multimem.red GPU Virtual ISA NVIDIA

multimem.red Data Movement and Conversion Instructions

// Integer type:
multimem.red{.redsem}{.scope}{.ss}.op.type [a], b;

The multimem.* operations operate on multimem addresses and accesses all of the multiple memory locations which the multimem address points to. Multimem addresses can be accessed only by multimem.* operations. Accessing a multimem address with ld, st or any other memory operations results in undefined behavior. Refer to CUDA programming guide for creation and management of the multimem addresses.

Encoding

PTX is a virtual instruction set. It has no single, stable native binary encoding - the compiler lowers this instruction to different native machine code depending on the selected NVIDIA target architecture (compute capability). This page intentionally shows no bit-diagram; see the target/version requirements below for what governs how this instruction compiles.
PTX ISA Version Introduced PTX ISA 8.1
Minimum Target sm_90

Syntax Forms

One mnemonic covers many type / state-space / scope / modifier combinations - each row below is an independently valid form.

Syntax Data Types State Space(s) Modifiers Min. Target Description
// Integer type: multimem.red{.redsem}{.scope}{.ss}.op.type [a], b; sm_90 Instruction multimem.ld_reduce performs the following operations: load operation on the multimem address a, which involves loading of data from all of the multiple memory locations pointed to by the m… (see the official PTX ISA docs for the full description)

Operands

  • Integer type:
    Operand

At a Glance

Data Types -

Related

More in Data Movement and Conversion Instructions

Reference

NVIDIA PTX ISA

Description

Instruction multimem.ld_reduce performs the following operations: load operation on the multimem address a, which involves loading of data from all of the multiple memory locations pointed to by the multimem address a, reduction operation specified by.op on the multiple data loaded from the multimem address a. The result of the reduction operation in returned in register d. Instruction multimem.st performs a store operation of the input operand b to all the memory locations pointed to by the multimem address a. Instruction multimem.red performs a reduction operation on all the memory locations pointed to by the multimem address a, with operand b. Instruction multimem. (see the official PTX ISA docs for the full description)

Examples

multimem.ld_reduce.and.b32                    val1_b32, [addr1];
multimem.ld_reduce.acquire.gpu.global.add.u32 val2_u32, [addr2];

multimem.st.relaxed.gpu.b32                [addr3], val3_b32;
multimem.st.release.cta.global.u32         [addr4], val4_u32;
// (truncated - see the official PTX ISA docs for the full example)

Reproduced from NVIDIA's official PTX ISA documentation for technical accuracy.

Sources