Emit FCCMP for AND of two float compares

Beginners
1

Hi Everyone! I was working on the issue: [AArch64] Missed FCCMP opportunity · Issue #60819 · llvm/llvm-project · GitHub . And I have created the patch for it: ⚙ D152714 [AArch64][Optimization]Emit FCCMP for AND of two float compares . But now I am unable to fix the test file i.e., select_fmf.ll. Can anyone please help me with that?

2

Looking at the generated code, it looks like what’s happening is that your new code is triggering on the testcase, but that blocks an existing transform that emits fccmp for a SELECT/SELECT_CC. You should be able to trace what’s happening with a debugger to be sure, but I expect the relevant code is getAArch64Cmp(). The code for lowering SELECT sees an AArch64ISD::CSINC, and doesn’t realize it can consume the flags directly.

I can think of a couple different ways to deal with this: one, you could walk the use list of the SETCC and see if it’s used by a SELECT. Or two, you could teach the SELECT lowering to look through the CSINC. The first is more conservative.

3

How can I see the use list of SETCC?

4

N->uses() returns the uses.

5 
// Transform and(fcmp(a, b), fcmp(c, d)) into fccmp(fcmp(a, b), c, d)
static SDValue performANDSETCCCombine(SDNode *N,
                                      TargetLowering::DAGCombinerInfo &DCI) {

  // This function performs an optimization on a specific pattern involving
  // an AND operation and SETCC (Set Condition Code) node.

  SDValue SetCC = N->getOperand(0);
  EVT VT = N->getValueType(0);
  SelectionDAG &DAG = DCI.DAG;
N->uses();
  // Check if the operand is a SETCC node with floating-point comparison
  if (SetCC.getOpcode() == ISD::SETCC &&
      SetCC.getOperand(0).getValueType() == MVT::f32) {

    SDValue Cmp;
    AArch64CC::CondCode CC;

Like I use it like this then where the SETCC uses will be printed?

6

If I write it like this:

dbgs()<<N->uses();

then it is giving me error.

7

Try something like for (auto U : N->uses()) U->dump();

8

Using this gives me the following error:

E:\llvm-project\llvm\lib\Target\AArch64\AArch64ISelLowering.cpp(16394,7): error C2146: syntax error: missing ')' before
 identifier 'U' [E:\llvm-build\lib\Target\AArch64\LLVMAArch64CodeGen.vcxproj]
E:\llvm-project\llvm\lib\Target\AArch64\AArch64ISelLowering.cpp(16394,7): error C2065: 'U': undeclared identifier [E:\l
lvm-build\lib\Target\AArch64\LLVMAArch64CodeGen.vcxproj]
E:\llvm-project\llvm\lib\Target\AArch64\AArch64ISelLowering.cpp(16394,7): error C3537: you cannot cast to a type that c
ontains 'auto' [E:\llvm-build\lib\Target\AArch64\LLVMAArch64CodeGen.vcxproj]
E:\llvm-project\llvm\lib\Target\AArch64\AArch64ISelLowering.cpp(16394,9): error C2143: syntax error: missing ';' before
 ':' [E:\llvm-build\lib\Target\AArch64\LLVMAArch64CodeGen.vcxproj]
E:\llvm-project\llvm\lib\Target\AArch64\AArch64ISelLowering.cpp(16394,9): error C2059: syntax error: ':' [E:\llvm-build
\lib\Target\AArch64\LLVMAArch64CodeGen.vcxproj]
E:\llvm-project\llvm\lib\Target\AArch64\AArch64ISelLowering.cpp(16394,20): error C2059: syntax error: ')' [E:\llvm-buil
d\lib\Target\AArch64\LLVMAArch64CodeGen.vcxproj]
9

Did you omit the second close paren in uses( ) ) ?

10

No. Still it is giving error for ).

11

As far as I can tell, you didn’t actually use the code I wrote. Please actually use the code I wrote, as I wrote it.

For anyone else following along, the following generates a similar sequence of error messages

void f(void*N) {
    1<<(auto U : N->uses()) U->dump();
}
12

Okay! It worked. Thank you very much.