1. Field of the Invention
Embodiments of the present invention generally relate to graphics processing, and more particularly to splitting a frame buffer between graphics adapters when multiple graphics adapters are installed in a system.
2. Description of the Related Art
Conventional graphics processing systems including a single graphics adapter typically have a single frame buffer that is accessed by an application program. In graphics processing systems including multiple graphics adapters, processing of an image is typically distributed across the multiple graphics adapters. The image is divided into portions corresponding to the distribution, and each portion is stored in a separate frame buffer coupled to one of the multiple graphics adapters. Consequently, the application program must access multiple frame buffers, each frame buffer storing a portion of the image, during the processing of the image. The application program must be aware of which frame buffer stores each portion of the image in order to correctly access the image.
Accordingly, it is desirable to present a unified frame buffer to an application program when multiple graphics adapters are installed in a graphics processing system. The unified frame buffer enables the application program to access the image without determining which of the multiple frame buffer stores each portion of the image.
The current invention involves new systems and methods for representing multiple prefetchable memory resources, such as frame buffers coupled to graphics devices, as a unified prefetchable memory space accessible by a software application. A graphics surface may be processed by multiple graphics devices, with portions of the surface residing in separate frame buffers, where each frame buffer is coupled to one of the multiple graphics devices. Within the unified prefetchable memory space, the one or more redirection regions may be specified. Accesses within a redirection region are transmitted to a prefetchable memory for a single graphics device. Accesses within the unified prefetchable memory space, but outside of any redirection region, may be simultaneously transmitted to one or more of the frame buffers of the multiple graphics devices.
Various embodiments of the invention include a bridge device, a first graphics device, and a second graphics device. The bridge device is configured to store values defining a redirection region within a prefetchable memory space. The first graphics device is coupled to the bridge device and has a first associated frame buffer which is configured to store a first portion of the redirection region. The second graphics device is coupled to the bridge device and has a second associated frame buffer which is configured to store a second portion of the redirection region.
Various embodiments of a method of the invention for accessing a frame buffer that is divided between multiple graphics devices as a unified frame buffer include determining whether the access is within a redirection region of a prefetchable memory space and outputting the access to one or more of the multiple graphics devices depending on whether an address associated with the access is either within the redirection region or within the prefetchable memory space and outside of the redirection region.
Various embodiments of the invention include a bridge device for interfacing between a host processor and multiple graphics devices. The bridge device includes a first register, redirection registers, and a memory access unit. The first register is configured to store a value specifying a prefetchable memory space. The redirection registers are configured to store a set of values specifying a redirection region within the prefetchable memory space. The memory access unit is configured to transmit accesses to one or more ports based on the first register and the redirection registers, wherein each of the one or more ports is coupled to a different one of the multiple graphics devices.
So that the manner in which the above recited features of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered limiting of its scope, for the invention may admit to other equally effective embodiments.
In the following description, numerous specific details are set forth to provide a more thorough understanding of the present invention. However, it will be apparent to one of skill in the art that the present invention may be practiced without one or more of these specific details. In other instances, well-known features have not been described in order to avoid obscuring the present invention.
A single device driver interfaces between an application program and multiple graphics devices. Therefore, unlike a conventional system where there is a one to one correspondence between device drivers and graphics devices, the application program does not need to communicate with multiple device drivers to distribute processing to the multiple graphics devices. Furthermore, a unified frame buffer may be presented to the application program when the frame buffer is distributed between the multiple graphics devices.
A single device driver, graphics driver 105, stored within main memory 110, configures the devices within graphics subsystem 180 and communicates between applications executed by host processor 120 and graphics adapters 165 and 164. In a conventional graphics processing system running the Windows® OS two device drivers are used, one for each graphics adapter installed in the system.
In some embodiments of system 100, chipset 130 may include a system memory switch and an input/output (I/O) switch that may include several interfaces such as, Advanced Technology Attachment (ATA) bus, Universal Serial Bus (USB), Peripheral component interface (PCI), or the like. Switch 160 provides an interface between chipset 130 and each of graphics adapter 165 and graphics adapter 164 when a first port and a second port of switch 160 are coupled to a connection 151 and a connection 141, respectively. In some embodiments of switch 160, switch 160 provides an indirect interface between graphics adapter 165 and graphics adapter 164 through the combination of connections 151 and 141. Connection 167 provides a direct connection between graphics adapter 165 and graphics adapter 164. In some embodiments of the present invention, connection 167 is omitted. Switch 160 may also include interfaces to other devices.
