Enum SmVersion 

pub enum SmVersion {
    Sm75,
    Sm80,
    Sm86,
    Sm89,
    Sm90,
    Sm100,
    Sm120,
}

GPU architecture version for occupancy estimation.

This is a local copy that avoids a dependency on oxicuda-ptx. Each variant encodes the SM architecture parameters needed for occupancy calculations (max warps per SM, register file size, etc.).

Variants

Sm75

Turing (compute capability 7.5).

Sm80

Ampere (compute capability 8.0).

Sm86

Ampere GA10x (compute capability 8.6).

Sm89

Ada Lovelace (compute capability 8.9).

Sm90

Hopper (compute capability 9.0).

Sm100

Blackwell (compute capability 10.0).

Sm120

Blackwell B200 (compute capability 12.0).

Implementations

impl SmVersion

pub const fn max_warps_per_sm(self) -> u32

Maximum number of warps that can reside on a single SM.

Ampere GA10x (Sm86) and Blackwell consumer (Sm120) share Ada Lovelace’s (Sm89) lower 1536-threads/SM ceiling (48 warps), unlike the datacenter parts (Sm80/Sm90/Sm100) which reach the full 2048 threads/SM (64 warps).

pub const fn max_blocks_per_sm(self) -> u32

Maximum number of thread blocks that can reside on a single SM.

pub const fn registers_per_sm(self) -> u32

Total number of 32-bit registers available per SM.

pub const fn max_registers_per_thread(self) -> u32

Maximum number of registers a single thread can use.

pub const fn max_shared_mem_per_sm(self) -> u32

Maximum shared memory per SM in bytes.

Note this is the per-SM pool, not the smaller per-block cap (e.g. Sm86/Sm89 allow up to 100 KiB per SM but only 99 KiB (101,376 bytes) opt-in per block).

pub const fn warp_size(self) -> u32

Warp size (always 32 for NVIDIA GPUs).

pub const fn register_alloc_granularity(self) -> u32

Register allocation granularity (in warps).

Registers are allocated to warps in chunks of this many registers per thread, rounded up to the nearest multiple.

pub const fn shared_mem_alloc_granularity(self) -> u32

Shared memory allocation granularity in bytes.

Trait Implementations

impl Clone for SmVersion

fn clone(&self) -> SmVersion

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for SmVersion

impl Debug for SmVersion

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

Formats the value using the given formatter.

impl Display for SmVersion

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

Formats the value using the given formatter.

impl Eq for SmVersion

impl Hash for SmVersion

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for SmVersion

fn eq(&self, other: &SmVersion) -> 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 StructuralPartialEq for SmVersion