Programmable streaming processor with mixed precision instruction execution

• US 8,633,936 B2
• Filed: 04/21/2008
• Issued: 01/21/2014
• Est. Priority Date: 04/21/2008
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Du, Yun, Yu, Chun, Jiao, Guofang, Molloy, Stephen
• Primary Examiner(s): Richer, Joni
• Application Number:
  US12/106,654
• Publication Number:
  US 20090265528A1
• Time in Patent Office: 2,101 Days
• Field of Search: 345/426, 345/501, 345/502, 345/522, 345/530, 345/536, 345/559, 345/561
• US Class Current: 345/522
• CPC Class Codes: G06F 8/47 Retargetable compilers
• CPC Class Codes: G06F 9/30014 with variable precision
• CPC Class Codes: G06F 9/30036 Instructions to perform operations on packed data, e.g. vector, tile or matrix operations
• CPC Class Codes: G06F 9/3851 from multiple instruction streams, e.g. multistreaming
• CPC Class Codes: G06F 9/3885 using a plurality of independent parallel functional units
• CPC Class Codes: G06F 9/3887 controlled by a single instruction for multiple data lanes [SIMD]
• CPC Class Codes: G06T 15/005 General purpose rendering architectures

Abstract

The disclosure relates to a programmable streaming processor that is capable of executing mixed-precision (e.g., full-precision, half-precision) instructions using different execution units. The various execution units are each capable of using graphics data to execute instructions at a particular precision level. An exemplary programmable shader processor includes a controller and multiple execution units. The controller is configured to receive an instruction for execution and to receive an indication of a data precision for execution of the instruction. The controller is also configured to receive a separate conversion instruction that, when executed, converts graphics data associated with the instruction to the indicated data precision. When operable, the controller selects one of the execution units based on the indicated data precision. The controller then causes the selected execution unit to execute the instruction with the indicated data precision using the graphics data associated with the instruction.

Claims

1. A method comprising:

receiving a graphics instruction for execution within a programmable streaming processor;

receiving an indication of a data precision for execution of the graphics instruction, wherein the indication of the data precision is contained within the graphics instruction, wherein the graphics instruction is a first executable instruction generated by a compiler that compiles graphics application instructions;

receiving a conversion instruction that, when executed by the programmable streaming processor, converts graphics data, associated with the graphics instruction, from a first data precision to converted graphics data having the indicated data precision, and wherein the conversion instruction is different than the graphics instruction, wherein the conversion instruction is generated by the compiler;

selecting one of a plurality of execution units within the processor based on the indicated data precision; and

using the selected execution unit to execute the graphics instruction with the indicated data precision using the converted graphics data associated with the graphics instruction.

2. The method of claim 1, further comprising:

receiving the graphics data associated with the graphics instruction;

generating a computation result with the indicated data precision during execution of the graphics instruction by the selected execution unit; and

providing the computation result as output.

3. The method of claim 1, wherein selecting one of the plurality of execution units comprises:

selecting one of a first set of one or more execution units within the processor that each execute instructions with the first data precision using the graphics data when the indicated data precision is the first data precision; and

selecting one of a second set of one or more execution units within the processor that each execute instructions with a second data precision using the graphics data when the indicated data precision is the second data precision, the second data precision being different than the first data precision.

4. The method of claim 3, wherein the first data precision comprises a full data precision, and wherein the second data precision comprises a half data precision.

5. The method of claim 1, wherein the execution units include a first set of one or more execution units within the processor that each execute instructions with the first data precision using the graphics data, and further include a second set of one or more execution units within the processor that each execute instructions with a second data precision different than the first data precision using the graphics data.

6. The method of claim 5, wherein:

selecting one of the plurality of execution units within the processor based on the indicated data precision comprises selecting one of the execution units in the first set; and

using the selected execution unit to execute the graphics instruction comprises using the selected execution unit in the first set to execute the graphics instruction with the indicated data precision using the graphics data associated with the graphics instruction.

