Source

`tests/purrc/ndarray_sys_test.cpp`

2// Original work; see ACKNOWLEDGMENTS.md for the open-source ideas we build upon.

3// System-level (whole-program) test for the `ndarray` stdlib module. Unlike the

4// per-function compile-run tests (tests/purrc/ndarray_cr_test.cpp), this drives a

5// single cohesive numeric program through EVERY purr-callable ndarray function and

6// asserts its exact stdout, so the functions are exercised together (factories feed

7// reshape, reshape feeds elementwise ops, those feed reductions / indexing).

8//

9// Coverage — every purr-callable function in stdlib/ndarray/ndarray.hpp:

10// array, scalar, zeros, ones, full, arange, reshape, add, sub, mul, divide,

11// sum, mean, get, shape_of, size_of, to_string.

12//

13// Skipped (not callable from .purr, same as the cr test):

14// - broadcast_to / broadcast_shapes: take std::vector<std::size_t>, but cheatah

15// `list<int>` lowers to std::vector<long long>, which doesn't convert.

16// - the NDArray class methods (shape/strides/ndim/size/at/buffer/offset/ctors)

17// are C++-side internals reached only through the free functions above.

18#include "e2e_harness.hpp"

20TEST(StdlibE2E, Ndarray) {

21 e2e::expect_e2e("ndarray_sys", R"PURR(import io

22import ndarray

24# --- factories ---

25let a = ndarray.array([1.0, 2.0, 3.0, 4.0, 5.0, 6.0])

26let s = ndarray.scalar(2.0)

27let z = ndarray.zeros([2, 3])

28let o = ndarray.ones([2, 3])

29let f = ndarray.full([2, 3], 4.0)

30let r = ndarray.arange(0.0, 6.0, 1.0)

32# --- reshape into a 2x3 matrix ---

33let m = ndarray.reshape(a, [2, 3])

34io.print(ndarray.to_string(m))

36# --- elementwise ops (broadcasting against scalar / same shape) ---

37let summ = ndarray.add(m, o)

38let diff = ndarray.sub(m, s)

39let prod = ndarray.mul(m, s)

40let quot = ndarray.divide(m, s)

41io.print(ndarray.to_string(summ))

42io.print(ndarray.to_string(diff))

43io.print(ndarray.to_string(prod))

44io.print(ndarray.to_string(quot))

46# combine zeros / full / arange

47let combo = ndarray.add(ndarray.add(z, f), ndarray.reshape(r, [2, 3]))

48io.print(ndarray.to_string(combo))

50# --- reductions ---

51io.print(ndarray.sum(m))

52io.print(ndarray.mean(m))

54# --- access / shape introspection ---

55io.print(ndarray.get(m, [1, 2]))

56let sh = ndarray.shape_of(m)

57io.print(sh[0], sh[1])

58io.print(ndarray.size_of(m))

59)PURR",

60 "[[1, 2, 3], [4, 5, 6]]\n"

61 "[[2, 3, 4], [5, 6, 7]]\n"

62 "[[-1, 0, 1], [2, 3, 4]]\n"

63 "[[2, 4, 6], [8, 10, 12]]\n"

64 "[[0.5, 1, 1.5], [2, 2.5, 3]]\n"

65 "[[4, 5, 6], [7, 8, 9]]\n"

66 "21\n"

67 "3.5\n"

68 "6\n"

69 "2 3\n"

70 "6\n");

71}

73// Complex support (complex/real/imag/conj) exercised together end-to-end: build a

74// complex vector from real & imaginary parts, pull the parts back out, conjugate it.

75TEST(StdlibE2E, NdarrayComplex) {

76 e2e::expect_e2e("ndarray_complex_sys", R"PURR(import io

77import ndarray

79let re = ndarray.array([0.0, 1.0, 2.0])

80let im = ndarray.array([1.0, 0.0, -3.0])

81let z = ndarray.complex(re, im)

83io.print(ndarray.to_string(z))

84io.print(ndarray.to_string(ndarray.conj(z)))

85io.print(ndarray.to_string(ndarray.real(z)))

86io.print(ndarray.to_string(ndarray.imag(z)))

87)PURR",

88 "[0+1j, 1+0j, 2-3j]\n"

89 "[0-1j, 1+0j, 2+3j]\n"

90 "[0, 1, 2]\n"

91 "[1, 0, -3]\n");

92}

94// Element-wise math ufuncs exercised together (the array forms of the math module).

95TEST(StdlibE2E, NdarrayMath) {

96 e2e::expect_e2e("ndarray_math_sys", R"PURR(import io

97import ndarray

99let a = ndarray.array([1.0, 4.0, 9.0, 16.0])

100io.print(ndarray.to_string(ndarray.sqrt(a)))

101io.print(ndarray.to_string(ndarray.cbrt(ndarray.array([1.0, 8.0, 27.0]))))

102io.print(ndarray.to_string(ndarray.abs(ndarray.array([-2.0, 3.0, -4.0]))))

103io.print(ndarray.to_string(ndarray.sin(ndarray.array([0.0]))))

104)PURR",

105 "[1, 2, 3, 4]\n"

106 "[1, 2, 3]\n"

107 "[2, 3, 4]\n"

108 "[0]\n");

109}