * | 2000-02-28 | 2003-03-20 | Sun Microsystems Inc | Supporting inter-process communication through a conditional trap instruction |
US6732363B1 (en) | 2000-02-28 | 2004-05-04 | Sun Microsystems, Inc. | Supporting inter-process communication through a conditional trap instruction |
Also Published As
Publication number | Publication date |
---|---|
WO1990014629A3 (en) | 1991-02-07 |
EP0473714A1 (en) | 1992-03-11 |
US5499349A (en) | 1996-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US5353418A (en) | System storing thread descriptor identifying one of plural threads of computation in storage only when all data for operating on thread is ready and independently of resultant imperative processing of thread | |
US5499349A (en) | Pipelined processor with fork, join, and start instructions using tokens to indicate the next instruction for each of multiple threads of execution | |
JP3120152B2 (en) | Computer system | |
US5560029A (en) | Data processing system with synchronization coprocessor for multiple threads | |
Nikhil et al. | T: A multithreaded massively parallel architecture | |
US7020871B2 (en) | Breakpoint method for parallel hardware threads in multithreaded processor | |
US5006980A (en) | Pipelined digital CPU with deadlock resolution | |
JP3461704B2 (en) | Instruction processing system and computer using condition codes | |
US5404552A (en) | Pipeline risc processing unit with improved efficiency when handling data dependency | |
US6021489A (en) | Apparatus and method for sharing a branch prediction unit in a microprocessor implementing a two instruction set architecture | |
US6671827B2 (en) | Journaling for parallel hardware threads in multithreaded processor | |
EP0357188B1 (en) | Pipelined processor | |
IE74215B1 (en) | Pipelined Processor | |
US4819155A (en) | Apparatus for reading to and writing from memory streams of data while concurrently executing a plurality of data processing operations | |
US6725365B1 (en) | Branching in a computer system | |
JP3797570B2 (en) | Apparatus and method using semaphore buffer for semaphore instructions | |
JP2535252B2 (en) | Parallel processor | |
JPH09152973A (en) | Method and device for support of speculative execution of count / link register change instruction | |
US5590293A (en) | Dynamic microbranching with programmable hold on condition, to programmable dynamic microbranching delay minimization | |
Kawano et al. | Fine-grain multi-thread processor architecture for massively parallel processing | |
US5737562A (en) | CPU pipeline having queuing stage to facilitate branch instructions | |
JP2916605B2 (en) | Computer processor | |
US6988121B1 (en) | Efficient implementation of multiprecision arithmetic | |
EP1050805B1 (en) | Transfer of guard values in a computer system | |
JPH10283178A (en) | Method and system for issuing instruction |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Kind code of ref document: A2 Designated state(s): JP |
|
AL | Designated countries for regional patents |
Kind code of ref document: A2 Designated state(s): AT BE CH DE DK ES FR GB IT LU NL SE |
|
AK | Designated states |
Kind code of ref document: A3 Designated state(s): JP |
|
AL | Designated countries for regional patents |
Kind code of ref document: A3 Designated state(s): AT BE CH DE DK ES FR GB IT LU NL SE |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1990909259 Country of ref document: EP |
|
WWP | Wipo information: published in national office |
Ref document number: 1990909259 Country of ref document: EP |
|
WWR | Wipo information: refused in national office |
Ref document number: 1990909259 Country of ref document: EP |
|
WWW | Wipo information: withdrawn in national office |
Ref document number: 1990909259 Country of ref document: EP |