diff options
author | Bill Schmidt <wschmidt@linux.vnet.ibm.com> | 2012-10-10 21:25:01 +0000 |
---|---|---|
committer | Bill Schmidt <wschmidt@linux.vnet.ibm.com> | 2012-10-10 21:25:01 +0000 |
commit | 26160f4e6403f3ddd7c4599422578cb46777103d (patch) | |
tree | 932119198bde25a7908b03c96ca07b03a12b427f | |
parent | a5d0ab555384baa293b06686bec5a01fb9638ca3 (diff) |
When generating spill and reload code for vector registers on PowerPC,
the compiler makes use of GPR0. However, there are two flavors of
GPR0 defined by the target: the 32-bit GPR0 (R0) and the 64-bit GPR0
(X0). The spill/reload code makes use of R0 regardless of whether we
are generating 32- or 64-bit code.
This patch corrects the problem in the obvious manner, using X0 and
ADDI8 for 64-bit and R0 and ADDI for 32-bit.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@165658 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r-- | lib/Target/PowerPC/PPCInstrInfo.cpp | 18 | ||||
-rw-r--r-- | test/CodeGen/PowerPC/vrspill.ll | 19 |
2 files changed, 31 insertions, 6 deletions
diff --git a/lib/Target/PowerPC/PPCInstrInfo.cpp b/lib/Target/PowerPC/PPCInstrInfo.cpp index d2df6645bb..d9d68446f5 100644 --- a/lib/Target/PowerPC/PPCInstrInfo.cpp +++ b/lib/Target/PowerPC/PPCInstrInfo.cpp @@ -570,12 +570,15 @@ PPCInstrInfo::StoreRegToStackSlot(MachineFunction &MF, // STVX VAL, 0, R0 // // FIXME: We use R0 here, because it isn't available for RA. - NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::ADDI), PPC::R0), + bool Is64Bit = TM.getSubtargetImpl()->isPPC64(); + unsigned Instr = Is64Bit ? PPC::ADDI8 : PPC::ADDI; + unsigned GPR0 = Is64Bit ? PPC::X0 : PPC::R0; + NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Instr), GPR0), FrameIdx, 0, 0)); NewMIs.push_back(BuildMI(MF, DL, get(PPC::STVX)) .addReg(SrcReg, getKillRegState(isKill)) - .addReg(PPC::R0) - .addReg(PPC::R0)); + .addReg(GPR0) + .addReg(GPR0)); } else { llvm_unreachable("Unknown regclass!"); } @@ -707,10 +710,13 @@ PPCInstrInfo::LoadRegFromStackSlot(MachineFunction &MF, DebugLoc DL, // Dest = LVX 0, R0 // // FIXME: We use R0 here, because it isn't available for RA. - NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(PPC::ADDI), PPC::R0), + bool Is64Bit = TM.getSubtargetImpl()->isPPC64(); + unsigned Instr = Is64Bit ? PPC::ADDI8 : PPC::ADDI; + unsigned GPR0 = Is64Bit ? PPC::X0 : PPC::R0; + NewMIs.push_back(addFrameReference(BuildMI(MF, DL, get(Instr), GPR0), FrameIdx, 0, 0)); - NewMIs.push_back(BuildMI(MF, DL, get(PPC::LVX),DestReg).addReg(PPC::R0) - .addReg(PPC::R0)); + NewMIs.push_back(BuildMI(MF, DL, get(PPC::LVX),DestReg).addReg(GPR0) + .addReg(GPR0)); } else { llvm_unreachable("Unknown regclass!"); } diff --git a/test/CodeGen/PowerPC/vrspill.ll b/test/CodeGen/PowerPC/vrspill.ll new file mode 100644 index 0000000000..fdd200c103 --- /dev/null +++ b/test/CodeGen/PowerPC/vrspill.ll @@ -0,0 +1,19 @@ +; RUN: llc -O0 -mtriple=powerpc-unknown-linux-gnu -verify-machineinstrs < %s | FileCheck %s +; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -verify-machineinstrs < %s | FileCheck %s + +; This verifies that we generate correct spill/reload code for vector regs. + +define void @addrtaken(i32 %i, <4 x float> %w) nounwind { +entry: + %i.addr = alloca i32, align 4 + %w.addr = alloca <4 x float>, align 16 + store i32 %i, i32* %i.addr, align 4 + store <4 x float> %w, <4 x float>* %w.addr, align 16 + call void @foo(i32* %i.addr) + ret void +} + +; CHECK: stvx 2, 0, 0 +; CHECK: lvx 2, 0, 0 + +declare void @foo(i32*) |