Struct SassReport 

pub struct SassReport {
    pub kernel: String,
    pub instructions: usize,
    pub imma: usize,
    pub ldsm: usize,
    pub reuse_insns: usize,
    pub registers: usize,
    pub avg_stall_x100: u64,
}

What the COMPILER learns about the machine code ptxas produced for one of its kernels — the SASS-awareness the backend needs to reason about the ptxas ceiling (register pressure/occupancy, tensor-core utilisation, reuse-cache usage). Every field is read from the real cubin, not guessed.

Fields

kernel: String

instructions: usize

imma: usize

ldsm: usize

reuse_insns: usize

registers: usize

avg_stall_x100: u64

Implementations

impl SassReport

pub fn blocks_per_sm(&self, threads: usize) -> usize

Estimate resident blocks/SM from the register count (Ampere: 65,536 regs/SM) at threads per block — this is where a register-bound kernel loses to cuBLAS, and the compiler can now SEE it.

pub fn analyze(cubin_path: &Path, kernel: &str) -> Option<SassReport>

Analyse the cubin a compiler backend just produced: parse its machine code + control codes and correlate with the SASS mnemonics. The backend calls this to know how well ptxas scheduled its kernel.

Trait Implementations

impl Clone for SassReport

fn clone(&self) -> SassReport

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for SassReport

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for SassReport

fn eq(&self, other: &SassReport) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for SassReport

impl StructuralPartialEq for SassReport