US20010019331A1 - Unified memory architecture for use in computer system - Google Patents
Unified memory architecture for use in computer system Download PDFInfo
- Publication number
- US20010019331A1 US20010019331A1 US09/137,067 US13706798A US2001019331A1 US 20010019331 A1 US20010019331 A1 US 20010019331A1 US 13706798 A US13706798 A US 13706798A US 2001019331 A1 US2001019331 A1 US 2001019331A1
- Authority
- US
- United States
- Prior art keywords
- memory
- computer system
- memory controller
- implemented
- coupled
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/60—Memory management
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F12/00—Accessing, addressing or allocating within memory systems or architectures
- G06F12/02—Addressing or allocation; Relocation
- G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
- G06F12/10—Address translation
- G06F12/1081—Address translation for peripheral access to main memory, e.g. direct memory access [DMA]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F12/00—Accessing, addressing or allocating within memory systems or architectures
- G06F12/02—Addressing or allocation; Relocation
- G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
- G06F12/10—Address translation
- G06F12/1027—Address translation using associative or pseudo-associative address translation means, e.g. translation look-aside buffer [TLB]
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09G—ARRANGEMENTS OR CIRCUITS FOR CONTROL OF INDICATING DEVICES USING STATIC MEANS TO PRESENT VARIABLE INFORMATION
- G09G2360/00—Aspects of the architecture of display systems
- G09G2360/12—Frame memory handling
- G09G2360/125—Frame memory handling using unified memory architecture [UMA]
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09G—ARRANGEMENTS OR CIRCUITS FOR CONTROL OF INDICATING DEVICES USING STATIC MEANS TO PRESENT VARIABLE INFORMATION
- G09G5/00—Control arrangements or circuits for visual indicators common to cathode-ray tube indicators and other visual indicators
- G09G5/36—Control arrangements or circuits for visual indicators common to cathode-ray tube indicators and other visual indicators characterised by the display of a graphic pattern, e.g. using an all-points-addressable [APA] memory
- G09G5/39—Control of the bit-mapped memory
Definitions
- the present invention relates to the field of computer systems. Specifically, the present invention relates to a computer system architecture including dynamic memory allocation of pixel buffers for graphics and image processing.
- peripheral processors are used to render graphics images (synthesis) and peripheral image processors are used to perform image processing (analysis).
- CPU main memory is separate from peripheral memory units which can be dedicated to graphics rendering or image processing or other computational functions.
- the prior art computer graphics system 100 includes three separate memory units; a main memory 102 , a dedicated graphics memory 104 , and a dedicated image processing memory (image processor memory) 105 .
- Main memory 102 provides fast access to data for a CPU 106 and an input/output device 108 .
- the CPU 106 and input/output device 108 are connected to main memory 102 via a main memory controller 110 .
- Dedicated graphics memory 104 provides fast access to graphics data for a graphics processor 112 via a graphics memory controller 114 .
- Dedicated image processor memory 105 provides fast access to buffers of data used by an image processor 116 via an image processor memory controller 118 .
- CPU 106 has read/write access to main memory 102 but not to dedicated graphics memory 104 or dedicated image processor memory 105 .
- image processor 116 has read/write access to dedicated image processor memory 105 , but not to main memory 102 or dedicated graphics memory 104 .
- graphics processor 112 has read/write access to dedicated graphics memory 104 but not to main memory 102 or dedicated image processor memory 105 .
- Certain computer system applications require that data, stored in main memory 102 or in one of the dedicated memory units 104 , 105 , be operated upon by a processor other than the processor which has access to the memory unit in which the desired data is stored. Whenever data stored in one particular memory unit is to be processed by a designated processor other than the processor which has access to that particular memory unit, the data must be transferred to a memory unit for which the designated processor has access.
- certain image processing applications require that data, stored in main memory 102 or dedicated graphics memory 104 , be processed by the image processor 116 .
- Image processing is defined as any function(s) that apply to two dimensional blocks of pixels.
- pixels may be in the format of file system images, fields, or frames of video entering the prior art computer system 100 through video ports, mass storage devices such as CD-ROMs, fixed-disk subsystems and Local or Wide Area network ports.
- image processor 116 In order to enable image processor 116 to access data stored in main memory 102 or in dedicated graphics memory 104 , the data must be transferred or copied to dedicated image processor memory 105 .
- One problem with the prior art computer graphics system 100 is the cost of high performance peripheral dedicated memory systems such as the dedicated graphics memory unit 104 and dedicated image processor memory 105 . Another problem with the prior art computer graphics system 100 is the cost of high performance interconnects for multiple memory systems. Another problem with the prior art computer graphics system 100 is that the above discussed transfers of data between memory units require time and processing resources.
- the present invention pertains to a computer system providing dynamic memory allocation for graphics.
- the computer system includes a memory controller, a unified system memory, and memory clients each having access to the system memory via the memory controller.
- Memory clients can include a graphics rendering engine, a central processing unit (CPU), an image processor, a data compression/expansion device, an input/output device, and a graphics back end device.
- the rendering engine and the memory controller are implemented on a first integrated circuit (first IC) and the image processor and the data compression/expansion are implemented on a second IC.
- the computer system provides read/write access to the unified system memory, through the memory controller, for each of the memory clients.
- Translation hardware is included for mapping virtual addresses of pixel buffers to physical memory locations in the unified system memory. Pixel buffers are dynamically allocated as tiles of physically contiguous memory. Translation hardware, for mapping the virtual addresses of pixel buffers to physical memory locations in the unified system memory, is implemented in each of the computational devices which are included as memory clients in the computer system.
- the unified system memory is implemented using synchronous DRAM.
- tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of 128 pixels wherein each pixel is a 4 byte pixel.
- the present invention is also well suited to using tiles of other sizes.
- the dynamically allocated pixel buffers are comprised of n 2 tiles where n is an integer.
- the computer system of the present invention provides functional advantages for graphical display and image processing. There are no dedicated memory units in the computer system of the present invention aside from the unified system memory. Therefore, it is not necessary to transfer data from one dedicated memory unit to another when a peripheral processor is called upon to process data generated by the CPU or by another peripheral device.
- FIG. 1A is a circuit block diagram of a typical prior art computer system including peripheral processors and associated dedicated memory units.
- FIG. 2A is a circuit block diagram of an exemplary unified system memory computer architecture according to the present invention.
- FIG. 2B is an internal circuit block diagram of a graphics rendering and memory controller IC including a memory controller (MC)and a graphics rendering engine integrated therein.
- MC memory controller
- FIG. 3A is an illustration of an exemplary tile for dynamic allocation of pixel buffers according to the present invention.
- FIG. 3B is an illustration of an exemplary pixel buffer comprised of n 2 tiles according to the present invention.
- FIG. 3C is a block diagram of an address translation scheme according to the present invention.
- FIG. 4 is a block diagram of a memory controller according to the present invention.
- FIG. 5 is a timing diagram for memory client requests issued to the unified system memory according to the present invention.
- FIG. 6 is a timing diagram for memory client write data according to the present invention.
- FIG. 7 is a timing diagram for memory client read data according to the present invention.
- FIG. 8 is a timing diagram for an exemplary write to a new page performed by the unified system memory according to the present invention.
- FIG. 9 is a timing diagram for an exemplary read to a new page performed by the unified system memory according to the present invention.
- FIG. 10 shows external banks of the memory controller according to the present invention.
- FIG. 11 shows a flow diagram for bank state machines according to the present invention.
- Computer system 200 includes a unified system memory 204 which is shared by various memory system clients including a CPU 206 , a graphics rendering engine 208 , an input/output IC 210 , a graphics back end IC 212 , an image processor 214 , and a memory controller 204 .
- Computer system 201 includes the unified system memory 202 which is shared by various memory system clients including the CPU 206 , the input/output IC 210 , the graphics back end IC 212 , an image processing and compression and expansion IC 216 , and a graphics rendering and memory controller IC 218 .
- the image processing and compression and expansion IC 216 includes the image processor 214 , and a data compression and expansion unit 215 .
- GRMC IC 218 includes the graphics rendering engine (rendering engine) 208 and the memory controller 204 integrated therein.
- the graphics rendering and memory controller IC 218 is coupled to unified system memory 202 via a high bandwidth memory data bus (HBWMD BUS) 225 .
- HBWMD BUS 225 includes a demultiplexer (SD-MUX) 220 , a first BUS 222 coupled between the graphics rendering and memory controller IC 218 and SD-MUX 220 , and a second bus 224 coupled between SD-MUX 220 and unified system memory 202 .
- BUS 222 includes 144 lines cycled at 133 MHz and BUS 224 includes 288 lines cycled at 66 MHz.
- SD-MUX 220 demultiplexes the 144 lines of BUS 222 , which are cycled at 133 MHz, to double the number of lines, 288, of BUS 224 , which are cycled at half the frequency, 66 MHz.
- CPU 206 is coupled to the graphics rendering and memory controller IC 218 by a third bus 226 .
- BUS 226 is 64 bits wide and carries signals cycled at 100 MHz.
- the image processing and compression and expansion IC 216 is coupled to BUS 226 , by a third bus 228 .
- BUS 228 is 64 bits wide and carries signals cycled at 100 MHz.
- the graphics back end IC 212 is coupled to the graphics rendering and memory controller IC 218 by a fourth bus 230 .
- BUS 230 is 64 bits wide and carries signals cycled at 133 MHz.
- the input/output IC 210 is coupled to the graphics rendering and memory controller IC 218 by a fifth bus 232 .
- BUS 232 is 32 bits wide and carries signals cycled at 133 MHz.
- the input/output IC 210 of FIG. 2A contains all of the input/output interfaces including: keyboard & mouse, interval timers, serial, parallel, ic, audio, video in & out, and fast ethernet.
- the input/output IC 210 also contains an interface to an external 64-bit PCI expansion bus, BUS 231 , that supports five masters (two SCSI controllers and three expansion slots).
- FIG. 2B an internal circuit block diagram is shown of the graphics rendering and memory controller IC 218 according to an embodiment of the present invention.
- rendering engine 208 and memory controller 214 are integrated within the graphics rendering and memory controller IC 218 .
- the graphics rendering and memory controller IC 218 also includes a CPU/IPCE interface 234 , an input/output interface 236 , and a GBE interface 232 .
- GBE interface 232 buffers and transfers display data from unified system memory 202 to the graphics back end IC 212 in 16 ⁇ 32-byte bursts.
- GBE interface 232 buffers and transfers video capture data from the graphics back end IC 212 to unified system memory 202 in 16 ⁇ 32 -byte bursts.
- GBE interface 232 issues GBE interrupts to CPU/IPCE interface 234 .
- BUS 228 shown in both FIG. 2A and FIG. 2B, couples GBE interface 232 to the graphics back end IC 212 (FIG. 2A).
- the input/output interface 236 buffers and transfers data from unified system memory 202 to the input/output IC 210 in 8 ⁇ 32-byte bursts.
- the input/output interface 236 buffers and transfers data from the input/output IC 210 to unified system memory 202 in 8 ⁇ 32-byte bursts.
- the input/output interface 236 issues the input/output IC interrupts to CPU/IBCE interface 234 .
- BUS 230 shown in both FIG. 2A and FIG. 2B, couples the input/output interface 236 to the input/output IC 210 (FIG. 2A).
- a bus, BUS 224 provides coupling between CPU/IPCE interface 234 and CPU 206 and the image processing and compression and expansion IC 216 .
- the memory controller 214 is the interface between memory system clients (CPU 206 , rendering engine 208 , input/output IC 210 , graphics back end IC 212 , image processor 214 , and data compression/expansion device 215 ) and the unified system memory 202 .
- the memory controller 214 is coupled to unified system memory 202 via HBWMD BUS 225 which allows fast transfer of large amounts of data to and from unified system memory 202 .
- Memory clients make read and write requests to unified system memory 202 through the memory controller 214 .
- the memory controller 214 converts requests into the appropriate control sequences and passes data between memory clients and unified system memory 202 .
- the memory controller 214 contains two pipeline structures, one for commands and another for data.
- the request pipe has three stages, arbitration, decode and issue/state machine.
- the data pipe has only one stage, ECC. Requests and data flow through the pipes in the following manner. Clients place their requests in a queue.
- the arbitration logic looks at all of the requests at the top of the client queues and decides which request to start through the pipe. From the arbitration stage, the request flows to the decode stage. During the decode stage, information about the request is collected and passed onto an issue/state machine stage.
- the rendering engine 208 is a 2-D and 3-D graphics coprocessor which can accelerate rasterization.
- the rendering engine 208 is also cycled at 66 MHz and operates synchronously to the unified system memory 202 .
- the rendering engine 208 receives rendering parameters from the CPU 206 and renders directly to frame buffers stored in the unified system memory 202 (FIG. 2A).
- the rendering engine 208 issues memory access requests to the memory controller 214 . Since the rendering engine 208 shares the unified system memory 202 with other memory clients, the performance of the rendering engine 208 will vary as a function of the load on the unified system memory 202 .
- the rendering engine 208 is logically partitioned into four major functional units: a host interface, a pixel pipeline, a memory transfer engine, and a memory request unit.
- the host interface controls reading and writing from the host to programming interface registers.
- the pixel pipeline implements a rasterization and rendering pipeline to a frame buffer.
- the memory transfer engine performs memory bandwidth byte aligned clears and copies on both linear buffers and frame buffers.
- the memory request unit arbitrates between requests from the pixel pipeline and queues up memory requests to be issued to the memory controller 214 .
- the computer system 200 includes dynamic memory allocation of virtual pixel buffers in the unified system memory 202 .
- Pixel buffers include frame buffers, texture maps, video maps, image buffers, etc.
- Each pixel buffer can include multiple color buffers, a depth buffer, and a stencil buffer.
- pixel buffers are allocated in units of contiguous memory called tiles and address translation buffers are provided for dynamic allocation of pixel buffers.
- each tile 300 includes 64 kilobytes of physically contiguous memory.
- a 64 kilobyte tile size can be comprised of 128 ⁇ 128 pixels for 32 bit pixels, 256 ⁇ 128 pixels for 16 bit pixels, or 512 ⁇ 128 pixels for 8 bit pixels.
- tiles begin on 64 kilobyte aligned addresses.
