AMDGPU/SI: use S_MOV_B64 for larger copies in copyPhysReg

Reviewers: arsenm, tstellarAMD

Subscribers: arsenm, llvm-commits

Differential Revision: http://reviews.llvm.org/D15629

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@256073 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/SIInstrInfo.cpp b/lib/Target/AMDGPU/SIInstrInfo.cpp
index a37f8ecaaa417509d7a9b2086491d48c2eca74db..a08a5a8fed3612ab0c53868a9d3fc40be99caeaf 100644 (file)
--- a/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -326,15 +326,31 @@ SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
     AMDGPU::sub12, AMDGPU::sub13, AMDGPU::sub14, AMDGPU::sub15,
   };
 
+  static const int16_t Sub0_15_64[] = {
+    AMDGPU::sub0_sub1, AMDGPU::sub2_sub3,
+    AMDGPU::sub4_sub5, AMDGPU::sub6_sub7,
+    AMDGPU::sub8_sub9, AMDGPU::sub10_sub11,
+    AMDGPU::sub12_sub13, AMDGPU::sub14_sub15,
+  };
+
   static const int16_t Sub0_7[] = {
     AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3,
     AMDGPU::sub4, AMDGPU::sub5, AMDGPU::sub6, AMDGPU::sub7,
   };
 
+  static const int16_t Sub0_7_64[] = {
+    AMDGPU::sub0_sub1, AMDGPU::sub2_sub3,
+    AMDGPU::sub4_sub5, AMDGPU::sub6_sub7,
+  };
+
   static const int16_t Sub0_3[] = {
     AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2, AMDGPU::sub3,
   };
 
+  static const int16_t Sub0_3_64[] = {
+    AMDGPU::sub0_sub1, AMDGPU::sub2_sub3,
+  };
+
   static const int16_t Sub0_2[] = {
     AMDGPU::sub0, AMDGPU::sub1, AMDGPU::sub2,
   };
@@ -376,18 +392,18 @@ SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB,
 
   } else if (AMDGPU::SReg_128RegClass.contains(DestReg)) {
     assert(AMDGPU::SReg_128RegClass.contains(SrcReg));
-    Opcode = AMDGPU::S_MOV_B32;
-    SubIndices = Sub0_3;
+    Opcode = AMDGPU::S_MOV_B64;
+    SubIndices = Sub0_3_64;
 
   } else if (AMDGPU::SReg_256RegClass.contains(DestReg)) {
     assert(AMDGPU::SReg_256RegClass.contains(SrcReg));
-    Opcode = AMDGPU::S_MOV_B32;
-    SubIndices = Sub0_7;
+    Opcode = AMDGPU::S_MOV_B64;
+    SubIndices = Sub0_7_64;
 
   } else if (AMDGPU::SReg_512RegClass.contains(DestReg)) {
     assert(AMDGPU::SReg_512RegClass.contains(SrcReg));
-    Opcode = AMDGPU::S_MOV_B32;
-    SubIndices = Sub0_15;
+    Opcode = AMDGPU::S_MOV_B64;
+    SubIndices = Sub0_15_64;
 
   } else if (AMDGPU::VGPR_32RegClass.contains(DestReg)) {
     assert(AMDGPU::VGPR_32RegClass.contains(SrcReg) ||