[Clang] Pragma vectorize_predicate implies vectorize

[SjoerdMeijer]

Summary

New pragma "vectorize_predicate(enable)" now implies "vectorize(enable)",
and it is ignored when vectorization is disabled with e.g.
"vectorize(disable) vectorize_predicate(enable)".

[SjoerdMeijer]

Forgot that this requires a doc change too. Will add that once we're happy with the proposed behaviour.

[Meinersbur]

Mmmh, I would have expected this to work the same way as vectorize_width. According to the docs:

The following example implicitly enables vectorization and interleaving by specifying a vector width and interleaving count:
#pragma clang loop vectorize_width(2) interleave_count(2)
for(...) {
...
}

However, vectorize_width does not automatically set llvm.loop.vectorize.enable. Neither does llvm.loop.vectorize.width > 1 imply LoopVectorizeHints::getForce(). At some places they are checked together, such as LoopVectorizeHints::allowReordering(). Other places, notably LoopVectorizationCostModel::selectVectorizationFactor(), only queries getForce(). That is, vectorize_width(2) does not implicitly force vectorization in practice.

[SjoerdMeijer]

Thanks for pointing this all out!

I am not entirely sure yet what to think about all this as I am new to the loop pragma business, but I think it looks inconsistent to me!

I think I find allowReordering() a little bit ugly, because it is also checking getWidth() > 1, and I am probably expecting the allowReordering decision to be based on 1 thing (the Force). But in the current implementation that probably makes sense though if vectorizer_width doesn't set vectorizer.enable (again, which is not what I would expect). Thus, it makes sense to me that LoopVectorizationCostModel::selectVectorizationFactor() only checks getForce()

Mmmh, I would have expected this to work the same way as vectorize_width.

So I don't see at this moment how it could work in the same way. And by working in the same way I think you mean adding vectorize_predicate checks to different places where vectorize_width is also checked, but that doesn't sound ideal to me.

Could a conclusion be that I was lucky? Lucky, because without all this knowledge that I have now, this patch looks to do what we would expect?

[fhahn]

In D65776#1615834, @Meinersbur wrote:

Mmmh, I would have expected this to work the same way as vectorize_width. According to the docs:

The following example implicitly enables vectorization and interleaving by specifying a vector width and interleaving count:
#pragma clang loop vectorize_width(2) interleave_count(2)
for(...) {
...
}

However, vectorize_width does not automatically set llvm.loop.vectorize.enable. Neither does llvm.loop.vectorize.width > 1 imply LoopVectorizeHints::getForce(). At some places they are checked together, such as LoopVectorizeHints::allowReordering(). Other places, notably LoopVectorizationCostModel::selectVectorizationFactor(), only queries getForce(). That is, vectorize_width(2) does not implicitly force vectorization in practice.

The current behavior with respect to vectorize_width seems a bit counterintuitive. The docs say vectorize_width implicitly enables vectorization, I would expect it to behave as if vectorize(enable) was also passed. There's also a PR about that mismatch: https://bugs.llvm.org/show_bug.cgi?id=27643

IMO it would make sense to have the more specific pragmas imply vectorize(enable) here (or update the docs accordingly).

[SjoerdMeijer]

Hi Florian, thanks for your input!

IMO it would make sense to have the more specific pragmas imply vectorize(enable) here (or update the docs accordingly).

Yep, fully agree with that, as I also wrote in my previous comment. And thanks for digging up that PR. I think that supports this case and that the proposed behaviour in this patch is what one would expect. If we agree on that, I wouldn't mind fixing that PR too as that seems low hanging-fruit to me.

And to be honest, at the moment I have no clue what the meaning is, or could/should be, when a more specific pragma like vector_width or vector_predicate is set, without also setting/implying vectorize(enable). I need to look into that...

[Meinersbur]

Looking at the similar situation of unroll(enable)/unroll_count(4), unroll_count also does not set llvm.loop.unroll.enable, but it is handled by the LoopUnroll pass itself:

bool ExplicitUnroll = PragmaCount > 0 || PragmaFullUnroll ||
                      PragmaEnableUnroll || UserUnrollCount;

(LoopUnrollPass.cpp line 770f)

I do not know whether/how "setting a transformation option implicitly enables the transformation" should be implemented, maybe we should discuss this. It is currently inconsistent. Also consider that non-Clang frontends and .ll files in the wild might also expect a specific behavior.

[SjoerdMeijer]

I do not know whether/how "setting a transformation option implicitly enables the transformation" should be implemented, maybe we should discuss this.

Yep, agreed. I've sent a mail to the dev list:
http://lists.llvm.org/pipermail/cfe-dev/2019-August/063054.html

[SjoerdMeijer]

thanks @Meinersbur and @fhahn for the discussions on this, also on the dev list.
It looks like we are happy with this behaviour?

[fhahn]

LGTM, thanks! Please wait a few days with committing, in case any additional comments come in via the mailing list.

[SjoerdMeijer]

Thanks for your help! And I will wait a few days.