- An integer number of tiles can be allocated for each pixel buffer. For example, a 200 ⁇ 200 pixel buffer and a 256 ⁇ 256 pixel buffer would both require four (128 ⁇ 128) pixel tiles.
- FIG. 3B an illustration is shown of an exemplary pixel buffer 302 according to the present invention.
- translation hardware maps virtual addresses of pixel buffers 302 to physical memory locations in unified system memory 202 .
- Each of the computational units of the computer system 200 (image processing and compression and expansion IC, 212 , graphics back end IC 212 , The input/output IC 210 , and rendering engine 208 ) includes translation hardware for mapping virtual addresses of pixel buffers 302 to physical memory locations in unified system memory 202 .
- the rendering engine 208 supports a frame buffer address translation buffer (TLB) to translate frame buffer (x,y) addresses into physical memory addresses.
- TLB frame buffer address translation buffer
- This TLB is loaded by CPU 206 with the base physical memory addresses of the tiles which compose a color buffer and the stencil-depth buffer of a frame buffer.
- the frame buffer TLB has enough entries to hold the tile base physical memory addresses of a 2048 ⁇ 2048 pixel color buffer and a 2048 ⁇ 2048 pixel stencil-depth buffer. Therefore, the TLB has 256 entries for color buffer tiles and 256 entries for stencil-depth buffer tiles.
- Tiles provide a convenient unit for memory allocation. By allowing tiles to be scattered throughout memory, tiling makes the amount of memory which must be contiguously allocated manageable. Additionally, tiling provides a means of reducing the amount of system memory consumed by frame buffers. Rendering to tiles which do not contain any pixels pertinent for display, invisible tiles, can be easily clipped out and hence no memory needs to be allocated for these tiles. For example, a 1024 ⁇ 1024 virtual frame buffer consisting of front and back RGBA buffers and a depth buffer would consume 12 Mb of memory if fully resident. However, if each 1024 ⁇ 1024 buffer were partitioned into 64 (128 ⁇ 128) tiles of which only four tiles contained non-occluded pixels, only memory for those visible tiles would need to be allocated. In this case, only 3MB would be consumed.
- memory system clients e.g., CPU 206 , rendering engine 208 , input/output IC 210 , graphics back end IC 212 , image processor 214 , and data compression/expansion device 215 .
- memory system clients e.g., CPU 206 , rendering engine 208 , input/output IC 210 , graphics back end IC 212 , image processor 214 , and data compression/expansion device 215 .
- each memory system client has access to memory shared by each of the other memory system clients, there is no need for transferring data from one dedicated memory unit to another.
- data can be received by the input/output IC 210 , decompressed (or expanded) by the data compression/expansion device 215 , and stored in the unified system memory 202 .
- This data can then be accessed by the CPU 206 , the rendering engine 208 , the input/output IC 210 , the graphics back end IC 212 , or the image processor 214 .
- the CPU 206 , the rendering engine 208 , the input/output IC 210 , the graphics back end IC 212 , or the image processor 214 can use data generated by the CPU 206 , the rendering engine 208 , the input/output IC 210 , the graphics back end IC 212 , or the image processor 214 .
- Each of the computational units (CPU 206 , input/output IC 210 , the graphics back end IC 212 , the image processing and compression and expansion IC 216 , the graphics rendering and memory controller IC 218 , and the data compression/expansion device 215 ) has translation hardware for determining the physical addresses of pixel buffers as is discussed below.
- input/output IC 210 can bring in a compressed stream of video data which can be stored into unified system memory 202 .
- the input/output IC 210 can access the compressed data stored in unified system memory 220 , via a path through the graphics rendering and memory controller IC 218 .
- the input/output IC 210 can then decompress the accessed data and store the decompressed data into unified system memory 202 .
- the stored image data can then be used, for example, as a texture map by rendering engine 208 for mapping the stored image onto another image.
- the resultant image can then be stored into a pixel buffer which has been allocated dynamically in unified system memory 202 . If the resultant image is stored into a frame buffer, allocated dynamically in unified system memory 202 , then the resultant image can be displayed by the graphics back end IC 212 or the image can be captured by writing the image back to another pixel buffer which has been allocated dynamically in unified system memory 202 . Since there is no necessity of transferring data from one dedicated memory unit to another in computer system 200 , functionality is increased.
- unified system memory 202 of FIG. 2A is implemented using synchronous DRAM (SDRAM) accessed via a 256-bit wide memory data bus cycled at 66 MHz.
- SDRAM synchronous DRAM
- a SDRAM is made up of rows and columns of memory cells.
- a row of memory cells is referred to as a page.
- a memory cell is accessed with a row address and column address.
- unified system memory 202 provides a peak data bandwidth of 2.133 Gb/s.
- unified system memory 202 is made up of 8 slots. Each slot can hold one SDRAM DIMM.
- a SDRAM DIMM is constructed from 1M ⁇ 16 or 4M ⁇ 16 SDRAM components and populated on the front only or the front and back side of the DIMM. Two DIMMs are required to make an external SDRAM bank. 1M ⁇ 16 SDRAM components construct a 32 Mbyte external bank, while 4M ⁇ 16 SDRAM components construct a 128 Mbyte external bank.
- unified system memory 202 can range in size from 32 Mbytes to 1 Gbyte.
- FIG. 3C shows a block diagram of an address translation scheme according to the present invention.
- FIG. 4 shows a block diagram of the memory controller 204 of the present invention.
- a memory client interface contains the signals listed in Table 1, below. TABLE 1 Memory client interface signals CRIME # of Signal Pin Name Bits Dir. Description clientreq.cmd internal 3 in type of request - only 1 - read 2 - write 4 - rmw clientreq.adr internal 25 in address of request only clientreq.msg internal 7 in message sent with request only clientreq.v internal 1 in 1 - valid only 0 - not valid clientreq.ecc internal 1 in 1 - ecc is valid only 0 - ecc not valid clientres.gnt internal 1 out 1 - room in client queue only 0 - no room clientres.wrrdy internal 1 out 1 - MC is ready for write data only 0 - MC not ready for write data clientres.rdrdy internal 1 out 1 - valid read data only 0 - not valid read data clientres.oe internal 1 out 1 - enable client driver only 0 - disable client driver client
- a memory client makes a request to the memory controller 204 by asserting clientreq.valid while setting the clientreq.adr, clientreq.msg, clientreq.cmd and clientreq.ecc lines to the appropriate values. If there is room in the queue, the request is latched into the memory client queue. Only two of the memory clients, the rendering engine 208 and the input/output IC 210 , use clientreq.msg. The message specifies which subsystem within the input/output IC 210 or the rendering engine 208 made the request. When an error occurs, this message is saved along with other pertinent information to aid in the debug process.
- the message is passed through the request pipe and returned with other pertinent information to aid in the debug process.
- the message is passed through the request pipe and returned with the clientreq.wrrdy signal for a write request or with the clientreq.rdrdy signal for a read request.
- the rendering engine 208 uses the information contained in the message to determine which rendering engine 208 queue to access.
- FIG. 6 a timing diagram for memory client write data is shown.
- the data for a write request is not latched with the address and request. Instead, the data, mask and message are latched when the memory controller 204 asserts the clientreq.wrrdy indicating that the request has reached the decode stage of the request pipe. Because the memory client queues are in front of the request pipe, there is not a simple relationship between the assertion clientreq.gnt and clientreq. wrrdy. Clientreq.msg is only valid for the rendering engine 208 and the input/output IC 210 .
- the memory controller 204 asserts the clientreq.oe signal at least one cycle before the assertion of clientreq.wrrdy. Clientreq.oe is latched locally by the memory client and is used to turn on the memory client's memory data bus drivers.
- FIG. 7 a timing diagram for memory client read data is shown.
- the read data is sent to the memory client over the memdata2client_out bus.
- clientres.rdrdy When clientres.rdrdy is asserted, the data and message are valid.
- a memory client interface contains the signals listed in Table 2, below. TABLE 2 Memory Interface Signals Crime Pin # of Signal Name Bits Dir. Description memwrite mem_dir 1 out controls direction of SDMUX chips- default to write memdata2- mem_data 256 out memory data from mem_out client going to unified system memory memdata2- internal only 256 out memory data from client_out main memory going to the memory client memmask_out mem_mask 32 out memory mask from client going to unified system memory memdataoe internal only 3 out enable memory data bus drivers ecc_out mem_ecc 32 out ecc going to unified system memory eccmask mem_eccmask 32 out ecc mask going to main memory mem_addr mem_addr 14 out memory address ras_n mem_ras_n 1 out row address strobe cas_n mem_cas_n 1 out column address strobe we_n mem_we_n 1 out write enable cs
- the data and mask are latched in the data pipe and flow out to the unified system memory 202 on memmask_out and memdata2mem_out.
- the ECC and ECC mask are generated and sent to the unified system memory 202 across eccmask and ecc_out.
- the memdataoe signal is used to turn on the memory bus drivers.
- Data and ECC from the unified system memory 202 come in on the memdata2client_in and ecc_in busses.
- the ECC is used to determine if the incoming data is correct. If there is a one bit error in the data, the error is corrected, and the corrected data is sent to the memory client. If there is more than one bit in error, the CPU 206 is interrupted, and incorrect data is returned to the memory client.
- Ras_n, cas_n, we_n and cs_n are control signals for the unified system memory 202 .
- a timing diagram is shown for an exemplary write to a new page performed by the unified system memory 202 .
- FIG. 9 a timing diagram is shown for an exemplary read to a new page performed by the unified system memory 202 .
- a read or write operation to the same SDRAM page is the same as the operation shown in FIGS. 8 and 9, except a same page operation does not need the precharge and activate cycles.
- a request pipe is the control center for the memory controller 204 .
- Memory client requests are placed in one end of the pipe and come out the other side as memory commands.
- the memory client queues are at the front of the pipe, followed by the arbitration, then the decode, and finally the issue/state machine. If there is room in their queue, a memory client can place a request in it.
- the arbitration logic looks at all of the requests at the top of the memory client queues and decides which request to start through the request pipe. From the arbitration stage, the request flows to the decode stage. During the decode stage, information about the request is collected and passed onto the issue/state machine stage. Based on this information, a state machine determines the proper sequence of commands for the unified system memory 202 .
- the later portion of the issue stage decodes the state of the state machine into control signals that are latched and then sent across to the unified system memory 202 .
- a request can sit in the issue stage for more than one cycle. While a request sits in the issue/state machine stage, the rest of the request pipe is stalled. Each stage of the request pipe is discussed herein.
- All of the memory clients have queues, except for refresh.
- a refresh request is guaranteed to retire before another request is issued, so a queue is not necessary.
- the five memory client queues are simple two-port structures with the memory clients on the write side and the arbitration logic on the read side. If there is space available in a memory client queue, indicated by the assertion of clientres.gnt, a memory client can place a request into its queue.
- a memory client request consists of an address, a command (read, write or read-modify-write), a message, an ECC valid and a request valid indication. If both clientreq.valid and clientres.gnt are asserted, the request is latched into the memory client queue.
- the arbitration logic looks at all of the requests at the top of the memory client queues and determines which request to pop off and pass to the decode stage of the requests at the top of the memory client queues and determines which request to pop off and pass to the decode stage of the request pipe.
- the clientres.gnt signal does not indicate that the request has retired. The request still needs to go through the arbitration process. To put it another way, memory client A might receive the clientres.gnt signal before memory client B, but if memory client B has a higher priority, its request might retire before the request from memory client A.
- the arbiter determines which memory client request to pass to the decode stage of the request pipe. This decision process has two steps. The first step is to determine if the arbitration slot for the current memory client is over or not. An arbitration slot is series of requests from the same memory client. The number and type of requests allowed in one arbitration slot varies. Table 3, below, lists what each memory client can do in an arbitration slot.
- the arbiter determines if the arbitration slot should end or not. If not, the request from the memory client who owns the current arbitration slot is passed to the decode stage. If the current arbitration slot is terminated, the arbiter uses the results from an arbitration algorithm to decide which request to pass to the decode stage. The arbitration algorithm to decide which request to pass to the decode stage.
- the arbitration algorithm ensures that the graphics back end IC 212 gets 1 ⁇ 2 of the arbitration slots, the input/output IC 210 gets 1 ⁇ 4, the image processing and compression and expansion IC 216 gets 1 ⁇ 8, the rendering engine 208 gets ⁇ fraction (1/16) ⁇ , the CPU 206 gets ⁇ fraction (1/32) ⁇ , and the refresh gets ⁇ fraction (1/64) ⁇ .
- the first step is to determine the maximum number of cycles that each memory client can use during an arbitration slot.
- Table 4, below shows the number of cycles associated with each type of operation. With reference to Table 4, below, “P” refers to precharge, “X” refers to a dead cycle, “A” refers to activate, “R0” refers to “read word 0”, “W0” refers to “write word 0”, and “Ref” refers to “refresh”.
- Table 5 refers to the maximum number of cycles for each of the memory clients. TABLE 5 Maximum # Cycles per Slot Memory Client Operation # of cycles Graphics Back End 16 memory word read or write 20 cycles CPU, Rendering Engine, 8 memory word read or write 12 cycles MACE IPCE 8 memory word read or write 18 cycles 1 page crossings REFRESH refresh 2 rows 14 cycles
- the decode logic receives requests from the arbiter. Based on state maintained from the previous requests and information contained in the current request, the decode logic determines which memory bank to select, which of the four state machines in the next stage will handle the request, and whether or not the current request is on the same page as the previous request. This information is passed to the issue/state machine stage.
- the unified system memory 202 is made up of 8 slots. Each slot can hold one SDRAM DIMM.
- a SDRAM DIMM is constructed from 1M ⁇ 16 or 4M ⁇ 16 SDRAM components and populated on the front only or the front and back side of the DIMM. Two DIMMs are required to make an external SDRAM bank. 1M ⁇ 16 SDRAM components construct a 32 Mbyte external bank, while 4M ⁇ 16 SDRAM components construct a 128 Mbyte external bank.
- the memory system can range in size from 32 Mbytes to 1 Gbyte.
- Each SDRAM component has two internal banks, hence two possible open pages.
- the maximum number of external banks is 8 and the maximum number of internal banks is 8 and the maximum number of internal banks is 16.
- the memory controller 204 only supports 4 open pages at a time. This issue will be discussed in detail later in this section.
- the decode logic is explained below in more detail.
