Method and Apparatus for Pseudo-Aligned Transfers of Data to Memory Wherein a Re-Alignment Is Performed Based on the Data Byte Control Header

1. Field of the Invention

The present invention relates in general to transfers of data in computer memory, and in particular to a method and apparatus for pseudo-aligned transfers of data to memory.

2. Description of Related Art

In computer systems, data is often transferred between processors, memories, and peripheral devices via an input/output (I/O) bus. Standardized buses such as the ISA, EISA, or Micro Channel.TM. buses have long been used in computer systems to provide a common I/O interface to peripherals across different platforms and different processors. However, there are a number of problems associated with data transfers across such buses.

One problem with data transfers across buses involves unaligned transfers of data. Unaligned transfers occur when a block of data bytes is not aligned to a word or double word boundary at the source and/or the destination. Such misalignment can incur significant processor or device overhead in shifting and/or masking the block of bytes during the transfer. For example, unaligned transfers in INTEL.RTM. 386 and 486 processors can be 50% to 250% slower than an equivalent aligned transfer of data. Thus, there is a need in the art for a method and apparatus for optimizing unaligned transfers of data to memory.

To overcome the limitations in the prior art described above, and to overcome other limitations that will become apparent upon reading and understanding the present specification, the present invention discloses a method and apparatus for pseudo-aligned transfers of data to memory. Alignment logic, typically coupled to a peripheral device, receives a plurality of data bytes from a processor. The alignment logic uses a control header transferred with the data bytes to determine whether the data bytes require re-alignment. To effect re-alignment, the alignment logic combines, rotates, and masks the data bytes as indicated by the control header.

Referring now to the drawings in which like reference numbers represent corresponding parts throughout:

FIG. 1 is an example that illustrates an aligned memory transfer between a processor and video memory;

FIG. 2 is an example that illustrates an unaligned memory transfer between a processor and video memory;

FIG. 3 is a block diagram illustrating the general hardware environment of the present invention;

FIG. 4 is a block diagram providing an expanded view of a video circuit in the present invention;

FIGS. 5A and 5B are a block diagram illustrating the components of alignment logic in the present invention; and

FIG. 6 is a block diagram illustrating the logic used by a software driver module in the present invention.

In the following description of the preferred embodiment, reference is made to the accompanying drawings which form a part hereof, and in which is shown by way of illustration a specific embodiment in which the invention may be practiced. It is to be understood that other embodiments may be used and structural changes may be made without departing from the scope of the present invention.

The most efficient INTEL.RTM. 386/486 instruction for performing data transfers is "REP MOVSD," which is a mnemonic for the instruction "REPeat MOVe String Double word." For its most efficient operation, this instruction requires that both the source and destination addresses be double word aligned, i.e., aligned on a 32-bit boundary. Byte (8-bit) and word (16-bit) transfers are also allowed and supported, but are less efficient because the I/O bus is 32 bits in width. However, many applications do not assure that data blocks are double word aligned, because they typically access the data blocks as a string of bytes.

The present invention provides for fast aligned transfers of data to memory, even in those cases where the data is unaligned. The present invention provides a method and apparatus for performing pseudo-aligned transfers of data to memory, wherein additional software and hardware perform unaligned data transfers as if they were aligned. The software generates a control header that instructs the hardware in a peripheral device to combine, rotate, and mask the bytes of an unaligned transfer to ensure that the bytes are properly aligned in the peripheral device.

FIGS. 1 and 2 are examples that illustrate the difference between aligned and unaligned data transfers. FIG. 1 shows an aligned transfer; and FIG. 2 shows an unaligned transfer. In both FIGS. 1 and 2, each box is a byte of data with double word aligned boundaries indicated by each horizontal group of four bytes.

