# US 2025/0156314 A1 · Non-uniform allocation of symmetric memory in parallel programs

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0156314 A1
- Document ID: US20250156314A1
- Status: Published application
- Kind: A1
- Application number: 18/507,659
- Filed: 2023-11-13
- Published: 2025-05-15
- Earliest claimed: 2023-11-13
- Assignees: NVIDIA Corporation
- Applicants: Nvidia Corporation
- Inventors: Akhil Langer; Sreeram Potluri; James Dinan
- Web page: https://patentgrep.com/patent/US20250156314A1

## CPC

- G06F 12/0246: {in block erasable memory, e.g. flash memory}
- G06F 9/5016: {the resource being the memory}
- G06F 12/0284: {Multiple user address space allocation, e.g. using different base addresses (interprocessor communication G06F15/163)}
- G06F 12/1027: using associative or pseudo-associative address translation means, e.g. translation look-aside buffer [TLB]
- G06F 12/109: for multiple virtual address spaces, e.g. segmentation (G06F12/1036 takes precedence)
- G06F 2212/656: Address space sharing
- G06F 2212/657: Virtual address space management
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems and methods herein are for at least one circuit to perform at least one process of different processes that may be associated with one or more applications, where the process may use one part of a virtual memory based on different requests by the different processes, where the virtual memory may include parts of equal allocations based on a maximum of different memory sizes in specifications associated with the different processes, where the one part of the virtual memory may be in a mapping with respect to one part of a physical memory of different allocated sizes, and where a translation for the mapping can occur using a start address and the maximum of the different memory sizes.

## Claims

### Claim 1

An interface to be associated with a plurality of applications, the interface to enable specifications of different memory sizes associated with different processes for the plurality of applications and to enable equal allocation of virtual memory to the different processes based on a maximum of the different memory sizes for the virtual memory, wherein at least one part of the virtual memory is associated with a mapping to one part of a physical memory that is comprised of different allocated sizes, wherein the mapping is associated with at least one process of the different processes based on a plurality of requests associated with the different processes, and wherein a translation for the mapping occurs using a start address and the maximum of the different memory sizes.

### Claim 2 (depends on 1)

The interface of claim 1, wherein the plurality of requests is associated with a first collective call, for the at least one process, using first arguments associated with a first one of the different memory sizes, and wherein the plurality of requests is further associated with a second collective call, for the at least one process, using second arguments associated with a second one of the different memory sizes.

### Claim 3 (depends on 1)

The interface of claim 1, wherein different parts of the physical memory are of different sizes and comprises the one part of the physical memory, and wherein the different parts of the physical memory are part of the mapping with respect to different parts of the virtual memory that comprises the one part of the virtual memory.

### Claim 4 (depends on 1)

The interface of claim 1, wherein the interface supports accumulation of the specifications of the different memory sizes and enables the equal allocation of the virtual memory on behalf of the different processes.

### Claim 5 (depends on 1)

The interface of claim 1, wherein the equal allocation of virtual memory is to use contiguous parts of the virtual memory for different requests, of the plurality of requests, in different collective calls associated with the at least one process.

### Claim 6 (depends on 1)

The interface of claim 1, wherein address translation is used to translate virtual addresses of the virtual memory to physical address of the physical memory at a destination process of the different processes, wherein the address translation enables the at least one part of the virtual memory to be part of the mapping to the one part of a physical memory.

### Claim 7 (depends on 1)

The interface of claim 1, wherein memory registration is used to enable the one part of the virtual memory to be part of the mapping to the one part of a physical memory, and wherein memory region (MR) keys are provided to confirm the mapping.

### Claim 8 (depends on 1)

The interface of claim 1, wherein on-demand paging is used in a network interface card (NIC) associated with the interface, the on-demand paging to provide a registration associated with the one part of the virtual memory being part of the mapping to the one part of a physical memory.

### Claim 9

A system comprising:
a plurality of circuits to support an interface with one or more applications and with different processes for the one or more applications, the interface to receive specifications of different memory sizes associated with the different processes and to enable equal allocation of virtual memory to the different processes based on a maximum of the different memory sizes for the virtual memory, wherein at least one part of the virtual memory is in a mapping with respect to one part of a physical memory that is of different allocated sizes, wherein the mapping is associated with at least one process of the different processes, and wherein a translation for the mapping occurs using a start address and the maximum of the different memory sizes.

### Claim 10 (depends on 9)

The system of claim 9, wherein the different requests are associated with a first collective call, for the at least one process, using first arguments associated with a first one of the different memory sizes, and wherein the different requests is further associated with a second collective call, for the at least one process, using second arguments associated with a second one of the different memory sizes.

### Claim 11 (depends on 9)

The system of claim 9, wherein the interface is adapted for accumulation of the specifications of the different memory sizes and enables the equal allocation of the virtual memory on behalf of the different processes.

### Claim 12 (depends on 9)

The system of claim 9, wherein the equal allocation of virtual memory is to use contiguous parts of the virtual memory for different requests in different collective calls associated with the at least one process.

### Claim 13 (depends on 9)

The system of claim 9, wherein address translation is used to translate virtual addresses of the virtual memory to physical address of the physical memory at a destination process of the different processes, wherein the address translation enables the at least one part of the virtual memory to be part of the mapping to the one part of a physical memory.

### Claim 14

A system comprising:
at least one circuit to perform at least one process of a plurality of different processes for one or more applications and to use one part of a virtual memory, based on plurality of requests associated with the plurality of different processes, the virtual memory comprising a plurality of parts of equal allocations based on a maximum of different memory sizes for the virtual memory provided in specifications associated with the plurality of different processes, wherein the one part of the virtual memory is in a mapping with respect to one part of a physical memory that is of different allocated sizes, and wherein a translation for the mapping occurs using a start address and the maximum of the different memory sizes.

### Claim 15 (depends on 14)

The system of claim 14, wherein different parts of the physical memory are of different sizes and comprises the one part of the physical memory, and wherein the different parts of the physical memory are part of the mapping with respect to different parts of the virtual memory that comprises the one part of the virtual memory.

### Claim 16 (depends on 14)

The system of claim 14, wherein the equal allocation of virtual memory is to use contiguous parts of the virtual memory for different requests, of the plurality of requests, in different collective calls associated with the at least one process.

### Claim 17 (depends on 14)

The system of claim 14, wherein address translation is used to translate virtual addresses of the virtual memory to physical address of the physical memory at a destination process of the different processes, wherein the address translation enables the at least one part of the virtual memory to be part of the mapping to the one part of a physical memory.

### Claim 18

A method for allocation of physical memory, comprising:
providing an interface to be associated with a plurality of applications and to enable specifications of different memory sizes associated with different processes for the plurality of applications;
enabling equal allocation of virtual memory to the different processes based on a maximum of the different memory sizes for the virtual memory;
associating at least one part of the virtual memory with a mapping to one part of a physical memory that is comprised of different allocated sizes; and
associating the mapping with at least one process of the different processes, wherein a translation for the mapping occurs using a start address and the maximum of the different memory sizes.

### Claim 19 (depends on 18)

The method of claim 18, further comprising:
providing, as part of a plurality of requests from the different processes or as a separate request, a first collective call for the at least one process using first arguments associated with a first one of the different memory sizes; and
providing, as part of the plurality of requests from the different processes or as another separate request, a second collective call for the at least one process using second arguments associated with a second one of the different memory sizes.

### Claim 20 (depends on 18)

The method of claim 18, further comprising:
using contiguous parts of the virtual memory for different requests, in different collective calls associated with the at least one process, the contiguous parts representing the equal allocation of virtual memory.