- software probes the memory to determine how many banks of memory are present and the size of each bank. Based on this information, the software programs the 8 bank control registers.
- Each bank control register (please refer to the register section) has one bit that indicates the size of the bank and 5 bits for the upper address bits of that bank.
- Software must place the 64 Mbit external banks in the lower address range followed by any 16 Mbit external banks. This is to prevent gaps in the memory.
- the decode logic compares the upper address bits of the incoming request to the 8 bank control registers to determine which external bank to select. The number of bits that are compared is dependent on the size of the bank.
- the decode logic compares bits 24:22 of the request address to bits 4:2 of the bank control register. If there is a match, that bank is selected. Each external bank has a separate chip select. If an incoming address matches more than one bank's control register, the bank with the lowest number is selected. If an incoming address does not match any of the bank control registers, a memory address error occurs. When an error occurs, pertinent information about the request is captured in error registers and the processor is interrupted—if the memory controller 204 interrupt is enabled.
- the request that caused the error is still sent to the next stage in the pipeline and is processed like a normal request, but the memory controller 204 deasserts all of the external bank selects so that the memory operation doesn't actually occur. Deasserting the external bank selects is also done when bit 6 of the rendering engine 208 message is set. The rendering engine 208 sets this bit when a request is generated using an invalid TLB entry.
- the memory controller 204 can handle any physical external bank configuration, we recommend that external bank 0 always be filled and that the external banks be placed in decreasing density order (for example a 64 Mbit external bank in bank 0 and a 16 Mbit external bank in bank 2 ).
- the previous paragraph describes how the decode logic determines what external bank to select. This paragraph describes the method for determining page crossings and which bank state machine to use in the next stage of the pipeline.
- the row address, along with the internal and external bank bits for previous requests, are kept in a set of registers which are referred to as the row registers.
- Each row register corresponds to a bank state machine.
- the decode logic compares the internal/external bank bits of the new request with the four row registers. If there is a match, then bank state machine corresponding to that row register is selected.
- the decode logic passes the request along with the external bank selects, state machine select and same page information to the issue/state machine stage.
- the selected bank state machine sequences through the proper states, while the issue logic decodes the state of the bank state machine into commands that are sent to the SDRAM DIMMS.
- the initialization/refresh state machine sequences through special states for initialization and refresh while the four bank state machines are forced to an idle state.
- the bank state machines and the initialization/refresh state machine are discussed in more detail in more detail in the following sections.
- the four bank state machines operate independently, subject only to conflicts for access to the control, address, and data signals.
- the bank state machines default to page mode operation. That is, the autoprecharge commands are not used, and the SDRAM bank must be explicitly precharged whenever there is a non-page-mode random reference.
- the decode state passes the request along with the page information to the selected state machine which sequences through the proper states. At certain states, interval timers are stated that inhibit the state machine from advancing to the next state until the SDRAM minimum interval requirements have been met.
- the bank state machines operate on one request at a time. That is, a request sequences through any required precharge and activation phases and then a read or write phase, at which point it is considered completed and the next request initiated. Finally, the state of the four bank state machines is decoded by the issue logic that generates the SDRAM control signals.
- Tr2rp and Tr2w are additional timing parameters that explicitly define the interval between successive read, write, and precharge commands. These parameters insure that successive commands do not cause conflicts on the data signals. While these parameters could be derived internally by a state machine sequencer, they are made explicit to simplify the state machines and use the same timer paradigm as the SDRAM parameters.
- Trc 7 Activate bank A to Activate bank A Tras 5 Activate bank A to Precharge bank A Trp 2 Precharge bank A to Activate bank A Trrd 2 Activate bank A to Activate bank B Trcd 2 Activate bank A to Read bank A Twp 1 Datain bank A to Precharge bank A Tr2rp 2 Read bank A to Read or Precharge bank C Tr2w 6 Read bank A to Write bank A
- Banks A and B are in the same external bank while Bank C is in a different external bank.
- Trp, Trrd and Trcd are enforced by design.
- the Trc, Tras, Tr2rp and Tr2w parameters have a timer for each of the four bank state machines.
- the Tr2rp and Tr2w timers are common to all of the four bank state machines, because they are used to prevent conflicts on the shared data lines.
- the initialization/refresh state machine has two functions, initialization and refresh.
- the initialization procedure is discussed first, followed by the refresh.
- the initialization/refresh state machine sequences through the SDRAM initialization procedure, which is a precharge to all banks, followed by a mode set.
- the issue stage decodes the state of the initialization/refresh state machine into commands that are sent the SDRAM.
- the mode set command programs the SDRAM mode set register to a CAS latency of 2, burst length of 1 and a sequential operation type.
- the SDRAM requires that 4096 refresh cycles occur every 64 ms.
- the timer sends out a signal every 27 microseconds which causes the refresh memory client to make a request to the arbiter.
- the arbiter treats refresh just like all the other memory clients.
- the arbiter determines that the time for the refresh slot has come, the arbiter passes the refresh request to the decode stage.
- the decode stage invalidates all of the row registers and passes the request onto the state machine/issue stage.
- a bank state machine sees that it is a refresh request, it goes to its idle state.
- the initialization/refresh state machine sequences through the refresh procedure which is a precharge to all banks followed by two refresh cycles.
- a refresh command puts the SDRAM in the automatic refresh mode.
- An address counter internal to the device, increments the word and bank address during the refresh cycle.
- the SDRAM is in the idle state, which means that all the pages are closed. This is why it is important that the bank state machines are forced to the idle state and the row registers are invalidated during a refresh request.
- the initialization/refresh state machine is very similar in structure to the bank state machines and has timers to enforce SDRAM parameters.
- a Trc timer is used to enforce the Trc requirement between refresh cycles, and the outputs from the bank Tras timers are used to ensure that the “precharge all” command does not violate Tras for any of the active banks.
- the main functions of the data pipe are to: (1) move data between a memory client and the unified system memory 202 , (2) perform ECC operations and (3) merge new byte from a memory client with old data from memory during a read-modify-write operation. Each of these functions is described below.
- the data pipe has one stage which is in lock-step with the last stage of the request pipe.
- the request pipe asserts clientres.wrrdy.
- the clientres.wrrdy signal indicates to the memory client that the data on the Memdata2mem_in bus has been latched into the ECC stage of the data pipe.
- the data is held in the ECC stage and flows out to the unified system memory 202 until the request is retired in the request pipe.
- Incoming read data is latched in the data pipe, flows through the ECC correction logic and then is latched again before going on the Memdata2client_out bus.
- the request pipe knows how many cycles the unified system memory 202 takes to return read responses data.
- the request pipe asserts clientres.rdrdy.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Image Input (AREA)
- Image Processing (AREA)
- Image Generation (AREA)
- Controls And Circuits For Display Device (AREA)
- Digital Computer Display Output (AREA)
- Memory System (AREA)
- Memory System Of A Hierarchy Structure (AREA)
Abstract
A computer systemprovides dynamic memory allocation for graphics. The computer system includes a memory controller, a unified system memory, and memory clients each having access to the system memory via the memory controller. Memory clients can include a graphics rendering engine, a CPU, an image processor, a data compression/expansion device, an input/output device, a graphics back end device. The computer system provides read/write access to the unified system memory, through the memory controller, for each of the memory clients. Translation hardware is included for mapping virtual addresses of pixel buffers to physical memory locations in the unified system memory. Pixel buffers are dynamically allocated as tiles of physically contiguous memory. Translation hardware is implemented in each of the computational devices, which are included as memory clients in the computer system, including primarily the rendering engine.
Description
- The present invention relates to the field of computer systems. Specifically, the present invention relates to a computer system architecture including dynamic memory allocation of pixel buffers for graphics and image processing.
- Typical prior art computer systems often rely on peripheral processors and dedicated peripheral memory units to perform various computational operations. For example, peripheral graphics display processors are used to render graphics images (synthesis) and peripheral image processors are used to perform image processing (analysis). In typical prior art computer systems, CPU main memory is separate from peripheral memory units which can be dedicated to graphics rendering or image processing or other computational functions.
- With reference to Prior Art FIG. 1, a prior art
computer graphics system 100 is shown. The prior artcomputer graphics system 100 includes three separate memory units; amain memory 102, adedicated graphics memory 104, and a dedicated image processing memory (image processor memory) 105.Main memory 102 provides fast access to data for aCPU 106 and an input/output device 108. TheCPU 106 and input/output device 108 are connected tomain memory 102 via amain memory controller 110. Dedicatedgraphics memory 104 provides fast access to graphics data for agraphics processor 112 via agraphics memory controller 114. Dedicatedimage processor memory 105 provides fast access to buffers of data used by animage processor 116 via an imageprocessor memory controller 118. In the prior artcomputer graphics system 100,CPU 106 has read/write access tomain memory 102 but not to dedicatedgraphics memory 104 or dedicatedimage processor memory 105. Likewise, theimage processor 116 has read/write access to dedicatedimage processor memory 105, but not tomain memory 102 ordedicated graphics memory 104. Similarly,graphics processor 112 has read/write access todedicated graphics memory 104 but not tomain memory 102 or dedicatedimage processor memory 105. - Certain computer system applications require that data, stored in
main memory 102 or in one of thededicated memory units main memory 102 ordedicated graphics memory 104, be processed by theimage processor 116. Image processing is defined as any function(s) that apply to two dimensional blocks of pixels. These pixels may be in the format of file system images, fields, or frames of video entering the priorart computer system 100 through video ports, mass storage devices such as CD-ROMs, fixed-disk subsystems and Local or Wide Area network ports. In order to enableimage processor 116 to access data stored inmain memory 102 or indedicated graphics memory 104, the data must be transferred or copied to dedicatedimage processor memory 105. - One problem with the prior art
computer graphics system 100 is the cost of high performance peripheral dedicated memory systems such as the dedicatedgraphics memory unit 104 and dedicatedimage processor memory 105. Another problem with the prior artcomputer graphics system 100 is the cost of high performance interconnects for multiple memory systems. Another problem with the prior artcomputer graphics system 100 is that the above discussed transfers of data between memory units require time and processing resources. - Thus, what is needed is a computer system architecture with a single unified memory system which can be shared by multiple processors in the computer system without transferring data between multiple dedicated memory units.
- The present invention pertains to a computer system providing dynamic memory allocation for graphics. The computer system includes a memory controller, a unified system memory, and memory clients each having access to the system memory via the memory controller. Memory clients can include a graphics rendering engine, a central processing unit (CPU), an image processor, a data compression/expansion device, an input/output device, and a graphics back end device. In a preferred embodiment, the rendering engine and the memory controller are implemented on a first integrated circuit (first IC) and the image processor and the data compression/expansion are implemented on a second IC. The computer system provides read/write access to the unified system memory, through the memory controller, for each of the memory clients. Translation hardware is included for mapping virtual addresses of pixel buffers to physical memory locations in the unified system memory. Pixel buffers are dynamically allocated as tiles of physically contiguous memory. Translation hardware, for mapping the virtual addresses of pixel buffers to physical memory locations in the unified system memory, is implemented in each of the computational devices which are included as memory clients in the computer system.
- In a preferred embodiment, the unified system memory is implemented using synchronous DRAM. Also in the preferred embodiment, tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of 128 pixels wherein each pixel is a 4 byte pixel. However, the present invention is also well suited to using tiles of other sizes. Also in the preferred embodiment, the dynamically allocated pixel buffers are comprised of n2 tiles where n is an integer.
- The computer system of the present invention provides functional advantages for graphical display and image processing. There are no dedicated memory units in the computer system of the present invention aside from the unified system memory. Therefore, it is not necessary to transfer data from one dedicated memory unit to another when a peripheral processor is called upon to process data generated by the CPU or by another peripheral device.
- The present invention is illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which like reference numerals refer to similar elements and in which:
- Prior Art FIG. 1A is a circuit block diagram of a typical prior art computer system including peripheral processors and associated dedicated memory units.
- FIG. 2A is a circuit block diagram of an exemplary unified system memory computer architecture according to the present invention.
- FIG. 2B is an internal circuit block diagram of a graphics rendering and memory controller IC including a memory controller (MC)and a graphics rendering engine integrated therein.
- FIG. 3A is an illustration of an exemplary tile for dynamic allocation of pixel buffers according to the present invention.
- FIG. 3B is an illustration of an exemplary pixel buffer comprised of n2 tiles according to the present invention.
- FIG. 3C is a block diagram of an address translation scheme according to the present invention.
- FIG. 4 is a block diagram of a memory controller according to the present invention.
- FIG. 5 is a timing diagram for memory client requests issued to the unified system memory according to the present invention.
- FIG. 6 is a timing diagram for memory client write data according to the present invention.
- FIG. 7 is a timing diagram for memory client read data according to the present invention.
- FIG. 8 is a timing diagram for an exemplary write to a new page performed by the unified system memory according to the present invention.
- FIG. 9 is a timing diagram for an exemplary read to a new page performed by the unified system memory according to the present invention.
- FIG. 10 shows external banks of the memory controller according to the present invention.
- FIG. 11 shows a flow diagram for bank state machines according to the present invention.
- In the following detailed description of the present invention, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be obvious to one skilled in the art that the present invention may be practiced without these specific details. In other instances well known methods, procedures, components, and circuits have not been described in detail as not to unnecessarily obscure aspects of the present invention.
- Reference will now be made in detail to the preferred embodiments of the present invention, a computer system architecture having dynamic memory allocation for graphics, examples of which are illustrated in the accompanying drawings. While the invention will be described in conjunction with the preferred embodiments, it will be understood that they are not intended to limit the invention to these embodiments. On the contrary, the invention is intended to cover alternatives, modifications and equivalents, which may be included within the spirit and scope of the invention as defined by the appended claims. Furthermore, in the following detailed description of the present invention, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, it will be obvious to one of ordinary skill in the art that the present invention may be practiced without these specific details. In other instances, well known methods, procedures, components, and circuits have not been described in detail as not to unnecessarily obscure aspects of the present invention.
