Contiguously packed data

• US 8,780,128 B2
• Filed: 12/17/2007
• Issued: 07/15/2014
• Est. Priority Date: 12/17/2007
• Status: Active Grant

• Current Assignee: NVIDIA Corporation
• Original Assignee: NVIDIA Corporation
• Inventors: Toksvig, Michael J. M., Mahan, Justin Michael, Mills, Christopher L.
• Primary Examiner(s): Tung, Kee M
• Assistant Examiner(s): Wu, Sing-Wai
• Application Number:
  US12/002,641
• Publication Number:
  US 20090157963A1
• Time in Patent Office: 2,402 Days
• Field of Search: 345/503
• US Class Current: 345/557
• CPC Class Codes: G06F 12/04 Addressing variable-length words or parts of words
• CPC Class Codes: G06F 12/0886 Variable-length word access
• CPC Class Codes: G06T 1/60 Memory management

Abstract

Data for data elements (e.g., pixels) can be stored in an addressable storage unit that can store a number of bits that is not a whole number multiple of the number of bits of data per data element. Similarly, a number of the data elements can be transferred per unit of time over a bus, where the width of the bus is not a whole number multiple of the number of bits of data per data element. Data for none of the data elements is stored in more than one of the storage units or transferred in more than one unit of time. Also, data for multiple data elements is packaged contiguously in the storage unit or across the width of the bus.

Claims

1. A computer-implemented method of managing data, said method comprising:

accessing data for a plurality of discrete data elements, wherein each of said data elements comprises a first number of bits of data; and

storing said data in a memory comprising a plurality of addressable storage units including a first storage unit with a capacity of at least 64 bytes, wherein 21 24-bit data elements are stored contiguously in said first storage unit.

2. The method of claim 1 wherein said first storage unit has a capacity of at least 256 bytes, wherein 85 24-bit data elements are stored contiguously in said first storage unit.

3. A computer-implemented method of managing data, said method comprising:

accessing data for a plurality of discrete data elements, wherein each of said data elements comprises a first number of bits of data; and

storing said data in a cache comprising a plurality of cache lines, each of said cache lines capable of storing a second number of bits that is not a whole number multiple of said first number of bits, wherein data for none of said data elements is stored in more than one of said cache lines, wherein data for multiple data elements is stored contiguously in a cache line, and wherein said second number of bits comprises 256 bits, wherein a first set of five 24-bit data elements is stored contiguously in a cache line of said plurality of cache lines and a second set of five 24-bit data elements is also stored contiguously in said cache line, said first set and said second set separated by a contiguous block of eight unused bits.

4. The method of claim 3 further comprising retrieving data for a selected data element from said cache by computing a location for said selected data element, said location computed by:

i) dividing a number associated with said selected data element by the maximum number of data elements that can be stored contiguously in a segment of bits to determine a floor and a remainder, ii) multiplying said floor by the number of bytes in said segment to determine a first result, iii) multiplying said remainder by the number of bytes per data element to determine a second result, and iv) adding said first and second results.

5. A computer-implemented method of managing data, said method comprising:

accessing data for a plurality of discrete data elements, wherein each of said data elements comprises a first number of bits of data; and

transferring a number of said data elements, comprising not more that said first number of bits, per unit of time over a bus, wherein said bus is capable of transferring a second number of bits per said unit of time, wherein said second number is not a whole number multiple of said first number of bits, wherein data for none of said data elements is transferred in more than one said unit of time, and wherein said second number of bits comprises 256 bits, wherein a first set of five 24-bit data elements is packaged contiguously in said bus during a first unit of time and a second set of five 24-bit data elements is also packaged contiguously in said bus during said first unit of time, said first set and said second set separated by a contiguous block of eight unused bits.