[llvm] [LV] Support generating masks for switch terminators. (PR #99808)

[Florian Hahn via llvm-commits]

================
@@ -204,9 +334,74 @@ define void @switch_multiple_common_dests(ptr %start, ptr %end) {
 ; FORCED-LABEL: define void @switch_multiple_common_dests(
 ; FORCED-SAME: ptr [[START:%.*]], ptr [[END:%.*]]) #[[ATTR0]] {
 ; FORCED-NEXT:  [[ENTRY:.*]]:
+; FORCED-NEXT:    [[START2:%.*]] = ptrtoint ptr [[START]] to i64
+; FORCED-NEXT:    [[END1:%.*]] = ptrtoint ptr [[END]] to i64
+; FORCED-NEXT:    [[TMP0:%.*]] = add i64 [[END1]], -8
+; FORCED-NEXT:    [[TMP1:%.*]] = sub i64 [[TMP0]], [[START2]]
+; FORCED-NEXT:    [[TMP2:%.*]] = lshr i64 [[TMP1]], 3
+; FORCED-NEXT:    [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
+; FORCED-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 8
+; FORCED-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; FORCED:       [[VECTOR_PH]]:
+; FORCED-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 8
+; FORCED-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
+; FORCED-NEXT:    [[TMP4:%.*]] = mul i64 [[N_VEC]], 8
+; FORCED-NEXT:    [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP4]]
+; FORCED-NEXT:    br label %[[VECTOR_BODY:.*]]
+; FORCED:       [[VECTOR_BODY]]:
+; FORCED-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; FORCED-NEXT:    [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 8
+; FORCED-NEXT:    [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 0
+; FORCED-NEXT:    [[TMP6:%.*]] = add i64 [[OFFSET_IDX]], 32
+; FORCED-NEXT:    [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]]
+; FORCED-NEXT:    [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP6]]
+; FORCED-NEXT:    [[TMP7:%.*]] = getelementptr i64, ptr [[NEXT_GEP]], i32 0
+; FORCED-NEXT:    [[TMP8:%.*]] = getelementptr i64, ptr [[NEXT_GEP]], i32 4
+; FORCED-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP7]], align 1
+; FORCED-NEXT:    [[WIDE_LOAD4:%.*]] = load <4 x i64>, ptr [[TMP8]], align 1
+; FORCED-NEXT:    [[TMP23:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], <i64 -12, i64 -12, i64 -12, i64 -12>
+; FORCED-NEXT:    [[TMP24:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD4]], <i64 -12, i64 -12, i64 -12, i64 -12>
+; FORCED-NEXT:    [[TMP25:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], zeroinitializer
+; FORCED-NEXT:    [[TMP26:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD4]], zeroinitializer
+; FORCED-NEXT:    [[TMP13:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], <i64 13, i64 13, i64 13, i64 13>
+; FORCED-NEXT:    [[TMP14:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD4]], <i64 13, i64 13, i64 13, i64 13>
+; FORCED-NEXT:    [[TMP17:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], <i64 14, i64 14, i64 14, i64 14>
+; FORCED-NEXT:    [[TMP18:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD4]], <i64 14, i64 14, i64 14, i64 14>
+; FORCED-NEXT:    [[TMP15:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD]], <i64 15, i64 15, i64 15, i64 15>
+; FORCED-NEXT:    [[TMP16:%.*]] = icmp eq <4 x i64> [[WIDE_LOAD4]], <i64 15, i64 15, i64 15, i64 15>
+; FORCED-NEXT:    [[TMP27:%.*]] = or <4 x i1> [[TMP23]], [[TMP25]]
+; FORCED-NEXT:    [[TMP28:%.*]] = or <4 x i1> [[TMP24]], [[TMP26]]
+; FORCED-NEXT:    [[TMP21:%.*]] = or <4 x i1> [[TMP13]], [[TMP17]]
+; FORCED-NEXT:    [[TMP22:%.*]] = or <4 x i1> [[TMP14]], [[TMP18]]
+; FORCED-NEXT:    [[TMP35:%.*]] = or <4 x i1> [[TMP21]], [[TMP15]]
+; FORCED-NEXT:    [[TMP36:%.*]] = or <4 x i1> [[TMP22]], [[TMP16]]
+; FORCED-NEXT:    [[TMP37:%.*]] = or <4 x i1> [[TMP27]], [[TMP35]]
+; FORCED-NEXT:    [[TMP38:%.*]] = or <4 x i1> [[TMP28]], [[TMP36]]
+; FORCED-NEXT:    [[TMP39:%.*]] = xor <4 x i1> [[TMP37]], <i1 true, i1 true, i1 true, i1 true>
+; FORCED-NEXT:    [[TMP40:%.*]] = xor <4 x i1> [[TMP38]], <i1 true, i1 true, i1 true, i1 true>
+; FORCED-NEXT:    [[TMP29:%.*]] = or <4 x i1> [[TMP35]], [[TMP35]]
+; FORCED-NEXT:    [[TMP30:%.*]] = or <4 x i1> [[TMP36]], [[TMP36]]
+; FORCED-NEXT:    [[TMP31:%.*]] = or <4 x i1> [[TMP29]], [[TMP35]]
+; FORCED-NEXT:    [[TMP32:%.*]] = or <4 x i1> [[TMP30]], [[TMP36]]
----------------
fhahn wrote:

Will improve generally as follow-up to createBlockInMask

https://github.com/llvm/llvm-project/pull/99808