- With reference to FIG. 2A, a computer system200, according to the present invention, is shown. Computer system 200 includes a
unified system memory 204 which is shared by various memory system clients including aCPU 206, agraphics rendering engine 208, an input/output IC 210, a graphics backend IC 212, animage processor 214, and amemory controller 204. - With reference to FIG. 2B, an
exemplary computer system 201, according to the present invention, is shown.Computer system 201 includes theunified system memory 202 which is shared by various memory system clients including theCPU 206, the input/output IC 210, the graphics backend IC 212, an image processing and compression andexpansion IC 216, and a graphics rendering andmemory controller IC 218. The image processing and compression andexpansion IC 216 includes theimage processor 214, and a data compression andexpansion unit 215.GRMC IC 218 includes the graphics rendering engine (rendering engine) 208 and thememory controller 204 integrated therein. The graphics rendering andmemory controller IC 218 is coupled tounified system memory 202 via a high bandwidth memory data bus (HBWMD BUS) 225. In a preferred embodiment of the present invention,HBWMD BUS 225 includes a demultiplexer (SD-MUX) 220, afirst BUS 222 coupled between the graphics rendering andmemory controller IC 218 and SD-MUX 220, and asecond bus 224 coupled between SD-MUX 220 andunified system memory 202. In the preferred embodiment of the present invention,BUS 222 includes 144 lines cycled at 133 MHz andBUS 224 includes 288 lines cycled at 66 MHz. SD-MUX 220 demultiplexes the 144 lines ofBUS 222, which are cycled at 133 MHz, to double the number of lines, 288, ofBUS 224, which are cycled at half the frequency, 66 MHz.CPU 206 is coupled to the graphics rendering andmemory controller IC 218 by athird bus 226. In the preferred embodiment of the present invention,BUS 226 is 64 bits wide and carries signals cycled at 100 MHz. The image processing and compression andexpansion IC 216 is coupled toBUS 226, by athird bus 228. In the preferred embodiment of the present invention,BUS 228 is 64 bits wide and carries signals cycled at 100 MHz. The graphics backend IC 212 is coupled to the graphics rendering andmemory controller IC 218 by afourth bus 230. In the preferred embodiment of the present invention,BUS 230 is 64 bits wide and carries signals cycled at 133 MHz. The input/output IC 210 is coupled to the graphics rendering andmemory controller IC 218 by afifth bus 232. In the preferred embodiment of the present invention,BUS 232 is 32 bits wide and carries signals cycled at 133 MHz. - The input/
output IC 210 of FIG. 2A contains all of the input/output interfaces including: keyboard & mouse, interval timers, serial, parallel, ic, audio, video in & out, and fast ethernet. The input/output IC 210 also contains an interface to an external 64-bit PCI expansion bus, BUS 231, that supports five masters (two SCSI controllers and three expansion slots). - With reference to FIG. 2B, an internal circuit block diagram is shown of the graphics rendering and
memory controller IC 218 according to an embodiment of the present invention. As previously mentioned,rendering engine 208 andmemory controller 214 are integrated within the graphics rendering andmemory controller IC 218. The graphics rendering andmemory controller IC 218 also includes a CPU/IPCE interface 234, an input/output interface 236, and aGBE interface 232. - With reference to FIGS. 2A and 2B,
GBE interface 232 buffers and transfers display data fromunified system memory 202 to the graphics backend IC 212 in 16×32-byte bursts.GBE interface 232 buffers and transfers video capture data from the graphics backend IC 212 tounified system memory 202 in 16×32 -byte bursts.GBE interface 232 issues GBE interrupts to CPU/IPCE interface 234.BUS 228, shown in both FIG. 2A and FIG. 2B, couplesGBE interface 232 to the graphics back end IC 212 (FIG. 2A). The input/output interface 236 buffers and transfers data fromunified system memory 202 to the input/output IC 210 in 8×32-byte bursts. The input/output interface 236 buffers and transfers data from the input/output IC 210 tounified system memory 202 in 8×32-byte bursts. The input/output interface 236 issues the input/output IC interrupts to CPU/IBCE interface 234.BUS 230, shown in both FIG. 2A and FIG. 2B, couples the input/output interface 236 to the input/output IC 210 (FIG. 2A). A bus,BUS 224, provides coupling between CPU/IPCE interface 234 andCPU 206 and the image processing and compression andexpansion IC 216. - With reference to FIG. 2A, the
memory controller 214 is the interface between memory system clients (CPU 206,rendering engine 208, input/output IC 210, graphics backend IC 212,image processor 214, and data compression/expansion device 215) and theunified system memory 202. As previously mentioned, thememory controller 214 is coupled tounified system memory 202 viaHBWMD BUS 225 which allows fast transfer of large amounts of data to and fromunified system memory 202. Memory clients make read and write requests tounified system memory 202 through thememory controller 214. Thememory controller 214 converts requests into the appropriate control sequences and passes data between memory clients andunified system memory 202. In the preferred embodiment of the present invention, thememory controller 214 contains two pipeline structures, one for commands and another for data. The request pipe has three stages, arbitration, decode and issue/state machine. The data pipe has only one stage, ECC. Requests and data flow through the pipes in the following manner. Clients place their requests in a queue. The arbitration logic looks at all of the requests at the top of the client queues and decides which request to start through the pipe. From the arbitration stage, the request flows to the decode stage. During the decode stage, information about the request is collected and passed onto an issue/state machine stage. - With reference to FIG. 2A, the
rendering engine 208 is a 2-D and 3-D graphics coprocessor which can accelerate rasterization. In a preferred embodiment of the present invention, therendering engine 208 is also cycled at 66 MHz and operates synchronously to theunified system memory 202. Therendering engine 208 receives rendering parameters from theCPU 206 and renders directly to frame buffers stored in the unified system memory 202 (FIG. 2A). Therendering engine 208 issues memory access requests to thememory controller 214. Since therendering engine 208 shares theunified system memory 202 with other memory clients, the performance of therendering engine 208 will vary as a function of the load on theunified system memory 202. Therendering engine 208 is logically partitioned into four major functional units: a host interface, a pixel pipeline, a memory transfer engine, and a memory request unit. The host interface controls reading and writing from the host to programming interface registers. The pixel pipeline implements a rasterization and rendering pipeline to a frame buffer. The memory transfer engine performs memory bandwidth byte aligned clears and copies on both linear buffers and frame buffers. The memory request unit arbitrates between requests from the pixel pipeline and queues up memory requests to be issued to thememory controller 214. - The computer system200 includes dynamic memory allocation of virtual pixel buffers in the
unified system memory 202. Pixel buffers include frame buffers, texture maps, video maps, image buffers, etc. Each pixel buffer can include multiple color buffers, a depth buffer, and a stencil buffer. In the present invention, pixel buffers are allocated in units of contiguous memory called tiles and address translation buffers are provided for dynamic allocation of pixel buffers. - With reference to FIG. 3A, an illustration is shown of an
exemplary tile 300 for dynamic allocation of pixel buffers according to the present invention. In a preferred embodiment of the present invention, eachtile 300 includes 64 kilobytes of physically contiguous memory. A 64 kilobyte tile size can be comprised of 128×128 pixels for 32 bit pixels, 256×128 pixels for 16 bit pixels, or 512×128 pixels for 8 bit pixels. In the present invention, tiles begin on 64 kilobyte aligned addresses. An integer number of tiles can be allocated for each pixel buffer. For example, a 200×200 pixel buffer and a 256×256 pixel buffer would both require four (128×128) pixel tiles. - With reference to FIG. 3B, an illustration is shown of an
exemplary pixel buffer 302 according to the present invention. In the computer system 200 of the present invention, translation hardware maps virtual addresses ofpixel buffers 302 to physical memory locations inunified system memory 202. Each of the computational units of the computer system 200 (image processing and compression and expansion IC, 212, graphics backend IC 212, The input/output IC 210, and rendering engine 208) includes translation hardware for mapping virtual addresses ofpixel buffers 302 to physical memory locations inunified system memory 202. Eachpixel buffer 302 is partitioned into n2 tiles 300, where n is an integer. In a preferred embodiment of the present invention, n=4. - The
rendering engine 208 supports a frame buffer address translation buffer (TLB) to translate frame buffer (x,y) addresses into physical memory addresses. This TLB is loaded byCPU 206 with the base physical memory addresses of the tiles which compose a color buffer and the stencil-depth buffer of a frame buffer. In a preferred embodiment of the present invention, the frame buffer TLB has enough entries to hold the tile base physical memory addresses of a 2048×2048 pixel color buffer and a 2048×2048 pixel stencil-depth buffer. Therefore, the TLB has 256 entries for color buffer tiles and 256 entries for stencil-depth buffer tiles. - Tiles provide a convenient unit for memory allocation. By allowing tiles to be scattered throughout memory, tiling makes the amount of memory which must be contiguously allocated manageable. Additionally, tiling provides a means of reducing the amount of system memory consumed by frame buffers. Rendering to tiles which do not contain any pixels pertinent for display, invisible tiles, can be easily clipped out and hence no memory needs to be allocated for these tiles. For example, a 1024×1024 virtual frame buffer consisting of front and back RGBA buffers and a depth buffer would consume 12 Mb of memory if fully resident. However, if each 1024×1024 buffer were partitioned into 64 (128×128) tiles of which only four tiles contained non-occluded pixels, only memory for those visible tiles would need to be allocated. In this case, only 3MB would be consumed.
- In the present invention, memory system clients (e.g.,
CPU 206,rendering engine 208, input/output IC 210, graphics backend IC 212,image processor 214, and data compression/expansion device 215) share theunified system memory 202. Since each memory system client has access to memory shared by each of the other memory system clients, there is no need for transferring data from one dedicated memory unit to another. For example, data can be received by the input/output IC 210, decompressed (or expanded) by the data compression/expansion device 215, and stored in theunified system memory 202. This data can then be accessed by theCPU 206, therendering engine 208, the input/output IC 210, the graphics backend IC 212, or theimage processor 214. As a second example, theCPU 206, therendering engine 208, the input/output IC 210, the graphics backend IC 212, or theimage processor 214 can use data generated by theCPU 206, therendering engine 208, the input/output IC 210, the graphics backend IC 212, or theimage processor 214. Each of the computational units (CPU 206, input/output IC 210, the graphics backend IC 212, the image processing and compression andexpansion IC 216, the graphics rendering andmemory controller IC 218, and the data compression/expansion device 215) has translation hardware for determining the physical addresses of pixel buffers as is discussed below. - There are numerous video applications for which the present invention computer system200 provides functional advantages over prior art computer system architectures. These applications range from video conferencing to video editing. There is significant variation in the processing required for the various applications, but a few processing steps are common to all applications: capture, filtering, scaling, compression, blending, and display. In operation of computer system 200, input/
output IC 210 can bring in a compressed stream of video data which can be stored intounified system memory 202. The input/output IC 210 can access the compressed data stored inunified system memory 220, via a path through the graphics rendering andmemory controller IC 218. The input/output IC 210 can then decompress the accessed data and store the decompressed data intounified system memory 202. The stored image data can then be used, for example, as a texture map byrendering engine 208 for mapping the stored image onto another image. The resultant image can then be stored into a pixel buffer which has been allocated dynamically inunified system memory 202. If the resultant image is stored into a frame buffer, allocated dynamically inunified system memory 202, then the resultant image can be displayed by the graphics backend IC 212 or the image can be captured by writing the image back to another pixel buffer which has been allocated dynamically inunified system memory 202. Since there is no necessity of transferring data from one dedicated memory unit to another in computer system 200, functionality is increased. - In the preferred embodiment of the present invention,
unified system memory 202 of FIG. 2A is implemented using synchronous DRAM (SDRAM) accessed via a 256-bit wide memory data bus cycled at 66 MHz. A SDRAM is made up of rows and columns of memory cells. A row of memory cells is referred to as a page. A memory cell is accessed with a row address and column address. When a row is accessed, the entire row is placed into latches, so that subsequent accesses to that row only require the column address. Accesses to the same row are referred to as page accesses. In a preferred embodiment of the present invention,unified system memory 202 provides a peak data bandwidth of 2.133 Gb/s. Also, in a preferred embodiment of the present invention,unified system memory 202 is made up of 8 slots. Each slot can hold one SDRAM DIMM. A SDRAM DIMM is constructed from 1M×16 or 4M×16 SDRAM components and populated on the front only or the front and back side of the DIMM. Two DIMMs are required to make an external SDRAM bank. 1M×16 SDRAM components construct a 32 Mbyte external bank, while 4M×16 SDRAM components construct a 128 Mbyte external bank.unified system memory 202 can range in size from 32 Mbytes to 1 Gbyte. - FIG. 3C shows a block diagram of an address translation scheme according to the present invention. FIG. 4 shows a block diagram of the
memory controller 204 of the present invention. - A memory client interface contains the signals listed in Table 1, below.
TABLE 1 Memory client interface signals CRIME # of Signal Pin Name Bits Dir. Description clientreq.cmd internal 3 in type of request - only 1 - read 2 - write 4 - rmw clientreq.adr internal 25 in address of request only clientreq.msg internal 7 in message sent with request only clientreq.v internal 1 in 1 - valid only 0 - not valid clientreq.ecc internal 1 in 1 - ecc is valid only 0 - ecc not valid clientres.gnt internal 1 out 1 - room in client queue only 0 - no room clientres.wrrdy internal 1 out 1 - MC is ready for write data only 0 - MC not ready for write data clientres.rdrdy internal 1 out 1 - valid read data only 0 - not valid read data clientres.oe internal 1 out 1 - enable client driver only 0 - disable client driver clientres.rdmsg internal 7 out read message sent with read only data clientres.wrmsg internal 7 out write message sent with wrrdy only memdata2- internal 256 out memory data from client mem_in only going to unified system memory memmask_in internal 32 in memory mask from client only going to unified system memory 0 - write byte 1 - don't write byte memmask_in (0) is matched with memdata2mem_in (7:0) and so on. memdata2- internal 256 out memory data from unified client_out only system memory going to the client - With reference to FIG. 5, a timing diagram for memory client requests is shown. A memory client makes a request to the
memory controller 204 by asserting clientreq.valid while setting the clientreq.adr, clientreq.msg, clientreq.cmd and clientreq.ecc lines to the appropriate values. If there is room in the queue, the request is latched into the memory client queue. Only two of the memory clients, therendering engine 208 and the input/output IC 210, use clientreq.msg. The message specifies which subsystem within the input/output IC 210 or therendering engine 208 made the request. When an error occurs, this message is saved along with other pertinent information to aid in the debug process. For therendering engine 208, the message is passed through the request pipe and returned with other pertinent information to aid in the debug process. For therendering engine 208, the message is passed through the request pipe and returned with the clientreq.wrrdy signal for a write request or with the clientreq.rdrdy signal for a read request. Therendering engine 208 uses the information contained in the message to determine whichrendering engine 208 queue to access. - With reference to FIG. 6, a timing diagram for memory client write data is shown. The data for a write request is not latched with the address and request. Instead, the data, mask and message are latched when the
memory controller 204 asserts the clientreq.wrrdy indicating that the request has reached the decode stage of the request pipe. Because the memory client queues are in front of the request pipe, there is not a simple relationship between the assertion clientreq.gnt and clientreq. wrrdy. Clientreq.msg is only valid for therendering engine 208 and the input/output IC 210. Thememory controller 204 asserts the clientreq.oe signal at least one cycle before the assertion of clientreq.wrrdy. Clientreq.oe is latched locally by the memory client and is used to turn on the memory client's memory data bus drivers. - With reference to FIG. 7, a timing diagram for memory client read data is shown. The read data is sent to the memory client over the memdata2client_out bus. When clientres.rdrdy is asserted, the data and message are valid.