7. The method of claim 6, further comprising:

receiving a second graphics instruction for execution within the processor;

receiving an indication of the second data precision for execution of the second graphics instruction;

receiving a second conversion instruction that, when executed by the processor, converts graphics data associated with the second graphics instruction to the indicated second data precision, the second conversion instruction being different than the second graphics instruction;

selecting one of the execution units in the second set based on the indicated second data precision; and

using the selected execution unit in the second set to execute the second graphics instruction with the indicated second data precision using the graphics data associated with the second graphics instruction.

8. The method of claim 1, wherein receiving the indication of the data precision for execution of the graphics instruction comprises decoding the graphics instruction to determine the data precision.

9. The method of claim 1, wherein the graphics data associated with the graphics instruction comprises at least one of vertex graphics data and pixel graphics data.

10. A non-transitory computer-readable storage medium comprising instructions for causing a programmable streaming processor to:

receive a graphics instruction for execution within the programmable streaming processor;

receive an indication of a data precision for execution of the graphics instruction, wherein the indication of the data precision is contained within the graphics instruction, wherein the graphics instruction is a first executable instruction generated by a compiler that compiles graphics application instructions;

receive a conversion instruction that, when executed by the processor, converts graphics data, associated with the graphics instruction, from a first data precision to converted graphics data having the indicated data precision, and wherein the conversion instruction is different than the graphics instruction, wherein the conversion instruction is generated by the compiler;

select one of a plurality of execution units within the processor based on the indicated data precision; and

use the selected execution unit to execute the graphics instruction with the indicated data precision using the converted graphics data associated with the graphics instruction.

11. The non-transitory computer-readable storage medium of claim 10, further comprising instructions for causing the processor to:

receive the graphics data associated with the graphics instruction;

generate a computation result with the indicated data precision during execution of the graphics instruction by the selected execution unit; and

provide the computation result as output.

12. The non-transitory computer-readable storage medium of claim 10, wherein the instructions for causing the processor to select one of the plurality of execution units comprise instructions for causing the processor to:

select one of a first set of one or more execution units within the processor that each execute instructions with the first data precision using the graphics data when the indicated data precision is the first data precision; and

select one of a second set of one or more execution units within the processor that each execute instructions with a second data precision using the graphics data when the indicated data precision is the second data precision, the second data precision being different than the first data precision.

13. The non-transitory computer-readable storage medium of claim 12, wherein the first data precision comprises a full data precision, and wherein the second data precision comprises a half data precision.

14. The non-transitory computer-readable storage medium of claim 10, wherein the execution units include a first set of one or more execution units within the processor that each execute instructions with the first data precision using the graphics data, and further include a second set of one or more execution units within the processor that each execute instructions with a second data precision different than the first data precision using the graphics data.

15. The non-transitory computer-readable storage medium of claim 14, wherein:

the instructions for causing the processor to select one of the plurality of execution units within the processor based on the indicated data precision comprise instructions for causing the processor to select one of the execution units in the first set; and

the instructions for causing the processor to use the selected execution unit to execute the instruction comprise instructions for causing the processor to use the selected execution unit in the first set to execute the graphics instruction with the indicated data precision using the graphics data associated with the graphics instruction.

16. The non-transitory computer-readable storage medium of claim 15, further comprising instructions for causing the processor to:

receive a second graphics instruction for execution within the processor;

receive an indication of the second data precision for execution of the second graphics instruction;

receive a second conversion instruction that, when executed by the processor, converts graphics data associated with the second graphics instruction to the indicated second data precision, the second conversion instruction being different than the second graphics instruction;

select one of the execution units in the second set based on the indicated second data precision; and

use the selected execution unit in the second set to execute the second graphics instruction with the indicated second data precision using the graphics data associated with the second graphics instruction.

17. The non-transitory computer-readable storage medium of claim 10, wherein the instructions for causing the processor to receive the indication of the data precision for execution of the graphics instruction comprise instructions for causing the processor to decode the graphics instruction to determine the data precision.

18. The non-transitory computer-readable storage medium of claim 10, wherein the graphics data associated with the graphics instruction comprises at least one of vertex graphics data and pixel graphics data.

