; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=instcombine -S | FileCheck %s

target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
target triple = "armv8-arm-none-eabi"

; We can turn a vtbl/vtbx intrinsic into a shufflevector instruction if the mask
; is constant and references 2 or fewer operands.

; Basic vtbl1 with all in-bounds indices should optimize to shufflevector.
define <8 x i8> @vtbl1_basic(<8 x i8> %a) {
; CHECK-LABEL: @vtbl1_basic(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %a, <8 x i8> <i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
  ret <8 x i8> %tbl
}

; vtbl2 with both operands the same should be optimized (1 unique source).
define <8 x i8> @vtbl2_duplicate_operands(<8 x i8> %a) {
; CHECK-LABEL: @vtbl2_duplicate_operands(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %a, <8 x i8> %a, <8 x i8> <i8 0, i8 1, i8 2, i8 3, i8 8, i8 9, i8 10, i8 11>)
  ret <8 x i8> %tbl
}

; vtbl3 referencing 2 unique operands should optimize.
define <8 x i8> @vtbl3_two_sources(<8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: @vtbl3_two_sources(
; CHECK-NEXT:    [[TBL:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 0, i32 1, i32 0, i32 0>
; CHECK-NEXT:    ret <8 x i8> [[TBL]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl3(<8 x i8> %a, <8 x i8> %b, <8 x i8> %a, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 0, i8 0>)
  ret <8 x i8> %tbl
}

; vtbl4 with alternating duplicate operands should optimize (2 unique sources).
define <8 x i8> @vtbl4_duplicate_operands(<8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: @vtbl4_duplicate_operands(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 0, i32 1, i32 8, i32 9>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> %a, <8 x i8> %b, <8 x i8> %a, <8 x i8> %b, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 24, i8 25>)
  ret <8 x i8> %tbl
}

; vtbl4 where mask only references first two operands should optimize.
define <8 x i8> @vtbl4_unused_operands(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d) {
; CHECK-LABEL: @vtbl4_unused_operands(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d, <8 x i8> <i8 0, i8 1, i8 2, i8 3, i8 8, i8 9, i8 10, i8 11>)
  ret <8 x i8> %tbl
}

; vtbl4 where mask only references one operand should optimize.
define <8 x i8> @vtbl4_single_operand_used(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d) {
; CHECK-LABEL: @vtbl4_single_operand_used(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d, <8 x i8> <i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
  ret <8 x i8> %tbl
}

; vtbl1 with some OOB indices should optimize (1 source + zero vector = 2 sources).
define <8 x i8> @vtbl1_with_oob(<8 x i8> %a) {
; CHECK-LABEL: @vtbl1_with_oob(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> <i8 0, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison>, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %a, <8 x i8> <i8 0, i8 1, i8 2, i8 3, i8 8, i8 9, i8 10, i8 11>)
  ret <8 x i8> %tbl
}

; vtbl2 with duplicate operands and OOB should optimize (1 unique source + zero vector = 2 sources).
define <8 x i8> @vtbl2_duplicate_with_oob(<8 x i8> %a) {
; CHECK-LABEL: @vtbl2_duplicate_with_oob(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> <i8 0, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison, i8 poison>, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 8, i32 8, i32 8, i32 8>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %a, <8 x i8> %a, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbl
}

; vtbl2 with OOB indices should NOT optimize (2 sources + zero vector = 3 sources).
define <8 x i8> @vtbl2_with_oob_bail(<8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: @vtbl2_with_oob_bail(
; CHECK-NEXT:    [[TBL:%.*]] = call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 99, i8 99, i8 99, i8 99>)
; CHECK-NEXT:    ret <8 x i8> [[TBL]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl2(<8 x i8> %a, <8 x i8> %b, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbl
}

; vtbl1 with all OOB indices should optimize to zero vector.
define <8 x i8> @vtbl1_all_oob(<8 x i8> %a) {
; CHECK-LABEL: @vtbl1_all_oob(
; CHECK-NEXT:    ret <8 x i8> zeroinitializer
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %a, <8 x i8> <i8 99, i8 99, i8 99, i8 99, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbl
}

; vtbl3 referencing all 3 operands should NOT optimize.
define <8 x i8> @vtbl3_three_sources_bail(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c) {
; CHECK-LABEL: @vtbl3_three_sources_bail(
; CHECK-NEXT:    [[TBL:%.*]] = call <8 x i8> @llvm.arm.neon.vtbl3(<8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i8> [[C:%.*]], <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 0, i8 0>)
; CHECK-NEXT:    ret <8 x i8> [[TBL]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl3(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 0, i8 0>)
  ret <8 x i8> %tbl
}

; vtbl4 referencing 3 unique operands should NOT optimize.
define <8 x i8> @vtbl4_three_sources_bail(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c) {
; CHECK-LABEL: @vtbl4_three_sources_bail(
; CHECK-NEXT:    [[TBL:%.*]] = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i8> [[C:%.*]], <8 x i8> [[A]], <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 24, i8 25>)
; CHECK-NEXT:    ret <8 x i8> [[TBL]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %a, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 24, i8 25>)
  ret <8 x i8> %tbl
}

; vtbl4 referencing all 4 unique operands should NOT optimize.
define <8 x i8> @vtbl4_four_sources_bail(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d) {
; CHECK-LABEL: @vtbl4_four_sources_bail(
; CHECK-NEXT:    [[TBL:%.*]] = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i8> [[C:%.*]], <8 x i8> [[D:%.*]], <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 24, i8 25>)
; CHECK-NEXT:    ret <8 x i8> [[TBL]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl4(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c, <8 x i8> %d, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 16, i8 17, i8 24, i8 25>)
  ret <8 x i8> %tbl
}

; vtbx1 with no OOB should optimize.
define <8 x i8> @vtbx1_no_oob(<8 x i8> %fallback, <8 x i8> %a) {
; CHECK-LABEL: @vtbx1_no_oob(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbx = call <8 x i8> @llvm.arm.neon.vtbx1(<8 x i8> %fallback, <8 x i8> %a, <8 x i8> <i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
  ret <8 x i8> %tbx
}

; vtbx2 where fallback == second source operand should optimize (deduplicated).
define <8 x i8> @vtbx2_fallback_equals_second_source(<8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: @vtbx2_fallback_equals_second_source(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 12, i32 13, i32 14, i32 15>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbx = call <8 x i8> @llvm.arm.neon.vtbx2(<8 x i8> %b, <8 x i8> %a, <8 x i8> %b, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbx
}

; vtbx1 with OOB where fallback == source should optimize (deduplicated).
define <8 x i8> @vtbx1_oob_fallback_same_as_source(<8 x i8> %a) {
; CHECK-LABEL: @vtbx1_oob_fallback_same_as_source(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbx = call <8 x i8> @llvm.arm.neon.vtbx1(<8 x i8> %a, <8 x i8> %a, <8 x i8> <i8 3, i8 2, i8 1, i8 0, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbx
}

; vtbx2 with OOB should NOT optimize (2 sources + fallback = 3 sources).
define <8 x i8> @vtbx2_with_oob_bail(<8 x i8> %fallback, <8 x i8> %a, <8 x i8> %b) {
; CHECK-LABEL: @vtbx2_with_oob_bail(
; CHECK-NEXT:    [[TBX:%.*]] = call <8 x i8> @llvm.arm.neon.vtbx2(<8 x i8> [[FALLBACK:%.*]], <8 x i8> [[A:%.*]], <8 x i8> [[B:%.*]], <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 99, i8 99, i8 99, i8 99>)
; CHECK-NEXT:    ret <8 x i8> [[TBX]]
;
  %tbx = call <8 x i8> @llvm.arm.neon.vtbx2(<8 x i8> %fallback, <8 x i8> %a, <8 x i8> %b, <8 x i8> <i8 0, i8 1, i8 8, i8 9, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbx
}

; vtbx1 with all OOB indices should optimize to fallback.
define <8 x i8> @vtbx1_all_oob(<8 x i8> %fallback, <8 x i8> %a) {
; CHECK-LABEL: @vtbx1_all_oob(
; CHECK-NEXT:    ret <8 x i8> [[FALLBACK:%.*]]
;
  %tbx = call <8 x i8> @llvm.arm.neon.vtbx1(<8 x i8> %fallback, <8 x i8> %a, <8 x i8> <i8 99, i8 99, i8 99, i8 99, i8 99, i8 99, i8 99, i8 99>)
  ret <8 x i8> %tbx
}

; Non-constant mask should NOT optimize.
define <8 x i8> @vtbl1_non_constant_mask(<8 x i8> %a, <8 x i8> %mask) {
; CHECK-LABEL: @vtbl1_non_constant_mask(
; CHECK-NEXT:    [[TBL:%.*]] = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> [[A:%.*]], <8 x i8> [[MASK:%.*]])
; CHECK-NEXT:    ret <8 x i8> [[TBL]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %a, <8 x i8> %mask)
  ret <8 x i8> %tbl
}

; Mask with some poison elements should optimize, with poison propagating to output.
define <8 x i8> @vtbl1_poison_mask_elements(<8 x i8> %a) {
; CHECK-LABEL: @vtbl1_poison_mask_elements(
; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x i8> [[A:%.*]], <8 x i8> poison, <8 x i32> <i32 0, i32 poison, i32 2, i32 poison, i32 4, i32 5, i32 6, i32 7>
; CHECK-NEXT:    ret <8 x i8> [[TMP1]]
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %a, <8 x i8> <i8 0, i8 poison, i8 2, i8 poison, i8 4, i8 5, i8 6, i8 7>)
  ret <8 x i8> %tbl
}

; Mask with all poison elements should optimize to poison.
define <8 x i8> @vtbl1_all_poison_mask(<8 x i8> %a) {
; CHECK-LABEL: @vtbl1_all_poison_mask(
; CHECK-NEXT:    ret <8 x i8> poison
;
  %tbl = call <8 x i8> @llvm.arm.neon.vtbl1(<8 x i8> %a, <8 x i8> poison)
  ret <8 x i8> %tbl
}