Enable runtime unrolling with unroll pragma metadata

[meheff]

Summary

Enable runtime unrolling for loops with unroll count metadata ("#pragma unroll N")
and a runtime trip count. Also, do not unroll loops with unroll full metadata if the
loop has a runtime loop count. Previously, such loops would be unrolled with a
very large threshold (pragma-unroll-threshold) if runtime unrolled happened to be
enabled resulting in a very large (and likely unwise) unroll factor.

[broune]

LGTM.

[hfinkel]

LGTM. But please don't commit this until my comment in D10857 is addressed.

[meheff]

Update unrolling behavior for unroll(full) directive as discussed in http://reviews.llvm.org/D10857.

[meheff]

Hal,

From your earlier comments it wasn't clear whether you were suggesting that
the front end should add unroll.full and unroll.disable.runtime metadata
with "#pragma unroll" OR if unroll.full metadata should imply no runtime
unrolling in the optimizer. I opted for the latter. It seemed more natural
for the front-end just to do a straight-forward map of the directives to
metadata, and then the optimizer has the logic to properly deal with funny
cases like pragma unroll on a runtime trip count loop.

Mark

[hfinkel]

In D10854#202586, @meheff wrote:

Hal,

From your earlier comments it wasn't clear whether you were suggesting that
the front end should add unroll.full and unroll.disable.runtime metadata
with "#pragma unroll" OR if unroll.full metadata should imply no runtime
unrolling in the optimizer. I opted for the latter. It seemed more natural
for the front-end just to do a straight-forward map of the directives to
metadata, and then the optimizer has the logic to properly deal with funny
cases like pragma unroll on a runtime trip count loop.

Yes, makes sense to me. Specifically requesting full unrolling on a loop with a non-static trip count does not make sense, and the optimizer should ignore it.

Mark

lib/Transforms/Scalar/LoopUnrollPass.cpp

bool LoopUnroll::runOnLoop(Loop *L, LPPassManager &LPM) {
   } else if (TripCount && Count < TripCount) {
     Unrolling = Partial;
   } else {
     Unrolling = Runtime;
   }
 
   // Reduce count based on the type of unrolling and the threshold values.
   unsigned OriginalCount = Count;
-  bool AllowRuntime = UserRuntime ? CurrentRuntime : UP.Runtime;
-  if (HasRuntimeUnrollDisablePragma(L)) {
+  bool AllowRuntime =
+      (PragmaCount > 0) || (UserRuntime ? CurrentRuntime : UP.Runtime);
+  // Don't unroll a runtime trip count loop with unroll full pragma.
+  if (HasRuntimeUnrollDisablePragma(L) || PragmaFullUnroll) {
     AllowRuntime = false;
   }
   if (Unrolling == Partial) {
     bool AllowPartial = UserAllowPartial ? CurrentAllowPartial : UP.Partial;
     if (!AllowPartial && !CountSetExplicitly) {
       DEBUG(dbgs() << "  will not try to unroll partially because "
                    << "-unroll-allow-partial not given\n");
       return false;

test/Transforms/LoopUnroll/unroll-pragmas.ll

 
 for.end:                                          ; preds = %for.body
   ret void
 }
 
 ; #pragma clang loop unroll(full)
 ; Loop should be fully unrolled.
 ;
-; CHECK-LABEL: @loop64_with_enable(
+; CHECK-LABEL: @loop64_with_full(
 ; CHECK-NOT: br i1
-define void @loop64_with_enable(i32* nocapture %a) {
+define void @loop64_with_full(i32* nocapture %a) {
 entry:
   br label %for.body
 
 for.body:                                         ; preds = %for.body, %entry
   %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
   %0 = load i32, i32* %arrayidx, align 4
   %inc = add nsw i32 %0, 1

 
 for.end:                                          ; preds = %for.body
   ret void
 }
 !5 = !{!5, !6}
 !6 = !{!"llvm.loop.unroll.count", i32 4}
 
 ; #pragma clang loop unroll(full)
-; Full unrolling is requested, but loop has a dynamic trip count so
+; Full unrolling is requested, but loop has a runtime trip count so
 ; no unrolling should occur.
 ;
-; CHECK-LABEL: @dynamic_loop_with_enable(
+; CHECK-LABEL: @runtime_loop_with_full(
 ; CHECK: store i32
 ; CHECK-NOT: store i32
-; CHECK: br i1
-define void @dynamic_loop_with_enable(i32* nocapture %a, i32 %b) {
+define void @runtime_loop_with_full(i32* nocapture %a, i32 %b) {
 entry:
   %cmp3 = icmp sgt i32 %b, 0
   br i1 %cmp3, label %for.body, label %for.end, !llvm.loop !8
 
 for.body:                                         ; preds = %entry, %for.body
   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
   %0 = load i32, i32* %arrayidx, align 4
   %inc = add nsw i32 %0, 1
   store i32 %inc, i32* %arrayidx, align 4
   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
   %lftr.wideiv = trunc i64 %indvars.iv.next to i32
   %exitcond = icmp eq i32 %lftr.wideiv, %b
   br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !8
 
 for.end:                                          ; preds = %for.body, %entry
   ret void
 }
 !8 = !{!8, !4}
 
 ; #pragma clang loop unroll_count(4)
-; Loop has a dynamic trip count.  Unrolling should occur, but no
-; conditional branches can be removed.
+; Loop has a runtime trip count.  Runtime unrolling should occur and loop
+; should be duplicated (original and 4x unrolled).
 ;
-; CHECK-LABEL: @dynamic_loop_with_count4(
+; CHECK-LABEL: @runtime_loop_with_count4(
+; CHECK: for.body.prol:
+; CHECK: store
 ; CHECK-NOT: store
 ; CHECK: br i1
+; CHECK: for.body
 ; CHECK: store
-; CHECK: br i1
 ; CHECK: store
-; CHECK: br i1
 ; CHECK: store
-; CHECK: br i1
 ; CHECK: store
+; CHECK-NOT: store
 ; CHECK: br i1
-; CHECK-NOT: br i1
-define void @dynamic_loop_with_count4(i32* nocapture %a, i32 %b) {
+define void @runtime_loop_with_count4(i32* nocapture %a, i32 %b) {
 entry:
   %cmp3 = icmp sgt i32 %b, 0
   br i1 %cmp3, label %for.body, label %for.end, !llvm.loop !9
 
 for.body:                                         ; preds = %entry, %for.body
   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
   %arrayidx = getelementptr inbounds i32, i32* %a, i64 %indvars.iv
   %0 = load i32, i32* %arrayidx, align 4