Struct PsaFa2Backward 

pub struct PsaFa2Backward<D: Float> {
    pub name: &'static str,
    pub id: String,
    pub head_dim: i32,
    pub kernel_source: String,
    pub entry_point_source: String,
    pub source: String,
    /* private fields */
}

Combined PSA Flash Attention 2 backward pass.

Each CTA handles one (n, bh) pair and computes both dQ_n (by iterating over all K rows) and dK_n / dV_n (by iterating over all Q rows). This is valid in PSA because N_CTX_Q == N_CTX_K == N (self-attention).

Atomicity: dq_ptr and dk_ptr are written with atomic_add because both FA2_lo and FA2_hi backward passes contribute to shared sections 0 and 1 of the d_packed buffer. dv_ptr uses a regular store (each FA2 call owns an exclusive V section so no overlap exists).

Grid: (N, BH, 1) — same shape as the FA2 forward pass.

Fields

name: &'static str

The kernel function’s name (e.g. "flash_attention2_forward").

id: String

Unique kernel identifier: fn_name + dtype(s) + const values joined by “__”.

head_dim: i32

Compile-time kernel constant, from the annotated fn’s const generics.

kernel_source: String

The original kernel function source.

entry_point_source: String

The Rust source of the generated C-ABI entry-point wrapper function.

source: String

Combined source (kernel_source + "\n\n" + entry_point_source); used by the Kernel trait.

Implementations

impl<D: Float> PsaFa2Backward<D>

pub fn new(head_dim: i32) -> Self

Constructs a new kernel instance for these compile-time parameters.

Trait Implementations

impl<D: Float> Kernel for PsaFa2Backward<D>

type Args<'__a> = (*mut D, *mut D, *mut D, *mut D, *mut D, *mut D, *mut D, *mut D, *mut D, i32, f32)

This kernel’s launch-argument tuple type.

fn id(&self) -> String

A content hash of this kernel’s source, used as a cache key.

fn name(&self) -> &str

This kernel’s name.

fn source(&self) -> &str

This kernel’s full source (DSL + kernel body + entry point wrapper).

fn kernel_source(&self) -> &str

This kernel’s body source, without the entry-point wrapper.

fn entry_point_source(&self) -> &str

Returns the Rust source of the generated C-ABI entry-point wrapper function.

fn entry_point_name(&self) -> String

Returns the PTX symbol name for this kernel: "{name}_entry_point".