Commit d39b590a authored by Hans Wennborg's avatar Hans Wennborg
Browse files

Merging r355227 and r355228: 

------------------------------------------------------------------------
r355227 | ctopper | 2019-03-01 22:02:34 +0100 (Fri, 01 Mar 2019) | 3 lines

[X86] Add test case for D58805. NFC

This demonstrates dead store elimination removing a store that may alias a gather that uses null as its base.
------------------------------------------------------------------------

------------------------------------------------------------------------
r355228 | ctopper | 2019-03-01 22:02:40 +0100 (Fri, 01 Mar 2019) | 7 lines

[X86] Remove IntrArgMemOnly from target specific gather/scatter intrinsics

IntrArgMemOnly implies that only memory pointed to by pointer typed arguments will be accessed. But these intrinsics allow you to pass null to the pointer argument and put the full address into the index argument. Other passes won't be able to understand this.

A colleague found that ISPC was creating gathers like this and then dead store elimination removed some stores because it didn't understand what the gather was doing since the pointer argument was null.

Differential Revision: https://reviews.llvm.org/D58805
------------------------------------------------------------------------

llvm-svn: 355383
parent d1786389

llvm/include/llvm/IR/IntrinsicsX86.td

+134 −120

File changed.

Preview size limit exceeded, changes collapsed.

llvm/test/Transforms/DeadStoreElimination/X86/gather-null-pointer.ll

0 → 100644
+21 −0
Original line number Diff line number Diff line 
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py

; RUN: opt < %s -dse -S | FileCheck %s



; Both stores should be emitted because we can't tell if the gather aliases.



define <4 x i32> @bar(<4 x i32> %arg, i32* %arg1) {

; CHECK-LABEL: @bar(

; CHECK-NEXT:  bb:

; CHECK-NEXT:    store i32 5, i32* [[ARG1:%.*]]

; CHECK-NEXT:    [[TMP:%.*]] = tail call <4 x i32> @llvm.x86.avx2.gather.d.d(<4 x i32> zeroinitializer, i8* null, <4 x i32> [[ARG:%.*]], <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, i8 1)

; CHECK-NEXT:    store i32 10, i32* [[ARG1]]

; CHECK-NEXT:    ret <4 x i32> [[TMP]]

;

bb:

  store i32 5, i32* %arg1

  %tmp = tail call <4 x i32> @llvm.x86.avx2.gather.d.d(<4 x i32> zeroinitializer, i8* null, <4 x i32> %arg, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>, i8 1)

  store i32 10, i32* %arg1

  ret <4 x i32> %tmp

}



declare <4 x i32> @llvm.x86.avx2.gather.d.d(<4 x i32>, i8*, <4 x i32>, <4 x i32>, i8)