R600/SI: Use correct dest register class for V_READFIRSTLANE_B32

This instructions writes to an 32-bit SGPR.  This change required adding
the 32-bit VCC_LO and VCC_HI registers, because the full VCC register
is 64 bits.

This fixes verifier errors on several of the indirect addressing piglit
tests.

Tested-by: Michel Dänzer <michel.daenzer@amd.com>
llvm-svn: 204055

llvm/lib/Target/R600/AMDGPUAsmPrinter.cpp

@@ -210,7 +210,8 @@ void AMDGPUAsmPrinter::findNumUsedRegistersSI(MachineFunction &MF,
           continue;
         }
         unsigned reg = MO.getReg();
-        if (reg == AMDGPU::VCC) {
+        if (reg == AMDGPU::VCC || reg == AMDGPU::VCC_LO ||
+	    reg == AMDGPU::VCC_HI) {
           VCCUsed = true;
           continue;
         }

llvm/lib/Target/R600/SIInstructions.td

@@ -631,7 +631,18 @@ let neverHasSideEffects = 1, isMoveImm = 1 in {
 defm V_MOV_B32 : VOP1_32 <0x00000001, "V_MOV_B32", []>;
 } // End neverHasSideEffects = 1, isMoveImm = 1
 
-defm V_READFIRSTLANE_B32 : VOP1_32 <0x00000002, "V_READFIRSTLANE_B32", []>;
+let Uses = [EXEC] in {
+
+def V_READFIRSTLANE_B32 : VOP1 <
+  0x00000002,
+  (outs SReg_32:$vdst),
+  (ins VReg_32:$src0),
+  "V_READFIRSTLANE_B32 $vdst, $src0",
+  []
+>;
+
+}
+
 defm V_CVT_I32_F64 : VOP1_32_64 <0x00000003, "V_CVT_I32_F64",
   [(set i32:$dst, (fp_to_sint f64:$src0))]
 >;

llvm/lib/Target/R600/SILowerControlFlow.cpp

@@ -345,12 +345,13 @@ void SILowerControlFlowPass::LoadM0(MachineInstr &MI, MachineInstr *MovRel) {
           .addReg(AMDGPU::EXEC);
 
   // Read the next variant into VCC (lower 32 bits) <- also loop target
-  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32_e32), AMDGPU::VCC)
+  BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_READFIRSTLANE_B32),
+          AMDGPU::VCC_LO)
           .addReg(Idx);
 
   // Move index from VCC into M0
   BuildMI(MBB, &MI, DL, TII->get(AMDGPU::S_MOV_B32), AMDGPU::M0)
-          .addReg(AMDGPU::VCC);
+          .addReg(AMDGPU::VCC_LO);
 
   // Compare the just read M0 value to all possible Idx values
   BuildMI(MBB, &MI, DL, TII->get(AMDGPU::V_CMP_EQ_U32_e32), AMDGPU::VCC)

llvm/lib/Target/R600/SIRegisterInfo.td

@@ -17,7 +17,16 @@ class SIReg <string n, bits<16> encoding = 0> : Register<n> {
 }
 
 // Special Registers
-def VCC : SIReg<"VCC", 106>;
+def VCC_LO : SIReg<"vcc_lo", 106>;
+def VCC_HI : SIReg<"vcc_hi", 107>;
+
+// VCC for 64-bit instructions
+def VCC : RegisterWithSubRegs<"VCC", [VCC_LO, VCC_HI]> {
+  let Namespace = "AMDGPU";
+  let SubRegIndices = [sub0, sub1];
+  let HWEncoding = 106;
+}
+
 def EXEC : SIReg<"EXEC", 126>;
 def SCC : SIReg<"SCC", 253>;
 def M0 : SIReg <"M0", 124>;
@@ -150,7 +159,7 @@ def M0Reg : RegisterClass<"AMDGPU", [i32], 32, (add M0)>;
 
 // Register class for all scalar registers (SGPRs + Special Registers)
 def SReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
-  (add SGPR_32, M0Reg)
+  (add SGPR_32, M0Reg, VCC_LO)
 >;
 
 def SGPR_64 : RegisterClass<"AMDGPU", [v2i32, i64], 64, (add SGPR_64Regs)>;

llvm/test/CodeGen/R600/private-memory.ll

@@ -13,10 +13,10 @@
 ; R600-CHECK-NOT: ALU clause
 ; R600-CHECK: 0 + AR.x
 
-; SI-CHECK: V_READFIRSTLANE
+; SI-CHECK: V_READFIRSTLANE_B32 vcc_lo
 ; SI-CHECK: V_MOVRELD
 ; SI-CHECK: S_CBRANCH
-; SI-CHECK: V_READFIRSTLANE
+; SI-CHECK: V_READFIRSTLANE_B32 vcc_lo
 ; SI-CHECK: V_MOVRELD
 ; SI-CHECK: S_CBRANCH
 define void @mova_same_clause(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) {