[llvm] [VPlan] Mark default value or ExtractLastActive as only first lane used. (PR #226153)
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Sat Sep 26 08:30:47 PDT 2026
https://github.com/fhahn updated https://github.com/llvm/llvm-project/pull/226153
>From a2aba8f2678a07b04ac5b13bf0da7a29e851f207 Mon Sep 17 00:00:00 2001
From: Florian Hahn <flo at fhahn.com>
Date: Thu, 24 Sep 2026 10:59:42 +0100
Subject: [PATCH] [VPlan] Mark default value or ExtractLastActive as only first
lane used.
The default value (operand 0) of ExtractLastActive is the scalar value
of @llvm.experimental.vector.extract.last.active. Only the first lane is
used, mark accordingly.
---
.../lib/Transforms/Vectorize/VPlanRecipes.cpp | 2 ++
...conditional-scalar-assignment-fold-tail.ll | 3 +-
.../AArch64/conditional-scalar-assignment.ll | 28 ++++++++-----------
...conditional-scalar-assignment-fold-tail.ll | 3 +-
.../RISCV/conditional-scalar-assignment.ll | 7 ++---
.../X86/conditional-scalar-assignment.ll | 3 +-
.../LoopVectorize/find-last-ptr-induction.ll | 7 ++---
.../iv-select-cmp-non-const-iv-start.ll | 10 +++----
.../LoopVectorize/iv-select-cmp-trunc.ll | 20 ++++++-------
.../Transforms/LoopVectorize/iv-select-cmp.ll | 10 +++----
10 files changed, 39 insertions(+), 54 deletions(-)
diff --git a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
index a79e8220a5b3d..650dfe21d267b 100644
--- a/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
+++ b/llvm/lib/Transforms/Vectorize/VPlanRecipes.cpp
@@ -1745,6 +1745,8 @@ bool VPInstruction::usesFirstLaneOnly(const VPValue *Op) const {
return Op == getOperand(1);
case Instruction::InsertElement:
return Op == getOperand(1) || Op == getOperand(2);
+ case VPInstruction::ExtractLastActive:
+ return Op == getOperand(0);
case Instruction::PHI:
return true;
case Instruction::FCmp:
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment-fold-tail.ll b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment-fold-tail.ll
index 8c98cd2e9663d..4515d331feb3c 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment-fold-tail.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment-fold-tail.ll
@@ -102,8 +102,7 @@ define i32 @non_speculatable_find_last_reduction(ptr noalias %a, ptr noalias %b,
; CHECK-NEXT: [[TMP16:%.*]] = xor i1 [[TMP15]], true
; CHECK-NEXT: br i1 [[TMP16]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: [[TMP17:%.*]] = extractelement <vscale x 4 x i32> [[BROADCAST_SPLAT2]], i64 0
-; CHECK-NEXT: [[TMP18:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP14]], <vscale x 4 x i1> [[TMP13]], i32 [[TMP17]])
+; CHECK-NEXT: [[TMP18:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP14]], <vscale x 4 x i1> [[TMP13]], i32 [[DEFAULT_VAL]])
; CHECK-NEXT: br label %[[EXIT:.*]]
; CHECK: [[EXIT]]:
; CHECK-NEXT: ret i32 [[TMP18]]
diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment.ll b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment.ll
index ee5b164eae8c3..d197d727eae66 100644
--- a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment.ll
+++ b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-scalar-assignment.ll
@@ -141,13 +141,12 @@ define ptr @simple_csa_ptr_select(i64 %N, ptr %data, i64 %a, ptr %init) {
; SVE-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; SVE-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[LOOP]], !llvm.loop [[LOOP4:![0-9]+]]
; SVE: [[MIDDLE_BLOCK]]:
-; SVE-NEXT: [[TMP13:%.*]] = extractelement <vscale x 2 x ptr> [[BROADCAST_SPLAT2]], i64 0
-; SVE-NEXT: [[TMP14:%.*]] = call ptr @llvm.experimental.vector.extract.last.active.nxv2p0(<vscale x 2 x ptr> [[TMP11]], <vscale x 2 x i1> [[TMP10]], ptr [[TMP13]])
