xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/pr52275.ll (revision 580210a0c938531ef9fd79f9ffedb93eeb2e66c2)

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=slp-vectorizer -S -mcpu=core-i7 | FileCheck %s
; RUN: opt < %s -passes=slp-vectorizer -slp-threshold=-100 -S -mcpu=core-i7 | FileCheck %s --check-prefix=FORCE_SLP

define  <4 x i8> @test(<4 x i8> %v, ptr %x) {
; CHECK-LABEL: @test(
; CHECK-NEXT:    [[X0:%.*]] = load i8, ptr [[X:%.*]], align 4
; CHECK-NEXT:    [[G1:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 1
; CHECK-NEXT:    [[X1:%.*]] = load i8, ptr [[G1]], align 4
; CHECK-NEXT:    [[V0:%.*]] = insertelement <4 x i8> [[V:%.*]], i8 [[X0]], i64 0
; CHECK-NEXT:    [[V1:%.*]] = insertelement <4 x i8> [[V0]], i8 [[X1]], i64 1
; CHECK-NEXT:    [[V2:%.*]] = add <4 x i8> [[V0]], [[V1]]
; CHECK-NEXT:    ret <4 x i8> [[V2]]
;
; FORCE_SLP-LABEL: @test(
; FORCE_SLP-NEXT:    [[X0:%.*]] = load i8, ptr [[X:%.*]], align 4
; FORCE_SLP-NEXT:    [[G1:%.*]] = getelementptr inbounds i8, ptr [[X]], i64 1
; FORCE_SLP-NEXT:    [[X1:%.*]] = load i8, ptr [[G1]], align 4
; FORCE_SLP-NEXT:    [[V0:%.*]] = insertelement <4 x i8> [[V:%.*]], i8 [[X0]], i64 0
; FORCE_SLP-NEXT:    [[V1:%.*]] = insertelement <4 x i8> [[V0]], i8 [[X1]], i64 1
; FORCE_SLP-NEXT:    [[V2:%.*]] = add <4 x i8> [[V0]], [[V1]]
; FORCE_SLP-NEXT:    ret <4 x i8> [[V2]]
;
  %x0 = load i8, ptr %x, align 4
  %g1 = getelementptr inbounds i8, ptr %x, i64 1
  %x1 = load i8, ptr %g1, align 4
  %v0 = insertelement <4 x i8> %v, i8 %x0, i64 0
  %v1 = insertelement <4 x i8> %v0, i8 %x1, i64 1
  %v2 = add <4 x i8> %v0, %v1
  ret <4 x i8> %v2
}

define  <2 x i8> @test2(<2 x i8> %t6, ptr %t1) {
; CHECK-LABEL: @test2(
; CHECK-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; CHECK-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; CHECK-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; CHECK-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; CHECK-NEXT:    [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 0
; CHECK-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; CHECK-NEXT:    [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 1
; CHECK-NEXT:    [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
; CHECK-NEXT:    ret <2 x i8> [[T11]]
;
; FORCE_SLP-LABEL: @test2(
; FORCE_SLP-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; FORCE_SLP-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; FORCE_SLP-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; FORCE_SLP-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; FORCE_SLP-NEXT:    [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 0
; FORCE_SLP-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; FORCE_SLP-NEXT:    [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 1
; FORCE_SLP-NEXT:    [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
; FORCE_SLP-NEXT:    ret <2 x i8> [[T11]]
;
  %t3 = load i32, ptr %t1, align 4
  %t4 = getelementptr inbounds i32, ptr %t1, i64 1
  %t5 = load i32, ptr %t4, align 4
  %t7 = trunc i32 %t3 to i8
  %t8 = insertelement <2 x i8> %t6, i8 %t7, i64 0
  %t9 = trunc i32 %t5 to i8
  %t10 = insertelement <2 x i8> %t8, i8 %t9, i64 1
  %t11 = add <2 x i8> %t10, %t8
  ret <2 x i8> %t11
}