- A memory client interface contains the signals listed in Table 2, below.
TABLE 2 Memory Interface Signals Crime Pin # of Signal Name Bits Dir. Description memwrite mem_dir 1 out controls direction of SDMUX chips- default to write memdata2- mem_data 256 out memory data from mem_out client going to unified system memory memdata2- internal only 256 out memory data from client_out main memory going to the memory client memmask_out mem_mask 32 out memory mask from client going to unified system memory memdataoe internal only 3 out enable memory data bus drivers ecc_out mem_ecc 32 out ecc going to unified system memory eccmask mem_eccmask 32 out ecc mask going to main memory mem_addr mem_addr 14 out memory address ras_n mem_ras_n 1 out row address strobe cas_n mem_cas_n 1 out column address strobe we_n mem_we_n 1 out write enable cs_n mem_cs(3:0)_n 8 out chip selects - The data and mask are latched in the data pipe and flow out to the
unified system memory 202 on memmask_out and memdata2mem_out. From the data and mask, the ECC and ECC mask are generated and sent to theunified system memory 202 across eccmask and ecc_out. The memdataoe signal is used to turn on the memory bus drivers. Data and ECC from theunified system memory 202 come in on the memdata2client_in and ecc_in busses. The ECC is used to determine if the incoming data is correct. If there is a one bit error in the data, the error is corrected, and the corrected data is sent to the memory client. If there is more than one bit in error, theCPU 206 is interrupted, and incorrect data is returned to the memory client. Ras_n, cas_n, we_n and cs_n are control signals for theunified system memory 202. - With reference to FIG. 8, a timing diagram is shown for an exemplary write to a new page performed by the
unified system memory 202. With reference to FIG. 9, a timing diagram is shown for an exemplary read to a new page performed by theunified system memory 202. A read or write operation to the same SDRAM page is the same as the operation shown in FIGS. 8 and 9, except a same page operation does not need the precharge and activate cycles. - A request pipe is the control center for the
memory controller 204. Memory client requests are placed in one end of the pipe and come out the other side as memory commands. The memory client queues are at the front of the pipe, followed by the arbitration, then the decode, and finally the issue/state machine. If there is room in their queue, a memory client can place a request in it. The arbitration logic looks at all of the requests at the top of the memory client queues and decides which request to start through the request pipe. From the arbitration stage, the request flows to the decode stage. During the decode stage, information about the request is collected and passed onto the issue/state machine stage. Based on this information, a state machine determines the proper sequence of commands for theunified system memory 202. The later portion of the issue stage decodes the state of the state machine into control signals that are latched and then sent across to theunified system memory 202. A request can sit in the issue stage for more than one cycle. While a request sits in the issue/state machine stage, the rest of the request pipe is stalled. Each stage of the request pipe is discussed herein. - All of the memory clients have queues, except for refresh. A refresh request is guaranteed to retire before another request is issued, so a queue is not necessary. The five memory client queues are simple two-port structures with the memory clients on the write side and the arbitration logic on the read side. If there is space available in a memory client queue, indicated by the assertion of clientres.gnt, a memory client can place a request into its queue. A memory client request consists of an address, a command (read, write or read-modify-write), a message, an ECC valid and a request valid indication. If both clientreq.valid and clientres.gnt are asserted, the request is latched into the memory client queue. If the pipeline is not stalled, the arbitration logic looks at all of the requests at the top of the memory client queues and determines which request to pop off and pass to the decode stage of the requests at the top of the memory client queues and determines which request to pop off and pass to the decode stage of the request pipe.
- Because there is a request queue between the memory client and the arbiter, the clientres.gnt signal does not indicate that the request has retired. The request still needs to go through the arbitration process. To put it another way, memory client A might receive the clientres.gnt signal before memory client B, but if memory client B has a higher priority, its request might retire before the request from memory client A.
- Arbiter
- As stated above, the arbiter determines which memory client request to pass to the decode stage of the request pipe. This decision process has two steps. The first step is to determine if the arbitration slot for the current memory client is over or not. An arbitration slot is series of requests from the same memory client. The number and type of requests allowed in one arbitration slot varies. Table 3, below, lists what each memory client can do in an arbitration slot.
TABLE 3 Requests allowed in an Arbitration Slot Client Possible Operations Graphics Back < = 16 memory word read with no page crossings End < = 16 memory word write with no page crossings IPCE IC < = 8 memory word read with 1 page crossings < = 8 memory word write with 1 page crossings 1 read-modify-write operation rendering < = 8 memory word read with no page crossings engine, CPU, < = 8 memory word write with no page crossings GRMC 1 read-modify-write operation REFRESH refresh 2 rows - Based on a state for the current arbitration slot and the next request from the current slot owner, the arbiter determines if the arbitration slot should end or not. If not, the request from the memory client who owns the current arbitration slot is passed to the decode stage. If the current arbitration slot is terminated, the arbiter uses the results from an arbitration algorithm to decide which request to pass to the decode stage. The arbitration algorithm to decide which request to pass to the decode stage. The arbitration algorithm ensures that the graphics back
end IC 212 gets ½ of the arbitration slots, the input/output IC 210 gets ¼, the image processing and compression andexpansion IC 216 gets ⅛, therendering engine 208 gets {fraction (1/16)}, theCPU 206 gets {fraction (1/32)}, and the refresh gets {fraction (1/64)}. - Predicting the average bandwidth for each memory client is difficult, but the worst-case slot frequency per memory client can be calculated. The first step is to determine the maximum number of cycles that each memory client can use during an arbitration slot. Table 4, below, shows the number of cycles associated with each type of operation. With reference to Table 4, below, “P” refers to precharge, “X” refers to a dead cycle, “A” refers to activate, “R0” refers to “read
word 0”, “W0” refers to “writeword 0”, and “Ref” refers to “refresh”.TABLE 4 Maximum Cycles for a Memory Operation Operation Command Sequence # of Cycles 8 Word Read P X A X R0 R1 R2 R3 R4 R5 R6 R7 12 8 Word Write P X A X W0 W1 W2 W3 W4 W5 12 W6 W7 Read-Modify-Write P X AX R0 X X X X X X W0 12 8 Word Vice Read P X A X R0 XX P X AX R1 R2 R3 18 with page crossing R4 R5 R6 R7 2 Row Refresh P X Ref X X X X X Ref X X X X X 14 - Table 5, below, refers to the maximum number of cycles for each of the memory clients.
TABLE 5 Maximum # Cycles per Slot Memory Client Operation # of cycles Graphics Back End 16 memory word read or write 20 cycles CPU, Rendering Engine, 8 memory word read or write 12 cycles MACE IPCE 8 memory word read or write 18 cycles 1 page crossings REFRESH refresh 2rows 14 cycles - Finally, slots per second for each memory client can be calculated. If all of the memory clients are requesting all of the time, every memory client will get a turn after 64 slots. This is referrred to as a “round”. In that round, the graphics back end gets 32 out of the 64 slots, the input/
output IC 210 gets 16 out of the 64 slots etc., so a round takes 32*20+8*18+4*12+2*12+14=1062 cycles.TABLE 6 Slot Frequency for Each Client Bandwidth f slot Client is fully utilized GBE 32 slots/15.93 us 1 GB/sec slot/0.50 us MACE slot/1.00 us 256 MB/sec VICE slot/2.00 us 128 MB/sec RE slot/4.00 64 MB/sec CPU slot/8.00 us 32 MB/sec Refresh slot/16.00 us NA - Decode Logic
- The decode logic receives requests from the arbiter. Based on state maintained from the previous requests and information contained in the current request, the decode logic determines which memory bank to select, which of the four state machines in the next stage will handle the request, and whether or not the current request is on the same page as the previous request. This information is passed to the issue/state machine stage.
- The
unified system memory 202 is made up of 8 slots. Each slot can hold one SDRAM DIMM. A SDRAM DIMM is constructed from 1M×16 or 4M×16 SDRAM components and populated on the front only or the front and back side of the DIMM. Two DIMMs are required to make an external SDRAM bank. 1M×16 SDRAM components construct a 32 Mbyte external bank, while 4M×16 SDRAM components construct a 128 Mbyte external bank. The memory system can range in size from 32 Mbytes to 1 Gbyte. - Each SDRAM component has two internal banks, hence two possible open pages. The maximum number of external banks is 8 and the maximum number of internal banks is 8 and the maximum number of internal banks is 16. The
memory controller 204 only supports 4 open pages at a time. This issue will be discussed in detail later in this section. - The decode logic is explained below in more detail. During initalization, software probes the memory to determine how many banks of memory are present and the size of each bank. Based on this information, the software programs the 8 bank control registers. Each bank control register (please refer to the register section) has one bit that indicates the size of the bank and 5 bits for the upper address bits of that bank. Software must place the 64 Mbit external banks in the lower address range followed by any 16 Mbit external banks. This is to prevent gaps in the memory. The decode logic compares the upper address bits of the incoming request to the 8 bank control registers to determine which external bank to select. The number of bits that are compared is dependent on the size of the bank. For example, if the bank size is 64 Mbit, the decode logic compares bits 24:22 of the request address to bits 4:2 of the bank control register. If there is a match, that bank is selected. Each external bank has a separate chip select. If an incoming address matches more than one bank's control register, the bank with the lowest number is selected. If an incoming address does not match any of the bank control registers, a memory address error occurs. When an error occurs, pertinent information about the request is captured in error registers and the processor is interrupted—if the
memory controller 204 interrupt is enabled. The request that caused the error is still sent to the next stage in the pipeline and is processed like a normal request, but thememory controller 204 deasserts all of the external bank selects so that the memory operation doesn't actually occur. Deasserting the external bank selects is also done whenbit 6 of therendering engine 208 message is set. Therendering engine 208 sets this bit when a request is generated using an invalid TLB entry. - With reference to FIG. 10, although the
memory controller 204 can handle any physical external bank configuration, we recommend thatexternal bank 0 always be filled and that the external banks be placed in decreasing density order (for example a 64 Mbit external bank inbank 0 and a 16 Mbit external bank in bank 2). - The previous paragraph describes how the decode logic determines what external bank to select. This paragraph describes the method for determining page crossings and which bank state machine to use in the next stage of the pipeline. The row address, along with the internal and external bank bits for previous requests, are kept in a set of registers which are referred to as the row registers. Each row register corresponds to a bank state machine. There are four row registers (hence four bank state machines), so the decode logic can kept track of up to four open pages. The decode logic compares the internal/external bank bits of the new request with the four row registers. If there is a match, then bank state machine corresponding to that row register is selected. If the new request does not match any of the row registers, one of the row registers, one of the row registers is selected and the register is updated with the new request information. If the internal/external bank bits match one of the row registers and the row bits of the new request match the row bits in that register, then request is on the same page otherwise it is not.
- State Machines and Issue Logic
- The decode logic passes the request along with the external bank selects, state machine select and same page information to the issue/state machine stage. The selected bank state machine sequences through the proper states, while the issue logic decodes the state of the bank state machine into commands that are sent to the SDRAM DIMMS. In addition to the four bank state machines, there is a state machine dedicated to refresh and initialization operations. The initialization/refresh state machine sequences through special states for initialization and refresh while the four bank state machines are forced to an idle state. The bank state machines and the initialization/refresh state machine are discussed in more detail in more detail in the following sections.
- Bank State Machines
- The four bank state machines operate independently, subject only to conflicts for access to the control, address, and data signals. The bank state machines default to page mode operation. That is, the autoprecharge commands are not used, and the SDRAM bank must be explicitly precharged whenever there is a non-page-mode random reference. The decode state passes the request along with the page information to the selected state machine which sequences through the proper states. At certain states, interval timers are stated that inhibit the state machine from advancing to the next state until the SDRAM minimum interval requirements have been met. The bank state machines operate on one request at a time. That is, a request sequences through any required precharge and activation phases and then a read or write phase, at which point it is considered completed and the next request initiated. Finally, the state of the four bank state machines is decoded by the issue logic that generates the SDRAM control signals.
- There are several SDRAM parameters that the state machines must obey. These parameters vary slightly from vendor to vendor, but to simplify the state machines, the most common parameters were chosen and hard coded into the interval timers. Any SDRAM that is not compatible with the parameters listed in the following table is not supported.
- Tr2rp and Tr2w are additional timing parameters that explicitly define the interval between successive read, write, and precharge commands. These parameters insure that successive commands do not cause conflicts on the data signals. While these parameters could be derived internally by a state machine sequencer, they are made explicit to simplify the state machines and use the same timer paradigm as the SDRAM parameters.
TABLE 7 SDRAM Parameters Parameter Value Description Trc 7 Activate bank A to Activate bank A Tras 5 Activate bank A to Precharge bank A Trp 2 Precharge bank A to Activate bank A Trrd 2 Activate bank A to Activate bank B Trcd 2 Activate bank A to Read bank A Twp 1 Datain bank A to Precharge bank A Tr2rp 2 Read bank A to Read or Precharge bank C Tr2w 6 Read bank A to Write bank A - With reference to Table 7, above, Banks A and B are in the same external bank while Bank C is in a different external bank.