+; SVE-NEXT: [[TMP13:%.*]] = call ptr @llvm.experimental.vector.extract.last.active.nxv2p0(<vscale x 2 x ptr> [[TMP11]], <vscale x 2 x i1> [[TMP10]], ptr [[INIT]])
; SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; SVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; SVE: [[SCALAR_PH]]:
; SVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
-; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi ptr [ [[TMP14]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ]
+; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi ptr [ [[TMP13]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ]
; SVE-NEXT: br label %[[LOOP1:.*]]
; SVE: [[LOOP1]]:
; SVE-NEXT: [[IV1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP1]] ]
@@ -161,7 +160,7 @@ define ptr @simple_csa_ptr_select(i64 %N, ptr %data, i64 %a, ptr %init) {
; SVE-NEXT: [[EXIT_CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; SVE-NEXT: br i1 [[EXIT_CMP]], label %[[EXIT]], label %[[LOOP1]], !llvm.loop [[LOOP5:![0-9]+]]
; SVE: [[EXIT]]:
-; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi ptr [ [[SELECT_DATA]], %[[LOOP1]] ], [ [[TMP14]], %[[MIDDLE_BLOCK]] ]
+; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi ptr [ [[SELECT_DATA]], %[[LOOP1]] ], [ [[TMP13]], %[[MIDDLE_BLOCK]] ]
; SVE-NEXT: ret ptr [[SELECT_DATA_LCSSA]]
;
entry:
@@ -1038,13 +1037,12 @@ define i32 @simple_csa_int_load(ptr noalias %a, ptr noalias %b, i32 %default_val
; SVE-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; SVE-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; SVE: [[MIDDLE_BLOCK]]:
-; SVE-NEXT: [[TMP13:%.*]] = extractelement <vscale x 4 x i32> [[BROADCAST_SPLAT2]], i64 0
-; SVE-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP11]], <vscale x 4 x i1> [[TMP10]], i32 [[TMP13]])
+; SVE-NEXT: [[TMP13:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP11]], <vscale x 4 x i1> [[TMP10]], i32 [[DEFAULT_VAL]])
; SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; SVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; SVE: [[SCALAR_PH]]:
; SVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
-; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP14]], %[[MIDDLE_BLOCK]] ], [ [[DEFAULT_VAL]], %[[ENTRY]] ]
+; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP13]], %[[MIDDLE_BLOCK]] ], [ [[DEFAULT_VAL]], %[[ENTRY]] ]
; SVE-NEXT: br label %[[LOOP:.*]]
; SVE: [[LOOP]]:
; SVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LATCH:.*]] ]
@@ -1063,7 +1061,7 @@ define i32 @simple_csa_int_load(ptr noalias %a, ptr noalias %b, i32 %default_val
; SVE-NEXT: [[EXIT_CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; SVE-NEXT: br i1 [[EXIT_CMP]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP17:![0-9]+]]
; SVE: [[EXIT]]:
-; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH]] ], [ [[TMP14]], %[[MIDDLE_BLOCK]] ]
+; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH]] ], [ [[TMP13]], %[[MIDDLE_BLOCK]] ]
; SVE-NEXT: ret i32 [[SELECT_DATA_LCSSA]]
;
entry:
@@ -1155,13 +1153,12 @@ define i32 @simple_csa_int_divide(ptr noalias %a, ptr noalias %b, i32 %default_v
; SVE-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; SVE-NEXT: br i1 [[TMP13]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]]
; SVE: [[MIDDLE_BLOCK]]:
-; SVE-NEXT: [[TMP14:%.*]] = extractelement <vscale x 4 x i32> [[BROADCAST_SPLAT2]], i64 0
-; SVE-NEXT: [[TMP15:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP12]], <vscale x 4 x i1> [[TMP11]], i32 [[TMP14]])
+; SVE-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP12]], <vscale x 4 x i1> [[TMP11]], i32 [[DEFAULT_VAL]])
; SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; SVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; SVE: [[SCALAR_PH]]:
; SVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
-; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP15]], %[[MIDDLE_BLOCK]] ], [ [[DEFAULT_VAL]], %[[ENTRY]] ]
+; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP14]], %[[MIDDLE_BLOCK]] ], [ [[DEFAULT_VAL]], %[[ENTRY]] ]
; SVE-NEXT: br label %[[LOOP:.*]]
; SVE: [[LOOP]]:
; SVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LATCH:.*]] ]
@@ -1179,7 +1176,7 @@ define i32 @simple_csa_int_divide(ptr noalias %a, ptr noalias %b, i32 %default_v
; SVE-NEXT: [[EXIT_CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; SVE-NEXT: br i1 [[EXIT_CMP]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP19:![0-9]+]]
; SVE: [[EXIT]]:
-; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH]] ], [ [[TMP15]], %[[MIDDLE_BLOCK]] ]
+; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH]] ], [ [[TMP14]], %[[MIDDLE_BLOCK]] ]
; SVE-NEXT: ret i32 [[SELECT_DATA_LCSSA]]
;
entry:
@@ -1283,13 +1280,12 @@ define i32 @csa_load_nested_ifs(ptr noalias %a, ptr noalias %b, i32 %default_val
; SVE-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; SVE-NEXT: br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; SVE: [[MIDDLE_BLOCK]]:
-; SVE-NEXT: [[TMP16:%.*]] = extractelement <vscale x 4 x i32> [[BROADCAST_SPLAT2]], i64 0
-; SVE-NEXT: [[TMP17:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP14]], <vscale x 4 x i1> [[TMP13]], i32 [[TMP16]])
+; SVE-NEXT: [[TMP16:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP14]], <vscale x 4 x i1> [[TMP13]], i32 [[DEFAULT_VAL]])
; SVE-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; SVE-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; SVE: [[SCALAR_PH]]:
; SVE-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
-; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP17]], %[[MIDDLE_BLOCK]] ], [ [[DEFAULT_VAL]], %[[ENTRY]] ]
+; SVE-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP16]], %[[MIDDLE_BLOCK]] ], [ [[DEFAULT_VAL]], %[[ENTRY]] ]
; SVE-NEXT: br label %[[LOOP:.*]]
; SVE: [[LOOP]]:
; SVE-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LATCH:.*]] ]
@@ -1311,7 +1307,7 @@ define i32 @csa_load_nested_ifs(ptr noalias %a, ptr noalias %b, i32 %default_val
; SVE-NEXT: [[EXIT_CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; SVE-NEXT: br i1 [[EXIT_CMP]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP21:![0-9]+]]
; SVE: [[EXIT]]:
-; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH]] ], [ [[TMP17]], %[[MIDDLE_BLOCK]] ]
+; SVE-NEXT: [[SELECT_DATA_LCSSA:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH]] ], [ [[TMP16]], %[[MIDDLE_BLOCK]] ]
; SVE-NEXT: ret i32 [[SELECT_DATA_LCSSA]]
;
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment-fold-tail.ll b/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment-fold-tail.ll
index 06758467221c5..0e4e110ca1997 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment-fold-tail.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment-fold-tail.ll
@@ -96,8 +96,7 @@ define i32 @non_speculatable_find_last_reduction(ptr noalias %a, ptr noalias %b,
; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[AVL_NEXT]], 0
; CHECK-NEXT: br i1 [[TMP14]], label %[[IF_THEN:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
; CHECK: [[IF_THEN]]:
-; CHECK-NEXT: [[TMP15:%.*]] = extractelement <vscale x 4 x i32> [[BROADCAST_SPLAT2]], i64 0
-; CHECK-NEXT: [[SELECT_DATA_LCSSA:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP12]], <vscale x 4 x i1> [[TMP11]], i32 [[TMP15]])