define  <2 x i8> @test_reorder(<2 x i8> %t6, ptr %t1) {
; CHECK-LABEL: @test_reorder(
; CHECK-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; CHECK-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; CHECK-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; CHECK-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; CHECK-NEXT:    [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 1
; CHECK-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; CHECK-NEXT:    [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 0
; CHECK-NEXT:    [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
; CHECK-NEXT:    ret <2 x i8> [[T11]]
;
; FORCE_SLP-LABEL: @test_reorder(
; FORCE_SLP-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; FORCE_SLP-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; FORCE_SLP-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; FORCE_SLP-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; FORCE_SLP-NEXT:    [[T8:%.*]] = insertelement <2 x i8> [[T6:%.*]], i8 [[T7]], i64 1
; FORCE_SLP-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; FORCE_SLP-NEXT:    [[T10:%.*]] = insertelement <2 x i8> [[T8]], i8 [[T9]], i64 0
; FORCE_SLP-NEXT:    [[T11:%.*]] = add <2 x i8> [[T10]], [[T8]]
; FORCE_SLP-NEXT:    ret <2 x i8> [[T11]]
;
  %t3 = load i32, ptr %t1, align 4
  %t4 = getelementptr inbounds i32, ptr %t1, i64 1
  %t5 = load i32, ptr %t4, align 4
  %t7 = trunc i32 %t3 to i8
  %t8 = insertelement <2 x i8> %t6, i8 %t7, i64 1
  %t9 = trunc i32 %t5 to i8
  %t10 = insertelement <2 x i8> %t8, i8 %t9, i64 0
  %t11 = add <2 x i8> %t10, %t8
  ret <2 x i8> %t11
}

define  <4 x i8> @test_subvector(<4 x i8> %t6, ptr %t1) {
; CHECK-LABEL: @test_subvector(
; CHECK-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; CHECK-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; CHECK-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; CHECK-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; CHECK-NEXT:    [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 0
; CHECK-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; CHECK-NEXT:    [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 1
; CHECK-NEXT:    [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
; CHECK-NEXT:    ret <4 x i8> [[T11]]
;
; FORCE_SLP-LABEL: @test_subvector(
; FORCE_SLP-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; FORCE_SLP-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; FORCE_SLP-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; FORCE_SLP-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; FORCE_SLP-NEXT:    [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 0
; FORCE_SLP-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; FORCE_SLP-NEXT:    [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 1
; FORCE_SLP-NEXT:    [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
; FORCE_SLP-NEXT:    ret <4 x i8> [[T11]]
;
  %t3 = load i32, ptr %t1, align 4
  %t4 = getelementptr inbounds i32, ptr %t1, i64 1
  %t5 = load i32, ptr %t4, align 4
  %t7 = trunc i32 %t3 to i8
  %t8 = insertelement <4 x i8> %t6, i8 %t7, i64 0
  %t9 = trunc i32 %t5 to i8
  %t10 = insertelement <4 x i8> %t8, i8 %t9, i64 1
  %t11 = add <4 x i8> %t10, %t8
  ret <4 x i8> %t11
}

define  <4 x i8> @test_subvector_reorder(<4 x i8> %t6, ptr %t1) {
; CHECK-LABEL: @test_subvector_reorder(
; CHECK-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; CHECK-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; CHECK-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; CHECK-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; CHECK-NEXT:    [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 3
; CHECK-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; CHECK-NEXT:    [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 2
; CHECK-NEXT:    [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
; CHECK-NEXT:    ret <4 x i8> [[T11]]
;
; FORCE_SLP-LABEL: @test_subvector_reorder(
; FORCE_SLP-NEXT:    [[T3:%.*]] = load i32, ptr [[T1:%.*]], align 4
; FORCE_SLP-NEXT:    [[T4:%.*]] = getelementptr inbounds i32, ptr [[T1]], i64 1
; FORCE_SLP-NEXT:    [[T5:%.*]] = load i32, ptr [[T4]], align 4
; FORCE_SLP-NEXT:    [[T7:%.*]] = trunc i32 [[T3]] to i8
; FORCE_SLP-NEXT:    [[T8:%.*]] = insertelement <4 x i8> [[T6:%.*]], i8 [[T7]], i64 3
; FORCE_SLP-NEXT:    [[T9:%.*]] = trunc i32 [[T5]] to i8
; FORCE_SLP-NEXT:    [[T10:%.*]] = insertelement <4 x i8> [[T8]], i8 [[T9]], i64 2
; FORCE_SLP-NEXT:    [[T11:%.*]] = add <4 x i8> [[T10]], [[T8]]
; FORCE_SLP-NEXT:    ret <4 x i8> [[T11]]
;
  %t3 = load i32, ptr %t1, align 4
  %t4 = getelementptr inbounds i32, ptr %t1, i64 1
  %t5 = load i32, ptr %t4, align 4
  %t7 = trunc i32 %t3 to i8
  %t8 = insertelement <4 x i8> %t6, i8 %t7, i64 3
  %t9 = trunc i32 %t5 to i8
  %t10 = insertelement <4 x i8> %t8, i8 %t9, i64 2
  %t11 = add <4 x i8> %t10, %t8
  ret <4 x i8> %t11
}