- With reference to FIG. 11, a flow diagram for the bank state machines is shown. As shown in FIG. 11, Trp, Trrd and Trcd are enforced by design. The Trc, Tras, Tr2rp and Tr2w parameters have a timer for each of the four bank state machines. The Tr2rp and Tr2w timers are common to all of the four bank state machines, because they are used to prevent conflicts on the shared data lines.
- The initialization/refresh state machine has two functions, initialization and refresh. The initialization procedure is discussed first, followed by the refresh. After a reset, the initialization/refresh state machine sequences through the SDRAM initialization procedure, which is a precharge to all banks, followed by a mode set. The issue stage decodes the state of the initialization/refresh state machine into commands that are sent the SDRAM. After the mode set command programs the SDRAM mode set register to a CAS latency of 2, burst length of 1 and a sequential operation type.
- The SDRAM requires that 4096 refresh cycles occur every 64 ms. In order to comply with this requirement, there is a refresh memory client with a timer. The timer sends out a signal every 27 microseconds which causes the refresh memory client to make a request to the arbiter. The arbiter treats refresh just like all the other memory clients. When the arbiter determines that the time for the refresh slot has come, the arbiter passes the refresh request to the decode stage. The decode stage invalidates all of the row registers and passes the request onto the state machine/issue stage. When a bank state machine sees that it is a refresh request, it goes to its idle state. The initialization/refresh state machine sequences through the refresh procedure which is a precharge to all banks followed by two refresh cycles. A refresh command puts the SDRAM in the automatic refresh mode. An address counter, internal to the device, increments the word and bank address during the refresh cycle. After a refresh cycle, the SDRAM is in the idle state, which means that all the pages are closed. This is why it is important that the bank state machines are forced to the idle state and the row registers are invalidated during a refresh request.
- The initialization/refresh state machine is very similar in structure to the bank state machines and has timers to enforce SDRAM parameters. A Trc timer is used to enforce the Trc requirement between refresh cycles, and the outputs from the bank Tras timers are used to ensure that the “precharge all” command does not violate Tras for any of the active banks.
- Data Pipe:
- The main functions of the data pipe are to: (1) move data between a memory client and the
unified system memory 202, (2) perform ECC operations and (3) merge new byte from a memory client with old data from memory during a read-modify-write operation. Each of these functions is described below. - Data Flow:
- With reference to FIG. 4, the data pipe has one stage which is in lock-step with the last stage of the request pipe. When a write request reaches the decode stage, the request pipe asserts clientres.wrrdy. The clientres.wrrdy signal indicates to the memory client that the data on the Memdata2mem_in bus has been latched into the ECC stage of the data pipe. The data is held in the ECC stage and flows out to the
unified system memory 202 until the request is retired in the request pipe. - Incoming read data is latched in the data pipe, flows through the ECC correction logic and then is latched again before going on the Memdata2client_out bus. The request pipe knows how many cycles the
unified system memory 202 takes to return read responses data. When the read response data is on the Memdata2client_out bus, the request pipe asserts clientres.rdrdy. - The preferred embodiment of the present invention, a computer system architecture featuring dynamic memory allocation for graphics, is thus described. While the present invention has been described in particular embodiments, it should be appreciated that the present invention should not be construed as limited by such embodiments, but rather construed according to the below claims.
Claims (29)
1. A computer system comprising:
a memory controller;
a graphics rendering engine coupled to said memory controller;
a CPU coupled to said memory controller;
an image processor coupled to said memory controller;
a data compression/expansion device coupled to said memory controller;
an input/output device coupled to said memory controller;
a graphics back end device coupled to said memory controller;
a system memory, coupled to said memory controller via a high bandwidth data bus, said system memory providing read/write access, through said memory controller, for memory clients including said CPU, said input/output device, said graphics back end device, said image processor, said data compression/expansion device, said rendering engine, and said memory controller, wherein said memory controller is the interface between said memory clients and said system memory; and
translation hardware for mapping virtual addresses of pixel buffers to physical memory locations in said system memory wherein said pixel buffers are dynamically allocated as tiles of physically contiguous memory.
2. The computer system of wherein said translation hardware is implemented in said rendering engine.
claim 1
3. The computer system of wherein said translation hardware is implemented in each of said rendering engine, said memory controller, said image processor, said data compression/expansion, said graphics back end IC, and said input/output IC.
claim 1
4. The computer system of wherein said system memory is implemented using synchronous DRAM.
claim 1
5. The computer system of wherein said system memory is implemented using synchronous DRAM (SDRAM) accessed via a 256-bit wide memory data bus cycled at 66 MHz.
claim 1
6. The computer system of wherein said tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of pixels.
claim 1
7. The computer system of wherein said tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of 128 pixels wherein each pixel is a 4 byte pixel.
claim 1
8. The computer system of wherein said rend ering engine and said memory controller are implemented on a first IC.
claim 1
9. The computer system of wherein said rendering engine and said memory controller are implemented on a first IC and said image processor and said data compression/expansion device are implemented on a second IC.
claim 1
10. The computer system of wherein said dynamically allocated pixel buffers are comprised of n2 tiles where n is an integer.
claim 1
11. A computer system comprising:
a graphics rendering engine and a memory controller implemented on a first IC;
a CPU coupled to said first IC;
an image processor coupled to said first IC;
a data compression/expansion device coupled to said first IC;
an input/output device coupled to said first IC;
a graphics back end device coupled to said first IC;
a system memory, coupled to said first IC via a high bandwidth data bus, said system memory providing read/write access, through said first IC, for memory clients including said CPU, said input/output device, said graphics back end device, said image processor, said data compression/expansion device, said rendering engine, and said memory controller, wherein said memory controller is the interface between said memory clients and said system memory; and
translation hardware for mapping virtual addresses of pixel buffers to physical memory locations in said system memory wherein said pixel buffers are dynamically allocated as tiles of physically contiguous memory.
12. The computer system of wherein said translation hardware is implemented in said rendering engine.
claim 11
13. The computer system of wherein said translation hardware is implemented in each of said rendering engine, said memory controller, said image processor, said data compression/expansion, said graphics back end IC, and said input/output IC.
claim 11
14. The computer system of wherein said system memory is implemented using synchronous DRAM.
claim 11
15. The computer system of wherein said system memory is implemented using synchronous DRAM (SDRAM) accessed via a 256-bit wide memory data bus cycled at 66 MHz.
claim 11
16. The computer system of wherein said tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of pixels.
claim 11
17. The computer system of wherein said tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of 128 pixels wherein each pixel is a 4 byte pixel.
claim 11
18. The computer system of wherein said image processor and said data compression/expansion are implemented on a second IC.
claim 11
19. The computer system of wherein said first IC is coupled to said system memory by a demultiplexing bus comprising a first bus, coupled to said first IC and having 144 lines cycled at 133 MHz, a second bus, coupled to said system memory and having 288 lines cycled at 66 MHz, and a demultiplexer for demultiplexing signals propagating between said first bus and said second bus.
claim 11
20. The computer system of cliam 11 wherein said dynamically allocated pixel buffers are comprised of n2 tiles where n is an integer.
21. A computer system comprising:
a CPU;
an input/output device;
a graphics back end unit;
a first IC including an image processor and a data compression and expansion device integrated therein;
a second IC including a graphics rendering engine and a memory controller device integrated therein;
a system memory which allows read/write access for memory clients including said CPU, said input/output device, said graphics back end device, said image processor, said data compression/expansion device, said rendering engine, and said memory controller, wherein said memory controller is the interface between said memory clients and said system memory;
a high bandwidth data bus for transferring data between said system memory and said second IC; and
translation hardware for mapping virtual addresses of pixel buffers to physical memory locations in said system memory wherein said pixel buffers are dynamically allocated as tiles of physically contiguous memory.
22. The computer system of wherein said translation hardware is implemented in said rendering engine.
claim 21
23. The computer system of wherein said translation hardware is implemented in each of said rendering engine, said memory controller, said image processor, said data compression/expansion, said graphics back end IC, and said input/output IC.
claim 21
24. The computer system of wherein said system memory is implemented using synchronous DRAM.
claim 21
25. The computer system of wherein said system memory is implemented using synchronous DRAM (SDRAM) accessed via a 256-bit wide memory data bus cycled at 66 MHz.
claim 21
26. The computer system of wherein said tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of pixels.
claim 21
27. The computer system of wherein said tiles are comprised of 64 kilobytes of physically contiguous memory arranged as 128 rows of 128 pixels wherein each pixel is a 4 byte pixel.
claim 21
28. The computer system of wherein said second IC is coupled to said system memory by a demultiplexing bus comprising a first bus, coupled to said second IC and having 144 lines cycled at 133 MHz, a second bus, coupled to said system memory and having 288 lines cycled at 66 MHz, and a demultiplexer for demultiplexing signals propagating between said second bus and said system memory.
claim 21
30. The computer system of wherein said dynamically allocated pixel buffers are comprised of n2 tiles where n is an integer.
claim 21
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US09/137,067 US20010019331A1 (en) | 1996-09-13 | 1998-08-20 | Unified memory architecture for use in computer system |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US08/713,779 US6104417A (en) | 1996-09-13 | 1996-09-13 | Unified memory computer architecture with dynamic graphics memory allocation |
US09/137,067 US20010019331A1 (en) | 1996-09-13 | 1998-08-20 | Unified memory architecture for use in computer system |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US08/713,779 Continuation US6104417A (en) | 1996-09-13 | 1996-09-13 | Unified memory computer architecture with dynamic graphics memory allocation |
Publications (1)
Publication Number | Publication Date |
---|---|
US20010019331A1 true US20010019331A1 (en) | 2001-09-06 |
Family
ID=24867506
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US08/713,779 Expired - Lifetime US6104417A (en) | 1996-09-13 | 1996-09-13 | Unified memory computer architecture with dynamic graphics memory allocation |
US09/137,067 Abandoned US20010019331A1 (en) | 1996-09-13 | 1998-08-20 | Unified memory architecture for use in computer system |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US08/713,779 Expired - Lifetime US6104417A (en) | 1996-09-13 | 1996-09-13 | Unified memory computer architecture with dynamic graphics memory allocation |
Country Status (5)
Country | Link |
---|---|
US (2) | US6104417A (en) |
EP (1) | EP0829820B1 (en) |
JP (1) | JPH10247138A (en) |
CA (1) | CA2214868C (en) |
DE (1) | DE69722117T2 (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20040183808A1 (en) * | 2001-10-09 | 2004-09-23 | William Radke | Embedded memory system and method including data error correction |
US20050024367A1 (en) * | 2000-12-13 | 2005-02-03 | William Radke | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US20060098021A1 (en) * | 2004-11-11 | 2006-05-11 | Samsung Electronics Co., Ltd. | Graphics system and memory device for three-dimensional graphics acceleration and method for three dimensional graphics processing |
US20060103658A1 (en) * | 2004-11-12 | 2006-05-18 | Via Technologies, Inc. | Color compression using multiple planes in a multi-sample anti-aliasing scheme |
US20060177122A1 (en) * | 2005-02-07 | 2006-08-10 | Sony Computer Entertainment Inc. | Method and apparatus for particle manipulation using graphics processing |
US20080278513A1 (en) * | 2004-04-15 | 2008-11-13 | Junichi Naoi | Plotting Apparatus, Plotting Method, Information Processing Apparatus, and Information Processing Method |
US20090160857A1 (en) * | 2007-12-20 | 2009-06-25 | Jim Rasmusson | Unified Compression/Decompression Graphics Architecture |
US20100030980A1 (en) * | 2006-12-25 | 2010-02-04 | Panasonic Corporation | Memory control device, memory device, and memory control method |
US20100118935A1 (en) * | 2004-04-23 | 2010-05-13 | Sumitomo Electric Industries, Ltd. | Coding method for motion-image data, decoding method, terminal equipment executing these, and two-way interactive system |
