[Chapter8] Added more TODOs and placeholders

dendibakh · Mar 14, 2024 · 8f0d3b9 · 8f0d3b9
1 parent c23864a
commit 8f0d3b9
Show file tree

Hide file tree

Showing 8 changed files with 12 additions and 1 deletion.
diff --git a/chapters/8-Optimizing-Memory-Accesses/8-2 Cache-Friendly Data Structures copy.md b/chapters/8-Optimizing-Memory-Accesses/8-2 Cache-Friendly Data Structures copy.md
@@ -142,6 +142,12 @@ One of the most important areas for alignment considerations is the SIMD code. W
 __m512 * ptr = new __m512[N];
 ```
 
+[TODO]: other data structure reorganization techniques:
+* structure splitting
+* field reordering
+* pointer inlining 
+
+
 [TODO]: Trim footnotes
 
 [^5]: The same applies to memory deallocation.

diff --git a/chapters/8-Optimizing-Memory-Accesses/8-3 Dynamic Memory Allocation.md b/chapters/8-Optimizing-Memory-Accesses/8-3 Dynamic Memory Allocation.md
@@ -9,7 +9,6 @@ Secondly, it is possible to speed up allocations using custom allocators, for ex
 ### Tune the Code for Memory Hierarchy.
 
 [TODO]: Elaborate more
-[TODO]: Discuss what to do when memory bandwidth is a limitation
 
 The performance of some applications depends on the size of the cache on a particular level. The most famous example here is improving matrix multiplication with [loop blocking](https://en.wikipedia.org/wiki/Loop_nest_optimization) (tiling). The idea is to break the working size of the matrix into smaller pieces (tiles) such that each tile will fit in the L2 cache.[^9] Most of the architectures provide `CPUID`-like instruction,[^11] which allows us to query the size of caches. Alternatively, one can use [cache-oblivious algorithms](https://en.wikipedia.org/wiki/Cache-oblivious_algorithm)[^19] whose goal is to work reasonably well for any size of the cache.
 

diff --git a/...ers/8-Optimizing-Memory-Accesses/8-4 Workaround Memory Bandwidth Limitations.md b/...ers/8-Optimizing-Memory-Accesses/8-4 Workaround Memory Bandwidth Limitations.md
@@ -0,0 +1,4 @@
+## Workaround Memory Bandwidth Limitations
+
+[TODO]: Discuss what to do when memory bandwidth is a limitation
+Use smaller data types, e.g. fp16, or buy a better HW.
diff --git a/...Memory-Accesses/8-4 Memory Prefetching.md → ...Memory-Accesses/8-5 Memory Prefetching.md b/...Memory-Accesses/8-4 Memory Prefetching.md → ...Memory-Accesses/8-5 Memory Prefetching.md
diff --git a/chapters/8-Optimizing-Memory-Accesses/8-6 Non-temporal stores.md b/chapters/8-Optimizing-Memory-Accesses/8-6 Non-temporal stores.md
@@ -0,0 +1,2 @@
+## Non-temporal stores
+
diff --git a/...mory-Accesses/8-5 Reducing DTLB misses.md → ...mory-Accesses/8-7 Reducing DTLB misses.md b/...mory-Accesses/8-5 Reducing DTLB misses.md → ...mory-Accesses/8-7 Reducing DTLB misses.md
diff --git a/...emory-Accesses/8-6 Questions-Exercises.md → ...emory-Accesses/8-8 Questions-Exercises.md b/...emory-Accesses/8-6 Questions-Exercises.md → ...emory-Accesses/8-8 Questions-Exercises.md
diff --git a/...ng-Memory-Accesses/8-7 Chapter Summary.md → ...ng-Memory-Accesses/8-9 Chapter Summary.md b/...ng-Memory-Accesses/8-7 Chapter Summary.md → ...ng-Memory-Accesses/8-9 Chapter Summary.md