[llvm] [VPlan] Compute cost of some VPWidenIntOrFpInductions in VPlan (PR #202232)

[Florian Hahn via llvm-commits]

================
@@ -680,49 +680,85 @@ exit:
 define void @wombat(i32 %arg, ptr %dst) #1 {
 ; CHECK-LABEL: define void @wombat(
 ; CHECK-SAME: i32 [[ARG:%.*]], ptr [[DST:%.*]]) #[[ATTR1:[0-9]+]] {
-; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:  [[ENTRY:.*]]:
 ; CHECK-NEXT:    [[MUL:%.*]] = mul i32 [[ARG]], 3
 ; CHECK-NEXT:    [[ZEXT:%.*]] = zext i32 [[ARG]] to i64
-; CHECK-NEXT:    br label %[[VECTOR_PH:.*]]
+; CHECK-NEXT:    br i1 false, label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MAIN_LOOP_ENTRY:.*]]
+; CHECK:       [[VECTOR_MAIN_LOOP_ENTRY]]:
+; CHECK-NEXT:    br i1 false, label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
 ; CHECK:       [[VECTOR_PH]]:
-; CHECK-NEXT:    [[TMP0:%.*]] = mul i32 56, [[ARG]]
-; CHECK-NEXT:    [[IND_END:%.*]] = add i32 [[MUL]], [[TMP0]]
-; CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <8 x i32> poison, i32 [[MUL]], i64 0
-; CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <8 x i32> [[DOTSPLATINSERT]], <8 x i32> poison, <8 x i32> zeroinitializer
-; CHECK-NEXT:    [[DOTSPLATINSERT1:%.*]] = insertelement <8 x i32> poison, i32 [[ARG]], i64 0
-; CHECK-NEXT:    [[DOTSPLAT2:%.*]] = shufflevector <8 x i32> [[DOTSPLATINSERT1]], <8 x i32> poison, <8 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP1:%.*]] = mul <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, [[DOTSPLAT2]]
-; CHECK-NEXT:    [[INDUCTION:%.*]] = add <8 x i32> [[DOTSPLAT]], [[TMP1]]
-; CHECK-NEXT:    [[TMP2:%.*]] = shl i32 [[ARG]], 3
-; CHECK-NEXT:    [[DOTSPLATINSERT3:%.*]] = insertelement <8 x i32> poison, i32 [[TMP2]], i64 0
-; CHECK-NEXT:    [[DOTSPLAT4:%.*]] = shufflevector <8 x i32> [[DOTSPLATINSERT3]], <8 x i32> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i32> poison, i32 [[ARG]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT]], <8 x i32> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP0:%.*]] = mul i32 48, [[ARG]]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[MUL]], [[TMP0]]
+; CHECK-NEXT:    [[TMP2:%.*]] = shl <8 x i32> [[BROADCAST_SPLAT]], splat (i32 3)
+; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <8 x i32> poison, i32 [[MUL]], i64 0
+; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <8 x i32> [[BROADCAST_SPLATINSERT1]], <8 x i32> poison, <8 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP3:%.*]] = mul <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>, [[BROADCAST_SPLAT]]
+; CHECK-NEXT:    [[INDUCTION:%.*]] = add <8 x i32> [[BROADCAST_SPLAT2]], [[TMP3]]
 ; 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 <8 x i32> [ [[INDUCTION]], %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
-; CHECK-NEXT:    [[OFFSET_IDX:%.*]] = add i64 4, [[INDEX]]
-; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i32, ptr [[DST]], i64 [[OFFSET_IDX]]
-; CHECK-NEXT:    [[TMP5:%.*]] = and <8 x i32> [[VEC_IND]], splat (i32 12)
-; CHECK-NEXT:    store <8 x i32> [[TMP5]], ptr [[TMP4]], align 4
-; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
-; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <8 x i32> [[VEC_IND]], [[DOTSPLAT4]]
-; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 56
-; CHECK-NEXT:    br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP28:![0-9]+]]
+; CHECK-NEXT:    [[STEP_ADD:%.*]] = add <8 x i32> [[VEC_IND]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = add i64 4, [[INDEX]]
+; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i32, ptr [[DST]], i64 [[TMP4]]
+; CHECK-NEXT:    [[TMP6:%.*]] = and <8 x i32> [[VEC_IND]], splat (i32 12)
+; CHECK-NEXT:    [[TMP7:%.*]] = and <8 x i32> [[STEP_ADD]], splat (i32 12)
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i32, ptr [[TMP5]], i64 8
+; CHECK-NEXT:    store <8 x i32> [[TMP6]], ptr [[TMP5]], align 4
----------------
fhahn wrote:

Yep, the cheaper loop is now interleaved

https://github.com/llvm/llvm-project/pull/202232