Add unit tests for template optimization helpers #3610

tenpercent · 2026-01-19T22:36:33Z

Summary

Add Google Tests for the optimized template utilities from the PR stack to ensure correctness and prevent regressions.

Tests Added

Test File	Tests	Coverage
`unit_sequence.cpp`	4 new	`sequence_gen` with custom functors
`unit_sequence_helper.cpp`	17 new	`generate_identity_sequences`, `find_in_tuple_of_sequences`, `sequence_find_value`
`unit_container_helper.cpp`	12 new	`container_concat`, `make_uniform_tuple`
`unit_tensor_descriptor_helper.cpp`	10 new	`compute_element_space_size`

Total: 43 new tests

Test Coverage by PR

Optimize sequence_gen and uniform_sequence_gen to reduce template instantiation depth #3585: sequence_gen with __make_integer_seq
Add generate_identity_sequences helper for common pattern #3588: generate_identity_sequences helper
Add container and tuple optimization helpers #3590: container_concat, make_uniform_tuple
Rewrite O(N) recursive templates with O(1) pack expansion #3596: compute_element_space_size
Replace nested static_for lambdas with compile-time search helper #3600: find_in_tuple_of_sequences, sequence_find_value

PR Stack

#	PR	Description
1	#3585	sequence_gen with `__make_integer_seq`
2	#3588	generate_identity_sequences helper
3	#3589	Named functors in transform_tensor_descriptor
4	#3590	container_concat optimization
5	#3596	O(1) pack expansion rewrites
6	#3600	TensorDescriptor/TensorAdaptor lambda elimination
7	#3608	Build time optimization documentation
8	This PR	Unit tests for optimization helpers

Tracking issue: #3575

Add Google Tests for optimized template utilities: - sequence_gen: Tests with custom functors (4 tests) - generate_identity_sequences: Tuple of identity sequences (4 tests) - find_in_tuple_of_sequences: O(1) sequence search (6 tests) - sequence_find_value: Value lookup in sequences (5 tests) - container_concat: Tuple/array concatenation (5 tests) - make_uniform_tuple: Repeated value tuples (4 tests) - compute_element_space_size: Fold expression (8 tests) - unpack_and_merge_sequences: Sequence merging (2 tests) Total: 43 new tests across 4 test files.

shumway · 2026-01-20T04:43:25Z

test/util/unit_tensor_descriptor_helper.cpp

+    // element_space_size = 1 + (4-1)*4 + (4-1)*1 = 1 + 12 + 3 = 16
+    constexpr auto lengths = make_tuple(Number<4>{}, Number<4>{});
+    constexpr auto strides = make_tuple(Number<4>{}, Number<1>{});
+    constexpr auto result  = detail::compute_element_space_size(lengths, strides, Sequence<0, 1>{});


Why are we testing an implementation detail?

shumway · 2026-01-20T04:47:04Z

test/util/unit_tensor_descriptor_helper.cpp

+    EXPECT_EQ(result, 5);
+}
+
+// Test make_naive_tensor_descriptor uses compute_element_space_size correctly


This is the right way to test the element space size.

shumway · 2026-01-20T04:49:11Z

test/util/unit_tensor_descriptor_helper.cpp

+// Test make_naive_tensor_descriptor uses compute_element_space_size correctly
+TEST(MakeNaiveTensorDescriptor, ElementSpaceSize2D)
+{
+    constexpr auto lengths = make_tuple(Number<4>{}, Number<4>{});


Can we avoid duplicate values unless we need to? Maybe use lengths 5, 4 with strides 4, 1?

shumway · 2026-01-20T04:56:45Z

test/util/unit_tensor_descriptor_helper.cpp

+{
+    // 4x4 tensor with padded strides [8, 1] (8 elements per row instead of 4)
+    // element_space_size = 1 + (4-1)*8 + (4-1)*1 = 1 + 24 + 3 = 28
+    constexpr auto lengths = make_tuple(Number<4>{}, Number<4>{});


With padding we can make all the different (without common factors, maybe prime numbers), e.g., lengths 11, 7, 3 strides 2, 97, 23,

This way the integer math is more likely to catch bad index.
This example is a valid stride, but doesn't have a simple meaning (some tiling ops might not work, but it's a good edge case for testing the element space size).

shumway · 2026-01-20T04:57:29Z

test/util/unit_tensor_descriptor_helper.cpp

+    EXPECT_EQ(desc.GetElementSpaceSize(), 5);
+}
+
+// Test with runtime values (index_t instead of Number<>)


These are good runtime value tests, but again let's avoid square shapes.

