[VPlan] Ignore poison incoming values when creating blend (#180005)

We have an optimization in VPPredicator when creating blends where if
all the incoming values are the same, we just return that value.

This extends it to handle cases like "phi [%x, %x, poison, %x]" by
ignoring poison values.

This is split off from #176143 to prevent regressions when maintaining
SSA by adding PHIs with a poison incoming value.

3 files changed, 66 insertions, 34 deletions

diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/pr87378-vpinstruction-or-drop-poison-generating-flags.ll b/llvm/test/Transforms/LoopVectorize/RISCV/pr87378-vpinstruction-or-drop-poison-generating-flags.ll
index 7495a215d020..19a2424a9ee7 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/pr87378-vpinstruction-or-drop-poison-generating-flags.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/pr87378-vpinstruction-or-drop-poison-generating-flags.ll
@@ -37,9 +37,7 @@ define void @pr87378_vpinstruction_or_drop_poison_generating_flags(ptr %arg, i64
 ; CHECK-NEXT:    [[TMP20:%.*]] = xor <vscale x 8 x i1> [[TMP14]], splat (i1 true)
 ; CHECK-NEXT:    [[TMP21:%.*]] = select <vscale x 8 x i1> [[TMP13]], <vscale x 8 x i1> [[TMP20]], <vscale x 8 x i1> zeroinitializer
 ; CHECK-NEXT:    [[TMP22:%.*]] = or <vscale x 8 x i1> [[TMP19]], [[TMP21]]
-; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <vscale x 8 x i1> [[TMP21]], i32 0
-; CHECK-NEXT:    [[PREDPHI:%.*]] = select i1 [[TMP23]], i64 poison, i64 [[INDEX]]
-; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr i16, ptr [[ARG]], i64 [[PREDPHI]]
+; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr i16, ptr [[ARG]], i64 [[INDEX]]
 ; CHECK-NEXT:    call void @llvm.vp.store.nxv8i16.p0(<vscale x 8 x i16> zeroinitializer, ptr align 2 [[TMP24]], <vscale x 8 x i1> [[TMP22]], i32 [[TMP25]])
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[TMP8]], [[INDEX]]
 ; CHECK-NEXT:    [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP8]]
@@ -71,7 +69,7 @@ then.2:
   br label %merge
 
 merge:
-  %idx = phi i64 [ poison, %then.1 ], [ %iv, %then.2 ]
+  %idx = phi i64 [ %iv, %then.1 ], [ %iv, %then.2 ]
   %getelementptr = getelementptr i16, ptr %arg, i64 %idx
   store i16 0, ptr %getelementptr, align 2
   br label %loop.latch
diff --git a/llvm/test/Transforms/LoopVectorize/uniform-blend.ll b/llvm/test/Transforms/LoopVectorize/uniform-blend.ll
index 24dc182fe24a..38bdd782257a 100644
--- a/llvm/test/Transforms/LoopVectorize/uniform-blend.ll
+++ b/llvm/test/Transforms/LoopVectorize/uniform-blend.ll
@@ -13,8 +13,7 @@ define void @blend_uniform_iv_trunc(i1 %c) {
 ; CHECK:       [[VECTOR_BODY]]:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP0:%.*]] = trunc i64 [[INDEX]] to i16
-; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[C]], i16 [[TMP0]], i16 poison
-; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i16 [[TMP6]]
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i16 [[TMP0]]
 ; CHECK-NEXT:    store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
 ; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
@@ -36,7 +35,7 @@ loop.next:                                        ; preds = %loop.header
   br label %loop.latch
 
 loop.latch:                                       ; preds = %loop.next, %loop.header
-  %blend = phi i16 [ poison, %loop.header ], [ %iv.trunc.2, %loop.next ]
+  %blend = phi i16 [ %iv.trunc.2, %loop.header ], [ %iv.trunc.2, %loop.next ]
   %dst.ptr = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i16 %blend
   store i16 0, ptr %dst.ptr
   %iv.next = add nuw nsw i64 %iv, 1
@@ -55,11 +54,10 @@ define void @blend_uniform_iv(i1 %c) {
 ; 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:    [[TMP6:%.*]] = select i1 [[C]], i64 [[INDEX]], i64 poison
+; CHECK-NEXT:    [[TMP6:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP6]]
 ; CHECK-NEXT:    store <4 x i16> zeroinitializer, ptr [[TMP7]], align 2
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[TMP6]], 4
 ; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
 ; CHECK-NEXT:    br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
@@ -79,7 +77,7 @@ loop.next:                                        ; preds = %loop.header
   br label %loop.latch
 
