dp2a GPU Virtual ISA NVIDIA

dp2a Integer Arithmetic Instructions

dp2a.mode.atype.btype d, a, b, c;

Two-way 16-bit to 8-bit dot product which is accumulated in 32-bit result.

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 5.0
Minimum Target sm_61

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
dp2a.mode.atype.btype d, a, b, c; sm_61 Two-way 16-bit to 8-bit dot product which is accumulated in 32-bit result. Operand a and b are 32-bit inputs. (see the official PTX ISA docs for the full description)

Operands

  • d
    Destination register
  • a
    Source operand
  • b
    Source operand
  • c
    Source operand

At a Glance

Data Types -

Related AMDGPU Concepts

2-Way Dot Product (Accumulate) ↗
equivalent with restrictions
v_dot2_i32_i16 (AMDGPU)
v_dot2_u32_u16 (AMDGPU)

Related

More in Integer Arithmetic Instructions

Reference

NVIDIA PTX ISA

Description

Two-way 16-bit to 8-bit dot product which is accumulated in 32-bit result. Operand a and b are 32-bit inputs. Operand a holds two 16-bits inputs in packed form and operand b holds 4 byte inputs in packed form for dot product. Depending on the.mode specified, either lower half or upper half of operand b will be used for dot product. Operand c has type.u32 if both.atype and.btype are.u32 else operand c has type.s32.

Semantics

d = c; // Extract two 16-bit values from a 32-bit input and sign or zero extend // based on input type. Va = extractAndSignOrZeroExt_2(a, .atype); // Extract four 8-bit values from a 32-bit input and sign or zer extend // based on input type. Vb = extractAndSignOrZeroExt_4(b, .btype); b_select = (.mode == .lo) ? 0 : 2; for (i = 0; i < 2; ++i) { d += Va[i] * Vb[b_select + i]; }

Examples

dp2a.lo.u32.u32           d0, a0, b0, c0;
dp2a.hi.u32.s32           d1, a1, b1, c1;

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

Sources