In some embodiments the present invention, transfers over connections 141 and 151 are performed by switch 160 using an industry standard protocol such as PCI-Express,™ and, in such cases, switch 160, graphics adapter 165 and graphics adapter 164, each include an interface unit corresponding to the industry standard protocol.
A primary graphics processor 140 within graphics adapter 164 outputs image data to a display 170. Display 170 may include one or more display devices, such as a cathode ray tube (CRT), flat panel display, or the like. Primary graphics processor 140 within graphics adapter 164 is also coupled to a primary frame buffer 145, which may be used to store graphics data, image data, and program instructions. At least a portion of primary frame buffer 145 is designated as frame buffer memory and is reported to a BIOS 102 during bootup of system 100 by graphics adapter 164 as a prefetchable memory space requirement. A graphics processor 150 within graphics adapter 165 is coupled to a frame buffer 155, which may also be used to store graphics data, image data, and program instructions. Likewise, at least a portion of frame buffer 155 is designated as frame buffer memory and is reported to BIOS 102 by graphics adapter 165 as a prefetchable memory space requirement.
Graphics driver 105 may configure graphics processor 150 and primary graphics processor 140 such that the graphics processing workload performed by system 100 is divided between graphics processor 150 and primary graphics processor 140 to produce the image data. For example, graphics processor 150 may process a larger portion of an image than primary graphics processor 140. In some embodiments of the present invention, graphics processor 150 may process the entire image and primary graphics processor 140 may receive the image data from graphics processor 150 via switch 160 or via connection 167. In other embodiments of the present invention, host processor 120 controls the transfer of the image data from graphics processor 150 to primary graphics processor 140.
Although system 100 as shown is a graphics processing system, alternate embodiments of system 100 may process other types of data, such as audio data, multi-media data, or the like. In those alternate embodiments, graphics processor 150 and primary graphics processor 140 would be replaced with other appropriate data processing devices. Likewise, graphics driver 105 would be replaced with a device driver corresponding to the data processing device.
An example of a multi-adapter configuration is a scalable link interface (SLI) configuration that permits multiple graphics devices to produce and combine image data for a single display device. In some embodiments of the present invention, multi-adapter configuration is fixed when system 100 is manufactured. For example, the multi-adapter configuration information may be fixed by providing hardwired inputs to each graphics processor. In other embodiments of the present invention, the multi-adapter configuration information is dynamic and is updated when a second or additional graphics adapter is installed in system 100. The multi-adapter configuration information, provided by each graphics device, may include an indication that one of the multiple graphics devices, such as graphics adapter 164 is a primary, e.g., master, graphics device that generates an output to display 170.
In step 210 the multiple graphics devices provide memory space requirements to BIOS 102. The reported memory space requirement for a graphics processor may include a register space that is non-prefetchable and other memory space, such as frame buffer memory, that is prefetchable. In some embodiments of the present invention, BIOS 102 may read registers within graphics processor 150 and primary graphics processor 140 to determine memory space requirements for each graphics processor. The amount of memory space reported to graphics driver 105 by each graphics processor via the registers read by BIOS 102, may vary dependent on the multi-adapter configuration.
A graphics processor, such as graphics processor primary graphics processor 140 may report a bloated register space including the register space needed by graphics processor 150, as described in conjunction with
In step 220 switch 160 is programmed by graphics driver 105 to translate accesses within the device address space into accesses to a single graphics device or into accesses to the multiple graphics devices. Specifically, switch 160 is programmed to remap accesses within a bloated memory space, (register space or frame buffer memory space) to a single graphics device. For example, within the bloated space a remapping aperture is specified and accesses lying within the remapping aperture are directed to graphics adapter 165 and accesses lying outside of the remapping aperture are directed to graphics adapter 164.
Graphics adapter 164 may be configured to report a bloated register space based on fixed or dynamic multi-adapter configuration information. For example, signals provided by pull-down or pull-up components, programmable registers, fuses, jumpers, switches, or the like, may be detected on at least one input to primary graphics processor 140. The signals may specific register space or memory space and whether or not the register or memory space should be reported to BIOS 102 as bloated. Likewise, signals may be detected on at least one input to graphics processor 150 which configure graphics adapter 165 to report register space or memory space that is not bloated.
Register space 310 may be bounded by addresses A0 and A1 and register space 311 may be bounded by addresses A1 and A2. Accesses received by switch 160 with addresses lying within register space 311 are redirected to graphics adapter 165. For example, in some embodiments of the present invention, addresses greater than or equal to A1 and less than A2 are translated into corresponding addresses within register space 321 which is bounded by addresses B0 and B1. In some embodiments of the present invention, register spaces 321 and 310 are used for configuration registers and are programmed by an application program via graphics driver 105.
