From 39536470af646dab4be3a8c840b789ec9e6a9ba8 Mon Sep 17 00:00:00 2001
From: LoserCheems <losercheems@gmail.com>
Date: Fri, 19 Dec 2025 18:14:42 +0800
Subject: [PATCH 1/2] Adds GEMM benchmark tests

Introduces parameterized GEMM benchmarks that compare multiple backend implementations across devices and dtypes to validate performance regressions early
---
 tests/test_gemm.py | 95 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 95 insertions(+)
 create mode 100644 tests/test_gemm.py

diff --git a/tests/test_gemm.py b/tests/test_gemm.py
new file mode 100644
index 0000000..908aff1
--- /dev/null
+++ b/tests/test_gemm.py
@@ -0,0 +1,95 @@
+import pytest
+import torch
+
+from kernel_course import testing
+from kernel_course.python_ops import gemm as python_gemm
+
+try:
+    from kernel_course.pytorch_ops import gemm as pytorch_gemm
+
+    HAS_PYTORCH = True
+except Exception:
+    pytorch_gemm = None
+    HAS_PYTORCH = False
+
+try:
+    from kernel_course.triton_ops import gemm as triton_gemm
+
+    HAS_TRITON = True
+except Exception:
+    triton_gemm = None
+    HAS_TRITON = False
+
+try:
+    from kernel_course.cute_ops import gemm as cute_gemm
+
+    HAS_CUTE = True
+except Exception:
+    cute_gemm = None
+    HAS_CUTE = False
+
+
+def factory(
+    MNK: tuple[int, int, int],
+    device: torch.device,
+    dtype: torch.dtype = torch.float32,
+):
+    M, N, K = MNK
+    A = torch.linspace(0.0, 1.0, steps=M * K, device=device, dtype=dtype).view(M, K)
+    B = torch.linspace(0.0, 1.0, steps=K * N, device=device, dtype=dtype).view(K, N)
+    C = torch.linspace(0.0, 1.0, steps=M * N, device=device, dtype=dtype).view(M, N)
+    alpha = 1.14
+    beta = 5.14
+    return (A, B, C, alpha, beta), {}
+
+
+@pytest.mark.parametrize(
+    "device",
+    [
+        pytest.param(
+            torch.device("cuda"),
+            marks=pytest.mark.skipif(
+                not torch.cuda.is_available(), reason="requires CUDA"
+            ),
+        ),
+        pytest.param(
+            torch.device("mps"),
+            marks=pytest.mark.skipif(
+                not torch.backends.mps.is_available(), reason="requires MPS"
+            ),
+        ),
+    ],
+)
+@pytest.mark.parametrize(
+    "dtype",
+    [torch.float32, torch.float16, torch.bfloat16],
+)
+@pytest.mark.parametrize(
+    "MNK",
+    [
+        (1 << 4, 1 << 4, 1 << 4),
+        (1 << 8, 1 << 8, 1 << 8),
+    ],
+)
+def test_gemm_benchmark(
+    device: torch.device,
+    dtype: torch.dtype,
+    MNK: tuple[int, int, int],
+) -> None:
+    impls = testing.get_impls(
+        python_impl=python_gemm.gemm,
+        pytorch_impl=pytorch_gemm.gemm if HAS_PYTORCH else None,
+        triton_impl=triton_gemm.gemm if HAS_TRITON else None,
+        cute_impl=cute_gemm.gemm if HAS_CUTE else None,
+    )
+
+    # Benchmark each implementation
+    config = testing.BenchmarkConfig(warmup=3, repeat=100)
+    results = testing.run_benchmarks(
+        impls,
+        lambda: factory(MNK, device, dtype),
+        flops=2 * MNK[0] * MNK[1] * MNK[2],
+        config=config,
+    )
+
+    testing.show_benchmarks(results)

From 1c44286f63c59c38cdbd77708c304bff06696f38 Mon Sep 17 00:00:00 2001
From: LoserCheems <losercheems@gmail.com>
Date: Fri, 19 Dec 2025 18:14:48 +0800
Subject: [PATCH 2/2] Adds test entry for GEMM implementation in README

---
 README.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/README.md b/README.md
index b54c4ab..4d0f140 100644
--- a/README.md
+++ b/README.md
@@ -24,7 +24,7 @@ The following common BLAS kernels have been implemented in multiple frameworks.
 | [dot](./docs/dot.md) | dot product | $z = x^\top y$ | $2n$ | $2n$ | [✅](./kernel_course/python_ops/dot.py) | [✅](./kernel_course/pytorch_ops/dot.py) | [✅](./kernel_course/triton_ops/dot.py) | ❌ | [✅](./tests/test_dot.py) |
 | [gemv](./docs/gemv.md) | general matrix-vector multiply | $y = \alpha A x + \beta y$ | $2mn$ | $mn + n + 2m$ | [✅](./kernel_course/python_ops/gemv.py) | [✅](./kernel_course/pytorch_ops/gemv.py) | [✅](./kernel_course/triton_ops/gemv.py) | ❌ | [✅](./tests/test_gemv.py) |
 | [geru](./docs/geru.md) | general rank-1 update | $A = A + \alpha x y^\top$ | $2mn$ | $2mn + m + n$ | [✅](./kernel_course/python_ops/geru.py) | [✅](./kernel_course/pytorch_ops/geru.py) | [✅](./kernel_course/triton_ops/geru.py) | ❌ | [✅](./tests/test_geru.py) |
-| [gemm](./docs/gemm.md) | general matrix-matrix multiply | $C = \alpha A B + \beta C$ | $2mnk$ | $mk + nk + 2mn$ | [✅](./kernel_course/python_ops/gemm.py) | [✅](./kernel_course/pytorch_ops/gemm.py) | [✅](./kernel_course/triton_ops/gemm.py) | ❌ | ❌ |
+| [gemm](./docs/gemm.md) | general matrix-matrix multiply | $C = \alpha A B + \beta C$ | $2mnk$ | $mk + nk + 2mn$ | [✅](./kernel_course/python_ops/gemm.py) | [✅](./kernel_course/pytorch_ops/gemm.py) | [✅](./kernel_course/triton_ops/gemm.py) | ❌ | [✅](./tests/test_gemm.py) |
 
 
 ## Transformer Modules