; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes="default<O3>" -S            < %s | FileCheck %s --check-prefix=SSE
; RUN: opt -passes="default<O3>" -S -mattr=avx < %s | FileCheck %s --check-prefix=AVX

; This test is based on https://github.com/llvm/llvm-project/issues/50778
; It's the unoptimized IR passed through -passes=mem2reg to remove obvious noise.
; This should show cooperation between instcombine, unrolling, inlining,
; and SLP to create the target-optimal vector math+logic ops.

target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

define noundef i64 @foo(i64 noundef %0) {
; CHECK-LABEL: @foo(
; CHECK-NEXT:    [[TMP2:%.*]] = shl i64 [[TMP0:%.*]], 44
; CHECK-NEXT:    [[TMP3:%.*]] = sub nuw nsw i64 -17592186044416, [[TMP2]]
; CHECK-NEXT:    ret i64 [[TMP3]]
;
; SSE-LABEL: @foo(
; SSE-NEXT:    [[TMP2:%.*]] = xor i64 [[TMP0:%.*]], -1
; SSE-NEXT:    [[TMP3:%.*]] = shl i64 [[TMP2]], 44
; SSE-NEXT:    ret i64 [[TMP3]]
;
; AVX-LABEL: @foo(
; AVX-NEXT:    [[TMP2:%.*]] = xor i64 [[TMP0:%.*]], -1
; AVX-NEXT:    [[TMP3:%.*]] = shl i64 [[TMP2]], 44
; AVX-NEXT:    ret i64 [[TMP3]]
;
  %2 = sub i64 1048575, %0
  %3 = shl i64 %2, 44
  ret i64 %3
}

define void @bar(ptr noundef %0) {
; SSE-LABEL: @bar(
; SSE-NEXT:    [[TMP2:%.*]] = load <2 x i64>, ptr [[TMP0:%.*]], align 8
; SSE-NEXT:    [[TMP3:%.*]] = xor <2 x i64> [[TMP2]], splat (i64 -1)
; SSE-NEXT:    [[TMP4:%.*]] = shl <2 x i64> [[TMP3]], splat (i64 44)
; SSE-NEXT:    store <2 x i64> [[TMP4]], ptr [[TMP0]], align 8
; SSE-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 16
; SSE-NEXT:    [[TMP6:%.*]] = load <2 x i64>, ptr [[TMP5]], align 8
; SSE-NEXT:    [[TMP7:%.*]] = xor <2 x i64> [[TMP6]], splat (i64 -1)
; SSE-NEXT:    [[TMP8:%.*]] = shl <2 x i64> [[TMP7]], splat (i64 44)
; SSE-NEXT:    store <2 x i64> [[TMP8]], ptr [[TMP5]], align 8
; SSE-NEXT:    [[TMP9:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 32
; SSE-NEXT:    [[TMP10:%.*]] = load <2 x i64>, ptr [[TMP9]], align 8
; SSE-NEXT:    [[TMP11:%.*]] = xor <2 x i64> [[TMP10]], splat (i64 -1)
; SSE-NEXT:    [[TMP12:%.*]] = shl <2 x i64> [[TMP11]], splat (i64 44)
; SSE-NEXT:    store <2 x i64> [[TMP12]], ptr [[TMP9]], align 8
; SSE-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 48
; SSE-NEXT:    [[TMP14:%.*]] = load <2 x i64>, ptr [[TMP13]], align 8
; SSE-NEXT:    [[TMP15:%.*]] = xor <2 x i64> [[TMP14]], splat (i64 -1)
; SSE-NEXT:    [[TMP16:%.*]] = shl <2 x i64> [[TMP15]], splat (i64 44)
; SSE-NEXT:    store <2 x i64> [[TMP16]], ptr [[TMP13]], align 8
; SSE-NEXT:    ret void
;
; AVX-LABEL: @bar(
; AVX-NEXT:    [[TMP2:%.*]] = load <4 x i64>, ptr [[TMP0:%.*]], align 8
; AVX-NEXT:    [[TMP3:%.*]] = xor <4 x i64> [[TMP2]], splat (i64 -1)
; AVX-NEXT:    [[TMP4:%.*]] = shl <4 x i64> [[TMP3]], splat (i64 44)
; AVX-NEXT:    store <4 x i64> [[TMP4]], ptr [[TMP0]], align 8
; AVX-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP0]], i64 32
; AVX-NEXT:    [[TMP6:%.*]] = load <4 x i64>, ptr [[TMP5]], align 8
; AVX-NEXT:    [[TMP7:%.*]] = xor <4 x i64> [[TMP6]], splat (i64 -1)
; AVX-NEXT:    [[TMP8:%.*]] = shl <4 x i64> [[TMP7]], splat (i64 44)
; AVX-NEXT:    store <4 x i64> [[TMP8]], ptr [[TMP5]], align 8
; AVX-NEXT:    ret void
;
  br label %2

2:                                                ; preds = %12, %1
  %.0 = phi i32 [ 0, %1 ], [ %13, %12 ]
  %3 = icmp slt i32 %.0, 8
  br i1 %3, label %5, label %4

4:                                                ; preds = %2
  br label %14

5:                                                ; preds = %2
  %6 = sext i32 %.0 to i64
  %7 = getelementptr inbounds i64, ptr %0, i64 %6
  %8 = load i64, ptr %7, align 8
  %9 = call noundef i64 @foo(i64 noundef %8)
  %10 = sext i32 %.0 to i64
  %11 = getelementptr inbounds i64, ptr %0, i64 %10
  store i64 %9, ptr %11, align 8
  br label %12

12:                                               ; preds = %5
  %13 = add nsw i32 %.0, 1
  br label %2

14:                                               ; preds = %4
  ret void
}