Graphics adapter 164 reports another bloated register space 306 including a register space 316 that is specific to graphics adapter 164 and a register space 317 that is specific to graphics adapter 165. Graphics adapter 165 reports a register space that is not bloated, register space 327. For example, register space 316 is bounded by addresses I0 and I1 and register space 317 is bounded by physical addresses I1 and I2. Accesses received by switch 160 with addresses lying within register space 317 are remapped to graphics adapter 165. In some embodiments of the present invention, additional register spaces are reported by graphics adapter 164 and graphics adapter 165. In some embodiments of the present invention, register spaces 316 and 327 are instance registers used to store the locations of data structures stored in primary frame buffer 145 and frame buffer 155, respectively.
Graphics adapter 164 reports a frame buffer memory space 315 that is not bloated. As previously described, the multi-adapter configuration information may specify whether or not to bloat register spaces separately from frame buffer memory spaces. Graphics adapter 165 reports a frame buffer memory space 325 that is also not bloated. Frame buffer memory space 315 may be configured as a broadcast memory space, such that all accesses to addresses within frame buffer memory space 315 are broadcast to corresponding addresses in both graphics adapter 164 and graphics adapter 165. Specifically, frame buffer memory space 315 may be configured to broadcast write accesses and not broadcast read accesses. In some embodiments of the present invention, switch 160 is configured to transmit accesses to addresses within frame buffer memory space 315 only to frame buffer memory space 325, e.g., graphics adapter 165. In other embodiments of the present invention, a portion of frame buffer memory space 315 is configured as a redirection region as described in conjunction with
As previously described in conjunction with
Accesses may also be redirected to another register space within a graphics device. For example, Accesses received by switch 160 with addresses lying within register space 362 are redirected to register space 366 within graphics adapter 164. In some embodiments of the present invention, addresses greater than or equal to A2 and less than A3 are translated into corresponding addresses within register space 366 which is bounded by addresses I0 and I1.
Graphics adapter 164 reports a memory space 365 that is not bloated. As previously described, the multi-adapter configuration information may specify whether or not to bloat register spaces separately from frame buffer memory spaces. Graphics adapter 165 reports a frame buffer memory space 375 that is also not bloated. All accesses to addresses within frame buffer memory space 365 may be broadcast to frame buffer memory space 375 so that accesses within frame buffer memory space 365 are transmitted to corresponding addresses in both graphics adapter 164 and graphics adapter 165.
A set of registers within remap registers 402 is programmed by graphics driver 105 to store a value defining a remapping aperture within a register memory space or a frame buffer memory space. Specifically, a first set of registers may store a value, such as A1, defining the remapping aperture within bloated register space 300 shown in
Memory access unit 405 determines if an access received from chipset 130 lies within a remapping aperture defined by one or more values stored in remap registers 402. In addition to a value defining a remapping aperture, each set of registers within remap registers 402 may also include a register storing a remap base address for each port of switch 160. The remap base address for a particular port is used by memory access unit 405 to compute a remapped address for the particular port, as described in conjunction with
In step 410 routing unit 400 determines if the address is within a remapping aperture. Specifically, memory access unit 405 compares the address with one or more remapping apertures defined by values stored in remap registers 402. If, in step 410 memory access unit 405 determines that the address is not within a remapping aperture, then in step 411 switch 160 outputs the access to the graphics device corresponding to the memory space. If, in step 410 memory access unit 405 determines that the address is within a remapping aperture, then memory access unit 405 proceeds to access remap steps 450 and in step 412 memory access unit 405 determines if the access is a write access.
If, in step 412 memory access unit 405 determines the access is not a write access, then in step 414 memory access unit 405 determines if port0 is read enabled, i.e., if read accesses are enabled for port0. If, in step 414 memory access unit 405 determines that port0 is read enabled, then in step 416 memory access unit 405 remaps the access received in step 407 to port0.
In some embodiments of the present invention, memory access unit 405 computes a remapped address for port0, port0_address, using the following equation:
port0_address=port0 base+address−remap base address,
where port0 base is the physical base address of the port0 memory address space, address is the address associated with the access received in step 407, and remap base address is the base address of the remapping aperture specified by a set of registers within remap registers 402. In some embodiments of the present invention, the high bits of port0_address, such as bits 32 through 63, are equal to the high bits of port base 255 and the low bits of port0_address, e.g., [31:0], are computed using the previous equation. In some embodiments of the present invention, port0 base is defined for each remap aperture.
