https://github.com/lukel97 updated https://github.com/llvm/llvm-project/pull/237121
>From acc50c3cf623e7c52b4fe15bff255e7f03ff59ac Mon Sep 17 00:00:00 2001 From: Luke Lau <[email protected]> Date: Fri, 30 Oct 2026 14:36:59 +0800 Subject: [PATCH 1/2] Precommit tests --- .../LoopVectorize/VPlan/predicator.ll | 83 +++++++++++++++++++ .../Transforms/LoopVectorize/predicator.ll | 75 ++++++++++++++++- 2 files changed, 157 insertions(+), 1 deletion(-) diff --git a/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll b/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll index 483ab8ad94b50..e8a3654b89c6d 100644 --- a/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll +++ b/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll @@ -751,6 +751,89 @@ exit: ret void } +define void @blend_poison_incoming(ptr noalias %a, ptr noalias %dst) { +; CHECK-LABEL: VPlan for loop in 'blend_poison_incoming' +; CHECK-NEXT: <x1> vector loop: { +; CHECK-NEXT: vp<[[VP3:%[0-9]+]]> = CANONICAL-IV +; CHECK-EMPTY: +; CHECK-NEXT: vector.body: +; CHECK-NEXT: ir<%iv> = WIDEN-INDUCTION nuw nsw ir<0>, ir<1>, vp<[[VP0:%[0-9]+]]> +; CHECK-NEXT: EMIT ir<%gep.a> = getelementptr ir<%a>, ir<%iv> +; CHECK-NEXT: EMIT-SCALAR ir<%x> = load ir<%gep.a> +; CHECK-NEXT: EMIT ir<%c0> = icmp sgt ir<%x>, ir<0> +; CHECK-NEXT: Successor(s): bb2 +; CHECK-EMPTY: +; CHECK-NEXT: bb2: +; CHECK-NEXT: EMIT vp<[[VP4:%[0-9]+]]> = not ir<%c0> +; CHECK-NEXT: EMIT ir<%mul> = mul ir<%x>, ir<3>, vp<[[VP4]]> (!vplan.execution.frequency 6917529027641081855 (37.5%, estimated)) +; CHECK-NEXT: Successor(s): bb1 +; CHECK-EMPTY: +; CHECK-NEXT: bb1: +; CHECK-NEXT: EMIT ir<%c1> = icmp slt ir<%x>, ir<100>, ir<%c0> (!vplan.execution.frequency 11529215046068469760 (62.5%, estimated)) +; CHECK-NEXT: EMIT ir<%add> = add ir<%x>, ir<1>, ir<%c0> (!vplan.execution.frequency 11529215046068469760 (62.5%, estimated)) +; CHECK-NEXT: Successor(s): bb3 +; CHECK-EMPTY: +; CHECK-NEXT: bb3: +; CHECK-NEXT: EMIT vp<[[VP5:%[0-9]+]]> = logical-and ir<%c0>, ir<%c1> +; CHECK-NEXT: Successor(s): bb4 +; CHECK-EMPTY: +; CHECK-NEXT: bb4: +; CHECK-NEXT: EMIT vp<[[VP6:%[0-9]+]]> = not ir<%c1> +; CHECK-NEXT: EMIT vp<[[VP7:%[0-9]+]]> = logical-and ir<%c0>, vp<[[VP6]]> +; CHECK-NEXT: BLEND ir<%phi> = ir<%mul>/vp<[[VP4]]> ir<poison>/vp<[[VP5]]> ir<%add>/vp<[[VP7]]> +; CHECK-NEXT: EMIT ir<%gep.dst> = getelementptr ir<%dst>, ir<%iv> +; CHECK-NEXT: EMIT store ir<%phi>, ir<%gep.dst> +; CHECK-NEXT: EMIT ir<%iv.next> = add nuw nsw ir<%iv>, ir<1> +; CHECK-NEXT: EMIT ir<%ec> = icmp eq ir<%iv.next>, ir<128> +; CHECK-NEXT: EMIT vp<%index.next> = add nuw vp<[[VP3]]>, vp<[[VP1:%[0-9]+]]> +; CHECK-NEXT: EMIT branch-on-count vp<%index.next>, vp<[[VP2:%[0-9]+]]> +; CHECK-NEXT: No successors +; CHECK-NEXT: } +; CHECK-NEXT: Successor(s): middle.block +; +entry: + br label %bb0 + +bb0: +; bb0 +; / \ +; bb1 bb2 +; | \ | +; bb3 | | +; \ | / +; bb4 +; Lanes that reach bb4 from bb3 can take any value, so the blend for %phi +; doesn't need the bb3->bb4 edge. + %iv = phi i64 [ 0, %entry ], [ %iv.next, %bb4 ] + %gep.a = getelementptr i32, ptr %a, i64 %iv + %x = load i32, ptr %gep.a + %c0 = icmp sgt i32 %x, 0 + br i1 %c0, label %bb1, label %bb2 + +bb1: + %c1 = icmp slt i32 %x, 100 + %add = add i32 %x, 1 + br i1 %c1, label %bb3, label %bb4 + +bb3: + br label %bb4 + +bb2: + %mul = mul i32 %x, 3 + br label %bb4 + +bb4: + %phi = phi i32 [ %add, %bb1 ], [ poison, %bb3 ], [ %mul, %bb2 ] + %gep.dst = getelementptr i32, ptr %dst, i64 %iv + store i32 %phi, ptr %gep.dst + %iv.next = add nsw nuw i64 %iv, 1 + %ec = icmp eq i64 %iv.next, 128 + br i1 %ec, label %exit, label %bb0 + +exit: + ret void +} + define void @simplifiable_blend(i1 %c1, i1 %c2, i1 %c3, i32 %x, i32 %y, ptr %p) { ; CHECK-LABEL: VPlan for loop in 'simplifiable_blend' ; CHECK-NEXT: <x1> vector loop: { diff --git a/llvm/test/Transforms/LoopVectorize/predicator.ll b/llvm/test/Transforms/LoopVectorize/predicator.ll index 57414ae62c341..1c609454085e6 100644 --- a/llvm/test/Transforms/LoopVectorize/predicator.ll +++ b/llvm/test/Transforms/LoopVectorize/predicator.ll @@ -232,6 +232,79 @@ exit: ret void } +define void @blend_poison_incoming(ptr noalias %a, ptr noalias %dst) { +; CHECK-LABEL: define void @blend_poison_incoming( +; CHECK-SAME: ptr noalias [[A:%.*]], ptr noalias [[DST:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: br label %[[VECTOR_PH:.*]] +; CHECK: [[VECTOR_PH]]: +; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] +; CHECK: [[VECTOR_BODY]]: +; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP0:%.*]] = getelementptr i32, ptr [[A]], i64 [[INDEX]] +; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4 +; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt <4 x i32> [[WIDE_LOAD]], zeroinitializer +; CHECK-NEXT: [[TMP2:%.*]] = mul <4 x i32> [[WIDE_LOAD]], splat (i32 3) +; CHECK-NEXT: [[TMP3:%.*]] = icmp slt <4 x i32> [[WIDE_LOAD]], splat (i32 100) +; CHECK-NEXT: [[TMP4:%.*]] = add <4 x i32> [[WIDE_LOAD]], splat (i32 1) +; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[TMP3]], <4 x i1> zeroinitializer +; CHECK-NEXT: [[TMP6:%.*]] = xor <4 x i1> [[TMP3]], splat (i1 true) +; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[TMP6]], <4 x i1> zeroinitializer +; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP5]], <4 x i32> poison, <4 x i32> [[TMP2]] +; CHECK-NEXT: [[PREDPHI1:%.*]] = select <4 x i1> [[TMP7]], <4 x i32> [[TMP4]], <4 x i32> [[PREDPHI]] +; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]] +; CHECK-NEXT: store <4 x i32> [[PREDPHI1]], ptr [[TMP8]], align 4 +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 +; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 128 +; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; CHECK: [[MIDDLE_BLOCK]]: +; CHECK-NEXT: br label %[[EXIT:.*]] +; CHECK: [[EXIT]]: +; CHECK-NEXT: ret void +; +entry: + br label %bb0 + +bb0: +; bb0 +; / \ +; bb1 bb2 +; | \ | +; bb3 | | +; \ | / +; bb4 +; Lanes that reach bb4 from bb3 can take any value, so the blend for %phi +; doesn't need the bb3->bb4 edge. + %iv = phi i64 [ 0, %entry ], [ %iv.next, %bb4 ] + %gep.a = getelementptr i32, ptr %a, i64 %iv + %x = load i32, ptr %gep.a + %c0 = icmp sgt i32 %x, 0 + br i1 %c0, label %bb1, label %bb2 + +bb1: + %c1 = icmp slt i32 %x, 100 + %add = add i32 %x, 1 + br i1 %c1, label %bb3, label %bb4 + +bb3: + br label %bb4 + +bb2: + %mul = mul i32 %x, 3 + br label %bb4 + +bb4: + %phi = phi i32 [ %add, %bb1 ], [ poison, %bb3 ], [ %mul, %bb2 ] + %gep.dst = getelementptr i32, ptr %dst, i64 %iv + store i32 %phi, ptr %gep.dst + %iv.next = add nsw nuw i64 %iv, 1 + %ec = icmp eq i64 %iv.next, 128 + br i1 %ec, label %exit, label %bb0 + +exit: + ret void +} + define void @simplifiable_blend(i1 %c1, i1 %c2, i1 %c3, i32 %x, i32 %y, ptr %p) { ; CHECK-LABEL: define void @simplifiable_blend( ; CHECK-SAME: i1 [[C1:%.*]], i1 [[C2:%.*]], i1 [[C3:%.*]], i32 [[X:%.*]], i32 [[Y:%.*]], ptr [[P:%.*]]) { @@ -250,7 +323,7 @@ define void @simplifiable_blend(i1 %c1, i1 %c2, i1 %c3, i32 %x, i32 %y, ptr %p) ; CHECK-NEXT: store <4 x i32> [[PREDPHI7]], ptr [[TMP3]], align 4 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 ; CHECK-NEXT: [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 128 -; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: ; CHECK-NEXT: br label %[[EXIT:.