FIG. 1 illustrates a typical double word aligned transfer of data. As shown, each double word S.sub.0 -S.sub.3, S.sub.4 -S.sub.7, S.sub.8 -S.sub.11, and S.sub.12 -S.sub.15 in the source memory remains intact as it is transferred to a corresponding double word D.sub.0 -D.sub.3, D.sub.4 -D.sub.7, D.sub.8 -D.sub.11, and D.sub.12 -D.sub.15 in the destination memory. Thus, this transfer maintains the double word alignment of the data as it is transferred.

FIG. 2 illustrates an example of an unaligned transfer of data. As shown, each byte S.sub.0 -S.sub.15 in the source memory changes its relative double word position as it is transferred to a corresponding byte D.sub.0 -D.sub.15 in the destination memory. In a typical case, combining, rotating and masking of bytes would be required to handle the change in alignment. However, these functions are usually performed by processors in a very inefficient manner.

FIG. 3 is a block diagram illustrating the general hardware environment of the present invention. A processor 10, e.g., an INTEL.RTM. 386 or 486 processor, is coupled to a video circuit 12 via an I/O bus 14 that transfers address, data, and control signals between the processor 10 and video circuit 12. The main purpose of the video circuit 12 is to control video memory 16 and CRT 18. The video memory 16 stores data from the processor 10, which can be later written to the CRT 18. The video circuit 12 arbitrates between data transfers from the processor 10 to the video memory 16, and from the video memory 16 to the CRT 18.

FIG. 4 is a block diagram providing an expanded view of the video circuit 12. The video circuit 12 includes arbitration logic 20, CRT controller logic 22, multiplexor 24, and alignment logic 26. The arbitration logic 20 controls the multiplexor 24 to arbitrate between the CRT controller logic 20 and the processor 10, and thus controls data transfers to and from the processor 10, video memory 16, and CRT 18. The CRT controller logic 22 provides the required addresses to retrieve data from the video memory 16 for display on the CRT 18. The alignment logic 26 ensures that the transfers from the processor 10 are correctly aligned in the video memory 16. With each transfer of data, the alignment logic 26 is used regardless of whether any alignment is actually required, although there is no combining, rotating or masking in an aligned transfer. This requirement that all data transfers go through the alignment logic 24 eliminates any timing problems and simplifies the control logic.

FIGS. 5A and 5B are a block diagram illustrating the components of the alignment logic 26 of the present invention. Data enters the alignment logic 26 on the four Data In (DIN) buses on the left side of FIGS. 5A and 5B, wherein each DIN bus comprises an 8-bit byte of data. Mask information enters the alignment logic 26 at the Mask In (MIN) bus on the left side of FIG. 5A, wherein each of the four signals in the MIN bus comprises a mask bit for the 8-bit byte of data on the corresponding DIN bus indicating whether the byte is part of the desired data or merely an extraneous byte carried along in the pseudo-aligned transfer. Control signals also enter the alignment logic 26 at the ALIGN.sub.-- LAT.sub.-- LO, DIR.sub.-- RIGHT, and ROTATE›1:0! signals on the left side of FIG. 5B.

Blocks 28a-28d are delay blocks for timing purposes. Blocks 30a-30d are 9-bit registers capable of latching each 8-bit byte of data from the respective DIN buses and a single mask bit from the MIN signals, when clocked by the ALIGN-LAT-LO control signal generated by the video circuit 12. Blocks 32a-32d are 2-to-1 multiplexors that select between the current data on the DIN buses and the output of blocks 30a-30d representing the previous data on the DIN buses, thereby allowing the selective combination of data bytes from two sequential double words into an output double word. The selection logic 34 for the blocks 32a-32d consists of NAND and NOR gates driven by the DIR.sub.-- RIGHT and ROTATE›1:0! control signals as described by the boolean logic in Tables IA-ID. Blocks 36a-36d are 9-bit registers that may be loaded from the outputs of any one of the four multiplexors 32a-32d to rotate the bytes within the output double word. The selection logic 38 for the blocks 36a-36d consists of NOR gates driven by the ROTATE›1:0! control signals as described by the boolean logic in Table II. The selection logic 38 for blocks 36a-36d perform the "barrel shifting" functions, wherein the bytes from each block 32a-32d may be stored in any of the blocks 36a-36d. The ROTATE›1:0! signals indicate the rotation of bytes, i.e., which bytes from blocks 32a-32d should be stored in each of the blocks 36a-36d.

