AMDGPU: Select s_buffer_load_dword with a non-constant SGPR offset

Summary:
Apps that benefit:
- alien isolation
- bioshock infinite
- civilization: beyond earth
- company of heroes 2
- dirt showdown
- dota 2
- F1 2015
- grid autosport
- hitman
- legend of grimrock
- serious sam 3: bfe
- shadow warrior
- talos principle
- total war: warhammer
- UE4 demos: effects cave, elemental, sun temple

Reviewers: arsenm, nhaehnle

Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye

Differential Revision: https://reviews.llvm.org/D38914

llvm-svn: 317038

lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

@@ -169,7 +169,6 @@ private:
   bool SelectSMRDSgpr(SDValue Addr, SDValue &SBase, SDValue &Offset) const;
   bool SelectSMRDBufferImm(SDValue Addr, SDValue &Offset) const;
   bool SelectSMRDBufferImm32(SDValue Addr, SDValue &Offset) const;
-  bool SelectSMRDBufferSgpr(SDValue Addr, SDValue &Offset) const;
   bool SelectMOVRELOffset(SDValue Index, SDValue &Base, SDValue &Offset) const;
 
   bool SelectVOP3Mods_NNaN(SDValue In, SDValue &Src, SDValue &SrcMods) const;
@@ -1466,13 +1465,6 @@ bool AMDGPUDAGToDAGISel::SelectSMRDBufferImm32(SDValue Addr,
   return !Imm && isa<ConstantSDNode>(Offset);
 }
 
-bool AMDGPUDAGToDAGISel::SelectSMRDBufferSgpr(SDValue Addr,
-                                              SDValue &Offset) const {
-  bool Imm;
-  return SelectSMRDOffset(Addr, Offset, Imm) && !Imm &&
-         !isa<ConstantSDNode>(Offset);
-}
-
 bool AMDGPUDAGToDAGISel::SelectMOVRELOffset(SDValue Index,
                                             SDValue &Base,
                                             SDValue &Offset) const {

lib/Target/AMDGPU/BUFInstructions.td

@@ -878,13 +878,6 @@ def BUFFER_WBINVL1_VOL : MUBUF_Invalidate <"buffer_wbinvl1_vol",
 // MUBUF Patterns
 //===----------------------------------------------------------------------===//
 
-// Offset in an 32-bit VGPR
-def : GCNPat <
-  (SIload_constant v4i32:$sbase, i32:$voff),
-  (BUFFER_LOAD_DWORD_OFFEN $voff, $sbase, (i32 0), 0, 0, 0, 0)
->;
-
-
 //===----------------------------------------------------------------------===//
 // buffer_load/store_format patterns
 //===----------------------------------------------------------------------===//

lib/Target/AMDGPU/SIInstrInfo.cpp

@@ -3709,6 +3709,27 @@ void SIInstrInfo::moveToVALU(MachineInstr &TopInst) const {
       splitScalar64BitBinaryOp(Worklist, Inst, AMDGPU::S_XNOR_B32);
       Inst.eraseFromParent();
       continue;
+
+    case AMDGPU::S_BUFFER_LOAD_DWORD_SGPR: {
+      unsigned VDst = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+
+      BuildMI(*MBB, Inst, Inst.getDebugLoc(),
+              get(AMDGPU::BUFFER_LOAD_DWORD_OFFEN), VDst)
+        .add(*getNamedOperand(Inst, AMDGPU::OpName::soff)) // vaddr
+        .add(*getNamedOperand(Inst, AMDGPU::OpName::sbase)) // srsrc
+        .addImm(0) // soffset
+        .addImm(0) // offset
+        .addImm(getNamedOperand(Inst, AMDGPU::OpName::glc)->getImm())
+        .addImm(0) // slc
+        .addImm(0) // tfe
+        .setMemRefs(Inst.memoperands_begin(), Inst.memoperands_end());
+
+      MRI.replaceRegWith(getNamedOperand(Inst, AMDGPU::OpName::sdst)->getReg(),
+                         VDst);
+      addUsersToMoveToVALUWorklist(VDst, MRI, Worklist);
+      Inst.eraseFromParent();
+      continue;
+    }
     }
 
     if (NewOpcode == AMDGPU::INSTRUCTION_LIST_END) {

lib/Target/AMDGPU/SMInstructions.td

@@ -239,7 +239,6 @@ def SMRDImm32       : ComplexPattern<i64, 2, "SelectSMRDImm32">;
 def SMRDSgpr        : ComplexPattern<i64, 2, "SelectSMRDSgpr">;
 def SMRDBufferImm   : ComplexPattern<i32, 1, "SelectSMRDBufferImm">;
 def SMRDBufferImm32 : ComplexPattern<i32, 1, "SelectSMRDBufferImm32">;
-def SMRDBufferSgpr  : ComplexPattern<i32, 1, "SelectSMRDBufferSgpr">;
 
 multiclass SMRD_Pattern <string Instr, ValueType vt> {
 
@@ -282,7 +281,7 @@ def SM_LOAD_PATTERN : GCNPat <  // name this pattern to reuse AddedComplexity on
 
 // 2. Offset loaded in an 32bit SGPR
 def : GCNPat <
-  (SIload_constant v4i32:$sbase, (SMRDBufferSgpr i32:$offset)),
+  (SIload_constant v4i32:$sbase, i32:$offset),
   (S_BUFFER_LOAD_DWORD_SGPR $sbase, $offset, 0)
 >;
 

test/CodeGen/AMDGPU/smrd.ll

@@ -175,6 +175,22 @@ main_body:
   ret void
 }
 
+; GCN-LABEL: {{^}}smrd_sgpr_offset:
+; GCN: s_buffer_load_dword s{{[0-9]}}, s[0:3], s4
+define amdgpu_ps float @smrd_sgpr_offset(<4 x i32> inreg %desc, i32 inreg %offset) #0 {
+main_body:
+  %r = call float @llvm.SI.load.const.v4i32(<4 x i32> %desc, i32 %offset)
+  ret float %r
+}
+
+; GCN-LABEL: {{^}}smrd_vgpr_offset:
+; GCN: buffer_load_dword v{{[0-9]}}, v0, s[0:3], 0 offen ;
+define amdgpu_ps float @smrd_vgpr_offset(<4 x i32> inreg %desc, i32 %offset) #0 {
+main_body:
+  %r = call float @llvm.SI.load.const.v4i32(<4 x i32> %desc, i32 %offset)
+  ret float %r
+}
+
 declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #0
 declare float @llvm.SI.load.const.v4i32(<4 x i32>, i32) #1