If, in step 414 memory access unit 405 determines that port0 is not read enabled, then in step 418 memory access unit 405 determines if port1 is read enabled. If in step 418 memory access unit 405 determines that port1 is read enabled, then in step 426 memory access unit 405 remaps the access to port1. If, in step 418 memory access unit 405 determines that port1 is not read enabled, then in step 428 the access is not remapped to port1.
As previously described in conjunction with step 416 regarding port0, in some embodiments of the present invention, memory access unit 405 computes a remapped address for port1, port1_address, using the following equation:
port1_address=port1 base+address−remap base address,
where port1 base is the physical base address of the port1 memory address space, address is the address associated with the access received in step 407, and remap base address is the base address of the remapping aperture specified by a set of registers within remap registers 402. In some embodiments of the present invention, the high bits of port1_address, such as bits 32 through 63, are equal to the high bits of port base 255 and the low bits of port1_address, e.g., [31:0], are computed using the previous equation. In some embodiments of the present invention, port1 base is defined for each remap aperture.
If, in step 412 memory access unit 405 determines the access is a write access, then in step 420 memory access unit 405 determines if port0 is write enabled i.e., if write accesses are enabled for port0. If, in step 420 memory access unit 405 determines that port0 is write enabled, then in step 422 memory access unit 405 remaps the access to port0 and proceeds to step 424. If, in step 420 memory access unit 405 determines that port0 is not write enabled, then in step 424 memory access unit 405 determines if port1 is write enabled. If, in step 424 memory access unit 405 determines that port1 is write enabled, then in step 426 memory access unit 405 remaps the access to port1. If, in step 424 memory access unit 405 determines that port1 is not write enabled, then in step 428 the access is not remapped to port1.
Therefore, accesses to a bloated memory space may be transmitted by switch 160 to one or more of the graphics devices dependent on write and read enables specified for each of the graphics devices. Specifically, accesses within a remapping aperture may be redirected to one graphics device, such as graphics adapter 165, and accesses outside of the remapping aperture are output to another graphics device, the graphics device reporting the bloated register space requirement. While read accesses are remapped to a single graphics device, write accesses within a remapping aperture, may be broadcast to all of the graphics devices by enabling write access for all of the graphics devices.
The device address space includes a non-prefetchable memory space, e.g., register memory, and a prefetchable memory space, e.g., frame buffer memory. In addition to remapping apertures, the prefetchable memory space may include one or more redirection regions. Accesses within a redirection region of the prefetchable memory space are only transmitted to a single graphics device. Switch 160 may be programmed to specify redirection regions within the prefetchable memory space to permit write accesses to only one of the multiple graphics devices while write accesses to portions of the prefetchable memory space lying outside of any of the redirection regions and within a remap aperture may be broadcast to all of the multiple graphics devices. Therefore, a single write access may be broadcast to write each frame buffer, thereby using less system bandwidth compared with separately writing each frame buffer. However, regions of the prefetchable memory may be identified for redirection to a single graphics device so that multiple graphics devices may each access a single surface memory space.
Within prefetchable memory space 500, redirection regions may be specified, such as surface memory space 510, 511, 512, and 513. Data stored in a surface memory space is typically image data intended for display on a display device. An access that lies within a redirection region is only transmitted to a single graphics device by switch 160. A redirection region may be further divided into portions and each portion may correspond to a particular graphics device, as described below in conjunction with
Within prefetchable memory space 500 one or more remap memory spaces, such as remap memory space 501, may be defined. Remap memory space 501 is bounded by remap base 505 and remap limit 505. An unused memory space 502 includes any memory space within prefetchable memory 500 that is not defined as remap memory space or surface memory space.
An access that lies within remap memory space 501 and outside of a surface memory space may be transmitted, i.e., broadcast, to each graphics device coupled to an enabled port of switch 160. Performing write operations within remap memory space 501 is an efficient mechanism for transferring graphics data, such as texture map data or program instructions, to all of the graphics devices. Each graphics device, such as graphics processor 150 or primary graphics processor 140, stores graphics data that is broadcast to a local frame buffer, such as frame buffer 155 or primary frame buffer 145, respectively.
Although surface 520 is split into two contiguous portions, in other embodiments of the present invention, surface 520 may be divided into different portions. For example, one frame buffer may provide the even scanline image data and another frame buffer may provide the odd scanline image data. Alternatively, each frame buffer may provide a portion of the image data defined by specific x,y pixel positions. Furthermore, each frame buffer does not have to provide an equal portion of the image data; rather processing of the image data may be distributed between the multiple graphics devices based on several factors, including the processing performance of each particular graphics device and the complexity of different portions of the image.
