[llvm-branch-commits] [llvm-branch] r167124 - in /llvm/branches/R600/lib/Target/AMDGPU: SIISelLowering.cpp SILowerFlowControl.cpp

[Tom Stellard]

Author: tstellar
Date: Wed Oct 31 10:26:11 2012
New Revision: 167124

URL: http://llvm.org/viewvc/llvm-project?rev=167124&view=rev
Log:
SI: Alternative handling of EXEC register for control flow

This version handles the EXEC register being modified in the if/else blocks,
e.g. for pixel discard.

Patch by: Michel Dänzer

Reviewed-by: Tom Stellard <thomas.stellar at amd.com>
Reviewed-by: Christian König <christian.koenig at amd.com>

Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>

Modified:
    llvm/branches/R600/lib/Target/AMDGPU/SIISelLowering.cpp
    llvm/branches/R600/lib/Target/AMDGPU/SILowerFlowControl.cpp

Modified: llvm/branches/R600/lib/Target/AMDGPU/SIISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/R600/lib/Target/AMDGPU/SIISelLowering.cpp?rev=167124&r1=167123&r2=167124&view=diff
==============================================================================
--- llvm/branches/R600/lib/Target/AMDGPU/SIISelLowering.cpp (original)
+++ llvm/branches/R600/lib/Target/AMDGPU/SIISelLowering.cpp Wed Oct 31 10:26:11 2012
@@ -228,26 +228,6 @@
           .addReg(AMDGPU::SREG_LIT_0)
           .addOperand(MI->getOperand(0));
 
-  // If the exec mask is non-zero, skip the next two instructions
-  BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDGPU::S_CBRANCH_EXECNZ))
-          .addImm(3)
-          .addReg(AMDGPU::EXEC);
-
-  // Exec mask is zero: Export to NULL target...
-  BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDGPU::EXP))
-          .addImm(0)
-          .addImm(0x09) // V_008DFC_SQ_EXP_NULL
-          .addImm(0)
-          .addImm(1)
-          .addImm(1)
-          .addReg(AMDGPU::SREG_LIT_0)
-          .addReg(AMDGPU::SREG_LIT_0)
-          .addReg(AMDGPU::SREG_LIT_0)
-          .addReg(AMDGPU::SREG_LIT_0);
-
-  // ... and terminate wavefront
-  BuildMI(BB, I, BB.findDebugLoc(I), TII->get(AMDGPU::S_ENDPGM));
-
   MI->eraseFromParent();
 }
 

Modified: llvm/branches/R600/lib/Target/AMDGPU/SILowerFlowControl.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/branches/R600/lib/Target/AMDGPU/SILowerFlowControl.cpp?rev=167124&r1=167123&r2=167124&view=diff
==============================================================================
--- llvm/branches/R600/lib/Target/AMDGPU/SILowerFlowControl.cpp (original)
+++ llvm/branches/R600/lib/Target/AMDGPU/SILowerFlowControl.cpp Wed Oct 31 10:26:11 2012
@@ -50,6 +50,7 @@
 
 #include "AMDGPU.h"
 #include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
@@ -117,20 +118,48 @@
                   AMDGPU::EXEC)
                   .addOperand(MI.getOperand(0)) // VCC
                   .addReg(AMDGPU::EXEC);
+          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_XOR_B64),
+                  PredicateStack.back())
+                  .addReg(PredicateStack.back())
+                  .addReg(AMDGPU::EXEC);
           MI.eraseFromParent();
           break;
         case AMDGPU::ELSE:
-          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_NOT_B64),
-                  AMDGPU::EXEC)
+          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_MOV_B64),
+                  UnusedRegisters.back())
                   .addReg(AMDGPU::EXEC);
-          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_AND_B64),
+          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_MOV_B64),
                   AMDGPU::EXEC)
-                  .addReg(PredicateStack.back())
-                  .addReg(AMDGPU::EXEC);
+                  .addReg(PredicateStack.back());
+          BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_MOV_B64),
+                  PredicateStack.back())
+                  .addReg(UnusedRegisters.back());
           MI.eraseFromParent();
           break;
         case AMDGPU::ENDIF:
           popExecMask(MBB, I);
+	  if (MF.getInfo<SIMachineFunctionInfo>()->ShaderType == ShaderType::PIXEL &&
+	      PredicateStack.empty()) {
+            // If the exec mask is non-zero, skip the next two instructions
+            BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_CBRANCH_EXECNZ))
+                    .addImm(3)
+                    .addReg(AMDGPU::EXEC);
+
+            // Exec mask is zero: Export to NULL target...
+            BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::EXP))
+                    .addImm(0)
+                    .addImm(0x09) // V_008DFC_SQ_EXP_NULL
+                    .addImm(0)
+                    .addImm(1)
+                    .addImm(1)
+                    .addReg(AMDGPU::SREG_LIT_0)
+                    .addReg(AMDGPU::SREG_LIT_0)
+                    .addReg(AMDGPU::SREG_LIT_0)
+                    .addReg(AMDGPU::SREG_LIT_0);
+
+            // ... and terminate wavefront
+            BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_ENDPGM));
+	  }
           MI.eraseFromParent();
           break;
       }
@@ -156,7 +185,8 @@
   unsigned StackReg = PredicateStack.back();
   PredicateStack.pop_back();
   UnusedRegisters.push_back(StackReg);
-  BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_MOV_B64),
+  BuildMI(MBB, I, MBB.findDebugLoc(I), TII->get(AMDGPU::S_OR_B64),
           AMDGPU::EXEC)
+          .addReg(AMDGPU::EXEC)
           .addReg(StackReg);
 }