US20110066815A1 (en) * | 2009-09-15 | 2011-03-17 | Olympus Corporation | Memory access control device and memory access control method |
US7928989B1 (en) * | 2006-07-28 | 2011-04-19 | Nvidia Corporation | Feedback and record of transformed vertices in a graphics library |
US20120162264A1 (en) * | 2010-12-22 | 2012-06-28 | Hughes Gregory F | System level graphics manipulations on protected content |
US8271746B1 (en) * | 2006-11-03 | 2012-09-18 | Nvidia Corporation | Tiering of linear clients |
US10692171B2 (en) | 2015-11-17 | 2020-06-23 | Samsung Electronics Co., Ltd. | Method of operating virtual address generator and method of operating system including the same |
Families Citing this family (146)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW360823B (en) * | 1996-09-30 | 1999-06-11 | Hitachi Ltd | Data processor and graphic processor |
US6308248B1 (en) * | 1996-12-31 | 2001-10-23 | Compaq Computer Corporation | Method and system for allocating memory space using mapping controller, page table and frame numbers |
US9098297B2 (en) * | 1997-05-08 | 2015-08-04 | Nvidia Corporation | Hardware accelerator for an object-oriented programming language |
US6118462A (en) | 1997-07-01 | 2000-09-12 | Memtrax Llc | Computer system controller having internal memory and external memory control |
US6266753B1 (en) * | 1997-07-10 | 2001-07-24 | Cirrus Logic, Inc. | Memory manager for multi-media apparatus and method therefor |
US6075546A (en) * | 1997-11-10 | 2000-06-13 | Silicon Grahphics, Inc. | Packetized command interface to graphics processor |
US6275243B1 (en) * | 1998-04-08 | 2001-08-14 | Nvidia Corporation | Method and apparatus for accelerating the transfer of graphical images |
US6480205B1 (en) | 1998-07-22 | 2002-11-12 | Nvidia Corporation | Method and apparatus for occlusion culling in graphics systems |
US6393543B1 (en) * | 1998-11-12 | 2002-05-21 | Acuid Corporation Limited | System and a method for transformation of memory device addresses |
US6381683B1 (en) * | 1998-12-09 | 2002-04-30 | Advanced Micro Devices, Inc. | Method and system for destination-sensitive memory control and access in data processing systems |
US6219769B1 (en) | 1998-12-09 | 2001-04-17 | Advanced Micro Devices, Inc. | Method and system for origin-sensitive memory control and access in data processing systems |
US6260123B1 (en) * | 1998-12-09 | 2001-07-10 | Advanced Micro Devices, Inc. | Method and system for memory control and access in data processing systems |
US6510497B1 (en) | 1998-12-09 | 2003-01-21 | Advanced Micro Devices, Inc. | Method and system for page-state sensitive memory control and access in data processing systems |
US6546439B1 (en) | 1998-12-09 | 2003-04-08 | Advanced Micro Devices, Inc. | Method and system for improved data access |
US6226721B1 (en) | 1998-12-09 | 2001-05-01 | Advanced Micro Devices, Inc. | Method and system for generating and utilizing speculative memory access requests in data processing systems |
US6563506B1 (en) * | 1998-12-14 | 2003-05-13 | Ati International Srl | Method and apparatus for memory bandwith allocation and control in a video graphics system |
US6362826B1 (en) * | 1999-01-15 | 2002-03-26 | Intel Corporation | Method and apparatus for implementing dynamic display memory |
US6414688B1 (en) * | 1999-01-29 | 2002-07-02 | Micron Technology, Inc. | Programmable graphics memory method |
US6189082B1 (en) * | 1999-01-29 | 2001-02-13 | Neomagic Corp. | Burst access of registers at non-consecutive addresses using a mapping control word |
US6377268B1 (en) * | 1999-01-29 | 2002-04-23 | Micron Technology, Inc. | Programmable graphics memory apparatus |
US6288729B1 (en) * | 1999-02-26 | 2001-09-11 | Ati International Srl | Method and apparatus for a graphics controller to extend graphics memory |
TW457430B (en) * | 1999-03-02 | 2001-10-01 | Via Tech Inc | Memory access control device |
US6526583B1 (en) * | 1999-03-05 | 2003-02-25 | Teralogic, Inc. | Interactive set-top box having a unified memory architecture |
US6433785B1 (en) * | 1999-04-09 | 2002-08-13 | Intel Corporation | Method and apparatus for improving processor to graphics device throughput |
US6504549B1 (en) * | 1999-05-19 | 2003-01-07 | Ati International Srl | Apparatus to arbitrate among clients requesting memory access in a video system and method thereof |
TW550956B (en) * | 1999-05-26 | 2003-09-01 | Koninkl Philips Electronics Nv | Digital video-processing unit |
US6469703B1 (en) | 1999-07-02 | 2002-10-22 | Ati International Srl | System of accessing data in a graphics system and method thereof |
US6496192B1 (en) * | 1999-08-05 | 2002-12-17 | Matsushita Electric Industrial Co., Ltd. | Modular architecture for image transposition memory using synchronous DRAM |
US6618048B1 (en) | 1999-10-28 | 2003-09-09 | Nintendo Co., Ltd. | 3D graphics rendering system for performing Z value clamping in near-Z range to maximize scene resolution of visually important Z components |
US6717577B1 (en) | 1999-10-28 | 2004-04-06 | Nintendo Co., Ltd. | Vertex cache for 3D computer graphics |
US6198488B1 (en) * | 1999-12-06 | 2001-03-06 | Nvidia | Transform, lighting and rasterization system embodied on a single semiconductor platform |
US6452595B1 (en) * | 1999-12-06 | 2002-09-17 | Nvidia Corporation | Integrated graphics processing unit with antialiasing |
US7209140B1 (en) | 1999-12-06 | 2007-04-24 | Nvidia Corporation | System, method and article of manufacture for a programmable vertex processing model with instruction set |
US6844880B1 (en) | 1999-12-06 | 2005-01-18 | Nvidia Corporation | System, method and computer program product for an improved programmable vertex processing model with instruction set |
US6643752B1 (en) * | 1999-12-09 | 2003-11-04 | Rambus Inc. | Transceiver with latency alignment circuitry |
US20050010737A1 (en) * | 2000-01-05 | 2005-01-13 | Fred Ware | Configurable width buffered module having splitter elements |
US7010642B2 (en) * | 2000-01-05 | 2006-03-07 | Rambus Inc. | System featuring a controller device and a memory module that includes an integrated circuit buffer device and a plurality of integrated circuit memory devices |
US6502161B1 (en) | 2000-01-05 | 2002-12-31 | Rambus Inc. | Memory system including a point-to-point linked memory subsystem |
US7266634B2 (en) * | 2000-01-05 | 2007-09-04 | Rambus Inc. | Configurable width buffered module having flyby elements |
US7356639B2 (en) * | 2000-01-05 | 2008-04-08 | Rambus Inc. | Configurable width buffered module having a bypass circuit |
US7404032B2 (en) * | 2000-01-05 | 2008-07-22 | Rambus Inc. | Configurable width buffered module having switch elements |
US7363422B2 (en) * | 2000-01-05 | 2008-04-22 | Rambus Inc. | Configurable width buffered module |
US7196710B1 (en) | 2000-08-23 | 2007-03-27 | Nintendo Co., Ltd. | Method and apparatus for buffering graphics data in a graphics system |
US6700586B1 (en) | 2000-08-23 | 2004-03-02 | Nintendo Co., Ltd. | Low cost graphics with stitching processing hardware support for skeletal animation |
US6707458B1 (en) | 2000-08-23 | 2004-03-16 | Nintendo Co., Ltd. | Method and apparatus for texture tiling in a graphics system |
US6811489B1 (en) | 2000-08-23 | 2004-11-02 | Nintendo Co., Ltd. | Controller interface for a graphics system |
US6636214B1 (en) | 2000-08-23 | 2003-10-21 | Nintendo Co., Ltd. | Method and apparatus for dynamically reconfiguring the order of hidden surface processing based on rendering mode |
US7538772B1 (en) | 2000-08-23 | 2009-05-26 | Nintendo Co., Ltd. | Graphics processing system with enhanced memory controller |
US7576748B2 (en) | 2000-11-28 | 2009-08-18 | Nintendo Co. Ltd. | Graphics system with embedded frame butter having reconfigurable pixel formats |
JP4042088B2 (en) * | 2000-08-25 | 2008-02-06 | 株式会社ルネサステクノロジ | Memory access method |
US6885378B1 (en) * | 2000-09-28 | 2005-04-26 | Intel Corporation | Method and apparatus for the implementation of full-scene anti-aliasing supersampling |
US8692844B1 (en) * | 2000-09-28 | 2014-04-08 | Nvidia Corporation | Method and system for efficient antialiased rendering |
US6859208B1 (en) * | 2000-09-29 | 2005-02-22 | Intel Corporation | Shared translation address caching |
US6853382B1 (en) | 2000-10-13 | 2005-02-08 | Nvidia Corporation | Controller for a memory system having multiple partitions |
US6753873B2 (en) * | 2001-01-31 | 2004-06-22 | General Electric Company | Shared memory control between detector framing node and processor |
US6864896B2 (en) * | 2001-05-15 | 2005-03-08 | Rambus Inc. | Scalable unified memory architecture |
GB2378108B (en) | 2001-07-24 | 2005-08-17 | Imagination Tech Ltd | Three dimensional graphics system |
EP1296237A1 (en) * | 2001-09-25 | 2003-03-26 | Texas Instruments Incorporated | Data transfer controlled by task attributes |
US6906720B2 (en) * | 2002-03-12 | 2005-06-14 | Sun Microsystems, Inc. | Multipurpose memory system for use in a graphics system |
US7508398B1 (en) | 2002-08-27 | 2009-03-24 | Nvidia Corporation | Transparent antialiased memory access |
US7085910B2 (en) * | 2003-08-27 | 2006-08-01 | Lsi Logic Corporation | Memory window manager for control structure access |
US8775997B2 (en) | 2003-09-15 | 2014-07-08 | Nvidia Corporation | System and method for testing and configuring semiconductor functional circuits |
US8768642B2 (en) | 2003-09-15 | 2014-07-01 | Nvidia Corporation | System and method for remotely configuring semiconductor functional circuits |
US8732644B1 (en) | 2003-09-15 | 2014-05-20 | Nvidia Corporation | Micro electro mechanical switch system and method for testing and configuring semiconductor functional circuits |
US7286134B1 (en) | 2003-12-17 | 2007-10-23 | Nvidia Corporation | System and method for packing data in a tiled graphics memory |
US7420568B1 (en) | 2003-12-17 | 2008-09-02 | Nvidia Corporation | System and method for packing data in different formats in a tiled graphics memory |
US8711161B1 (en) | 2003-12-18 | 2014-04-29 | Nvidia Corporation | Functional component compensation reconfiguration system and method |
US6999088B1 (en) | 2003-12-23 | 2006-02-14 | Nvidia Corporation | Memory system having multiple subpartitions |
US7296139B1 (en) | 2004-01-30 | 2007-11-13 | Nvidia Corporation | In-memory table structure for virtual address translation system with translation units of variable range size |
US7334108B1 (en) * | 2004-01-30 | 2008-02-19 | Nvidia Corporation | Multi-client virtual address translation system with translation units of variable-range size |
US7278008B1 (en) | 2004-01-30 | 2007-10-02 | Nvidia Corporation | Virtual address translation system with caching of variable-range translation clusters |
US20050237329A1 (en) * | 2004-04-27 | 2005-10-27 | Nvidia Corporation | GPU rendering to system memory |
US20050246502A1 (en) * | 2004-04-28 | 2005-11-03 | Texas Instruments Incorporated | Dynamic memory mapping |
US7227548B2 (en) * | 2004-05-07 | 2007-06-05 | Valve Corporation | Method and system for determining illumination of models using an ambient cube |
US7221369B1 (en) | 2004-07-29 | 2007-05-22 | Nvidia Corporation | Apparatus, system, and method for delivering data to multiple memory clients via a unitary buffer |
US7277098B2 (en) * | 2004-08-23 | 2007-10-02 | Via Technologies, Inc. | Apparatus and method of an improved stencil shadow volume operation |
US8723231B1 (en) | 2004-09-15 | 2014-05-13 | Nvidia Corporation | Semiconductor die micro electro-mechanical switch management system and method |
US8711156B1 (en) | 2004-09-30 | 2014-04-29 | Nvidia Corporation | Method and system for remapping processing elements in a pipeline of a graphics processing unit |
US8427496B1 (en) | 2005-05-13 | 2013-04-23 | Nvidia Corporation | Method and system for implementing compression across a graphics bus interconnect |
US20060282604A1 (en) | 2005-05-27 | 2006-12-14 | Ati Technologies, Inc. | Methods and apparatus for processing graphics data using multiple processing circuits |
US7493520B2 (en) * | 2005-06-07 | 2009-02-17 | Microsoft Corporation | System and method for validating the graphical output of an updated software module |
US7562271B2 (en) | 2005-09-26 | 2009-07-14 | Rambus Inc. | Memory system topologies including a buffer device and an integrated circuit memory device |
US11328764B2 (en) | 2005-09-26 | 2022-05-10 | Rambus Inc. | Memory system topologies including a memory die stack |
US7464225B2 (en) * | 2005-09-26 | 2008-12-09 | Rambus Inc. | Memory module including a plurality of integrated circuit memory devices and a plurality of buffer devices in a matrix topology |
US8212832B2 (en) * | 2005-12-08 | 2012-07-03 | Ati Technologies Ulc | Method and apparatus with dynamic graphics surface memory allocation |
US8698811B1 (en) | 2005-12-15 | 2014-04-15 | Nvidia Corporation | Nested boustrophedonic patterns for rasterization |
US8390645B1 (en) | 2005-12-19 | 2013-03-05 | Nvidia Corporation | Method and system for rendering connecting antialiased line segments |
US9117309B1 (en) | 2005-12-19 | 2015-08-25 | Nvidia Corporation | Method and system for rendering polygons with a bounding box in a graphics processor unit |
US7791617B2 (en) * | 2005-12-19 | 2010-09-07 | Nvidia Corporation | Method and system for rendering polygons having abutting edges |
US8218091B2 (en) * | 2006-04-18 | 2012-07-10 | Marvell World Trade Ltd. | Shared memory multi video channel display apparatus and methods |
US8284322B2 (en) * | 2006-04-18 | 2012-10-09 | Marvell World Trade Ltd. | Shared memory multi video channel display apparatus and methods |
US8264610B2 (en) * | 2006-04-18 | 2012-09-11 | Marvell World Trade Ltd. | Shared memory multi video channel display apparatus and methods |
US8928676B2 (en) * | 2006-06-23 | 2015-01-06 | Nvidia Corporation | Method for parallel fine rasterization in a raster stage of a graphics pipeline |
US8477134B1 (en) | 2006-06-30 | 2013-07-02 | Nvidia Corporation | Conservative triage of polygon status using low precision edge evaluation and high precision edge evaluation |
US20080028181A1 (en) * | 2006-07-31 | 2008-01-31 | Nvidia Corporation | Dedicated mechanism for page mapping in a gpu |
GB2449399B (en) * | 2006-09-29 | 2009-05-06 | Imagination Tech Ltd | Improvements in memory management for systems for generating 3-dimensional computer images |
JP2008090673A (en) * | 2006-10-03 | 2008-04-17 | Mitsubishi Electric Corp | Cache memory control device |
US7986327B1 (en) * | 2006-10-23 | 2011-07-26 | Nvidia Corporation | Systems for efficient retrieval from tiled memory surface to linear memory display |
US7805587B1 (en) | 2006-11-01 | 2010-09-28 | Nvidia Corporation | Memory addressing controlled by PTE fields |