*]] ; CHECK: [[EXIT]]: >From 1fab03a2c8160173618c435f2a525bec8a5db460 Mon Sep 17 00:00:00 2001 From: Luke Lau <[email protected]> Date: Wed, 28 Oct 2026 17:52:14 +0800 Subject: [PATCH 2/2] [VPlan] Skip poison incoming values computeBlendEdges A poison value can be replaced with anything, so we don't need to create an incoming value for it when converting phis to blends. Needed to avoid regressions in an upcoming PR to keep SSA with early exits. --- llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp | 2 ++ llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll | 2 +- llvm/test/Transforms/LoopVectorize/predicator.ll | 7 ++----- 3 files changed, 5 insertions(+), 6 deletions(-) diff --git a/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp b/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp index a66471e1d5bb7..49845e3d598a3 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanPredicator.cpp @@ -310,6 +310,8 @@ VPPredicator::computeBlendEdges(VPPhi *Phi) { } } + Edges.remove_if([](auto X) { return match(X.second, m_Poison()); }); + return Edges; } diff --git a/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll b/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll index e8a3654b89c6d..46887ec305304 100644 --- a/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll +++ b/llvm/test/Transforms/LoopVectorize/VPlan/predicator.ll @@ -780,7 +780,7 @@ define void @blend_poison_incoming(ptr noalias %a, ptr noalias %dst) { ; CHECK-NEXT: bb4: ; CHECK-NEXT: EMIT vp<[[VP6:%[0-9]+]]> = not ir<%c1> ; CHECK-NEXT: EMIT vp<[[VP7:%[0-9]+]]> = logical-and ir<%c0>, vp<[[VP6]]> -; CHECK-NEXT: BLEND ir<%phi> = ir<%mul>/vp<[[VP4]]> ir<poison>/vp<[[VP5]]> ir<%add>/vp<[[VP7]]> +; CHECK-NEXT: BLEND ir<%phi> = ir<%mul>/vp<[[VP4]]> ir<%add>/vp<[[VP7]]> ; CHECK-NEXT: EMIT ir<%gep.dst> = getelementptr ir<%dst>, ir<%iv> ; CHECK-NEXT: EMIT store ir<%phi>, ir<%gep.dst> ; CHECK-NEXT: EMIT ir<%iv.next> = add nuw nsw ir<%iv>, ir<1> diff --git a/llvm/test/Transforms/LoopVectorize/predicator.ll b/llvm/test/Transforms/LoopVectorize/predicator.ll index 1c609454085e6..8a7b4fb2f7181 100644 --- a/llvm/test/Transforms/LoopVectorize/predicator.ll +++ b/llvm/test/Transforms/LoopVectorize/predicator.ll @@ -245,13 +245,10 @@ define void @blend_poison_incoming(ptr noalias %a, ptr noalias %dst) { ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP0]], align 4 ; CHECK-NEXT: [[TMP1:%.*]] = icmp sgt <4 x i32> [[WIDE_LOAD]], zeroinitializer ; CHECK-NEXT: [[TMP2:%.*]] = mul <4 x i32> [[WIDE_LOAD]], splat (i32 3) -; CHECK-NEXT: [[TMP3:%.*]] = icmp slt <4 x i32> [[WIDE_LOAD]], splat (i32 100) +; CHECK-NEXT: [[TMP6:%.*]] = icmp sge <4 x i32> [[WIDE_LOAD]], splat (i32 100) ; CHECK-NEXT: [[TMP4:%.*]] = add <4 x i32> [[WIDE_LOAD]], splat (i32 1) -; CHECK-NEXT: [[TMP5:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[TMP3]], <4 x i1> zeroinitializer -; CHECK-NEXT: [[TMP6:%.*]] = xor <4 x i1> [[TMP3]], splat (i1 true) ; CHECK-NEXT: [[TMP7:%.*]] = select <4 x i1> [[TMP1]], <4 x i1> [[TMP6]], <4 x i1> zeroinitializer -; CHECK-NEXT: [[PREDPHI:%.*]] = select <4 x i1> [[TMP5]], <4 x i32> poison, <4 x i32> [[TMP2]] -; CHECK-NEXT: [[PREDPHI1:%.*]] = select <4 x i1> [[TMP7]], <4 x i32> [[TMP4]], <4 x i32> [[PREDPHI]] +; CHECK-NEXT: [[PREDPHI1:%.*]] = select <4 x i1> [[TMP7]], <4 x i32> [[TMP4]], <4 x i32> [[TMP2]] ; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[DST]], i64 [[INDEX]] ; CHECK-NEXT: store <4 x i32> [[PREDPHI1]], ptr [[TMP8]], align 4 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 _______________________________________________ llvm-branch-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