19. A device comprising:

a controller configured to receive a graphics instruction for execution within a programmable streaming processor, wherein the indication of the data precision is contained within the graphics instruction and wherein the graphics instruction is a first executable instruction generated by a compiler that compiles graphics application instructions, to receive an indication of a data precision for execution of the graphics instruction, and to receive a conversion instruction that, when executed by the programmable streaming processor, converts graphics data associated, with the graphics instruction, from a first data precision to converted graphics data having a second data precision, wherein the conversion instruction is different than the graphics instruction and wherein the conversion instruction is generated by the compiler; and

a plurality of execution units within the processor,

wherein the controller is configured to select one of the execution units based on the indicated data precision and cause the selected execution unit to execute the graphics instruction with the indicated data precision using the converted graphics data associated with the graphics instruction.

20. The device of claim 19, wherein the plurality of execution units includes a first execution unit configured to execute instructions with the indicated data precision and a second execution unit configured to execute instructions with a second data precision that is different from the indicated data precision, and wherein the controller is configured to select the first execution unit to execute the graphics instruction with the indicated data precision using the graphics data.

21. The device of claim 19, wherein the plurality of execution units includes one or more full-precision execution units and at least four half-precision execution units.

22. The device of claim 21, wherein when the indicated data precision for execution of the graphics instruction comprises a full precision, the controller is configured to select one of the full-precision execution units to execute the graphics instruction using the graphics data.

23. The device of claim 21, wherein when the indicated data precision for execution of the graphics instruction comprises a half precision, the controller is configured to select one of the half-precision execution units to execute the graphics instruction using the graphics data.

24. The device of claim 21, further comprising:

at least one full-precision register bank to store computation results when the at least one full-precision execution unit executes instructions; and

at least four half-precision register banks to store computation results when the at least four half-precision execution units execute instructions.

25. The device of claim 19, wherein the plurality of execution units includes at least one full-precision execution unit and at least one half-precision execution unit, and wherein when the indicated data precision for execution of the graphics instruction comprises a half precision, the controller is configured to shut down power to the at least one full-precision execution unit and cause the at least one half-precision execution unit to execute the graphics instruction using the graphics data.

26. The device of claim 19, wherein the processor comprises a shader processor.

27. The device of claim 19, wherein the device comprises a wireless communication device handset.

28. The device of claim 19, wherein the device comprises one or more integrated circuit devices.

29. A device comprising:

means for receiving a graphics instruction for execution within a programmable streaming processor;

means for receiving an indication of a data precision for execution of the graphics instruction, wherein the indication of the data precision is contained within the graphics instruction, wherein the graphics instruction is a first executable instruction generated by a compiler that compiles graphics application instructions;

means for receiving a conversion instruction that, when executed by the programmable streaming processor, converts graphics data associated, with the graphics instruction, from a first data precision to converted graphics data having the indicated data precision, and wherein the conversion instruction is different than the graphics instruction, wherein the conversion instruction is generated by the compiler;

means for selecting one of a plurality of execution units within the processor based on the indicated data precision; and

means for using the selected execution unit to execute the graphics instruction with the indicated data precision using the converted graphics data associated with the graphics instruction.

30. The device of claim 29, further comprising:

means for receiving the graphics data associated with the graphics instruction;

means for generating a computation result with the indicated data precision during execution of the graphics instruction by the selected execution unit; and

means for providing the computation result as output.

31. The device of claim 29, wherein the means for selecting one of the plurality of execution units comprises:

means for selecting one of a first set of one or more execution units within the processor that each execute instructions with the first data precision using the graphics data when the indicated data precision is the first data precision; and

means for selecting one of a second set of one or more execution units within the processor that each execute instructions with a second data precision using the graphics data when the indicated data precision is the second data precision, the second data precision being different than the first data precision.

32. The device of claim 31, wherein the first data precision comprises a full data precision, and wherein the second data precision comprises a half data precision.

33. The device of claim 29, wherein the execution units include a first set of one or more execution units within the processor that each execute instructions with the first data precision using the graphics data, and further include a second set of one or more execution units within the processor that each execute instructions with a second data precision different than the first data precision using the graphics data.

34. The device of claim 33, wherein:

the means for selecting one of the plurality of execution units within the processor based on the indicated data precision comprises means for selecting one of the execution units in the first set; and

the means for using the selected execution unit to execute the graphics instruction comprises means for using the selected execution unit in the first set to execute the graphics instruction with the indicated data precision using the graphics data associated with the graphics instruction.

