ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-19 22:39:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Remove concrete performance numbers from BUILD_TIME_OPTIMIZATION.md (#3702)

Browse Source 
Replace specific benchmark numbers with qualitative descriptions since
measurements vary across environments and may become outdated.

Co-authored-by: Claude <noreply@anthropic.com>
This commit is contained in:
Max Podkorytov
2026-02-03 02:54:18 -08:00
committed by GitHub
parent 8b56ffb6ae
commit 3f04d27b68
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
include/ck/BUILD_TIME_OPTIMIZATION.md
View File

@@ -105,7 +105,7 @@ struct generate_identity_sequence
generate_tuple(generate_identity_sequence{}, Number<N>{});
```
This reduced `transform_tensor_descriptor` instantiations from 388 to 32 (92% reduction).
This significantly reduces template instantiations for `transform_tensor_descriptor`.
**Example: container_concat**
@@ -135,7 +135,7 @@ __host__ __device__ constexpr auto container_concat(const Tuple<X...>& tx, const
}
```
This reduced `container_concat` instantiations from 186 to 93 (50% reduction).
This reduces `container_concat` template instantiations.
**Example: make_uniform_tuple**
@@ -192,7 +192,7 @@ __host__ __device__ constexpr index_t find_source_index(Sequence<Is...>)
}
```
This reduced `sequence_map_inverse` instantiations from 45 to 10 (78% reduction) and wall-clock time by 95%.
This significantly reduces `sequence_map_inverse` instantiations and compile time.
### 4. Use Fold Expressions for Accumulation
@@ -222,4 +222,4 @@ __host__ __device__ constexpr auto compute_element_space_size(
}
```
This reduced `calculate_element_space_size` instantiations from 24 to 10 (58% reduction) and wall-clock time by 73%.
This reduces `calculate_element_space_size` instantiations and compile time.