Testing Framework

Overview

The sneppx-alg test suite uses the CTest framework (via CMake) with custom test harness macros defined in tests/test_harness.h. Tests cover all 5 supported architectures and 5 sneppx primitives.

Test organization

tests/
├── test_harness.h              # Shared macros (ASSERT_EQ, TEST, etc.)
├── test_context_extension.c    # 128K context extension (all architectures)
├── test_mha_forward.c          # MHA forward pass (all architectures)
├── test_model_zoo.c            # Model zoo presets, JSON, weights
├── test_llama.c / test_llama3.c
├── test_mistral.c / test_qwen2.c / test_deepseek_v2.c
├── test_hss.c / test_ser.c / test_arc.c / test_npe.c / test_fm.c
└── ...

Running tests

cmake --preset release
cmake --build build --config Release
cd build && ctest -C Release --output-on-failure

Test macros

Macro	Description
ASSERT_EQ(a, b)	Assert a == b
ASSERT_NEAR(a, b, eps)	Assert
ASSERT_NULL(p)	Assert p == NULL
ASSERT_NOT_NULL(p)	Assert p != NULL
ASSERT_STREQ(s1, s2)	Assert string equality
ASSERT_SUCCESS(ret)	Assert ret == 0
ASSERT_FAILURE(ret)	Assert ret != 0
ASSERT_TRUE(cond)	Assert cond is truthy
ASSERT_FALSE(cond)	Assert cond is falsy

Architecture test pattern

Tests for each architecture follow an ARCH_test_* naming convention with automatic fixture setup/teardown:

static SNEPPXLlamaConfig LLLAMA_CFG;

// Called before each Llama test
void LLAMA_setup(void) {
    SNEPPX_llm_config_from_name("llama3", "8B", &LLAMA_CFG);
}

void LLAMA_test_forward(void) {
    // ... test body ...
}

What's tested

Feature	File	Architectures
128K context extension	testcontextextension.c	LLaMA2/3, Mistral, Qwen2, DeepSeek
MHA forward pass	testmhaforward.c	LLaMA2/3, Mistral, Qwen2, DeepSeek
Model zoo presets	testmodelzoo.c	LLaMA2/3, Mistral, Qwen2, DeepSeek
JSON serialization	testmodelzoo.c	All
Weight name mapping	testmodelzoo.c	All