35. The device of claim 34, further comprising:

means for receiving a second graphics instruction for execution within the processor;

means for receiving an indication of the second data precision for execution of the second graphics instruction;

means for receiving a second conversion instruction that, when executed by the processor, converts graphics data associated with the second graphics instruction to the indicated second data precision, the second conversion instruction being different than the second graphics instruction;

means for selecting one of the execution units in the second set based on the indicated second data precision; and

means for using the selected execution unit in the second set to execute the second graphics instruction with the indicated second data precision using the graphics data associated with the second graphics instruction.

36. The device of claim 29, wherein the means for receiving the indication of the data precision for execution of the graphics instruction comprises means for decoding the graphics instruction to determine the data precision.

37. The device of claim 29, wherein the graphics data associated with the graphics instruction comprises at least one of vertex graphics data and pixel graphics data.

38. A device comprising:

a programmable streaming processor; and

at least one memory module coupled to the programmable streaming processor,

wherein the programmable streaming processor comprises;

a controller configured to receive a graphics instruction for execution from the at least one memory module, to receive an indication of a data precision for execution of the graphics instruction, wherein the indication of the data precision is contained within the graphics instruction and wherein the graphics instruction is a first executable instruction generated by a compiler that compiles graphics application instructions, and to receive a conversion instruction that, when executed by the processor, converts graphics data, associated with the graphics instruction, to converted graphics data, wherein the graphics data has a first data precision and the converted graphics data has the indicated data precision, and wherein the conversion instruction is different than the graphics instruction and wherein the conversion instruction is generated by the compiler; and

a plurality of execution units that are configured to execute instructions,

wherein the controller is configured to select one of the execution units based on the indicated data precision and cause the selected execution unit to execute the graphics instruction with the indicated data precision using the converted graphics data associated with the graphics instruction.

39. The device of claim 38, further comprising at least one graphics engine coupled to the processor.

40. The device of claim 38, wherein the plurality of execution units includes a first execution unit configured to execute instructions with the indicated data precision and a second execution unit configured to execute instructions with a second data precision that is different from the indicated data precision, and wherein the controller is configured to select the first execution unit to execute the graphics instruction with the indicated data precision using the graphics data.

41. The device of claim 38, wherein the plurality of execution units includes one or more full-precision execution units and at least four half-precision execution units.

42. The device of claim 41, wherein when the indicated data precision for execution of the graphics instruction comprises a full precision, the controller is configured to select one of the full-precision execution units to execute the graphics instruction using the graphics data.

43. The device of claim 41, wherein when the indicated data precision for execution of the graphics instruction comprises a half precision, the controller is configured to select one of the half-precision execution units to execute the graphics instruction using the graphics data.

44. The device of claim 41, wherein the processor further comprises:

at least one full-precision register bank to store computation results when the at least one full-precision execution unit executes instructions; and

at least four half-precision register banks to store computation results when the at least four half-precision execution units execute instructions.

45. The device of claim 38, wherein the plurality of execution units includes at least one full-precision execution unit and at least one half-precision execution unit, and wherein when the indicated data precision for execution of the graphics instruction comprises a half precision, the controller is configured to shut down power to the at least one full-precision execution unit and cause the at least one half-precision execution unit to execute the graphics instruction using the graphics data.

46. The device of claim 38, wherein the processor comprises a shader processor.

47. The device of claim 38, wherein the device comprises a wireless communication device handset.

48. The device of claim 38, wherein the device comprises one or more integrated circuit devices.

49. A method, comprising:

analyzing, by a compiler executed by a processor, a plurality of application instructions for a graphics application;

for each application instruction that specifies a first data precision level for its execution, generating, by the compiler, one or more corresponding compiled instructions that each indicate the first data precision level for its execution, wherein the first precision level comprises a full data precision level; and

generating, by the compiler, one or more conversion instructions to convert graphics data from a second, different data precision level to the first data precision level when the one or more compiled instructions are executed.

50. The method of claim 49, wherein the second data precision level comprises a half data precision level.

51. The method of claim 49, wherein generating the one or more compiled instructions comprises generating one or more compiled instructions that each indicate a full data precision level when a corresponding application instruction specifies the full data precision level for its execution.

