Enum ProfileTarget 

pub enum ProfileTarget {
    Kernel {
        name: String,
        size: u32,
        roofline: bool,
        metrics: Option<String>,
    },
    Cublas {
        op: String,
        size: u32,
    },
    Wgpu {
        shader: String,
        dispatch: Option<String>,
        target: Option<String>,
    },
    Metal {
        shader: String,
        dispatch: Option<u32>,
    },
    Simd {
        function: String,
        size: u32,
        arch: String,
    },
    Wasm {
        function: String,
        size: u32,
    },
    Quant {
        kernel: Option<String>,
        size: Option<String>,
        all: bool,
    },
    Scalar {
        function: String,
        size: u32,
    },
    Parallel {
        function: String,
        size: u32,
        threads: Option<String>,
    },
    Compare {
        kernel: String,
        size: u32,
        backends: Vec<String>,
    },
    Scaling {
        size: u32,
        max_threads: Option<usize>,
        runs: usize,
    },
    Binary {
        path: String,
        kernel_filter: Option<String>,
        trace: bool,
        duration: Option<String>,
    },
    Python {
        args: Vec<String>,
    },
    Library {
        so: String,
        symbol: String,
        args: Option<String>,
    },
}

Targets accepted by cgp profile.

Variants

Kernel

Profile a CUDA PTX kernel via ncu + CUPTI

Fields

name: String

Kernel name

size: u32

Problem size (e.g., 512 for square matrix)

roofline: bool

Generate roofline overlay

metrics: Option<String>

Specific ncu metrics to collect

Cublas

Profile cuBLAS/cuBLASLt operations

Fields

op: String

Operation (gemm_f16, gemm_f32, etc.)

size: u32

Problem size

Wgpu

Profile wgpu compute shaders

Fields

shader: String

WGSL shader path

dispatch: Option<String>

Dispatch dimensions (e.g., 256,256,1)

target: Option<String>

Target (native or web)

Metal

Profile Apple Metal compute kernels

Fields

shader: String

Metal shader name

dispatch: Option<u32>

Dispatch size

Simd

Profile CPU SIMD functions

Fields

function: String

Function name

size: u32

Problem size

arch: String

Target architecture (avx2, avx512, neon, sse2)

Wasm

Profile WASM SIMD128 via wasmtime

Fields

function: String

Function name

size: u32

Problem size

Quant

Profile quantized CPU kernels (Q4K/Q6K)

Fields

kernel: Option<String>

Kernel name (q4k_gemv, q6k_gemv, q5k_gemv, q8_gemv, nf4_gemv)

size: Option<String>

Dimensions (MxNxK format)

all: bool

Profile all standard LLM layer sizes (ffn_up, ffn_down, attn_qkv, generic_4K)

Scalar

Profile scalar baseline

Fields

function: String

Function name

size: u32

Problem size

Parallel

Profile Rayon parallel workloads

Fields

function: String

Function name

size: u32

Problem size

threads: Option<String>

Thread count (or “auto”)

Compare

Cross-backend comparison

Fields

kernel: String

Kernel name

size: u32

Problem size

backends: Vec<String>

Backends to compare (comma-separated)

#2583: this was a free-form String, split on , and matched in analysis::compare::measure_backend, whose other => arm warns to stderr and continues. So --backends avx2,cudaa exited 0 and printed a comparison table silently missing the CUDA row — and under --json the omitted row is the only signal. Rejecting the typo at parse time is the same fix apr serve run --backend got.

Scaling

Parallel scaling sweep (thread count vs throughput)

Fields

size: u32

Problem size

max_threads: Option<usize>

Max threads to test (default: num_cpus)

runs: usize

Runs per thread count for min-of-N timing

Binary

Profile an arbitrary binary

Fields

path: String

Binary path

kernel_filter: Option<String>

Kernel name filter

trace: bool

Enable system trace

duration: Option<String>

Trace duration

Python

Profile a Python script

Fields

args: Vec<String>

Arguments after –

Library

Profile a shared library function

Fields

so: String

Path to .so file

symbol: String

Symbol name

args: Option<String>

Arguments (key=value pairs)

Trait Implementations

impl Clone for ProfileTarget

fn clone(&self) -> ProfileTarget

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ProfileTarget

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

Formats the value using the given formatter.

impl FromArgMatches for ProfileTarget

fn from_arg_matches(__clap_arg_matches: &ArgMatches) -> Result<Self, Error>

Instantiate Self from ArgMatches, parsing the arguments as needed.

fn from_arg_matches_mut( __clap_arg_matches: &mut ArgMatches, ) -> Result<Self, Error>

Instantiate Self from ArgMatches, parsing the arguments as needed.

fn update_from_arg_matches( &mut self, __clap_arg_matches: &ArgMatches, ) -> Result<(), Error>

Assign values from ArgMatches to self.

fn update_from_arg_matches_mut<'b>( &mut self, __clap_arg_matches: &mut ArgMatches, ) -> Result<(), Error>

Assign values from ArgMatches to self.

impl Subcommand for ProfileTarget

fn augment_subcommands<'b>(__clap_app: Command) -> Command

Append to Command so it can instantiate Self via FromArgMatches::from_arg_matches_mut

fn augment_subcommands_for_update<'b>(__clap_app: Command) -> Command

Append to Command so it can instantiate self via FromArgMatches::update_from_arg_matches_mut

fn has_subcommand(__clap_name: &str) -> bool

Test whether Self can parse a specific subcommand