[X86] Refine speed up checking clobbered FP/BP to make IPRA work.

[FreddyLeaf]

X86 IPRA had below issue: https://gcc.godbolt.org/z/6hh88xv9r
This patch resolved it based on spillFPBP

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Freddy Ye (FreddyLeaf)

Changes

0d471b3 resolved IPRA issue on X86 before, refined a little to make it still work after speed up checking.

---

Full diff: https://github.com/llvm/llvm-project/pull/109246.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86FrameLowering.cpp (+7-2)
• (added) llvm/test/CodeGen/X86/ipra-local-linkage-2.ll (+21)

diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp
index 4f83267c999e4a..dfdf0554ef1f27 100644
--- a/llvm/lib/Target/X86/X86FrameLowering.cpp
+++ b/llvm/lib/Target/X86/X86FrameLowering.cpp
@@ -4495,7 +4495,10 @@ void X86FrameLowering::spillFPBP(MachineFunction &MF) const {
 
   // Currently only inline asm and function call can clobbers fp/bp. So we can
   // do some quick test and return early.
-  if (!MF.hasInlineAsm()) {
+  // And when IPRA is on, callee may also clobber fp/bp.
+  // FIXME: setF/BPClobberedByCall(true) in RegUsageInfoPropagation::setRegMask
+  // so we don't need the condition of IPRA here.
+  if (!MF.hasInlineAsm() && !MF.getTarget().Options.EnableIPRA) {
     X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>();
     if (!X86FI->getFPClobberedByCall())
       FP = 0;
@@ -4567,7 +4570,9 @@ void X86FrameLowering::spillFPBP(MachineFunction &MF) const {
 
       // If the bp is clobbered by a call, we should save and restore outside of
       // the frame setup instructions.
-      if (KillMI->isCall() && DefMI != ME) {
+      // When IPRA is enabled, we could skip this step.
+      if (KillMI->isCall() && DefMI != ME && !MF.hasInlineAsm() &&
+          !MF.getTarget().Options.EnableIPRA) {
         auto FrameSetup = std::next(DefMI);
         // Look for frame setup instruction toward the start of the BB.
         // If we reach another call instruction, it means no frame setup
diff --git a/llvm/test/CodeGen/X86/ipra-local-linkage-2.ll b/llvm/test/CodeGen/X86/ipra-local-linkage-2.ll
new file mode 100644
index 00000000000000..aad2b5c6c85417
--- /dev/null
+++ b/llvm/test/CodeGen/X86/ipra-local-linkage-2.ll
@@ -0,0 +1,21 @@
+; RUN: llc -enable-ipra < %s | FileCheck %s
+
+target triple = "x86_64--"
+
+define internal void @foo() norecurse {
+  call void asm sideeffect "xor	%ebp, %ebp", "~{ebp}"()
+  ret void
+}
+
+define void @bar(i32 %X) "frame-pointer"="all" {
+  ; When $rbp is clobbered by foo() and IPRA is enabled, $rbp should be spill/reload before/after call foo
+  ; CHECK-LABEL: bar:
+  ; CHECK: pushq %rbp
+  ; CHECK: callq foo
+  ; CHECK: popq %rbp
+  ; CHECK: movl $5, -4(%rbp)
+  call void @foo()
+  %addr = alloca i32, align 4
+  store i32 5, ptr %addr, align 4
+  ret void
+}

[phoebewang]

LGTM.

[FreddyLeaf]

Sorry, this PR is still not correct. set as draft first.

[FreddyLeaf]

After more investigation, I realized spillFPBP has to be done before argument passing even IPRA is enabled. Considering there is a function containing e.g. 9 pointers as arguments, if there is a push $rbp after argument passing and before call function, the argument inside the function will not be passed correctly as it requires stack passing for the 7th, 8th, ... arguments. Closed,

[FreddyLeaf]

Created #109597 as a new fix for the issue here.