 loop.latch:                                       ; preds = %loop.next, %loop.header
-  %blend = phi i64 [ poison, %loop.header ], [ %iv, %loop.next ]
+  %blend = phi i64 [ %iv, %loop.header ], [ %iv, %loop.next ]
   %dst.ptr = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 %blend
   store i16 0, ptr %dst.ptr
   %iv.next = add nuw nsw i64 %iv, 1
@@ -99,12 +97,10 @@ define void @blend_chain_iv(i1 %c) {
 ; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
 ; CHECK:       [[VECTOR_BODY]]:
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[PREDPHI1:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[PREDPHI2:%.*]] = select i1 [[C]], <4 x i64> [[PREDPHI1]], <4 x i64> poison
-; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i64> [[PREDPHI2]], i32 0
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i64> [[PREDPHI2]], i32 1
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i64> [[PREDPHI2]], i32 2
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i64> [[PREDPHI2]], i32 3
+; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT:    [[TMP5:%.*]] = add i64 [[INDEX]], 2
+; CHECK-NEXT:    [[TMP7:%.*]] = add i64 [[INDEX]], 3
 ; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP1]]
 ; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP3]]
 ; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP5]]
@@ -114,7 +110,6 @@ define void @blend_chain_iv(i1 %c) {
 ; CHECK-NEXT:    store i16 0, ptr [[TMP6]], align 2
 ; CHECK-NEXT:    store i16 0, ptr [[TMP8]], align 2
 ; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
-; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[PREDPHI1]], splat (i64 4)
 ; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
 ; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
 ; CHECK:       [[MIDDLE_BLOCK]]:
@@ -137,11 +132,11 @@ loop.next.2:
   br label %loop.next.3
 
 loop.next.3:
-  %blend.1 = phi i64 [ poison, %loop.next ], [ %iv, %loop.next.2 ]
+  %blend.1 = phi i64 [ %iv, %loop.next ], [ %iv, %loop.next.2 ]
   br label %loop.latch
 
 loop.latch:                                       ; preds = %loop.next, %loop.header
-  %blend = phi i64 [ poison, %loop.header ], [ %blend.1, %loop.next.3 ]
+  %blend = phi i64 [ %iv, %loop.header ], [ %blend.1, %loop.next.3 ]
   %dst.ptr = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 %blend
   store i16 0, ptr %dst.ptr
   %iv.next = add nuw nsw i64 %iv, 1
@@ -251,10 +246,57 @@ exit:
   ret void
 }
 
+; Test that blend is optimized away when other incoming values are poison.
+
+define void @blend_poison(ptr %p, i1 %c) {
+; CHECK-LABEL: define void @blend_poison(
+; CHECK-SAME: ptr [[P:%.*]], i1 [[C:%.*]]) {
+; 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:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i64, ptr [[P]], i64 [[INDEX]]
+; CHECK-NEXT:    store <4 x i64> [[VEC_IND]], ptr [[TMP0]], align 4
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add nuw nsw <4 x i64> [[VEC_IND]], splat (i64 4)
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
+; CHECK-NEXT:    br i1 [[TMP1]], 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 %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %latch ]
+  br i1 %c, label %if, label %latch
+
+if:
+  br label %latch
+
+latch:
+  %blend = phi i64 [ poison, %loop ], [ %iv, %if ]
+  %gep = getelementptr i64, ptr %p, i64 %iv
+  store i64 %blend, ptr %gep
+  %iv.next = add nuw nsw i64 %iv, 1
+  %ec = icmp ult i64 %iv, 31
+  br i1 %ec, label %loop, label %exit
+
+exit:
+  ret void
+}
+
+
 ;.
 ; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
 ; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
 ; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
 ; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]], [[META2]]}
 ; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
+; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]], [[META2]]}
 ;.
diff --git a/llvm/test/Transforms/LoopVectorize/unused-blend-mask-for-first-operand.ll b/llvm/test/Transforms/LoopVectorize/unused-blend-mask-for-first-operand.ll
index 5d07341263bc..feab77583a59 100644
--- a/llvm/test/Transforms/LoopVectorize/unused-blend-mask-for-first-operand.ll
+++ b/llvm/test/Transforms/LoopVectorize/unused-blend-mask-for-first-operand.ll
@@ -4,9 +4,9 @@
 target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n32:64-S128"
 
