gpu: generic: implement SYCL reduction

sgeor255 · Dec 12, 2024 · def0794 · def0794
1 parent 80ee714
commit def0794
Show file tree

Hide file tree

Showing 9 changed files with 1,373 additions and 2 deletions.
diff --git a/src/common/memory_tracking.hpp b/src/common/memory_tracking.hpp
@@ -281,6 +281,7 @@ enum {
     key_reducer_space_bctx,
     key_reduction,
     key_reduction_1,
+    key_reduction_out,
     key_reorder_cross_space,
     key_reorder_space,
     key_reorder_src_scales,

diff --git a/src/gpu/generic/sycl/README.md b/src/gpu/generic/sycl/README.md
@@ -166,3 +166,8 @@ The implementation supports both forward and backward propagations.
 
 * Supported formats: plain formats with up to 7 dimensions
 * Supported data types: `f32`, `bf16`, `f16`, `s8`, `u8`
+
+## Reduction
+
+* Supported formats: plain formats with up to 6 dimensions
+* Supported data types: `f32`, `bf16`, `f16`, `s8`, `u8`