MiniTensor – A Lightweight C++20 Tensor Library

Made it so that i can use it for my other projects

Tiny, header-friendly tensor library with NumPy-compatible strides, zero dependencies at runtime, and 100 % test coverage via GoogleTest.

Current Features

Feature	Status	Notes
Zero-copy creation	✅	Takes unique_ptr<float[]> – no extra memcpy
Scalar → 8-D tensors	✅	MAX_RANK = 8, shape {0} = 0-D scalar
NumPy strides	✅	Row-major, calculated exactly like NumPy
Element-wise ops	✅	+, -, * (shape must match, no broadcast yet)
Bounds checking	✅	operator() throws std::out_of_range
Factory helpers	✅	createZeros, createOnes, createScalar
Alignment	✅	All data 64-byte aligned for SIMD / cache lines
Modern CMake	✅	FetchContent pulls GoogleTest, sanitizers in Debug
Compiler warnings	✅	-Wall -Wextra -Werror (GCC/Clang) and /W4 /WX (MSVC)
UB & AddressSanitizer	✅	Auto-enabled in Debug builds

---

What the 42 Tests Prove

Suite	Tests	Highlights
TensorTest (30)	Shape & stride correctness	1-D … 8-D, shapes with ones, negative/overflow indices
	Element access	Flat index vs. multi-index equivalence
	Memory safety	64-byte alignment, move semantics, unique ownership
	Performance	1 000 000-element zero-init < 50 ms
TensorOpsTest (12)	Scalar & 2-D ops	+, -, * with exact element verification
	Error handling	Mismatched shapes throw invalid_argument
	IEEE-754 edge cases	inf - inf = nan, nan + x = nan
	Alias safety	t*t does not mutate original tensor

All tests pass in 31 ms on a cold run (Release build).

---

Quick Start

git clone https://github.com/yourname/minitensor.git
cd minitensor

cmake -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build --parallel
ctest --test-dir build --output-on-failure

or

mkdir build && cd build
cmake ..
make

Run any program you want