Nvidia Corporation

Samsung Electronics Co., Ltd.

Mediatek Singapore Pte. Ltd.

Srinivasan Iyer

David Conrad Tannenbaum

Stuart F Oberman

Mitchell K Alsup

Ming Yiu Siu

Colin Sprinkle

Michael Alan Fetterman

Carsten Aagaard Pedersen

Christopher James Goodman

Derek Lentz

Ian Chi Yan Kwong

Tejash M Shah

Ian-Chi Yan Kwong

Ming (Michael) Y Siu

A system and method are described for providing hints to a processing unit that subsequent operations are likely. Responsively, the processing unit takes steps to prepare for the likely subsequent operations. Where the hints are more likely than not to be correct, the processing unit operates more efficiently. For example, in an embodiment, the processing unit consumes less power. In another embodiment, subsequent operations are performed more quickly because the processing unit is prepared to efficiently handle the subsequent operations.

Providing hints to an execution unit to prepare for predicted subsequent arithmetic operations

A method and apparatus are provided. The method includes executing a plurality of threads in a temporal dimension, executing a plurality of threads in a spatial dimension, determining a branch target address for each of the plurality of threads in the temporal dimension and the plurality of threads in the spatial dimension, and comparing each of the branch target addresses to determine a minimum branch target address, wherein the minimum branch target address is a minimum value among branch target addresses of each of the plurality of threads.

Power saving branch modes in hardware

A system and method for binding instructions to a graphical processing unit (GPU) includes a GPU configured to receive bindlessly compiled instructions and interpret the bindlessly compiled instruction at runtime to identify a needed conversion The GPU generates a conversion information based on the bindlessly compiled instruction and needed conversion and converts the bindlessly compiled instruction according to the conversion information to generate a bound format instruction. The GPU may then execute the bound format instruction.

Efficient interface and transport mechanism for binding bindless shader programs to run-time specified graphics pipeline configurations and objects

According to one general aspect, an apparatus may include a network of node circuits and a central arbiter circuit. The network of node circuits is within an integrated circuit, wherein the network includes a plurality of segments. The central arbiter circuit may be configured to schedule a routing of a message between a pair of node circuits in the network, wherein the routing includes a guaranteed latency between the pair of node circuits.

Central arbitration scheme for a highly efficient interconnection topology in a GPU

A graphics processing unit may include a register file memory, a processing element (PE) and a load-store unit (LSU). The register file memory includes a plurality of registers. The PE is coupled to the register file memory and processes at least one thread of a vector of threads of a graphical application. Each thread in the vector of threads are processed in a non-stalling manner. The PE stores data in a first predetermined set of the plurality of registers in the register file memory that has been generated by processing the at least one thread and that is to be routed to a first stallable logic unit that is external to the PE. The LSU is coupled to the register file memory, and the LSU accesses the data in the first predetermined set of the plurality of registers and routes to the first stallable logic unit.

System and method for maintaining data in a low-power structure

An approach is provided for enabling power reduction in floating-point operations. In one example, a system receives floating-point numbers of a fused multiply-add instruction. The system determines the fused multiply-add instruction does not require compliance with a standard of precision for floating-point numbers. The system generates gating signals for an integrated circuit that is configured to perform operations of the fused multiply-add instruction. The system then sends the gating signals to the integrated circuit to turn off a plurality of logic gates included in the integrated circuit.

Approach to power reduction in floating-point operations

One embodiment of the present invention includes a method for performing arithmetic operations on arbitrary width integers using fixed width elements. The method includes receiving a plurality of input operands, segmenting each input operand into multiple sectors, performing a plurality of multiply-add operations based on the multiple sectors to generate a plurality of multiply-add operation results, and combining the multiply-add operation results to generate a final result. One advantage of the disclosed embodiments is that, by using a common fused floating point multiply-add unit to perform arithmetic operations on integers of arbitrary width, the method avoids the area and power penalty of having additional dedicated integer units.

Technique for performing arbitrary width integer arithmetic operations using fixed width elements

A fused floating-point multiply-add element includes a multiplier that generates a product, and a shifter that shifts an addend within a narrow range. Interpreting logic analyzes the magnitude of the addend relative to the product and then causes logic arrays to position the shifted addend within the left, center, or right portions of a composite register depending in the magnitude of the addend relative to the product. The interpreting logic also forces other portions of the composite register to zero. When the addend is zero, the interpreting logic forces all portions of the composite register to zero. Final combining logic then adds the contents of the composite register to the product.

FFMA operations using a multi-step approach to data shifting

A system and method are provided for performing 32-bit or dual 16-bit floating-point arithmetic operations using logic circuitry. An operating mode that specifies an operating mode for a multiplication operation is received, where the operating mode is one of a 32-bit floating-point mode and a dual 16-bit floating-point mode. Based on the operating mode, nine recoding terms for a mantissa of at least one floating-point input operand are determined. A dual-mode multiplier array circuit that is configurable to generate partial products for either one 32-bit floating-point result or for two 16-bit floating-point results computes the partial products based on the nine recoding terms. The partial products are processed to generate an output based on the operating mode.

Logic circuitry configurable to perform 32-bit or dual 16-bit floating-point operations

A system and method are provided for dynamically reducing power consumption of floating-point logic. A disable control signal that is based on a characteristic of a floating-point format input operand is received and a portion of a logic circuit is disabled based on the disable control signal. The logic circuit processes the floating-point format input operand to generate an output.

System and method for dynamically reducing power consumption of floating-point logic

A processor for calculating a convolution of a first input sequence of numbers with a second input sequence of numbers to generate an output sequence is provided. The processor includes multipliers, each multiplying two real numbers to generate an output; multiplexers to direct the numbers in the first and second input sequences or parts of the numbers to the multipliers; and control circuitry to control the multiplexers to direct the first and second input sequences of numbers to the multipliers dependent on whether the numbers are complex or real. An accumulator adds partial products from multiplications performed by the multipliers to calculate the convolution.

Growing community of inventors

Austin, TX, United States of America

Srinivasan Iyer