Allow loop-idiom to run on multiple BB loops, but still only scan the loop
header for now for memset/memcpy opportunities. It turns out that loop-rotate
is successfully rotating loops, but *DOESN'T MERGE THE BLOCKS*, turning "for
loops" into 2 basic block loops that loop-idiom was ignoring.
With this fix, we form many *many* more memcpy and memsets than before, including
on the "history" loops in the viterbi benchmark, which look like this:
for (j=0; j<MAX_history; ++j) {
history_new[i][j+1] = history[2*i][j];
}
Transforming these loops into memcpy's speeds up the viterbi benchmark from
11.98s to 3.55s on my machine. Woo.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122685 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/test/Transforms/LoopIdiom/basic.ll b/test/Transforms/LoopIdiom/basic.ll
index 8929fe4..589fea4 100644
--- a/test/Transforms/LoopIdiom/basic.ll
+++ b/test/Transforms/LoopIdiom/basic.ll
@@ -21,6 +21,30 @@
; CHECK-NOT: store
}
+; This is a loop that was rotated but where the blocks weren't merged. This
+; shouldn't perturb us.
+define void @test1a(i8* %Base, i64 %Size) nounwind ssp {
+bb.nph: ; preds = %entry
+ br label %for.body
+
+for.body: ; preds = %bb.nph, %for.body
+ %indvar = phi i64 [ 0, %bb.nph ], [ %indvar.next, %for.body.cont ]
+ %I.0.014 = getelementptr i8* %Base, i64 %indvar
+ store i8 0, i8* %I.0.014, align 1
+ %indvar.next = add i64 %indvar, 1
+ br label %for.body.cont
+for.body.cont:
+ %exitcond = icmp eq i64 %indvar.next, %Size
+ br i1 %exitcond, label %for.end, label %for.body
+
+for.end: ; preds = %for.body, %entry
+ ret void
+; CHECK: @test1a
+; CHECK: call void @llvm.memset.p0i8.i64(i8* %Base, i8 0, i64 %Size, i32 1, i1 false)
+; CHECK-NOT: store
+}
+
+
define void @test2(i32* %Base, i64 %Size) nounwind ssp {
entry:
%cmp10 = icmp eq i64 %Size, 0