diff options
author | Owen Anderson <resistor@mac.com> | 2010-09-08 23:10:07 +0000 |
---|---|---|
committer | Owen Anderson <resistor@mac.com> | 2010-09-08 23:10:07 +0000 |
commit | 4975447d94bc5a3fcfdd3fd530e77163d4dab11e (patch) | |
tree | 302d6650be2016088a5da0518a2b319d7c973ac5 /test | |
parent | a88d8577e696e35bf72bbc2e5ab1f7e2002b8cc4 (diff) | |
download | llvm-4975447d94bc5a3fcfdd3fd530e77163d4dab11e.tar.gz llvm-4975447d94bc5a3fcfdd3fd530e77163d4dab11e.tar.bz2 llvm-4975447d94bc5a3fcfdd3fd530e77163d4dab11e.tar.xz |
Relax the "don't unroll loops containing calls" rule. Instead, when a loop contains a call, lower the
unrolling threshold to the optimize-for-size threshold. Basically, for loops containing calls, unrolling
can still be profitable as long as the loop is REALLY small.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@113439 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'test')
-rw-r--r-- | test/Transforms/LoopUnroll/call.ll | 51 |
1 files changed, 51 insertions, 0 deletions
diff --git a/test/Transforms/LoopUnroll/call.ll b/test/Transforms/LoopUnroll/call.ll new file mode 100644 index 0000000000..5dcb3bc2af --- /dev/null +++ b/test/Transforms/LoopUnroll/call.ll @@ -0,0 +1,51 @@ +; RUN: opt < %s -S -loop-unroll | FileCheck %s + +@id = internal global i32 0 +@val = internal global [4 x i32] zeroinitializer, align 16 + +; CHECK: @test +define i32 @test(i32 %k) nounwind ssp { +; CHECK-NOT: call i32 @test(i32 %t.06) +; CHECK: call i32 @test(i32 0) +; CHECK-NOT: call i32 @test(i32 %t.06) +; CHECK: call i32 @test(i32 1) +; CHECK-NOT: call i32 @test(i32 %t.06) +; CHECK: call i32 @test(i32 2) +; CHECK-NOT: call i32 @test(i32 %t.06) +; CHECK: call i32 @test(i32 3) +; CHECK-NOT: call i32 @test(i32 %t.06) + +bb.nph: + %0 = load i32* @id, align 4 + %1 = add nsw i32 %0, 1 + store i32 %1, i32* @id, align 4 + %2 = sext i32 %k to i64 + %3 = getelementptr inbounds [4 x i32]* @val, i64 0, i64 %2 + store i32 %1, i32* %3, align 4 + br label %bb + +bb: ; preds = %bb2, %bb.nph + %indvar = phi i64 [ 0, %bb.nph ], [ %indvar.next, %bb2 ] + %scevgep = getelementptr [4 x i32]* @val, i64 0, i64 %indvar + %4 = load i32* %scevgep, align 4 + %5 = icmp eq i32 %4, 0 + br i1 %5, label %bb1, label %bb2 + +bb1: ; preds = %bb + %t.06 = trunc i64 %indvar to i32 + %6 = tail call i32 @test(i32 %t.06) nounwind + br label %bb2 + +bb2: ; preds = %bb1, %bb + %indvar.next = add i64 %indvar, 1 + %exitcond = icmp eq i64 %indvar.next, 4 + br i1 %exitcond, label %bb4, label %bb + +bb4: ; preds = %bb2 + %.pre = load i32* @id, align 4 + %7 = add nsw i32 %.pre, -1 + store i32 %7, i32* @id, align 4 + store i32 0, i32* %3, align 4 + ret i32 undef +; CHECK: } +} |