[llvm] r315852 - [MergeFunctions] Replace all uses of unnamed_addr functions.

Wed Oct 18 11:45:17 PDT 2017

I tried the second patch (since you said it is more efficient). That fixes the problem. Can you get it committed soon?

> On Oct 18, 2017, at 3:46 AM, whitequark <whitequark at whitequark.org> wrote:
> 
> On 2017-10-17 23:53, Bob Wilson wrote:
>> This change is causing an assertion failure for the Swift compiler
>> (https://ci.swift.org/view/swift-master-next/job/oss-swift-incremental-RA-osx-master-next/).
>> Assertion failed: (isa<KeySansPointerT>(new_key) && "Invalid RAUW on
>> key of ValueMap<>"), function allUsesReplacedWith, file
>> /Users/buildnode/jenkins/workspace/oss-swift-incremental-RA-osx-master-next/llvm/include/llvm/IR/ValueMap.h,
>> line 275.
>> The FunctionComparator utility used in MergeFunctions defines a
>> GlobalNumberState that uses a ValueMap to map GlobalValues to
>> integers. The ConstantExpr bitcast created here is not a GlobalValue.
>> I investigated this and found that the assertion fails when the value
>> is one of those new bitcast instructions. As an experiment, I changed
>> the ValueMap in the FunctionComparator to operate on Values instead of
>> GlobalValues and that fixed the assertion. I’m not sure that is the
>> right solution, though.
> 
> I don't think this is the right solution. Could you please try one
> of the attached patches? Either should work but the second is more
> efficient.
> 
>> Unfortunately, I don’t have a small testcase for this.
>>> On Oct 15, 2017, at 5:29 AM, whitequark via llvm-commits <llvm-commits at lists.llvm.org> wrote:
>>> Author: whitequark
>>> Date: Sun Oct 15 05:29:01 2017
>>> New Revision: 315852
>>> URL: http://llvm.org/viewvc/llvm-project?rev=315852&view=rev
>>> Log:
>>> [MergeFunctions] Replace all uses of unnamed_addr functions.
>>> This reduces code size for constructs like vtables or interrupt
>>> tables that refer to functions in global initializers.
>>> Differential Revision: https://reviews.llvm.org/D34805
>>> Added:
>>>   llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr-bitcast.ll
>>>   llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr.ll
>>> Modified:
>>>   llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp
>>> Modified: llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp?rev=315852&r1=315851&r2=315852&view=diff
>>> ==============================================================================
>>> --- llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp (original)
>>> +++ llvm/trunk/lib/Transforms/IPO/MergeFunctions.cpp Sun Oct 15 05:29:01 2017
>>> @@ -628,9 +628,15 @@ void MergeFunctions::filterInstsUnrelate
>>> // call sites to point to F even when within the same translation unit.
>>> void MergeFunctions::writeThunk(Function *F, Function *G) {
>>>  if (!G->isInterposable() && !MergeFunctionsPDI) {
>>> -    // Redirect direct callers of G to F. (See note on MergeFunctionsPDI
>>> -    // above).
>>> -    replaceDirectCallers(G, F);
>>> +    if (G->hasGlobalUnnamedAddr()) {
>>> +      // If G's address is not significant, replace it entirely.
>>> +      Constant *BitcastF = ConstantExpr::getBitCast(F, G->getType());
>>> +      G->replaceAllUsesWith(BitcastF);
>>> +    } else {
>>> +      // Redirect direct callers of G to F. (See note on MergeFunctionsPDI
>>> +      // above).
>>> +      replaceDirectCallers(G, F);
>>> +    }
>>>  }
>>>  // If G was internal then we may have replaced all uses of G with F. If so,
>>> Added: llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr-bitcast.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr-bitcast.ll?rev=315852&view=auto
>>> ==============================================================================
>>> --- llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr-bitcast.ll (added)
>>> +++ llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr-bitcast.ll Sun Oct 15 05:29:01 2017
>>> @@ -0,0 +1,30 @@
>>> +; RUN: opt -S -mergefunc < %s | FileCheck %s
>>> +
>>> +%A = type { i32 }
>>> +%B = type { i32 }
>>> +
>>> +; CHECK-NOT: @b
>>> +
>>> + at x = constant { i32 (i32)*, i32 (i32)* }
>>> +  { i32 (i32)* bitcast (i32 (%A)* @a to i32 (i32)*),
>>> +    i32 (i32)* bitcast (i32 (%B)* @b to i32 (i32)*) }
>>> +; CHECK: { i32 (i32)* bitcast (i32 (%A)* @a to i32 (i32)*), i32 (i32)* bitcast (i32 (%A)* @a to i32 (i32)*) }
>>> +
>>> +define internal i32 @a(%A) unnamed_addr {
>>> +  extractvalue %A %0, 0
>>> +  xor i32 %2, 0
>>> +  ret i32 %3
>>> +}
>>> +
>>> +define internal i32 @b(%B) unnamed_addr {
>>> +  extractvalue %B %0, 0
>>> +  xor i32 %2, 0
>>> +  ret i32 %3
>>> +}
>>> +
>>> +define i32 @c(i32) {
>>> +  insertvalue %B undef, i32 %0, 0
>>> +  call i32 @b(%B %2)
>>> +; CHECK: call i32 bitcast (i32 (%A)* @a to i32 (%B)*)(%B %2)
>>> +  ret i32 %3
>>> +}
>>> Added: llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr.ll
>>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr.ll?rev=315852&view=auto
>>> ==============================================================================
>>> --- llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr.ll (added)
>>> +++ llvm/trunk/test/Transforms/MergeFunc/merge-unnamed-addr.ll Sun Oct 15 05:29:01 2017
>>> @@ -0,0 +1,18 @@
>>> +; RUN: opt -S -mergefunc < %s | FileCheck %s
>>> +
>>> +; CHECK-NOT: @b
>>> +
>>> + at x = constant { i32 (i32)*, i32 (i32)* } { i32 (i32)* @a, i32 (i32)* @b }
>>> +; CHECK: { i32 (i32)* @a, i32 (i32)* @a }
>>> +
>>> +define internal i32 @a(i32 %a) unnamed_addr {
>>> +  %b = xor i32 %a, 0
>>> +  %c = xor i32 %b, 0
>>> +  ret i32 %c
>>> +}
>>> +
>>> +define internal i32 @b(i32 %a) unnamed_addr {
>>> +  %b = xor i32 %a, 0
>>> +  %c = xor i32 %b, 0
>>> +  ret i32 %c
>>> +}
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at lists.llvm.org
>>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits
> 
> -- 
> whitequark<fix-mergefunc-2.patch><fix-mergefunc-1.patch>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171018/abe58cb8/attachment.html>