Addresses for a redirection limit 535, redirection base 537, and crossover 536 are programmed by graphics driver 105 into switch 160. Switch 160 determines to which port(s) to output each access based on the address associated with the access, as described in further detail below in conjunction with
Port base 553 and port limit 555 are determined by BIOS 102 based on memory configuration information reported by a graphics device, such as graphics adapter 164 or 165, coupled to switch 160. Switch 160 is programmed with physical address information to translate addresses within prefetchable memory space 500 to physical addresses within one or more port memory spaces, such as port memory space 550.
Each surface memory space within port memory 550 corresponds to a specific redirection region within prefetchable memory space 500. For example, in one embodiment of the present invention, surface memory spaces 560, 562, 563, and 564 correspond to surface memory spaces 510, 511, 512, and 513, respectively. Likewise, remap memory space 561 and unused memory space 565 within port memory 550 correspond to remap memory space 501 and unused memory space 502, respectively, within prefetchable memory space 500. In some embodiments of the present invention, the organization of surface and/or remap memory spaces within port memory 550 may be arbitrary because each surface and/or remap memory space within port memory 550 has a specific base address.
Graphics driver 105 programs values stored in redirection registers 602 and translation registers 603. Redirection registers 602 store values, e.g., addresses, specifying one or more redirection regions within the prefetchable memory space. In some embodiments of the present invention, redirection registers 602 stores 32 sets of values, e.g., base, limit, and at least one crossover, to support 32 application programs. Translation registers 603 store one or more values, e.g., addresses, each value specifying a physical address, e.g. port base, for one port of switch 160. For example, a port0 base address value may correspond to the base address of the frame buffer coupled to port0 of switch 160 via a graphics device.
A memory access unit 605 uses the one or more values programmed in BAR1 601, remap registers 402, and redirection registers 602 to determine whether an access received by switch 160 from chipset 130 should be output to any, all, or only one of port0 or port1, as described below in conjunction with
If, in step 612, memory access unit 605 determines the address is within the non-prefetchable memory space, then in step 618 memory access unit 605 determines if the address is within a remapping aperture. If, in step 618 memory access unit 650 determines that the address is within a remapping aperture, then in step 620 memory access unit 650 performs the access remap steps 450 as described in conjunction with
If, in step 614, memory access unit 605 determines the address is within the prefetchable memory space, then in step 616 memory access unit 605 determines if the address is within a redirection region specified by a set of values stored in redirection registers 602. For example, memory access unit 605 may compare high bits of the address, such as bits 32 through 63, i.e., [63:32], to the high bits of source limit 517 and determine that the address is within the redirection region when those bits are equal. In some embodiments of the present invention, the high bits of source limit 517 are the same as the high bits of source base 515.
If, in step 616 memory access unit 605 determines the address is not within any redirection region, then memory access unit 605 proceeds to step 618 to determine if the address is within a remapping aperture. If, in step 616 memory access unit 605 determines the address is within a particular redirection region, then in step 624 memory access unit 605 determines is access is enabled for the particular redirection region. In some embodiments of the present invention access may be independently enabled or disabled for each of the 32 programmable redirection regions. If, in step 624 memory access unit 605 determines is access is not enabled for the particular redirection region, then in step 626 memory access unit 605 ignores the access and does not transmit the access to either port of switch 160.
If, in step 624 memory access unit 605 determines is access is enabled for the particular redirection region, then in step 628 memory access unit 605 determines if the address associated with the access is less than a crossover value specified for the particular redirection region. If, in step 628 memory access unit 605 determines the address associated with the access is less than the crossover value, then in step 630 memory access unit 605 reads a port0 base address value stored in translation registers 603 and computes a physical address for port0. In step 632, memory access unit 605 outputs the access, including the port0_address, to port0 of switch 160. In some embodiments of the present invention, memory access unit 605 computes an address for port0, port0_address, using the following equation:
port0_address=port0 base+address−source base,
where port0 lower base is the physical base address of the port0 prefetchable memory address space, address is the address associated with the access received in step 610, and source base is the base address of the source prefetchable memory address space specified by BAR1 601. In some embodiments of the present invention, the high bits of port0_address, such as bits 32 through 63, are equal to the high bits of port base 553 and the low bits of port0_address, e.g., [31:0], are computed using the previous equation.