+; CHECK-NEXT: [[SELECT_DATA_LCSSA:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP12]], <vscale x 4 x i1> [[TMP11]], i32 [[DEFAULT_VAL]])
; CHECK-NEXT: br label %[[LATCH:.*]]
; CHECK: [[LATCH]]:
; CHECK-NEXT: ret i32 [[SELECT_DATA_LCSSA]]
diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment.ll b/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment.ll
index f4b3bf607fa9d..d685a238e7d9b 100644
--- a/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment.ll
+++ b/llvm/test/Transforms/LoopVectorize/RISCV/conditional-scalar-assignment.ll
@@ -114,13 +114,12 @@ define i32 @non_speculatable_find_last_reduction(ptr noalias %a, ptr noalias %b,
; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP13]], label %[[IF_THEN:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[IF_THEN]]:
-; CHECK-NEXT: [[TMP15:%.*]] = extractelement <vscale x 4 x i32> [[BROADCAST_SPLAT2]], i64 0
-; CHECK-NEXT: [[SELECT_DATA_LCSSA:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP12]], <vscale x 4 x i1> [[TMP11]], i32 [[TMP15]])
+; CHECK-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.nxv4i32(<vscale x 4 x i32> [[TMP12]], <vscale x 4 x i1> [[TMP11]], i32 [[DEFAULT_VAL]])
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT1:.*]], label %[[SCALAR_PH1]]
; CHECK: [[SCALAR_PH1]]:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[IF_THEN]] ], [ 0, %[[SCALAR_PH]] ]
-; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[SELECT_DATA_LCSSA]], %[[IF_THEN]] ], [ [[DEFAULT_VAL]], %[[SCALAR_PH]] ]
+; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP14]], %[[IF_THEN]] ], [ [[DEFAULT_VAL]], %[[SCALAR_PH]] ]
; CHECK-NEXT: br label %[[LATCH:.*]]
; CHECK: [[LATCH]]:
; CHECK-NEXT: [[IV1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH1]] ], [ [[IV_NEXT:%.*]], %[[LATCH1:.*]] ]
@@ -139,7 +138,7 @@ define i32 @non_speculatable_find_last_reduction(ptr noalias %a, ptr noalias %b,
; CHECK-NEXT: [[EXIT_CMP:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; CHECK-NEXT: br i1 [[EXIT_CMP]], label %[[EXIT1]], label %[[LATCH]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: [[EXIT1]]:
-; CHECK-NEXT: [[SELECT_DATA_LCSSA1:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH1]] ], [ [[SELECT_DATA_LCSSA]], %[[IF_THEN]] ]
+; CHECK-NEXT: [[SELECT_DATA_LCSSA1:%.*]] = phi i32 [ [[SELECT_DATA]], %[[LATCH1]] ], [ [[TMP14]], %[[IF_THEN]] ]
; CHECK-NEXT: ret i32 [[SELECT_DATA_LCSSA1]]
;
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/X86/conditional-scalar-assignment.ll b/llvm/test/Transforms/LoopVectorize/X86/conditional-scalar-assignment.ll
index f6bd69acb775b..f5c51e4c1b071 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/conditional-scalar-assignment.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/conditional-scalar-assignment.ll
@@ -167,8 +167,7 @@ define ptr @simple_csa_ptr_select(i64 %N, ptr %data, i64 %a, ptr %init) {
; AVX512-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; AVX512-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[LOOP]], !llvm.loop [[LOOP4:![0-9]+]]
; AVX512: [[MIDDLE_BLOCK]]:
-; AVX512-NEXT: [[TMP9:%.*]] = extractelement <8 x ptr> [[BROADCAST_SPLAT2]], i64 0
-; AVX512-NEXT: [[TMP10:%.*]] = call ptr @llvm.experimental.vector.extract.last.active.v8p0(<8 x ptr> [[TMP7]], <8 x i1> [[TMP6]], ptr [[TMP9]])
+; AVX512-NEXT: [[TMP10:%.*]] = call ptr @llvm.experimental.vector.extract.last.active.v8p0(<8 x ptr> [[TMP7]], <8 x i1> [[TMP6]], ptr [[INIT]])
; AVX512-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; AVX512-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; AVX512: [[SCALAR_PH]]:
diff --git a/llvm/test/Transforms/LoopVectorize/find-last-ptr-induction.ll b/llvm/test/Transforms/LoopVectorize/find-last-ptr-induction.ll
index 2cc0581e6b629..815452573858b 100644
--- a/llvm/test/Transforms/LoopVectorize/find-last-ptr-induction.ll
+++ b/llvm/test/Transforms/LoopVectorize/find-last-ptr-induction.ll
@@ -110,14 +110,13 @@ define ptr @find_last_ptr_induction_separate_array(ptr %a, ptr %b, ptr %init, i6
; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
-; CHECK-NEXT: [[TMP11:%.*]] = extractelement <4 x ptr> [[BROADCAST_SPLAT]], i64 0
-; CHECK-NEXT: [[TMP12:%.*]] = call ptr @llvm.experimental.vector.extract.last.active.v4p0(<4 x ptr> [[TMP9]], <4 x i1> [[TMP8]], ptr [[TMP11]])
+; CHECK-NEXT: [[TMP11:%.*]] = call ptr @llvm.experimental.vector.extract.last.active.v4p0(<4 x ptr> [[TMP9]], <4 x i1> [[TMP8]], ptr [[INIT]])
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; CHECK: [[SCALAR_PH]]:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[TMP2]], %[[MIDDLE_BLOCK]] ], [ [[B]], %[[ENTRY]] ]
-; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi ptr [ [[TMP12]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ]
+; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi ptr [ [[TMP11]], %[[MIDDLE_BLOCK]] ], [ [[INIT]], %[[ENTRY]] ]
; CHECK-NEXT: br label %[[LOOP:.*]]
; CHECK: [[LOOP]]:
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ]
@@ -132,7 +131,7 @@ define ptr @find_last_ptr_induction_separate_array(ptr %a, ptr %b, ptr %init, i6
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: [[EXIT]]:
-; CHECK-NEXT: [[R:%.*]] = phi ptr [ [[LAST_NEXT]], %[[LOOP]] ], [ [[TMP12]], %[[MIDDLE_BLOCK]] ]
+; CHECK-NEXT: [[R:%.*]] = phi ptr [ [[LAST_NEXT]], %[[LOOP]] ], [ [[TMP11]], %[[MIDDLE_BLOCK]] ]
; CHECK-NEXT: ret ptr [[R]]
;
entry:
diff --git a/llvm/test/Transforms/LoopVectorize/iv-select-cmp-non-const-iv-start.ll b/llvm/test/Transforms/LoopVectorize/iv-select-cmp-non-const-iv-start.ll
index 967c70f094346..ad84f30810955 100644
--- a/llvm/test/Transforms/LoopVectorize/iv-select-cmp-non-const-iv-start.ll
+++ b/llvm/test/Transforms/LoopVectorize/iv-select-cmp-non-const-iv-start.ll
@@ -297,13 +297,12 @@ define i32 @select_trunc_non_const_iv_start_signed_guard(ptr %a, i32 %rdx_start,
; CHECK-VF4IC1-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-VF4IC1: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC1-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC1-NEXT: [[TMP11:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP8]], <4 x i1> [[TMP7]], i32 [[TMP10]])
+; CHECK-VF4IC1-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP8]], <4 x i1> [[TMP7]], i32 [[RDX_START]])
; CHECK-VF4IC1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP1]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[CMP_N]], label %[[EXIT_LOOPEXIT:.*]], label %[[SCALAR_PH]]
; CHECK-VF4IC1: [[SCALAR_PH]]:
; CHECK-VF4IC1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP2]], %[[MIDDLE_BLOCK]] ], [ [[TMP0]], %[[FOR_BODY_PREHEADER]] ]
-; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP11]], %[[MIDDLE_BLOCK]] ], [ [[RDX_START]], %[[FOR_BODY_PREHEADER]] ]
+; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP14]], %[[MIDDLE_BLOCK]] ], [ [[RDX_START]], %[[FOR_BODY_PREHEADER]] ]
; CHECK-VF4IC1-NEXT: br label %[[FOR_BODY:.*]]
; CHECK-VF4IC1: [[FOR_BODY]]:
; CHECK-VF4IC1-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[FOR_BODY]] ]
@@ -317,7 +316,7 @@ define i32 @select_trunc_non_const_iv_start_signed_guard(ptr %a, i32 %rdx_start,
; CHECK-VF4IC1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[WIDE_TRIP_COUNT]]
; CHECK-VF4IC1-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK-VF4IC1: [[EXIT_LOOPEXIT]]:
-; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ], [ [[TMP11]], %[[MIDDLE_BLOCK]] ]
+; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ], [ [[TMP14]], %[[MIDDLE_BLOCK]] ]
; CHECK-VF4IC1-NEXT: br label %[[EXIT]]
; CHECK-VF4IC1: [[EXIT]]:
; CHECK-VF4IC1-NEXT: [[IDX_0_LCSSA:%.*]] = phi i32 [ [[RDX_START]], %[[ENTRY]] ], [ [[COND_LCSSA]], %[[EXIT_LOOPEXIT]] ]
@@ -392,8 +391,7 @@ define i32 @select_trunc_non_const_iv_start_signed_guard(ptr %a, i32 %rdx_start,
; CHECK-VF4IC4-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC4-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK-VF4IC4: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC4-NEXT: [[TMP10:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC4-NEXT: [[TMP11:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP8]], <4 x i1> [[TMP7]], i32 [[TMP10]])
+; CHECK-VF4IC4-NEXT: [[TMP11:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP8]], <4 x i1> [[TMP7]], i32 [[RDX_START]])
; CHECK-VF4IC4-NEXT: [[TMP34:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP28]], <4 x i1> [[TMP24]], i32 [[TMP11]])
; CHECK-VF4IC4-NEXT: [[TMP35:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP29]], <4 x i1> [[TMP25]], i32 [[TMP34]])
; CHECK-VF4IC4-NEXT: [[TMP36:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP30]], <4 x i1> [[TMP26]], i32 [[TMP35]])
diff --git a/llvm/test/Transforms/LoopVectorize/iv-select-cmp-trunc.ll b/llvm/test/Transforms/LoopVectorize/iv-select-cmp-trunc.ll
index ede9f3eb7ce04..c0dfbbb7ffca9 100644
--- a/llvm/test/Transforms/LoopVectorize/iv-select-cmp-trunc.ll
+++ b/llvm/test/Transforms/LoopVectorize/iv-select-cmp-trunc.ll
@@ -1413,13 +1413,12 @@ define i32 @not_vectorized_select_iv_icmp_no_guard(ptr %a, ptr %b, i32 %start, i
; CHECK-VF4IC1-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK-VF4IC1: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC1-NEXT: [[TMP8:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC1-NEXT: [[TMP9:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP6]], <4 x i1> [[TMP5]], i32 [[TMP8]])
+; CHECK-VF4IC1-NEXT: [[TMP12:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP6]], <4 x i1> [[TMP5]], i32 [[START]])
; CHECK-VF4IC1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; CHECK-VF4IC1: [[SCALAR_PH]]:
; CHECK-VF4IC1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
-; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ [[START]], %[[ENTRY]] ]
+; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP12]], %[[MIDDLE_BLOCK]] ], [ [[START]], %[[ENTRY]] ]
; CHECK-VF4IC1-NEXT: br label %[[FOR_BODY:.*]]
; CHECK-VF4IC1: [[FOR_BODY]]:
; CHECK-VF4IC1-NEXT: [[IV1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
@@ -1435,7 +1434,7 @@ define i32 @not_vectorized_select_iv_icmp_no_guard(ptr %a, ptr %b, i32 %start, i
; CHECK-VF4IC1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[WIDE_TRIP_COUNT]]
; CHECK-VF4IC1-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]]
; CHECK-VF4IC1: [[EXIT]]:
-; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ]
+; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi i32 [ [[COND]], %[[FOR_BODY]] ], [ [[TMP12]], %[[MIDDLE_BLOCK]] ]
; CHECK-VF4IC1-NEXT: ret i32 [[COND_LCSSA]]
;
; CHECK-VF4IC4-LABEL: define i32 @not_vectorized_select_iv_icmp_no_guard(
@@ -1505,8 +1504,7 @@ define i32 @not_vectorized_select_iv_icmp_no_guard(ptr %a, ptr %b, i32 %start, i
; CHECK-VF4IC4-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC4-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
