[RISCV] Add areInlineCompatible for riscv target

[jacquesguan]

Inline a callee if its target-features are a subset of the callers target-features.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-backend-risc-v

Author: Jianjian Guan (jacquesguan)

Changes

Inline a callee if its target-features are a subset of the callers target-features.

---

Full diff: https://github.com/llvm/llvm-project/pull/86639.diff

4 Files Affected:

• (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp (+14)
• (modified) llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h (+3)
• (added) llvm/test/Transforms/Inline/RISCV/inline-target-features.ll (+34)
• (added) llvm/test/Transforms/Inline/RISCV/lit.local.cfg (+2)

diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
index f75b3d3caa62f2..b8f3c13e4508b0 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp
@@ -1675,3 +1675,17 @@ bool RISCVTTIImpl::isLegalMaskedCompressStore(Type *DataTy, Align Alignment) {
     return false;
   return true;
 }
+
+bool RISCVTTIImpl::areInlineCompatible(const Function *Caller,
+                                       const Function *Callee) const {
+  const TargetMachine &TM = getTLI()->getTargetMachine();
+
+  const FeatureBitset &CallerBits =
+      TM.getSubtargetImpl(*Caller)->getFeatureBits();
+  const FeatureBitset &CalleeBits =
+      TM.getSubtargetImpl(*Callee)->getFeatureBits();
+
+  // Inline a callee if its target-features are a subset of the callers
+  // target-features.
+  return (CallerBits & CalleeBits) == CalleeBits;
+}
diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
index 8daf6845dc8bc9..ac32aea4ce2b8f 100644
--- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
+++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.h
@@ -60,6 +60,9 @@ class RISCVTTIImpl : public BasicTTIImplBase<RISCVTTIImpl> {
       : BaseT(TM, F.getParent()->getDataLayout()), ST(TM->getSubtargetImpl(F)),
         TLI(ST->getTargetLowering()) {}
 
+  bool areInlineCompatible(const Function *Caller,
+                           const Function *Callee) const;
+
   /// Return the cost of materializing an immediate for a value operand of
   /// a store instruction.
   InstructionCost getStoreImmCost(Type *VecTy, TTI::OperandValueInfo OpInfo,
diff --git a/llvm/test/Transforms/Inline/RISCV/inline-target-features.ll b/llvm/test/Transforms/Inline/RISCV/inline-target-features.ll
new file mode 100644
index 00000000000000..b626a229a737d5
--- /dev/null
+++ b/llvm/test/Transforms/Inline/RISCV/inline-target-features.ll
@@ -0,0 +1,34 @@
+; RUN: opt < %s -mtriple=riscv64-unknown-linux-gnu -S -passes=inline | FileCheck %s
+; RUN: opt < %s -mtriple=riscv64-unknown-linux-gnu -S -passes='cgscc(inline)' | FileCheck %s
+; Check that we only inline when we have compatible target attributes.
+
+target datalayout = "e-m:e-p:64:64-i64:64-i128:128-n64-S128"
+target triple = "riscv64-unknown-linux-gnu"
+
+define i32 @foo() #0 {
+entry:
+  %call = call i32 (...) @baz()
+  ret i32 %call
+; CHECK-LABEL: foo
+; CHECK: call i32 (...) @baz()
+}
+declare i32 @baz(...) #0
+
+define i32 @bar() #1 {
+entry:
+  %call = call i32 @foo()
+  ret i32 %call
+; CHECK-LABEL: bar
+; CHECK: call i32 (...) @baz()
+}
+
+define i32 @qux() #0 {
+entry:
+  %call = call i32 @bar()
+  ret i32 %call
+; CHECK-LABEL: qux
+; CHECK: call i32 @bar()
+}
+
+attributes #0 = { "target-cpu"="generic-rv64" "target-features"="+f,+d" }
+attributes #1 = { "target-cpu"="generic-rv64" "target-features"="+f,+d,+m,+v" }
diff --git a/llvm/test/Transforms/Inline/RISCV/lit.local.cfg b/llvm/test/Transforms/Inline/RISCV/lit.local.cfg
new file mode 100644
index 00000000000000..17351748513d98
--- /dev/null
+++ b/llvm/test/Transforms/Inline/RISCV/lit.local.cfg
@@ -0,0 +1,2 @@
+if not "RISCV" in config.root.targets:
+    config.unsupported = True

[lukel97]

Does this mean we'll inline a function even if it we disabled an extension via an overriding target attribute arch string? e.g.

__attribute__((target("arch=rv64i")))
float foo(float x) {
  return x + 42;
}

float bar(float x) {
  return foo(x);
}

foo will now be inlined if we compile with -march=rv64if. But I presume that's fine given that this seems to be inline with what other targets do

[jacquesguan]

Does this mean we'll inline a function even if it we disabled an extension via an overriding target attribute arch string? e.g.

__attribute__((target("arch=rv64i")))
float foo(float x) {
  return x + 42;
}

float bar(float x) {
  return foo(x);
}

foo will now be inlined if we compile with -march=rv64if. But I presume that's fine given that this seems to be inline with what other targets do

Yes, in my work, there are some similar cases. I have some common tool functions and want to inline them into some users with
different target attributes.

[topperc]

LGTM

[lukel97]

LGTM