Data exits the alignment logic 26 at the four Data Out (DOUT) buses on the right side of FIGS. 5A and 5B, wherein each DOUT bus comprises an 8-bit byte of data. Mask information also exits the alignment logic 26 at the Mask Out (MOUT) bus on the right side of FIG. 5B, wherein each of the four signals in the MOUT bus indicates whether the 8-bit byte of data on the corresponding DOUT bus should be stored in the video memory 16.

FIG. 6 is a block diagram illustrating the logic used by the software driver module of the present invention. The software driver module is executed on the processor 10, and controls the pseudo-aligned transfer of data from the processor 10. Block 40 represents the driver module receiving an application request to move a block of data from the processor 10 to the video memory 16, wherein the block is identified by its starting byte address and byte length.

Block 42 represents the driver module calculating and formatting a control header for the transfer containing the source address, destination address, byte length, and alignment control words. Each alignment control word comprises seven bits representing the ROTATE›1:0!, DIR-RIGHT, and MIN›3:0! signals for a corresponding double word. As described above, the ROTATE›1:0!, DIR.sub.-- RIGHT, and MIN›3:0! bits in the control header inform the alignment logic 26 of the true nature of the transfer, i.e., whether the bytes should be rotated, the direction of the rotation, and the masking of extraneous bytes. For example, the transfer in FIG. 2 would transfer five sequential double words, and would require a rotation of one byte to the right. Moreover, the two extraneous bytes to the left of S.sub.15 and the two extraneous bytes to the right of S.sub.0 would be masked.

Block 44 represents the driver module modifying the source address, destination address, and byte length presented to the processor 10 to ensure that aligned transfers are performed. Block 46 represents the driver module instructing the processor 10 to transfer the control header and the pseudo-aligned data block to the video circuit 12.

The video circuit 12 includes state machines and FIFOs that buffer up transfers from the processor 10 and then drive the DIN and MIN buses, as well as the ALIGN.sub.-- LAT.sub.-- LO, DIR.sub.-- RIGHT, and ROTATE›1:0! control signals into the alignment logic 26. The design of such state machines and FIFOs are well known in the art, and thus are not described further herein.

In summary, the present invention discloses a method and apparatus for pseudo-aligned transfers of data to memory. Alignment logic, typically coupled to a peripheral device, receives a plurality of data bytes from a processor. The alignment logic uses a control header transferred with the data bytes to determine whether the data bytes require re-alignment. To effect re-alignment, the alignment logic combines, rotates, and masks the data bytes as indicated by the control header.

The foregoing description of the preferred embodiment of the present invention has been presented only for the purposes of illustration and description. The foregoing description is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the teaching herein. For example, the following paragraphs describe some alternatives in accomplishing the same invention.

Those skilled in the art will recognize that the present invention is applicable to any device that has a memory and is not limited to video controllers. The video controller cited in the present specification is for illustrative purposes only and is not intended to be exhaustive or to limit the invention to the precise form disclosed.

Those skilled in the art will recognize that the present invention is applicable to systems with different configurations of devices and components. The example configurations of devices and components cited in the present specification are for illustrative purposes only and are not intended to be exhaustive or to limit the invention to the precise form disclosed.

Those skilled in the art will recognize that the present invention is applicable to I/O buses of different bit widths. The bit widths cited in the present specification are for illustrative purposes only and are not intended to be exhaustive or to limit the invention to the precise form disclosed.

Those skilled in the art will recognize that the present invention is applicable to different processors with different instruction sets. The processors and instruction sets cited in the present specification are for illustrative purposes only and are not intended to be exhaustive or to limit the invention to the precise form disclosed.

It is intended that the scope of the invention be limited not by this detailed description, but rather by the claims appended hereto.