vector-bfloat16-spec.adoc

Intrinsics for BFloat16 (Brain Float 16) instruction set extensions

The RISC-V vector C intrinsics supports intrinsics that exposes the control of BFloat16 (Brain Float 16) instruction set extensions ³¹.

Naming scheme

The BFloat16 intrinsics follows the naming scheme defined under [naming-scheme], with bf as the abbreviation for BFloat16 types in the function suffix.

Control of the vector extension programming model

The BFloat16 intrinsics follows provides the same control of the vector programming model defined under [control-of-vector-programming-model]. Intrinsics that represents BFloat16 instructions that are affected by frm (vfncvtbf16.f.f.w and vfwmaccbf16) follow what is defined under [control-of-frm] and provides variants of [implicit-frm] and [explicit-frm].

Type system

Floating-point types have EEW and EMUL encoded into the type. The first row describes the EMUL and the first column describes the data type and element width of the scalar type.

Floating-point types with element widths of 16 (Types=__bf16) require the zfbfmin and zvfbfmin extension to be specified in the architecture.

Note	Although C++23 introduces <stdfloat> for fixed-width floating-point types, this latest standard is not yet supported in the upstream RISC-V compiler. The specification (along with the prototype lists in appendix) uses __bf16 to represent the BFloat16 floating-point type.

Table 1. BFloat16 types

Types	EMUL=1/8	EMUL=1/4	EMUL=1/ 2	EMUL=1	EMUL=2	EMUL=4	EMUL=8
__bf16	N/A	vbfloat16mf4_t	vbfloat16mf2_t	vbfloat16m1_t	vbfloat16m2_t	vbfloat16m4_t	vbfloat16m8_t

Table 2. Tuple types

Non-tuple Types (NFILED=1)	NFIELD=2	NFIELD=3	NFIELD=4	NFIELD=5	NFIELD=6	NFIELD=7	NFIELD=8
vbfloat16mf4_t	vbfloat16mf4x2_t	vbfloat16mf4x3_t	vbfloat16mf4x4_t	vbfloat16mf4x5_t	vbfloat16mf4x6_t	vbfloat16mf4x7_t	vbfloat16mf4x8_t
vbfloat16mf2_t	vbfloat16mf2x2_t	vbfloat16mf2x3_t	vbfloat16mf2x4_t	vbfloat16mf2x5_t	vbfloat16mf2x6_t	vbfloat16mf2x7_t	vbfloat16mf2x8_t
vbfloat16m1_t	vbfloat16m1x2_t	vbfloat16m1x3_t	vbfloat16m1x4_t	vbfloat16m1x5_t	vbfloat16m1x6_t	vbfloat16m1x7_t	vbfloat16m1x8_t
vbfloat16m2_t	vbfloat16m2x2_t	vbfloat16m2x3_t	vbfloat16m2x4_t	N/A	N/A	N/A	N/A
vbfloat16m4_t	vbfloat16m4x2_t	N/A	N/A	N/A	N/A	N/A	N/A

Psuedo intrinsics

The RISC-V vector BFloat16 types (provided under Type system) also have pseudo intrinsics variants from [pseudo-intrinsics] to help variable declaration and manipulation across intrinsic types.