[PATCH] D12499: [WIP] Replace ScalarEvolution based domain generation

[James Molloy via llvm-commits]

Hi,

Sorry to jump in on this change, but I'm starting to get confused with all
the Polly traffic mentioning core LLVM stuff like ScalarEvolution in its
title. Would it be possible for such changes to have a [Polly] tag added,
like many of the other subprojects/areas of LLVM?

Cheers,

James

On Mon, 31 Aug 2015 at 20:51 Johannes Doerfert via llvm-commits <
llvm-commits at lists.llvm.org> wrote:

> jdoerfert added a comment.
>
> Some comments to understand the test changes.
>
>
> ================
> Comment at:
> test/Isl/CodeGen/OpenMP/loop-body-references-outer-values-3.ll:12
> @@ -11,3 +12,3 @@
>  ; AST: #pragma omp parallel for
> -; AST: for (int c0 = 0; c0 < cols; c0 += 1)
>  ; AST:   Stmt_for_body(c0);
> ----------------
> There was an if (cols >= 2) around this loop and a Stmt_for_body(0) in the
> else case.
> The conditional is gone in the new version.
>
> ================
> Comment at:
> test/Isl/CodeGen/OpenMP/loop-body-references-outer-values-3.ll:42
> @@ -41,3 +43,3 @@
>  ; AST: #pragma omp parallel for
> -; AST: for (int c0 = 0; c0 < cols; c0 += 1)
>  ; AST:   Stmt_for_body(c0);
> ----------------
> There was an if (cols >= 2) around this loop and a Stmt_for_body(0) in the
> else case.
> The conditional is gone in the new version.
>
> ================
> Comment at: test/Isl/CodeGen/OpenMP/reference-preceeding-loop.ll:17
> @@ -16,3 @@
> -; AST-NEXT:   #pragma omp parallel for
> -; AST-NEXT:   for (int c0 = 0; c0 <= p_0 + p_2; c0 += 1)
> -; AST-NEXT:     Stmt_while_body(c0);
> ----------------
> This test was funny, p_0 is actually -p_2, hence the loop had one
> iteration only, namely Stmt_while_body(0)
>
> ================
> Comment at: test/Isl/CodeGen/two-scops-in-row.ll:11
> @@ -10,3 +11,1 @@
> -; SCALAR:       if (Scalar0.val >= 100)
> -; SCALAR:         Stmt_for_1(0);
>  ; SCALAR:     }
> ----------------
> Similar to the case above, Stmt_for_1(0) is always executed here but it is
> explicit in the new version
>
> ================
> Comment at: test/ScopInfo/isl_trip_count_01.ll:3
> @@ -2,3 +2,3 @@
>  ;
> -; CHECK: [M, N] -> { Stmt_while_body[i0] : i0 >= 0 and 4i0 <= -M + N }
>  ;
> ----------------
> Here the original test case is actually broken. If you look at the code
> but ignore the entry node (it is __not__ in the SCoP!) than you see that
> for N=0 and M=1 we would execute while_body but the domain says otherwise.
>
> ================
> Comment at: test/ScopInfo/loop_affine_bound_0.ll:67
> @@ -66,3 +66,1 @@
> -; CHECK-DAG:          and
> -; CHECK-DAG:             N >= 0
>  ; CHECK:               }
> ----------------
> This is redundant as it can be extracted from the rest.
>
> ================
> Comment at: test/ScopInfo/loop_affine_bound_1.ll:19
> @@ -18,3 +18,3 @@
>    %4 = icmp sgt i64 %3, 0                         ; <i1> [#uses=1]
> -  br i1 %4, label %bb.nph8, label %return
> +  br i1 true, label %bb.nph8, label %return
>
> ----------------
> This was just to make clear that the original code did not use the entry
> condition but just generates the wrong domain (similar to the
> isl_trip_count_01.ll above).
>
> ================
> Comment at: test/ScopInfo/loop_affine_bound_1.ll:66
> @@ -65,3 +67,3 @@
>  ; CHECK-DAG:          and
> -; CHECK-DAG:             i0 <= 1 + 5N
>  ; CHECK:               }
> ----------------
> Again a broken test for M=-1 and N=0.
>
> ================
> Comment at: test/ScopInfo/loop_affine_bound_2.ll:77
> @@ -76,3 +76,1 @@
> -; CHECK-DAG:                and
> -; CHECK-DAG:                  4i0 <= 10 + 5N - 6M
>  ; CHECK-DAG:                }
> ----------------
> Again redundant.
>
>
> http://reviews.llvm.org/D12499
>
>
>
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20150831/7bb3619d/attachment.html>