[PATCH] D92132: [LV] Support widened induction variables in epilogue vectorization.

[Venkataramanan Kumar via Phabricator via llvm-commits]

venkataramanan.kumar.llvm added a comment.

Hi Florian,
--Snip--
void foo(double * restrict  a,  double * restrict b, int N) {
int k=0;
for(int i = 0 ,  k =2; i < N; ++i, ++k)

  a[i] = sin(i)+k;

}
--Snip--

I am able to see the resume value for induction "i" updated properly now after the patch .

--Snip--
33:                                               ; preds = %8, %28

  %34 = phi i64 [ %11, %28 ], [ 0, %8 ]
  %35 = and i64 %6, 4294967294
  %36 = trunc i64 %35 to i32
  %37 = add i32 %36, 2
  %38 = insertelement <2 x i64> poison, i64 %34, i32 0
  %39 = shufflevector <2 x i64> %38, <2 x i64> poison, <2 x i32> zeroinitializer
  %40 = or <2 x i64> %39, <i64 0, i64 1>
  %41 = trunc i64 %34 to i32
  %42 = insertelement <2 x i32> poison, i32 %41, i32 0
  %43 = shufflevector <2 x i32> %42, <2 x i32> poison, <2 x i32> zeroinitializer
  %44 = or <2 x i32> %43, <i32 0, i32 1>
  br label %45

45:                                               ; preds = %45, %33

  %46 = phi i64 [ %34, %33 ], [ %55, %45 ]
  %47 = phi <2 x i64> [ %40, %33 ], [ %56, %45 ]   <== updated
  %48 = phi <2 x i32> [ %44, %33 ], [ %57, %45 ] <== updated

--Snip--

I was thinking if we could reuse the incremented widened induction results from the main vector loop.  But this seems cleaner and simpler code .

Note the second induction variable k is not updated properly as already pointed out by bmahjour.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D92132/new/

https://reviews.llvm.org/D92132