[llvm] r176487 - R600: Turn BUILD_VECTOR into Reg_Sequence

[Vincent Lejeune]

Author: vljn
Date: Tue Mar  5 09:04:49 2013
New Revision: 176487

URL: http://llvm.org/viewvc/llvm-project?rev=176487&view=rev
Log:
R600: Turn BUILD_VECTOR into Reg_Sequence

Reviewed-by: Tom Stellard <thomas.stellard at amd.com>

Modified:
    llvm/trunk/lib/Target/R600/AMDILISelDAGToDAG.cpp
    llvm/trunk/lib/Target/R600/CMakeLists.txt
    llvm/trunk/test/CodeGen/R600/fdiv.v4f32.ll

Modified: llvm/trunk/lib/Target/R600/AMDILISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/AMDILISelDAGToDAG.cpp?rev=176487&r1=176486&r2=176487&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/AMDILISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/R600/AMDILISelDAGToDAG.cpp Tue Mar  5 09:04:49 2013
@@ -162,6 +162,35 @@ SDNode *AMDGPUDAGToDAGISel::Select(SDNod
   }
   switch (Opc) {
   default: break;
+  case ISD::BUILD_VECTOR: {
+    const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
+    if (ST.device()->getGeneration() > AMDGPUDeviceInfo::HD6XXX) {
+      break;
+    }
+    // BUILD_VECTOR is usually lowered into an IMPLICIT_DEF + 4 INSERT_SUBREG
+    // that adds a 128 bits reg copy when going through TwoAddressInstructions
+    // pass. We want to avoid 128 bits copies as much as possible because they
+    // can't be bundled by our scheduler.
+    SDValue RegSeqArgs[9] = {
+      CurDAG->getTargetConstant(AMDGPU::R600_Reg128RegClassID, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub0, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub1, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub2, MVT::i32),
+      SDValue(), CurDAG->getTargetConstant(AMDGPU::sub3, MVT::i32)
+    };
+    bool IsRegSeq = true;
+    for (unsigned i = 0; i < N->getNumOperands(); i++) {
+      if (dyn_cast<RegisterSDNode>(N->getOperand(i))) {
+        IsRegSeq = false;
+        break;
+      }
+      RegSeqArgs[2 * i + 1] = N->getOperand(i);
+    }
+    if (!IsRegSeq)
+      break;
+    return CurDAG->SelectNodeTo(N, AMDGPU::REG_SEQUENCE, N->getVTList(),
+        RegSeqArgs, 2 * N->getNumOperands() + 1);
+  }
   case ISD::ConstantFP:
   case ISD::Constant: {
     const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();

Modified: llvm/trunk/lib/Target/R600/CMakeLists.txt
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/CMakeLists.txt?rev=176487&r1=176486&r2=176487&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/CMakeLists.txt (original)
+++ llvm/trunk/lib/Target/R600/CMakeLists.txt Tue Mar  5 09:04:49 2013
@@ -37,7 +37,6 @@ add_llvm_target(R600CodeGen
   R600ExpandSpecialInstrs.cpp
   R600InstrInfo.cpp
   R600ISelLowering.cpp
-  R600LowerConstCopy.cpp
   R600MachineFunctionInfo.cpp
   R600RegisterInfo.cpp
   SIAnnotateControlFlow.cpp

Modified: llvm/trunk/test/CodeGen/R600/fdiv.v4f32.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/R600/fdiv.v4f32.ll?rev=176487&r1=176486&r2=176487&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/R600/fdiv.v4f32.ll (original)
+++ llvm/trunk/test/CodeGen/R600/fdiv.v4f32.ll Tue Mar  5 09:04:49 2013
@@ -1,13 +1,13 @@
 ;RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
 
 ;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
-;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: RECIP_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 ;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
+;CHECK: MUL_IEEE T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
 
 define void @test(<4 x float> addrspace(1)* %out, <4 x float> addrspace(1)* %in) {
   %b_ptr = getelementptr <4 x float> addrspace(1)* %in, i32 1