[PATCH] D119271: clang: emit allocalign to LLVM for alloc_align attributes

[Augie Fackler via Phabricator via cfe-commits]

durin42 added inline comments.


================
Comment at: clang/test/CodeGen/catch-alignment-assumption-attribute-alloc_align-on-function-variable.cpp:55
   // CHECK-NEXT:                        %[[ALIGNMENT_RELOADED:.*]] = load i64, i64* %[[ALIGNMENT_ADDR]], align 8
-  // CHECK-NEXT:                        %[[X_RETURNED:.*]] = call noundef i8** @[[PASSTHROUGH]](i8** noundef %[[X_RELOADED]], i64 noundef %[[ALIGNMENT_RELOADED]])
-  // CHECK-SANITIZE-NEXT:               %[[PTRINT:.*]] = ptrtoint i8** %[[X_RETURNED]] to i64
-  // CHECK-SANITIZE-NEXT:               %[[MASK:.*]] = sub i64 %[[ALIGNMENT_RELOADED]], 1
-  // CHECK-SANITIZE-NEXT:               %[[MASKEDPTR:.*]] = and i64 %[[PTRINT]], %[[MASK]]
-  // CHECK-SANITIZE-NEXT:               %[[MASKCOND:.*]] = icmp eq i64 %[[MASKEDPTR]], 0
-  // CHECK-SANITIZE-NEXT:               %[[PTRINT_DUP:.*]] = ptrtoint i8** %[[X_RETURNED]] to i64, !nosanitize
-  // CHECK-SANITIZE-NEXT:               br i1 %[[MASKCOND]], label %[[CONT:.*]], label %[[HANDLER_ALIGNMENT_ASSUMPTION:[^,]+]],{{.*}} !nosanitize
-  // CHECK-SANITIZE:                  [[HANDLER_ALIGNMENT_ASSUMPTION]]:
-  // CHECK-SANITIZE-NORECOVER-NEXT:     call void @__ubsan_handle_alignment_assumption_abort(i8* bitcast ({ {{{.*}}}, {{{.*}}}, {{{.*}}}* }* @[[LINE_100_ALIGNMENT_ASSUMPTION]] to i8*), i64 %[[PTRINT_DUP]], i64 %[[ALIGNMENT_RELOADED]], i64 0){{.*}}, !nosanitize
-  // CHECK-SANITIZE-RECOVER-NEXT:       call void @__ubsan_handle_alignment_assumption(i8* bitcast ({ {{{.*}}}, {{{.*}}}, {{{.*}}}* }* @[[LINE_100_ALIGNMENT_ASSUMPTION]] to i8*), i64 %[[PTRINT_DUP]], i64 %[[ALIGNMENT_RELOADED]], i64 0){{.*}}, !nosanitize
-  // CHECK-SANITIZE-TRAP-NEXT:          call void @llvm.ubsantrap(i8 23){{.*}}, !nosanitize
-  // CHECK-SANITIZE-UNREACHABLE-NEXT:   unreachable, !nosanitize
-  // CHECK-SANITIZE:                  [[CONT]]:
-  // CHECK-NEXT:                        call void @llvm.assume(i1 true) [ "align"(i8** %[[X_RETURNED]], i64 %1) ]
+  // CHECK-NEXT:                        %[[X_RETURNED:.*]] = call noundef i8** @[[PASSTHROUGH]](i8** noundef %[[X_RELOADED]], i64 allocalign noundef %[[ALIGNMENT_RELOADED]])
   // CHECK-NEXT:                        ret i8** %[[X_RETURNED]]
----------------
lebedev.ri wrote:
> This is a regression, the old behavior was correct.
I don't think so: we now (as of D121629) check the alignment of the returned pointer in the callee instead of the caller, and LLVM knows about allocalign implying an alignment at the callsite. So I think it's correctly optimizing away the sanitizer checks here.

You or @jyknight will need to correct me if my understanding is incorrect here, but my understanding was that this change was what motivated D121629 in the first place.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D119271/new/

https://reviews.llvm.org/D119271