Gemm Universal unit tests for MainloopIntelW8A8 API

[rishi-yadav]

Unit tests for MainloopIntelW8A8 FP8 GEMM operations on Intel Xe

Comprehensive test coverage for Intel GPU FP8 GEMM kernels including:

• LLM workloads: LLaMA2-7B, Mistral-7B configurations
• Parallelization: Tensor/model parallel scenarios
• Batch sizes: Micro-batch (4x) to large batch (32x)
• Matrix shapes: Small (64²) to large (2048²), tall/wide matrices
• Edge cases: Large K/N dimensions

Tests validate MainloopIntelW8A8 dispatch policy with FP8→FP32 precision on Intel Xe XMX architecture.

[Copilot]

Pull Request Overview

This PR adds comprehensive unit test coverage for Intel Xe FP8 GEMM operations using the MainloopIntelW8A8 dispatch policy. The tests validate FP8→FP32 matrix multiplication across diverse workload scenarios including LLM model configurations (LLaMA2-7B, Mistral-7B), various parallelization strategies, different batch sizes, and edge cases with large matrix dimensions.

Key Changes:

• Added 14 test cases covering LLM workloads, parallelization patterns, batch processing, and matrix shape variations
• Extended TestXe helper function to support 4D problem shapes (MNKL) for batched GEMM operations
• Integrated new test file into CMake build system

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
gemm_universal_fp8_fp8_fp32_tensor_op_f32_xe_models.cpp	New test file with 14 unit tests for MainloopIntelW8A8 covering LLM models, parallelization, batching, and edge cases
gemm_testbed_3x.hpp	Added overloaded TestXe function to support batched GEMM testing with 4D problem shapes
CMakeLists.txt	Registered new test file in SYCL build configuration

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Antonyvance]

Don't we need tests based on new APIs? Do you plan to add them after #579 #573

[Copilot]

Pull Request Overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

[Copilot]

[nitpick] Using EXPECT_TRUE(false) is unconventional. Consider using ADD_FAILURE() or FAIL() instead for clearer test failure reporting.

Suggested change

	EXPECT_TRUE(false) << "TestXe: testbed.run threw an exception: " << e.what();
	throw;
	}
	catch (...) {
	EXPECT_TRUE(false) << "TestXe: testbed.run threw an unknown exception for MNKL = "
	ADD_FAILURE() << "TestXe: testbed.run threw an exception: " << e.what();
	throw;
	}
	catch (...) {
	ADD_FAILURE() << "TestXe: testbed.run threw an unknown exception for MNKL = "

[Copilot]

[nitpick] Using EXPECT_TRUE(false) is unconventional. Consider using ADD_FAILURE() or FAIL() instead for clearer test failure reporting.

Suggested change

	EXPECT_TRUE(false) << "TestXe: testbed.run threw an exception: " << e.what();
	throw;
	}
	catch (...) {
	EXPECT_TRUE(false) << "TestXe: testbed.run threw an unknown exception for MNKL = "
	ADD_FAILURE() << "TestXe: testbed.run threw an exception: " << e.what();
	throw;
	}
	catch (...) {
	ADD_FAILURE() << "TestXe: testbed.run threw an unknown exception for MNKL = "