 ; Test cases for https://github.com/llvm/llvm-project/issues/87410.
-define void @test_not_first_lane_only_constant(ptr %A, ptr noalias %B)  {
+define void @test_not_first_lane_only_constant(ptr %A, ptr noalias %B, ptr noalias %C)  {
 ; CHECK-LABEL: define void @test_not_first_lane_only_constant(
-; CHECK-SAME: ptr [[A:%.*]], ptr noalias [[B:%.*]]) {
+; CHECK-SAME: ptr [[A:%.*]], ptr noalias [[B:%.*]], ptr noalias [[C:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[VECTOR_PH:%.*]]
 ; CHECK:       vector.ph:
@@ -45,7 +45,7 @@ else.2:
   br label %loop.latch
 
 loop.latch:
-  %merge = phi ptr [ %B, %else.2 ], [ poison, %loop.header ]
+  %merge = phi ptr [ %B, %else.2 ], [ %C, %loop.header ]
   %l = load i16, ptr %merge, align 2
   %iv.next = add i16 %iv, 1
   store i16 %l, ptr %gep.A
@@ -67,11 +67,7 @@ define void @test_not_first_lane_only_wide_compare(ptr %A, ptr noalias %B, i16 %
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = trunc i32 [[INDEX]] to i16
 ; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[A]], i16 [[OFFSET_IDX]]
-; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i16>, ptr [[TMP1]], align 2
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i16> [[WIDE_LOAD]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i16 [[TMP3]], [[X]]
-; CHECK-NEXT:    [[TMP12:%.*]] = select i1 [[TMP4]], ptr poison, ptr [[B]]
-; CHECK-NEXT:    [[TMP13:%.*]] = load i16, ptr [[TMP12]], align 2
+; CHECK-NEXT:    [[TMP13:%.*]] = load i16, ptr [[B]], align 2
 ; CHECK-NEXT:    [[BROADCAST_SPLATINSERT5:%.*]] = insertelement <4 x i16> poison, i16 [[TMP13]], i64 0
 ; CHECK-NEXT:    [[BROADCAST_SPLAT6:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT5]], <4 x i16> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    store <4 x i16> [[BROADCAST_SPLAT6]], ptr [[TMP1]], align 2
@@ -104,7 +100,7 @@ else.2:
   br label %loop.latch
 
 loop.latch:
-  %merge = phi ptr [ %B, %else.2 ], [ poison, %loop.header ]
+  %merge = phi ptr [ %B, %else.2 ], [ %B, %loop.header ]
   %l = load i16, ptr %merge, align 2
   %iv.next = add i16 %iv, 1
   store i16 %l, ptr %gep.A
@@ -126,11 +122,7 @@ define void @test_not_first_lane_only_wide_compare_incoming_order_swapped(ptr %A
 ; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
 ; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = trunc i32 [[INDEX]] to i16
 ; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i16, ptr [[A]], i16 [[OFFSET_IDX]]
-; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i16>, ptr [[TMP1]], align 2
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i16> [[WIDE_LOAD]], i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i16 [[TMP3]], [[X]]
-; CHECK-NEXT:    [[PREDPHI:%.*]] = select i1 [[TMP4]], ptr poison, ptr [[B]]
-; CHECK-NEXT:    [[TMP12:%.*]] = load i16, ptr [[PREDPHI]], align 2
+; CHECK-NEXT:    [[TMP12:%.*]] = load i16, ptr [[B]], align 2
 ; CHECK-NEXT:    [[BROADCAST_SPLATINSERT3:%.*]] = insertelement <4 x i16> poison, i16 [[TMP12]], i64 0
 ; CHECK-NEXT:    [[BROADCAST_SPLAT4:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT3]], <4 x i16> poison, <4 x i32> zeroinitializer
 ; CHECK-NEXT:    store <4 x i16> [[BROADCAST_SPLAT4]], ptr [[TMP1]], align 2