52. The method of claim 49, wherein generating the one or more compiled instructions comprises generating one or more compiled instructions that each indicate a half data precision level when a corresponding application instruction specifies the half data precision level for its execution.

53. The method of claim 49, wherein the one or more compiled instructions each include a predefined field that includes information indicating the first data precision level when the corresponding application instruction specifies the first data precision level for its execution.

54. The method of claim 49, further comprising storing the one or more compiled instructions in memory for subsequent execution.

55. A non-transitory computer-readable storage medium comprising instructions for causing a processor to:

analyze, by a compiler executed by the processor, a plurality of application instructions for a graphics application;

for each application instruction that specifies a first data precision level for its execution, generate, by the compiler, one or more corresponding compiled instructions that each indicate the first data precision level for its execution, wherein the first precision level comprises a full data precision level; and

generate, by the compiler, one or more conversion instructions to convert graphics data from a second, different data precision level to the first data precision level when the one or more compiled instructions are executed.

56. The non-transitory computer-readable storage medium of claim 55, wherein the second data precision level comprises a half data precision level.

57. The non-transitory computer-readable storage medium of claim 55, wherein the instructions for causing the processor to generate the one or more compiled instructions comprise instructions for causing the processor to generate the one or more compiled instructions that each indicate a full data precision level when a corresponding application instruction specifies the full data precision level for its execution.

58. The non-transitory computer-readable storage medium of claim 55, wherein the instructions for causing the processor to generate the one or more compiled instructions comprise instructions for causing the processor to generate the one or more compiled instructions that each indicate a half data precision level when a corresponding application instruction specifies the half data precision level for its execution.

59. The non-transitory computer-readable storage medium of claim 55, wherein the one or more compiled instructions each include a predefined field that includes information indicating the first data precision level when the corresponding application instruction specifies the first data precision level for its execution.

60. The non-transitory computer-readable storage medium of claim 55, further comprising instructions for causing the processor to store the one or more compiled instructions in memory for subsequent execution.

61. An apparatus comprising:

means for analyzing a plurality of graphics application instructions;

for each graphics application instruction that specifies a first data precision level for its execution, means for generating one or more corresponding compiled instructions that each indicate the first data precision level for its execution, wherein the first precision level comprises a full data precision level; and

means for generating one or more conversion instructions to convert graphics data from a second, different data precision level to the first data precision level when the one or more compiled instructions are executed.

62. The apparatus of claim 61, wherein the second data precision level comprises a half data precision level.

63. The apparatus of claim 61, wherein the means for generating the one or more compiled instructions comprises means for generating the one or more compiled instructions that each indicate a full data precision level when a corresponding graphics application instruction specifies the full data precision level for its execution.

64. The apparatus of claim 61, wherein the means for generating the one or more compiled instructions comprises means for generating the one or more compiled instructions that each indicate a half data precision level when a corresponding graphics application instruction specifies the half data precision level for its execution.

65. The apparatus of claim 61, wherein the one or more compiled instructions each include a predefined field that includes information indicating the first data precision level when the corresponding graphics application instruction specifies the first data precision level for its execution.

66. The apparatus of claim 61, further comprising means for storing the one or more compiled instructions in memory for subsequent execution.

67. A non-transitory computer-readable data storage medium comprising:

one or more first executable instructions generated by a compiler, wherein the one or more first executable instructions, when executed by a programmable streaming processor, support one or more functions of a graphics application, wherein each of the first executable instructions indicates a first data precision level for its execution;

one or more second executable instructions generated by a compiler, wherein the one or more second executable instructions, when executed by the programmable streaming processor, support one or more functions of the graphics application, wherein each of the second executable instructions indicates a second data precision level different from the first data precision level for its execution, wherein the first precision level comprises a full data precision level; and

one or more third executable instructions generated by a compiler, wherein the one or more third executable instructions, when executed by the programmable streaming processor, support one or more functions of the graphics application, wherein each of the third executable instructions converts graphics data from the second data precision level to the first data precision level when the one or more first executable instructions are executed by a programmable streaming processor.

68. The non-transitory computer-readable data storage medium of claim 67, and wherein the second data precision level comprises a half data precision level.