If, in step 628 memory access unit 505 determines the address associated with the access is not less than the crossover value, then in step 634 memory access unit 605 reads a port1 base address value stored in translation registers 603 and computes a physical address for port1, port1_address. In step 636, memory access unit 605 outputs the access, including the port1_address, to port1 of switch 160. As previously described in conjunction with step 630 (regarding port0), in some embodiments of the present invention, memory access unit 605 computes a physical address for port1, port1_address, using the following equation:
port1_address=port1 base+address−source base,
where port1 base is the physical base address of the port1 prefetchable memory address space, address is the address associated with the access received in step 610, and source base is the base address of the source prefetchable memory address space specified by BAR1 601. In some embodiments of the present invention, the high bits of port1_address, such as bits 32 through 63, are equal to the high bits of port base 553 and the low bits of port1_address, e.g., [31:0], are computed using the previous equation. In other embodiments of the present invention, additional physical port addresses may be computed for additional ports. Furthermore, in other embodiments of the present invention, different techniques, known to those skilled in the art, may be used to convert addresses received from chipset 130 into physical addresses for each port of switch 160.
Read or write accesses within a redirection region of a prefetchable memory space are translated into accesses to a single port of switch 160, specifically to the single port coupled to the frame buffer storing image data for the portion of an image stored within the redirection region. Write accesses outside of any redirection region, but within a remap memory space, e.g., remapping aperture, may be broadcast to all of the ports of switch 160. Therefore, frame buffer writes to all of the graphics devices coupled to switch 160 may efficiently performed through the remap address spaces. A unified frame buffer is presented to an application program for all frame buffer accesses because switch 160 performs the address translation for and routing to one or all of the multiple graphics devices processing and storing image data in the multiple frame buffers.
In alternate embodiments of the present invention, the graphics processors may be replaced with other types of processors, such as audio processors, multi-media processors, or the like. Likewise, graphics driver 705 may be replaced with another driver corresponding to the other types of processors. Therefore, application programs need only communicate with a single driver to distribute processing over multiple processors.
Each of the multiple graphics devices within the graphics subsystems reports non-prefetchable memory space requirements, e.g., register space, and prefetchable memory space requirements, e.g., frame buffer memory space, to BIOS 702. Memory space is assigned to the multiple graphics devices, and remap registers 402, BAR1 601, redirection registers 602, and translation registers 603, and the like, within switch 760 and within each switch 160 are programmed to define the prefetchable memory space, non-prefetchable memory space, remapping apertures, redirection regions, and port base addresses for address translation. Graphics driver 705 communicates between the multiple graphics devices and application programs, in order to present the multiple frame buffers as a unified frame buffer.
Register spaces 812 and 813 are also included within bloated register space 800 and are specific to the second graphics subsystem (coupled to switch 760 via connection 751) in system 700. Physical storage for register space 812 is within a third graphics device within the second graphics subsystem. Physical storage for register space 813 is within a fourth graphics device that is also within the second graphics subsystem. The third graphics device reports a bloated register space 830 to BIOS 702. Bloated register space 830 includes a register space 832 corresponding to register space 812 reported by the first graphics device and a register space 833 corresponding to register space 813 reported by the first graphics device. The fourth graphics device reports a register space that is not bloated to BIOS 702, register space 843.
As previously described, each graphics device may be configured to report a bloated register space or frame buffer memory space based on fixed or dynamic configuration information. In the configuration shown in
Register spaces 812 and 813 within bloated register space 800 may be bounded by addresses A0 and A1 and register space 811 may be bounded by addresses A2 and A4. A first switch 160 within the first graphics subsystem, a second switch 160 within the second graphics subsystem, and switch 760, may each be programmed by a single graphics driver, graphics driver 705 to define at least one remapping aperture or redirection region. Accesses received by switch 760 with associated addresses lying within register spaces 812 and 813 are remapped to the second graphics subsystem and correspond to bloated register space 830. Accesses received by switch 760 with associated addresses lying within register spaces 810 and 811 are remapped to the first graphics subsystem.
Register space 810 may be bounded by addresses A0 and A1 and register space 811 may be bounded by addresses A1 and A2. Accesses received by first switch 160 from switch 760, with associated addresses lying within register space 811 are remapped to the second graphics device within the first graphics subsystem. For example, in some embodiments of the present invention, addresses greater than or equal to A1 and less than A2 are translated into corresponding addresses within register space 821 which is bounded by addresses B0 and B1. In some embodiments of the present invention, the address associated with the access is translated using values stored in registers within first switch 160 of the first graphics subsystem to produce a remapped address within register space 821.
