Code Samples

Warning

DRAFT DOCUMENTATION - This documentation is currently in draft status and subject to change.

Sample applications demonstrate how to use PTI SDK tracing and profiling capabilities. All samples are located in the samples/ directory of the repository.

Repository Location:

• Local: sdk/samples/

• GitHub: intel/pti-gpu

ptiView API Samples

Demonstrates how to use the ptiView API for tracing GPU operations, including kernel execution, memory copies, and API calls.

Sample	Description	PTI_VIEW_...
vector_sq_add	SYCL vector square-add kernel	DEVICE_GPU_KERNEL, DEVICE_GPU_MEM_COPY, DEVICE_GPU_MEM_FILL, RUNTIME_API, EXTERNAL_CORRELATION
dpc_gemm	DPC++ matrix multiplication with comprehensive tracing	DEVICE_GPU_KERNEL, DEVICE_GPU_MEM_COPY, DEVICE_GPU_MEM_FILL, RUNTIME_API, DRIVER_API, EXTERNAL_CORRELATION, COLLECTION_OVERHEAD
onemkl_gemm	Matrix multiplication using Intel(R) oneMKL library	DEVICE_GPU_KERNEL, DEVICE_GPU_MEM_COPY, DEVICE_GPU_MEM_FILL, RUNTIME_API
itt_ccl	Intel(R) oneCCL library operations tracing (Linux only)	COMMUNICATION
dpc_gemm_threaded	Multi-threaded DPC++ matrix multiplication	DEVICE_GPU_KERNEL, DEVICE_GPU_MEM_COPY, DEVICE_GPU_MEM_FILL, RUNTIME_API, DRIVER_API, COLLECTION_OVERHEAD
omp_vec_add	C-language OpenMP sample with GPU offload	DEVICE_GPU_KERNEL, DEVICE_GPU_MEM_COPY, DEVICE_GPU_MEM_FILL, DRIVER_API
view_record_versioned	Version-aware tracing that handles v1 and v2 view record structures using ptiVersion()	DEVICE_GPU_KERNEL, DEVICE_GPU_MEM_COPY, DEVICE_GPU_MEM_FILL, RUNTIME_API, DRIVER_API, COLLECTION_OVERHEAD

ptiCallback API Sample

callback

Demonstrates the experimental ptiCallback API for synchronous GPU operation notifications.

Warning

The ptiCallback API is experimental and subject to change. See PTI Callback API Reference (Experimental) for documentation.

GPU Hardware Metrics Collection Samples

metrics_scope

Per-kernel hardware metrics collection using ptiMetricsScope API.

metrics_perf

Device-level time-based metrics sampling using the ptiMetrics API. Use the --list-metrics option to output all available metrics sets and metrics on the system GPUs.

metrics_iso3dfd_dpcpp

Hardware metrics collection in ISO3DFD stencil computation workload.

Building Samples

Samples are built automatically with PTI SDK. See Building and Installing for build instructions.

From the SDK build directory:

Linux:

./build/bin/vector_sq_add
./build/bin/dpc_gemm
./build/bin/metrics_scope

Windows:

build\bin\vector_sq_add.exe
build\bin\dpc_gemm.exe
build\bin\metrics_scope.exe

Next Steps

• PTI View API Reference - PTI View API for tracing

• PTI Metrics Scope API Reference - Per-kernel metrics collection

• PTI Metrics API Reference - Device-level metrics collection

• PTI Callback API Reference (Experimental) - Callback API documentation