; CHECK-VF4IC4: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC4-NEXT: [[TMP8:%.*]] = extractelement <4 x i32> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC4-NEXT: [[TMP9:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP6]], <4 x i1> [[TMP5]], i32 [[TMP8]])
+; CHECK-VF4IC4-NEXT: [[TMP9:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP6]], <4 x i1> [[TMP5]], i32 [[START]])
; CHECK-VF4IC4-NEXT: [[TMP35:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP29]], <4 x i1> [[TMP25]], i32 [[TMP9]])
; CHECK-VF4IC4-NEXT: [[TMP36:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP30]], <4 x i1> [[TMP26]], i32 [[TMP35]])
; CHECK-VF4IC4-NEXT: [[TMP37:%.*]] = call i32 @llvm.experimental.vector.extract.last.active.v4i32(<4 x i32> [[TMP31]], <4 x i1> [[TMP27]], i32 [[TMP36]])
@@ -1900,13 +1898,12 @@ define i16 @not_vectorized_select_iv_icmp_overflow_unwidened_tripcount(ptr %a, p
; CHECK-VF4IC1-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK-VF4IC1: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC1-NEXT: [[TMP8:%.*]] = extractelement <4 x i16> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC1-NEXT: [[TMP9:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP6]], <4 x i1> [[TMP5]], i16 [[TMP8]])
+; CHECK-VF4IC1-NEXT: [[TMP12:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP6]], <4 x i1> [[TMP5]], i16 [[START]])
; CHECK-VF4IC1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[CMP_N]], label %[[EXIT_LOOPEXIT:.*]], label %[[SCALAR_PH]]
; CHECK-VF4IC1: [[SCALAR_PH]]:
; CHECK-VF4IC1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
-; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi i16 [ [[TMP9]], %[[MIDDLE_BLOCK]] ], [ [[START]], %[[FOR_BODY_PREHEADER]] ]
+; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi i16 [ [[TMP12]], %[[MIDDLE_BLOCK]] ], [ [[START]], %[[FOR_BODY_PREHEADER]] ]
; CHECK-VF4IC1-NEXT: br label %[[FOR_BODY:.*]]
; CHECK-VF4IC1: [[FOR_BODY]]:
; CHECK-VF4IC1-NEXT: [[IV1:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[INC:%.*]], %[[FOR_BODY]] ]
@@ -1922,7 +1919,7 @@ define i16 @not_vectorized_select_iv_icmp_overflow_unwidened_tripcount(ptr %a, p
; CHECK-VF4IC1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[WIDE_TRIP_COUNT]]
; CHECK-VF4IC1-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]]
; CHECK-VF4IC1: [[EXIT_LOOPEXIT]]:
-; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi i16 [ [[COND]], %[[FOR_BODY]] ], [ [[TMP9]], %[[MIDDLE_BLOCK]] ]
+; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi i16 [ [[COND]], %[[FOR_BODY]] ], [ [[TMP12]], %[[MIDDLE_BLOCK]] ]
; CHECK-VF4IC1-NEXT: br label %[[EXIT]]
; CHECK-VF4IC1: [[EXIT]]:
; CHECK-VF4IC1-NEXT: [[RDX_0_LCSSA:%.*]] = phi i16 [ [[START]], %[[ENTRY]] ], [ [[COND_LCSSA]], %[[EXIT_LOOPEXIT]] ]
@@ -1998,8 +1995,7 @@ define i16 @not_vectorized_select_iv_icmp_overflow_unwidened_tripcount(ptr %a, p
; CHECK-VF4IC4-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC4-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
; CHECK-VF4IC4: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC4-NEXT: [[TMP8:%.*]] = extractelement <4 x i16> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC4-NEXT: [[TMP9:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP6]], <4 x i1> [[TMP5]], i16 [[TMP8]])
+; CHECK-VF4IC4-NEXT: [[TMP9:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP6]], <4 x i1> [[TMP5]], i16 [[START]])
; CHECK-VF4IC4-NEXT: [[TMP35:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP29]], <4 x i1> [[TMP25]], i16 [[TMP9]])
; CHECK-VF4IC4-NEXT: [[TMP36:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP30]], <4 x i1> [[TMP26]], i16 [[TMP35]])