After this, I will look at that PR, will have a look at diagnostics, and then at the LLVM side of things.

clang/lib/CodeGen/CGLoopInfo.cpp

MDNode *Followup = createUnrollAndJamMetadata(Attrs, FollowupLoopProperties,
                                                 FollowupHasTransforms);
 
   SmallVector<Metadata *, 4> Args;
   TempMDTuple TempNode = MDNode::getTemporary(Ctx, None);
   Args.push_back(TempNode.get());
   Args.append(LoopProperties.begin(), LoopProperties.end());
 
   // Setting vectorize.predicate
-  if (Attrs.VectorizePredicateEnable != LoopAttributes::Unspecified) {
+  bool IsVectorPredicateEnabled = false;
+  if (Attrs.VectorizePredicateEnable != LoopAttributes::Unspecified &&
+      Attrs.VectorizeEnable != LoopAttributes::Disable &&
+      Attrs.VectorizeWidth < 1) {
+
+    IsVectorPredicateEnabled =
+        (Attrs.VectorizePredicateEnable == LoopAttributes::Enable);
+
     Metadata *Vals[] = {
         MDString::get(Ctx, "llvm.loop.vectorize.predicate.enable"),
-        ConstantAsMetadata::get(ConstantInt::get(
-            llvm::Type::getInt1Ty(Ctx),
-            (Attrs.VectorizePredicateEnable == LoopAttributes::Enable)))};
+        ConstantAsMetadata::get(ConstantInt::get(llvm::Type::getInt1Ty(Ctx),
+                                                 IsVectorPredicateEnabled))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
   // Setting vectorize.width
   if (Attrs.VectorizeWidth > 0) {
     Metadata *Vals[] = {
         MDString::get(Ctx, "llvm.loop.vectorize.width"),
         ConstantAsMetadata::get(ConstantInt::get(llvm::Type::getInt32Ty(Ctx),
                                                  Attrs.VectorizeWidth))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
   // Setting interleave.count
   if (Attrs.InterleaveCount > 0) {
     Metadata *Vals[] = {
         MDString::get(Ctx, "llvm.loop.interleave.count"),
         ConstantAsMetadata::get(ConstantInt::get(llvm::Type::getInt32Ty(Ctx),
                                                  Attrs.InterleaveCount))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
   // Setting vectorize.enable
-  if (Attrs.VectorizeEnable != LoopAttributes::Unspecified) {
+  if (Attrs.VectorizeEnable != LoopAttributes::Unspecified ||
+      IsVectorPredicateEnabled) {
     Metadata *Vals[] = {
         MDString::get(Ctx, "llvm.loop.vectorize.enable"),
         ConstantAsMetadata::get(ConstantInt::get(
             llvm::Type::getInt1Ty(Ctx),
-            (Attrs.VectorizeEnable == LoopAttributes::Enable)))};
+            IsVectorPredicateEnabled
+                ? true
+                : (Attrs.VectorizeEnable == LoopAttributes::Enable)))};
     Args.push_back(MDNode::get(Ctx, Vals));
   }
 
   if (FollowupHasTransforms)
     Args.push_back(MDNode::get(
         Ctx,
         {MDString::get(Ctx, "llvm.loop.vectorize.followup_all"), Followup}));
 

clang/test/CodeGenCXX/pragma-loop-predicate.cpp

 // CHECK-LABEL: @{{.*}}test2{{.*}}(
 // CHECK:       br label {{.*}}, !llvm.loop ![[LOOP2:.*]]
 
   #pragma clang loop vectorize(enable) vectorize_predicate(disable)
   for (int i = 0; i < Length; i++)
     List[i] = i * 2;
 }
 
+// vectorize_predicate(enable) implies vectorize(enable)
+void test3(int *List, int Length) {
+// CHECK-LABEL: @{{.*}}test3{{.*}}(
+// CHECK:       br label {{.*}}, !llvm.loop ![[LOOP3:.*]]
+
+  #pragma clang loop vectorize_predicate(enable)
+  for (int i = 0; i < Length; i++)
+    List[i] = i * 2;
+}
+
+// Check that disabling vectorization means a vectorization width of 1, and
+// also that vectorization_predicate isn't enabled.
+void test4(int *List, int Length) {
+// CHECK-LABEL: @{{.*}}test4{{.*}}(
+// CHECK: br label {{.*}}, !llvm.loop ![[LOOP4:.*]]
+
+  #pragma clang loop vectorize(disable)
+  for (int i = 0; i < Length; i++)
+    List[i] = i * 2;
+}
+
+// Check that vectorize and vectorize_predicate are disabled.
+void test5(int *List, int Length) {
+// CHECK-LABEL: @{{.*}}test5{{.*}}(
+// CHECK: br label {{.*}}, !llvm.loop ![[LOOP5:.*]]
+
+  #pragma clang loop vectorize(disable) vectorize_predicate(enable)
+  for (int i = 0; i < Length; i++)
+    List[i] = i * 2;
+}
+
+
 // CHECK:      ![[LOOP0]] = distinct !{![[LOOP0]], !3}
 // CHECK-NEXT: !3 = !{!"llvm.loop.vectorize.enable", i1 true}
+
 // CHECK-NEXT: ![[LOOP1]] = distinct !{![[LOOP1]], !5, !3}
 // CHECK-NEXT: !5 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}
+
 // CHECK-NEXT: ![[LOOP2]] = distinct !{![[LOOP2]], !7, !3}
 // CHECK-NEXT: !7 = !{!"llvm.loop.vectorize.predicate.enable", i1 false}
+
+// CHECK-NEXT: ![[LOOP3]] = distinct !{![[LOOP3]], !5, !3}
+
+// CHECK-NEXT: ![[LOOP4]] = distinct !{![[LOOP4]], !10}
+// CHECK-NEXT: !10 = !{!"llvm.loop.vectorize.width", i32 1}
+
+// CHECK-NEXT: ![[LOOP5]] = distinct !{![[LOOP5]], !10}