US8427487B1 (en) | 2006-11-02 | 2013-04-23 | Nvidia Corporation | Multiple tile output using interface compression in a raster stage |
US8482567B1 (en) | 2006-11-03 | 2013-07-09 | Nvidia Corporation | Line rasterization techniques |
US8120613B2 (en) * | 2006-11-29 | 2012-02-21 | Siemens Medical Solutions Usa, Inc. | Method and apparatus for real-time digital image acquisition, storage, and retrieval |
KR100867640B1 (en) * | 2007-02-06 | 2008-11-10 | 삼성전자주식회사 | System on chip including image processing memory with multiple access |
US8724483B2 (en) * | 2007-10-22 | 2014-05-13 | Nvidia Corporation | Loopback configuration for bi-directional interfaces |
US8063903B2 (en) * | 2007-11-09 | 2011-11-22 | Nvidia Corporation | Edge evaluation techniques for graphics hardware |
US9064333B2 (en) | 2007-12-17 | 2015-06-23 | Nvidia Corporation | Interrupt handling techniques in the rasterizer of a GPU |
US8780123B2 (en) | 2007-12-17 | 2014-07-15 | Nvidia Corporation | Interrupt handling techniques in the rasterizer of a GPU |
US8255783B2 (en) * | 2008-04-23 | 2012-08-28 | International Business Machines Corporation | Apparatus, system and method for providing error protection for data-masking bits |
US8681861B2 (en) | 2008-05-01 | 2014-03-25 | Nvidia Corporation | Multistandard hardware video encoder |
US8923385B2 (en) | 2008-05-01 | 2014-12-30 | Nvidia Corporation | Rewind-enabled hardware encoder |
JP5658430B2 (en) * | 2008-08-15 | 2015-01-28 | パナソニックIpマネジメント株式会社 | Image processing device |
US8392667B2 (en) * | 2008-12-12 | 2013-03-05 | Nvidia Corporation | Deadlock avoidance by marking CPU traffic as special |
GB0823254D0 (en) | 2008-12-19 | 2009-01-28 | Imagination Tech Ltd | Multi level display control list in tile based 3D computer graphics system |
US8330766B1 (en) | 2008-12-19 | 2012-12-11 | Nvidia Corporation | Zero-bandwidth clears |
US8319783B1 (en) | 2008-12-19 | 2012-11-27 | Nvidia Corporation | Index-based zero-bandwidth clears |
GB0823468D0 (en) | 2008-12-23 | 2009-01-28 | Imagination Tech Ltd | Display list control stream grouping in tile based 3D computer graphics systems |
US20110063309A1 (en) * | 2009-09-16 | 2011-03-17 | Nvidia Corporation | User interface for co-processing techniques on heterogeneous graphics processing units |
US10175990B2 (en) * | 2009-12-22 | 2019-01-08 | Intel Corporation | Gathering and scattering multiple data elements |
US9530189B2 (en) | 2009-12-31 | 2016-12-27 | Nvidia Corporation | Alternate reduction ratios and threshold mechanisms for framebuffer compression |
US9331869B2 (en) | 2010-03-04 | 2016-05-03 | Nvidia Corporation | Input/output request packet handling techniques by a device specific kernel mode driver |
US8495464B2 (en) * | 2010-06-28 | 2013-07-23 | Intel Corporation | Reliability support in memory systems without error correcting code support |
US9171350B2 (en) | 2010-10-28 | 2015-10-27 | Nvidia Corporation | Adaptive resolution DGPU rendering to provide constant framerate with free IGPU scale up |
US9477597B2 (en) | 2011-03-25 | 2016-10-25 | Nvidia Corporation | Techniques for different memory depths on different partitions |
US8701057B2 (en) | 2011-04-11 | 2014-04-15 | Nvidia Corporation | Design, layout, and manufacturing techniques for multivariant integrated circuits |
US9529712B2 (en) | 2011-07-26 | 2016-12-27 | Nvidia Corporation | Techniques for balancing accesses to memory having different memory types |
US9886312B2 (en) | 2011-09-28 | 2018-02-06 | Microsoft Technology Licensing, Llc | Dynamic provisioning of virtual video memory based on virtual video controller configuration |
JP2013242694A (en) * | 2012-05-21 | 2013-12-05 | Renesas Mobile Corp | Semiconductor device, electronic device, electronic system, and method of controlling electronic device |
US9373182B2 (en) | 2012-08-17 | 2016-06-21 | Intel Corporation | Memory sharing via a unified memory architecture |
KR20140060141A (en) * | 2012-11-09 | 2014-05-19 | 삼성전자주식회사 | Display apparatus and method for controlling thereof |
US9047198B2 (en) | 2012-11-29 | 2015-06-02 | Apple Inc. | Prefetching across page boundaries in hierarchically cached processors |
US9607407B2 (en) | 2012-12-31 | 2017-03-28 | Nvidia Corporation | Variable-width differential memory compression |
US9591309B2 (en) | 2012-12-31 | 2017-03-07 | Nvidia Corporation | Progressive lossy memory compression |
US9235905B2 (en) | 2013-03-13 | 2016-01-12 | Ologn Technologies Ag | Efficient screen image transfer |
US9607356B2 (en) | 2013-05-02 | 2017-03-28 | Arm Limited | Graphics processing systems |
US9070200B2 (en) * | 2013-05-02 | 2015-06-30 | Arm Limited | Graphics processing systems |
US9741089B2 (en) | 2013-05-02 | 2017-08-22 | Arm Limited | Graphics processing systems |
US9767595B2 (en) | 2013-05-02 | 2017-09-19 | Arm Limited | Graphics processing systems |
US9710894B2 (en) | 2013-06-04 | 2017-07-18 | Nvidia Corporation | System and method for enhanced multi-sample anti-aliasing |
US9098924B2 (en) * | 2013-07-15 | 2015-08-04 | Nvidia Corporation | Techniques for optimizing stencil buffers |
US9514563B2 (en) | 2013-08-30 | 2016-12-06 | Arm Limited | Graphics processing systems |
JP5928914B2 (en) * | 2014-03-17 | 2016-06-01 | 株式会社ソニー・インタラクティブエンタテインメント | Graphics processing apparatus and graphics processing method |
JP6313632B2 (en) * | 2014-03-31 | 2018-04-18 | キヤノン株式会社 | Image processing device |
US9832388B2 (en) | 2014-08-04 | 2017-11-28 | Nvidia Corporation | Deinterleaving interleaved high dynamic range image by using YUV interpolation |
US20180012327A1 (en) * | 2016-07-05 | 2018-01-11 | Ubitus Inc. | Overlaying multi-source media in vram |
US10438569B2 (en) | 2017-04-17 | 2019-10-08 | Intel Corporation | Consolidation of data compression using common sectored cache for graphics streams |
US11321804B1 (en) * | 2020-10-15 | 2022-05-03 | Qualcomm Incorporated | Techniques for flexible rendering operations |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0814803B2 (en) * | 1986-05-23 | 1996-02-14 | 株式会社日立製作所 | Address translation method |
DE3852457T2 (en) * | 1987-07-31 | 1995-06-14 | Qms Inc | Page printing system with virtual memory. |
US5640543A (en) * | 1992-06-19 | 1997-06-17 | Intel Corporation | Scalable multimedia platform architecture |
WO1994016391A1 (en) * | 1992-12-31 | 1994-07-21 | Intel Corporation | Bus to bus interface with address translation |
US5450542A (en) * | 1993-11-30 | 1995-09-12 | Vlsi Technology, Inc. | Bus interface with graphics and system paths for an integrated memory system |
-
1996
- 1996-09-13 US US08/713,779 patent/US6104417A/en not_active Expired - Lifetime
-
1997
- 1997-09-05 DE DE69722117T patent/DE69722117T2/en not_active Expired - Lifetime
- 1997-09-05 EP EP97306897A patent/EP0829820B1/en not_active Expired - Lifetime
- 1997-09-09 CA CA002214868A patent/CA2214868C/en not_active Expired - Fee Related
- 1997-09-16 JP JP9251134A patent/JPH10247138A/en active Pending
-
1998
- 1998-08-20 US US09/137,067 patent/US20010019331A1/en not_active Abandoned
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8194086B2 (en) | 2000-12-13 | 2012-06-05 | Round Rock Research, Llc | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US7724262B2 (en) | 2000-12-13 | 2010-05-25 | Round Rock Research, Llc | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US20100220103A1 (en) * | 2000-12-13 | 2010-09-02 | Round Rock Research, Llc | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US8446420B2 (en) | 2000-12-13 | 2013-05-21 | Round Rock Research, Llc | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US20110169846A1 (en) * | 2000-12-13 | 2011-07-14 | Round Rock Research, Llc | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US7916148B2 (en) | 2000-12-13 | 2011-03-29 | Round Rock Research, Llc | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US7379068B2 (en) | 2000-12-13 | 2008-05-27 | Micron Technology, Inc. | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US20080218525A1 (en) * | 2000-12-13 | 2008-09-11 | William Radke | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US20050024367A1 (en) * | 2000-12-13 | 2005-02-03 | William Radke | Memory system and method for improved utilization of read and write bandwidth of a graphics processing system |
US6956577B2 (en) * | 2001-10-09 | 2005-10-18 | Micron Technology, Inc. | Embedded memory system and method including data error correction |
US20040183808A1 (en) * | 2001-10-09 | 2004-09-23 | William Radke | Embedded memory system and method including data error correction |
US8203569B2 (en) * | 2004-04-15 | 2012-06-19 | Sony Computer Entertainment Inc. | Graphics processor, graphics processing method, information processor and information processing method |
US20080278513A1 (en) * | 2004-04-15 | 2008-11-13 | Junichi Naoi | Plotting Apparatus, Plotting Method, Information Processing Apparatus, and Information Processing Method |
US7983497B2 (en) * | 2004-04-23 | 2011-07-19 | Sumitomo Electric Industries, Ltd. | Coding method for motion-image data, decoding method, terminal equipment executing these, and two-way interactive system |
US20100118935A1 (en) * | 2004-04-23 | 2010-05-13 | Sumitomo Electric Industries, Ltd. | Coding method for motion-image data, decoding method, terminal equipment executing these, and two-way interactive system |
US20060098021A1 (en) * | 2004-11-11 | 2006-05-11 | Samsung Electronics Co., Ltd. | Graphics system and memory device for three-dimensional graphics acceleration and method for three dimensional graphics processing |
US20060103658A1 (en) * | 2004-11-12 | 2006-05-18 | Via Technologies, Inc. | Color compression using multiple planes in a multi-sample anti-aliasing scheme |
CN100357972C (en) * | 2004-11-12 | 2007-12-26 | 威盛电子股份有限公司 | Systems and methods for compressing computer graphics color data |
US7126615B2 (en) * | 2004-11-12 | 2006-10-24 | Via Technologies, Inc. | Color compression using multiple planes in a multi-sample anti-aliasing scheme |
US20060177122A1 (en) * | 2005-02-07 | 2006-08-10 | Sony Computer Entertainment Inc. | Method and apparatus for particle manipulation using graphics processing |
US7928989B1 (en) * | 2006-07-28 | 2011-04-19 | Nvidia Corporation | Feedback and record of transformed vertices in a graphics library |
US8271746B1 (en) * | 2006-11-03 | 2012-09-18 | Nvidia Corporation | Tiering of linear clients |
US20100030980A1 (en) * | 2006-12-25 | 2010-02-04 | Panasonic Corporation | Memory control device, memory device, and memory control method |
US8307190B2 (en) * | 2006-12-25 | 2012-11-06 | Panasonic Corporation | Memory control device, memory device, and memory control method |
US8738888B2 (en) | 2006-12-25 | 2014-05-27 | Panasonic Corporation | Memory control device, memory device, and memory control method |
US20090160857A1 (en) * | 2007-12-20 | 2009-06-25 | Jim Rasmusson | Unified Compression/Decompression Graphics Architecture |
US9665951B2 (en) * | 2007-12-20 | 2017-05-30 | Telefonaktiebolaget Lm Ericsson (Publ) | Unified compression/decompression graphics architecture |
US20110066815A1 (en) * | 2009-09-15 | 2011-03-17 | Olympus Corporation | Memory access control device and memory access control method |
US20120162264A1 (en) * | 2010-12-22 | 2012-06-28 | Hughes Gregory F | System level graphics manipulations on protected content |
US8836727B2 (en) * | 2010-12-22 | 2014-09-16 | Apple Inc. | System level graphics manipulations on protected content |
US10692171B2 (en) | 2015-11-17 | 2020-06-23 | Samsung Electronics Co., Ltd. | Method of operating virtual address generator and method of operating system including the same |
Also Published As
Publication number | Publication date |
---|---|
MX9706495A (en) | 1998-06-30 |
CA2214868C (en) | 2006-06-06 |
JPH10247138A (en) | 1998-09-14 |
EP0829820B1 (en) | 2003-05-21 |
CA2214868A1 (en) | 1998-03-13 |
EP0829820A2 (en) | 1998-03-18 |
DE69722117T2 (en) | 2004-02-05 |
EP0829820A3 (en) | 1998-11-18 |
US6104417A (en) | 2000-08-15 |
DE69722117D1 (en) | 2003-07-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US6104417A (en) | Unified memory computer architecture with dynamic graphics memory allocation | |
US6721864B2 (en) | Programmable memory controller | |
US6745309B2 (en) | Pipelined memory controller | |
US6026464A (en) | Memory control system and method utilizing distributed memory controllers for multibank memory | |
US6622228B2 (en) | System and method of processing memory requests in a pipelined memory controller | |
US5911149A (en) | Apparatus and method for implementing a programmable shared memory with dual bus architecture | |
JP3976342B2 (en) | Method and apparatus for enabling simultaneous access to shared memory from multiple agents | |
US6330645B1 (en) | Multi-stream coherent memory controller apparatus and method | |
US7707328B2 (en) | Memory access control circuit | |
US5506968A (en) | Terminating access of an agent to a shared resource when a timer, started after a low latency agent requests access, reaches a predetermined value | |
US6591323B2 (en) | Memory controller with arbitration among several strobe requests | |
JP7384806B2 (en) | Scheduling memory requests for ganged memory devices | |
EP1474747A1 (en) | Address space, bus system, memory controller and device system | |
US5822768A (en) | Dual ported memory for a unified memory architecture | |
US6272583B1 (en) | Microprocessor having built-in DRAM and internal data transfer paths wider and faster than independent external transfer paths | |
US5802581A (en) | SDRAM memory controller with multiple arbitration points during a memory cycle | |
US5802597A (en) | SDRAM memory controller while in burst four mode supporting single data accesses | |
US11360897B1 (en) | Adaptive memory access management | |
JP2003316642A (en) | Memory control circuit, dma request block and memory access system | |
US6425020B1 (en) | Systems and methods for passively transferring data across a selected single bus line independent of a control circuitry | |
JPH10144073A (en) | Access mechanism for synchronous dram | |
MXPA97006495A (en) | An unified memory architecture with dynamic allocation of grafi memory | |
JP2000172553A (en) | Data processor | |
JPH04199450A (en) | Direct memory access control circuit | |
KR20070098352A (en) | Apparatus for bus arbitration |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STCB | Information on status: application discontinuation |
Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION |
|
AS | Assignment |
Owner name: MICROSOFT TECHNOLOGY LICENSING, LLC, WASHINGTON Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:MICROSOFT CORPORATION;REEL/FRAME:034766/0001 Effective date: 20141014 |