R600/SI: handle all registers in copyPhysReg v2

v2: based on Michels patch, but now allows copying of all registers sizes.

Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176346 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 88 insertions, 16 deletions

diff --git a/lib/Target/R600/SIInstrInfo.cpp b/lib/Target/R600/SIInstrInfo.cpp
index d9dbd6a4ca..de2373b11a 100644
--- a/lib/Target/R600/SIInstrInfo.cpp
+++ b/lib/Target/R600/SIInstrInfo.cpp
@@ -33,36 +33,108 @@ const SIRegisterInfo &SIInstrInfo::getRegisterInfo() const {
 
 void
 SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
-                           MachineBasicBlock::iterator MI, DebugLoc DL,
-                           unsigned DestReg, unsigned SrcReg,
-                           bool KillSrc) const {
+                         MachineBasicBlock::iterator MI, DebugLoc DL,
+                         unsigned DestReg, unsigned SrcReg,
+                         bool KillSrc) const {
+
   // If we are trying to copy to or from SCC, there is a bug somewhere else in
   // the backend.  While it may be theoretically possible to do this, it should
   // never be necessary.
   assert(DestReg != AMDGPU::SCC && SrcReg != AMDGPU::SCC);
 
-  if (AMDGPU::VReg_64RegClass.contains(DestReg)) {
-    assert(AMDGPU::VReg_64RegClass.contains(SrcReg) ||
-	   AMDGPU::SReg_64RegClass.contains(SrcReg));
-    BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), RI.getSubReg(DestReg, AMDGPU::sub0))
-            .addReg(RI.getSubReg(SrcReg, AMDGPU::sub0), getKillRegState(KillSrc))
-            .addReg(DestReg, RegState::Define | RegState::Implicit);
-    BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), RI.getSubReg(DestReg, AMDGPU::sub1))
-            .addReg(RI.getSubReg(SrcReg, AMDGPU::sub1), getKillRegState(KillSrc));
+  const int16_t Sub0_15[] = {
+    AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3,
+    AMDGPU::sub4, AMDGPU::sub5, AMDGPU::sub6, AMDGPU::sub7,
+    AMDGPU::sub8, AMDGPU::sub9, AMDGPU::sub10, AMDGPU::sub11,
+    AMDGPU::sub12, AMDGPU::sub13, AMDGPU::sub14, AMDGPU::sub15, 0
+  };
+
+  const int16_t Sub0_7[] = {
+    AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3,
+    AMDGPU::sub4, AMDGPU::sub5, AMDGPU::sub6, AMDGPU::sub7, 0
+  };
+
+  const int16_t Sub0_3[] = {
+    AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3, 0
+  };
+
+  const int16_t Sub0_1[] = {
+    AMDGPU::sub0, AMDGPU::sub1, 0
+  };
+
+  unsigned Opcode;
+  const int16_t *SubIndices;
+
+  if (AMDGPU::SReg_32RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_32RegClass.contains(SrcReg));
+    BuildMI(MBB, MI, DL, get(AMDGPU::S_MOV_B32), DestReg)
+            .addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+
   } else if (AMDGPU::SReg_64RegClass.contains(DestReg)) {
     assert(AMDGPU::SReg_64RegClass.contains(SrcReg));
     BuildMI(MBB, MI, DL, get(AMDGPU::S_MOV_B64), DestReg)
             .addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+
+  } else if (AMDGPU::SReg_128RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_128RegClass.contains(SrcReg));
+    Opcode = AMDGPU::S_MOV_B32;
+    SubIndices = Sub0_3;
+
+  } else if (AMDGPU::SReg_256RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_256RegClass.contains(SrcReg));
+    Opcode = AMDGPU::S_MOV_B32;
+    SubIndices = Sub0_7;
+
+  } else if (AMDGPU::SReg_512RegClass.contains(DestReg)) {
+    assert(AMDGPU::SReg_512RegClass.contains(SrcReg));
+    Opcode = AMDGPU::S_MOV_B32;
+    SubIndices = Sub0_15;
+
   } else if (AMDGPU::VReg_32RegClass.contains(DestReg)) {
     assert(AMDGPU::VReg_32RegClass.contains(SrcReg) ||
-           AMDGPU::SReg_32RegClass.contains(SrcReg));
+	   AMDGPU::SReg_32RegClass.contains(SrcReg));
     BuildMI(MBB, MI, DL, get(AMDGPU::V_MOV_B32_e32), DestReg)
             .addReg(SrcReg, getKillRegState(KillSrc));
+    return;
+
+  } else if (AMDGPU::VReg_64RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_64RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_64RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_1;
+
+  } else if (AMDGPU::VReg_128RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_128RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_128RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_3;
+
+  } else if (AMDGPU::VReg_256RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_256RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_256RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_7;
+
+  } else if (AMDGPU::VReg_512RegClass.contains(DestReg)) {
+    assert(AMDGPU::VReg_512RegClass.contains(SrcReg) ||
+	   AMDGPU::SReg_512RegClass.contains(SrcReg));
+    Opcode = AMDGPU::V_MOV_B32_e32;
+    SubIndices = Sub0_15;
+
   } else {
-    assert(AMDGPU::SReg_32RegClass.contains(DestReg));
-    assert(AMDGPU::SReg_32RegClass.contains(SrcReg));
-    BuildMI(MBB, MI, DL, get(AMDGPU::S_MOV_B32), DestReg)
-            .addReg(SrcReg, getKillRegState(KillSrc));
+    llvm_unreachable("Can't copy register!");
+  }
+
+  while (unsigned SubIdx = *SubIndices++) {
+    MachineInstrBuilder Builder = BuildMI(MBB, MI, DL,
+      get(Opcode), RI.getSubReg(DestReg, SubIdx));
+
+    Builder.addReg(RI.getSubReg(SrcReg, SubIdx), getKillRegState(KillSrc));
+
+    if (*SubIndices)
+      Builder.addReg(DestReg, RegState::Define | RegState::Implicit);
   }
 }