@@ -50,8 +50,8 @@ Copyright (c) 2025 Intel Corporation. All rights reserved.
50
50
51
51
[width="40%",cols="25,25"]
52
52
|========================================
53
- | Last Modified Date | 2025-01-07
54
- | Revision | 1
53
+ | Last Modified Date | 2025-02-28
54
+ | Revision | 2
55
55
|========================================
56
56
57
57
== Dependencies
@@ -692,8 +692,9 @@ The following restrictions apply to the 2D block load, store and prefetch instru
692
692
** the per-subgroup source or destination base address is cache-line aligned (64 bytes).
693
693
** the per-invocation source or destination address is aligned to a multiple of the _Element Size_.
694
694
** the _Memory Width_ is greater than or equal to 64 bytes and less than or equal to 2^24^ bytes.
695
+ ** the _Memory Width_ is a multiple of four for 1-byte or 2-byte elements, or a multiple of the element size otherwise.
695
696
** the _Memory Height_ is greater than zero and less than or equal to 2^24^ rows.
696
- ** the _Memory Pitch_ is greater than or equal to the _Memory Width_ and a multiple of 8 bytes.
697
+ ** the _Memory Pitch_ is greater than or equal to the _Memory Width_ and a multiple of 16 bytes.
697
698
** the *SubgroupMaxSize* is a power of two.
698
699
** the *SubgroupSize* is equal to the *SubgroupMaxSize*; in other words, this is a full subgroup.
699
700
@@ -750,4 +751,5 @@ Specifically, out-of-bounds reads are assigned the value zero, and out-of-bounds
750
751
|========================================
751
752
|Rev|Date|Author|Changes
752
753
|1|2025-01-07|Ben Ashbaugh|Initial revision for publication
754
+ |2|2025-02-28|Ben Ashbaugh|Updated restrictions
753
755
|========================================
0 commit comments