[PATCH] D134281: [CGP] Update MemIntrinsic alignment if possible

[Phoebe Wang via Phabricator via llvm-commits]

pengfei added inline comments.


================
Comment at: llvm/lib/CodeGen/CodeGenPrepare.cpp:2224-2230
     for (auto &Arg : CI->args()) {
       // We want to align both objects whose address is used directly and
       // objects whose address is used in casts and GEPs, though it only makes
       // sense for GEPs if the offset is a multiple of the desired alignment and
       // if size - offset meets the size threshold.
       if (!Arg->getType()->isPointerTy())
         continue;
----------------
Re-format it.


================
Comment at: llvm/test/CodeGen/X86/memset-2.ll:8-18
+; CHECK-NEXT:    movaps %xmm0, 160
+; CHECK-NEXT:    movaps %xmm0, 144
+; CHECK-NEXT:    movaps %xmm0, 128
+; CHECK-NEXT:    movaps %xmm0, 112
+; CHECK-NEXT:    movaps %xmm0, 96
+; CHECK-NEXT:    movaps %xmm0, 80
+; CHECK-NEXT:    movaps %xmm0, 64
----------------
Not sure if the use of `movaps` is a good idea considering the test in below file.


================
Comment at: llvm/test/CodeGen/X86/memset64-on-x86-32.ll:17-34
 ; SLOW_32-LABEL: bork:
 ; SLOW_32:       # %bb.0:
-; SLOW_32-NEXT:    movl $0, 4
-; SLOW_32-NEXT:    movl $0, 0
-; SLOW_32-NEXT:    movl $0, 12
-; SLOW_32-NEXT:    movl $0, 8
-; SLOW_32-NEXT:    movl $0, 20
-; SLOW_32-NEXT:    movl $0, 16
-; SLOW_32-NEXT:    movl $0, 28
-; SLOW_32-NEXT:    movl $0, 24
-; SLOW_32-NEXT:    movl $0, 36
-; SLOW_32-NEXT:    movl $0, 32
-; SLOW_32-NEXT:    movl $0, 44
-; SLOW_32-NEXT:    movl $0, 40
-; SLOW_32-NEXT:    movl $0, 52
-; SLOW_32-NEXT:    movl $0, 48
-; SLOW_32-NEXT:    movl $0, 60
-; SLOW_32-NEXT:    movl $0, 56
-; SLOW_32-NEXT:    movl $0, 68
-; SLOW_32-NEXT:    movl $0, 64
-; SLOW_32-NEXT:    movl $0, 76
-; SLOW_32-NEXT:    movl $0, 72
+; SLOW_32-NEXT:    xorps %xmm0, %xmm0
+; SLOW_32-NEXT:    movaps %xmm0, 64
+; SLOW_32-NEXT:    movaps %xmm0, 48
+; SLOW_32-NEXT:    movaps %xmm0, 32
+; SLOW_32-NEXT:    movaps %xmm0, 16
----------------
Are they intent to use `movl/q` here?


================
Comment at: llvm/test/Transforms/CodeGenPrepare/X86/memset_chk-simplify-nobuiltin.ll:14
 ; CHECK-LABEL: @test_nobuiltin(
-; CHECK-NEXT:    call void @llvm.memset.p0i8.i64(i8* [[DST:%.*]], i8 0, i64 [[LEN:%.*]], i1 false) #1
+; CHECK-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 1 [[DST:%.*]], i8 0, i64 [[LEN:%.*]], i1 false) #[[ATTR1:[0-9]+]]
 ; CHECK-NEXT:    ret void
----------------
No `ATTR1` to match with it. Just removing the `#1` instead?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134281/new/

https://reviews.llvm.org/D134281