Register space 812 may be bounded by addresses A2 and A3 and register space 813 may be bounded by addresses A3 and A4. Switch 760 remaps accesses to register spaces 812 and 813 to the second graphics device. Specifically, register space 812 corresponds to register space 832 which may be bounded by addresses C0 and C1 and register space 813 corresponds to register space 833 which may be bounded by addresses C1 and C2. In some embodiments of the present invention, the addresses associated with an access that lies within register spaces 812 and 813 are translated using values stored in registers within switch 760 to produce remapped addresses within register space 830.
Accesses received by the second switch 160 from switch 760, with associated addresses lying within register space 833 are remapped to the second graphics device within the first graphics subsystem. For example, in some embodiments of the present invention, addresses greater than or equal to C1 and less than C2 are translated into corresponding addresses within register space 843 which is bounded by addresses D0 and D1. In some embodiments of the present invention, register spaces 843, 832, 821, and 810 are used for configuration registers and are programmed by an application program via graphics driver 705.
The first graphics device within the first graphics subsystem also reports a bloated register space 801 including a register space 816 that is specific to the first graphics device, a register space 817 that is specific to the second graphics device, a register space 818 that is specific to the third graphics device, a register space 819 that is specific to the fourth graphics device. The third graphics device within the second graphics subsystem reports a bloated register space 831, including register spaces 838 and 839 corresponding to register spaces 818 and 819, respectively. The first graphics device includes physical storage for register space 816, physical storage for register space 817 is within the second graphics device, physical storage for register spaces 818 and 819 is within the second graphics subsystem. The second graphics device and the fourth graphics device each report a register space that is not bloated, register spaces 827 and 849, respectively. In some embodiments of the present invention, additional register spaces are reported by graphics subsystems within system 700. In some embodiments of the present invention, register spaces 816, 827, 838, and 849 are instance registers used to store the locations of data structures stored in prefetchable memory, e.g., frame buffer memory.
The first and second graphics subsystems each report a frame buffer memory space that is not bloated, frame buffer memory spaces 815 and 835, respectively. As previously described, the multi-adapter configuration information may specify whether or not to bloat register spaces separately from frame buffer memory spaces. Switch 760 may be programmed to define one or more redirection regions within frame buffer memory space 815.
In some embodiments of the present invention, switch 760 is programmed to define a remapping aperture within frame buffer memory space 815 such that, all accesses within frame buffer memory space 815 are broadcast to frame buffer memory space 835 and frame buffer memory space 815. Likewise, first switch 160 within the first graphics subsystem may be programmed to define a remapping aperture within frame buffer memory space 815 such that all accesses within frame buffer memory space 815 are broadcast to frame buffer memory space 825 and frame buffer memory space 815. Similarly, second switch 160 within the second graphics subsystem may be programmed by graphics driver define a remapping aperture within frame buffer memory space 815 such that all accesses within frame buffer memory space 835 are broadcast to a frame buffer memory space 845 and frame buffer memory space 835. Therefore, accesses to frame buffer memory space 815 may be received by each graphics adapter in system 700. Furthermore, redirection regions may be defined within frame buffer memory space 815 such that each of the graphics devices may store a portion of a surface memory space.
In alternate embodiments of the present invention, the graphics processors may be replaced with other types of processors, such as audio processors, multi-media processors, or the like. Likewise, graphics driver 905 may be replaced with another driver corresponding to the other types of processors. Therefore, application programs need only communicate with a single driver to distribute processing over multiple processors.
If, in step 946 the memory access unit determines the address associated with the access is not less than the first crossover value, then in step 950 the memory access unit determines if the address associated with the access is less than a second crossover value. If, in step 950 the memory access unit determines the address associated with the access is less than the second crossover value, then in step 951 the memory access unit uses the address received in step 940 to compute a port specific physical address for port1 of switch 960. In step 952 the memory access unit outputs the access, including the computed physical address for port1 to graphics adapter 965 that is coupled to the port1 of switch 960.
If, in step 950 the memory access unit determines the address associated with the access is not less than the second crossover value, then in step 953 the memory access unit determines if the address associated with the access is less than a third crossover value. If, in step 953 the memory access unit determines the address associated with the access is less than the third crossover value, then in step 954 the memory access unit uses the address received in step 940 to compute a port specific physical address for port2 of switch 960. In step 955 the memory access unit outputs the access, including the computed physical address for port2 to graphics adapter 965 that is coupled to the port2 of switch 960.
If, in step 953 the memory access unit determines the address associated with the access is not less than the third crossover value, then in step 956 the memory access unit uses the address received in step 940 to compute a port specific physical address for port3 of switch 960. In step 957 the memory access unit outputs the access, including the computed physical address for port3 to graphics adapter 965 that is coupled to the port3 of switch 960.