Also don't test implementation details.

shumway

This looks good, thanks! It looks like CI failed, so we'll have to recheck. My only feedback is for the element space size:

1.Don't test implementation details (avoid namespace detail::)
2. Avoid square shapes, or at least favor rectangular cases.
3. Prime number or relatives primes can help catch more edge cases (avoid accidentally getting the right answer from a bad example

(using primes or relative primes can lead to big numbers, so you may want a simpler test case, too)

shumway · 2026-01-20T05:07:03Z

test/util/unit_tensor_descriptor_helper.cpp

+    EXPECT_EQ(result, 28);
+}
+
+TEST(ComputeElementSpaceSize, ColumnMajor)


This square gives the same answer for column major or row major, so it's not really testing column major. For packed arrays, ordering doesn't change the packed size.

shumway · 2026-01-20T05:16:56Z

test/util/unit_tensor_descriptor_helper.cpp

+    EXPECT_EQ(result, 16);
+}
+
+TEST(ComputeElementSpaceSizeRuntime, WithPadding)


There are two kinds of padding:

Padding where there is just space for extra elements. That is, you could add more elements without changing strides and get back to a packed array. These paddings often come up taking a stride slice or region out of a packed tensor. (example: shape 2,2,2 strides 1, 3, 9, which is like part of a 3,3,3 tensor)

Padding where there are funny offsets that don't correspond to any packed array. Those are probably worth testing just to verify there aren't any constraints on the implementation. (A simple example: shape 2,2,2 strides 1,3,7, so that elements are at offsets 0, 1, 3, 4, 7, 8, 10, 11)

- Remove tests of implementation details (detail::compute_element_space_size) - Use public API (make_naive_tensor_descriptor) for all tests - Avoid square/cube shapes that could hide row/column major bugs - Use prime numbers for padding tests to catch index calculation errors - Add two padding test cases: arbitrary offsets and stride slice

- Add tests for make_naive_tensor_descriptor_packed (1D, 2D, 3D) - Add tests for make_naive_tensor_descriptor_aligned (2D, 3D) - Add 1D tensor tests with explicit strides - Ensure all shapes use distinct, coprime dimensions

tenpercent requested review from Snektron, ThomasNing, afagaj, andriy-ca, aosewski, asleepzzz, bartekxk, carlushuang, cgmillette, coderfeli, geyyer, illsilin, poyenc, qianfengz, shumway, vidyasagar-amd and vpietila-amd as code owners January 19, 2026 22:36

tenpercent force-pushed the mpodkory/template-optimization-tests branch from 1e1fac9 to 8d2b954 Compare January 19, 2026 22:40

tenpercent force-pushed the mpodkory/build-time-docs branch 2 times, most recently from 9b5a7fa to 867fae3 Compare January 20, 2026 00:15

tenpercent force-pushed the mpodkory/template-optimization-tests branch 2 times, most recently from 69144aa to 0b35651 Compare January 20, 2026 00:31

tenpercent force-pushed the mpodkory/template-optimization-tests branch from 0b35651 to 0c81883 Compare January 20, 2026 00:35

tenpercent force-pushed the mpodkory/build-time-docs branch 6 times, most recently from 97983c9 to ecef7c8 Compare January 20, 2026 01:04

tenpercent force-pushed the mpodkory/build-time-docs branch 2 times, most recently from 80c4f97 to 71413bd Compare January 20, 2026 01:14

shumway reviewed Jan 20, 2026

View reviewed changes

tenpercent added 2 commits January 19, 2026 23:31

Expand tensor descriptor test coverage

97873bc

- Add tests for make_naive_tensor_descriptor_packed (1D, 2D, 3D) - Add tests for make_naive_tensor_descriptor_aligned (2D, 3D) - Add 1D tensor tests with explicit strides - Ensure all shapes use distinct, coprime dimensions

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add unit tests for template optimization helpers #3610

Add unit tests for template optimization helpers #3610

Uh oh!

tenpercent commented Jan 19, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway left a comment

Uh oh!

shumway Jan 20, 2026

Uh oh!

shumway Jan 20, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Add unit tests for template optimization helpers #3610

Are you sure you want to change the base?

Add unit tests for template optimization helpers #3610

Uh oh!

Conversation

tenpercent commented Jan 19, 2026

Summary

Tests Added

Test Coverage by PR

PR Stack

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

shumway left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants