llvm/test/Transforms/SLPVectorizer/X86/gep.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -passes=slp-vectorizer -S |FileCheck %s
; RUN: opt < %s -aa-pipeline=basic-aa -passes=slp-vectorizer -S |FileCheck %s
target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-unknown"

; Test if SLP can handle GEP expressions.
; The test perform the following action:
;   x->first  = y->first  + 16
;   x->second = y->second + 16

define void @foo1 (ptr noalias %x, ptr noalias %y) {
; CHECK-LABEL: @foo1(
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds { ptr, ptr }, ptr [[Y:%.*]], i64 0, i32 0
; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds { ptr, ptr }, ptr [[X:%.*]], i64 0, i32 0
; CHECK-NEXT:    [[TMP4:%.*]] = load <2 x ptr>, ptr [[TMP1]], align 8
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i32, <2 x ptr> [[TMP4]], <2 x i64> <i64 16, i64 16>
; CHECK-NEXT:    store <2 x ptr> [[TMP5]], ptr [[TMP2]], align 8
; CHECK-NEXT:    ret void
;
  %1 = getelementptr inbounds { ptr, ptr }, ptr %y, i64 0, i32 0
  %2 = load ptr, ptr %1, align 8
  %3 = getelementptr inbounds i32, ptr %2, i64 16
  %4 = getelementptr inbounds { ptr, ptr }, ptr %x, i64 0, i32 0
  store ptr %3, ptr %4, align 8
  %5 = getelementptr inbounds { ptr, ptr }, ptr %y, i64 0, i32 1
  %6 = load ptr, ptr %5, align 8
  %7 = getelementptr inbounds i32, ptr %6, i64 16
  %8 = getelementptr inbounds { ptr, ptr }, ptr %x, i64 0, i32 1
  store ptr %7, ptr %8, align 8
  ret void
}

; Test that we don't vectorize GEP expressions if indexes are not constants.
; We can't produce an efficient code in that case.
define void @foo2 (ptr noalias %x, ptr noalias %y, i32 %i) {
; CHECK-LABEL: @foo2(
; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds { ptr, ptr }, ptr [[Y:%.*]], i64 0, i32 0
; CHECK-NEXT:    [[TMP2:%.*]] = load ptr, ptr [[TMP1]], align 8
; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[TMP2]], i32 [[I:%.*]]
; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds { ptr, ptr }, ptr [[X:%.*]], i64 0, i32 0
; CHECK-NEXT:    store ptr [[TMP3]], ptr [[TMP4]], align 8
; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds { ptr, ptr }, ptr [[Y]], i64 0, i32 1
; CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[TMP5]], align 8
; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[TMP6]], i32 [[I]]
; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds { ptr, ptr }, ptr [[X]], i64 0, i32 1
; CHECK-NEXT:    store ptr [[TMP7]], ptr [[TMP8]], align 8
; CHECK-NEXT:    ret void
;
  %1 = getelementptr inbounds { ptr, ptr }, ptr %y, i64 0, i32 0
  %2 = load ptr, ptr %1, align 8
  %3 = getelementptr inbounds i32, ptr %2, i32 %i
  %4 = getelementptr inbounds { ptr, ptr }, ptr %x, i64 0, i32 0
  store ptr %3, ptr %4, align 8
  %5 = getelementptr inbounds { ptr, ptr }, ptr %y, i64 0, i32 1
  %6 = load ptr, ptr %5, align 8
  %7 = getelementptr inbounds i32, ptr %6, i32 %i
  %8 = getelementptr inbounds { ptr, ptr }, ptr %x, i64 0, i32 1
  store ptr %7, ptr %8, align 8
  ret void
}