[PATCH] D70866: Improve SLP code snippet

[Leonardo Sandoval via Phabricator via llvm-commits]

lsandov1 marked an inline comment as done.
lsandov1 added inline comments.


================
Comment at: llvm/docs/Vectorizers.rst:420
 
-  void foo(int a1, int a2, int b1, int b2, int *A) {
-    A[0] = a1*(a1 + b1)/b1 + 50*b1/a1;
-    A[1] = a2*(a2 + b2)/b2 + 50*b2/a2;
+  void foo(long a1, long a2, long b1, long b2, long *A) {
+    A[0] = a1*(a1 + b1);
----------------
spatel wrote:
> I understand that integer division is a poor example for vectorization because most targets can't handle it.
> But what is the intent of changing 'int' to 'long' in this example? That change appears to prevent vectorization (for most x86 targets at least).
I was biased by AVX2 in my example (4 longs fits a ymm register) but agree that 'int' is less intrusive and more generic for the proposed code snippet. I will change it.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70866/new/

https://reviews.llvm.org/D70866