; CHECK-VF4IC4-NEXT: [[TMP37:%.*]] = call i16 @llvm.experimental.vector.extract.last.active.v4i16(<4 x i16> [[TMP31]], <4 x i1> [[TMP27]], i16 [[TMP36]])
diff --git a/llvm/test/Transforms/LoopVectorize/iv-select-cmp.ll b/llvm/test/Transforms/LoopVectorize/iv-select-cmp.ll
index efbbe480e0041..c9c4615c36760 100644
--- a/llvm/test/Transforms/LoopVectorize/iv-select-cmp.ll
+++ b/llvm/test/Transforms/LoopVectorize/iv-select-cmp.ll
@@ -1892,14 +1892,13 @@ define float @not_vectorized_select_float_induction_icmp(ptr %a, ptr %b, float %
; CHECK-VF4IC1-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[FOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; CHECK-VF4IC1: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC1-NEXT: [[TMP11:%.*]] = extractelement <4 x float> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC1-NEXT: [[TMP12:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP9]], <4 x i1> [[TMP8]], float [[TMP11]])
+; CHECK-VF4IC1-NEXT: [[TMP14:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP9]], <4 x i1> [[TMP8]], float [[RDX_START]])
; CHECK-VF4IC1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
; CHECK-VF4IC1-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
; CHECK-VF4IC1: [[SCALAR_PH]]:
; CHECK-VF4IC1-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
; CHECK-VF4IC1-NEXT: [[BC_RESUME_VAL2:%.*]] = phi float [ [[TMP13]], %[[MIDDLE_BLOCK]] ], [ 0.000000e+00, %[[ENTRY]] ]
-; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP12]], %[[MIDDLE_BLOCK]] ], [ [[RDX_START]], %[[ENTRY]] ]
+; CHECK-VF4IC1-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP14]], %[[MIDDLE_BLOCK]] ], [ [[RDX_START]], %[[ENTRY]] ]
; CHECK-VF4IC1-NEXT: br label %[[FOR_BODY1:.*]]
; CHECK-VF4IC1: [[FOR_BODY1]]:
; CHECK-VF4IC1-NEXT: [[IV1:%.*]] = phi i64 [ [[INC:%.*]], %[[FOR_BODY1]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
@@ -1916,7 +1915,7 @@ define float @not_vectorized_select_float_induction_icmp(ptr %a, ptr %b, float %
; CHECK-VF4IC1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]]
; CHECK-VF4IC1-NEXT: br i1 [[EXITCOND_NOT]], label %[[EXIT]], label %[[FOR_BODY1]], !llvm.loop [[LOOP21:![0-9]+]]
; CHECK-VF4IC1: [[EXIT]]:
-; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi float [ [[COND]], %[[FOR_BODY1]] ], [ [[TMP12]], %[[MIDDLE_BLOCK]] ]
+; CHECK-VF4IC1-NEXT: [[COND_LCSSA:%.*]] = phi float [ [[COND]], %[[FOR_BODY1]] ], [ [[TMP14]], %[[MIDDLE_BLOCK]] ]
; CHECK-VF4IC1-NEXT: ret float [[COND_LCSSA]]
;
; CHECK-VF4IC4-LABEL: define float @not_vectorized_select_float_induction_icmp(
@@ -1988,8 +1987,7 @@ define float @not_vectorized_select_float_induction_icmp(ptr %a, ptr %b, float %
; CHECK-VF4IC4-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-VF4IC4-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[FOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]]
; CHECK-VF4IC4: [[MIDDLE_BLOCK]]:
-; CHECK-VF4IC4-NEXT: [[TMP11:%.*]] = extractelement <4 x float> [[BROADCAST_SPLAT]], i64 0
-; CHECK-VF4IC4-NEXT: [[TMP12:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP9]], <4 x i1> [[TMP8]], float [[TMP11]])
+; CHECK-VF4IC4-NEXT: [[TMP12:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP9]], <4 x i1> [[TMP8]], float [[RDX_START]])
; CHECK-VF4IC4-NEXT: [[TMP37:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP31]], <4 x i1> [[TMP27]], float [[TMP12]])
; CHECK-VF4IC4-NEXT: [[TMP38:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP32]], <4 x i1> [[TMP28]], float [[TMP37]])
; CHECK-VF4IC4-NEXT: [[TMP39:%.*]] = call float @llvm.experimental.vector.extract.last.active.v4f32(<4 x float> [[TMP33]], <4 x i1> [[TMP29]], float [[TMP38]])
More information about the llvm-commits
mailing list