Graphics adapters 965 and graphics adapter 964, process accesses received from switch 960 via port0, port1, port2, and port3, according to the steps described in conjunction with
Each of the multiple graphics devices reports memory space requirements for register space and frame buffer space and a single device driver, graphics driver 905 is used to communicate between an application program and the multiple graphics devices. In some embodiments of the present invention, graphics adapter 964 may report a bloated memory space requirement, including memory space for each of the graphics adapters 965. Furthermore, redirection regions may be defined within frame buffer memory so that the application program can distribute processing of an image between the multiple graphics devices. Image data for a surface may be processed by the multiple graphics devices, with portions of the surface residing in separate frame buffers, each frame buffer coupled to one of the multiple graphics devices.
While the foregoing is directed to embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof, and the scope thereof is determined by the claims that follow. Specifically, persons skilled in the art will recognize that the methods and systems described may be used for processing data other than graphics data where the data is used by processors in a multi-processing data processing system. The foregoing description and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. The listing of steps in method claims do not imply performing the steps in any particular order, unless explicitly stated in the claim.
All trademarks are the respective property of their owners.
Number | Name | Date | Kind |
---|---|---|---|
3492538 | Fergusson | Jan 1970 | A |
4647123 | Chin et al. | Mar 1987 | A |
5448655 | Yamaguchi | Sep 1995 | A |
5502808 | Goddard et al. | Mar 1996 | A |
5522027 | Matsumoto et al. | May 1996 | A |
5784628 | Reneris | Jul 1998 | A |
5794016 | Kelleher | Aug 1998 | A |
5936640 | Horan et al. | Aug 1999 | A |
5999198 | Horan et al. | Dec 1999 | A |
6023281 | Grigor et al. | Feb 2000 | A |
6111757 | Dell et al. | Aug 2000 | A |
6191800 | Arenburg et al. | Feb 2001 | B1 |
6195734 | Porterfield | Feb 2001 | B1 |
6253299 | Smith et al. | Jun 2001 | B1 |
6296493 | Michiya | Oct 2001 | B1 |
6326973 | Behrbaum et al. | Dec 2001 | B1 |
6329996 | Bowen et al. | Dec 2001 | B1 |
6473086 | Morein et al. | Oct 2002 | B1 |
6501999 | Cai | Dec 2002 | B1 |
6535939 | Arimilli et al. | Mar 2003 | B1 |
6593932 | Porterfield | Jul 2003 | B2 |
6631474 | Cai et al. | Oct 2003 | B1 |
6633296 | Laksono et al. | Oct 2003 | B1 |
6683614 | Walls et al. | Jan 2004 | B2 |
6711638 | Wu | Mar 2004 | B1 |
6750870 | Olarig | Jun 2004 | B2 |
6760031 | Langendorf et al. | Jul 2004 | B1 |
6882346 | LeFebvre et al. | Apr 2005 | B1 |
6902419 | Conway et al. | Jun 2005 | B2 |
6919896 | Sasaki et al. | Jul 2005 | B2 |
7030837 | Vong et al. | Apr 2006 | B1 |
7176847 | Loh | Feb 2007 | B2 |
7184003 | Cupps et al. | Feb 2007 | B2 |
20020047851 | Hirase et al. | Apr 2002 | A1 |
20020093507 | Olarig | Jul 2002 | A1 |
20020105523 | Behrbaum et al. | Aug 2002 | A1 |
20020118201 | Mukherjee et al. | Aug 2002 | A1 |
20030128216 | Walls et al. | Jul 2003 | A1 |
20030137483 | Callway | Jul 2003 | A1 |
20040072460 | Conway et al. | Apr 2004 | A1 |
20040104913 | Walls et al. | Jun 2004 | A1 |
20050017980 | Chang et al. | Jan 2005 | A1 |
20050088445 | Gonzalez et al. | Apr 2005 | A1 |
20050134588 | Aila et al. | Jun 2005 | A1 |
20050160212 | Caruk | Jul 2005 | A1 |
20050278559 | Sutardja et al. | Dec 2005 | A1 |
Number | Date | Country |
---|---|---|
0428277 | May 1991 | EP |
0571969 | Dec 1993 | EP |
2834097 | Jun 2003 | FR |
2839563 | Nov 2003 | FR |
5324583 | Dec 1993 | JP |
328392 | Mar 1998 | TW |
570243 | Jan 2004 | TW |
WO 03083636 | Oct 2003 | WO |