[PATCH] D17749: [PPC64] Use mfocrf in prologue when we only need to save 1 nonvolatile CR field

[Chuang-Yu Cheng via llvm-commits]

cycheng updated this revision to Diff 49469.
cycheng marked 2 inline comments as done.
cycheng added a comment.

- Also use MfcrOpcode checking in the CR/LR serializing case
- Only do this for ELFv2ABI
- Fix test case


http://reviews.llvm.org/D17749

Files:
  lib/Target/PowerPC/PPCFrameLowering.cpp
  test/CodeGen/PowerPC/crsave.ll
  test/CodeGen/PowerPC/pr26690.ll

Index: test/CodeGen/PowerPC/pr26690.ll
===================================================================
--- test/CodeGen/PowerPC/pr26690.ll
+++ test/CodeGen/PowerPC/pr26690.ll
@@ -101,7 +101,7 @@
   ret i32 2
 }
 
-; CHECK: mfcr {{[0-9]+}}
+; CHECK: mfocrf {{[0-9]+}}
 
 !llvm.ident = !{!0}
 
Index: test/CodeGen/PowerPC/crsave.ll
===================================================================
--- test/CodeGen/PowerPC/crsave.ll
+++ test/CodeGen/PowerPC/crsave.ll
@@ -1,5 +1,6 @@
 ; RUN: llc -O0 -disable-fp-elim -mtriple=powerpc-unknown-linux-gnu -mcpu=g5 < %s | FileCheck %s -check-prefix=PPC32
 ; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -mcpu=g5 < %s | FileCheck %s -check-prefix=PPC64
+; RUN: llc -O0 -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s -check-prefix=PPC64-ELFv2
 
 declare void @foo()
 
@@ -60,3 +61,22 @@
 ; PPC64: mtocrf 16, 12
 ; PPC64: mtocrf 8, 12
 
+; Generate mfocrf in prologue when we need to save 1 nonvolatile CR field
+define void @cloberOneNvCrField() {
+entry:
+  tail call void asm sideeffect "# clobbers", "~{cr2}"()
+  ret void
+
+; PPC64-ELFv2-LABEL: @cloberOneNvCrField
+; PPC64-ELFv2: mfocrf [[REG1:[0-9]+]], 32
+}
+
+; Generate mfcr in prologue when we need to save all nonvolatile CR field
+define void @cloberAllNvCrField() {
+entry:
+  tail call void asm sideeffect "# clobbers", "~{cr2},~{cr3},~{cr4}"()
+  ret void
+
+; PPC64-ELFv2-LABEL: @cloberAllNvCrField
+; PPC64-ELFv2: mfcr [[REG1:[0-9]+]]
+}
Index: lib/Target/PowerPC/PPCFrameLowering.cpp
===================================================================
--- lib/Target/PowerPC/PPCFrameLowering.cpp
+++ lib/Target/PowerPC/PPCFrameLowering.cpp
@@ -838,11 +838,15 @@
   // If we need to spill the CR and the LR but we don't have two separate
   // registers available, we must spill them one at a time
   if (MustSaveCR && SingleScratchReg && MustSaveLR) {
-    // FIXME: In the ELFv2 ABI, we are not required to save all CR fields.
-    // If only one or two CR fields are clobbered, it could be more
-    // efficient to use mfocrf to selectively save just those fields.
+    // In the ELFv2 ABI, we are not required to save all CR fields.
+    // If only one or two CR fields are clobbered, it is more efficient to use
+    // mfocrf to selectively save just those fields, because mfocrf has short
+    // latency compares to mfcr.
+    unsigned MfcrOpcode = PPC::MFCR8;
+    if (isELFv2ABI && MustSaveCRs.size() == 1)
+      MfcrOpcode = PPC::MFOCRF8;
     MachineInstrBuilder MIB =
-      BuildMI(MBB, MBBI, dl, TII.get(PPC::MFCR8), TempReg);
+      BuildMI(MBB, MBBI, dl, TII.get(MfcrOpcode), TempReg);
     for (unsigned i = 0, e = MustSaveCRs.size(); i != e; ++i)
       MIB.addReg(MustSaveCRs[i], RegState::ImplicitKill);
     BuildMI(MBB, MBBI, dl, TII.get(PPC::STW8))
@@ -856,11 +860,15 @@
 
   if (MustSaveCR &&
       !(SingleScratchReg && MustSaveLR)) { // will only occur for PPC64
-    // FIXME: In the ELFv2 ABI, we are not required to save all CR fields.
-    // If only one or two CR fields are clobbered, it could be more
-    // efficient to use mfocrf to selectively save just those fields.
+    // In the ELFv2 ABI, we are not required to save all CR fields.
+    // If only one or two CR fields are clobbered, it is more efficient to use
+    // mfocrf to selectively save just those fields, because mfocrf has short
+    // latency compares to mfcr.
+    unsigned MfcrOpcode = PPC::MFCR8;
+    if (isELFv2ABI && MustSaveCRs.size() == 1)
+      MfcrOpcode = PPC::MFOCRF8;
     MachineInstrBuilder MIB =
-      BuildMI(MBB, MBBI, dl, TII.get(PPC::MFCR8), TempReg);
+      BuildMI(MBB, MBBI, dl, TII.get(MfcrOpcode), TempReg);
     for (unsigned i = 0, e = MustSaveCRs.size(); i != e; ++i)
       MIB.addReg(MustSaveCRs[i], RegState::ImplicitKill);
   }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D17749.49469.patch
Type: text/x-patch
Size: 3805 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160301/4ec5fe76/attachment.bin>