Revert "[AMDGPU] Invert the handling of skip insertion."

This reverts commit 0dc6c24.

The commit is reported to cause a regression in piglit/bin/glsl-vs-loop for
Mesa.

Author: nhaehnle

llvm/lib/Target/AMDGPU/AMDGPU.h

@@ -156,9 +156,6 @@ extern char &SIWholeQuadModeID;
 void initializeSILowerControlFlowPass(PassRegistry &);
 extern char &SILowerControlFlowID;
 
-void initializeSIRemoveShortExecBranchesPass(PassRegistry &);
-extern char &SIRemoveShortExecBranchesID;
-
 void initializeSIInsertSkipsPass(PassRegistry &);
 extern char &SIInsertSkipsPassID;
 

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -228,7 +228,6 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
   initializeSIModeRegisterPass(*PR);
   initializeSIWholeQuadModePass(*PR);
   initializeSILowerControlFlowPass(*PR);
-  initializeSIRemoveShortExecBranchesPass(*PR);
   initializeSIInsertSkipsPass(*PR);
   initializeSIMemoryLegalizerPass(*PR);
   initializeSIOptimizeExecMaskingPass(*PR);
@@ -994,7 +993,6 @@ void GCNPassConfig::addPreEmitPass() {
   // be better for it to emit S_NOP <N> when possible.
   addPass(&PostRAHazardRecognizerID);
 
-  addPass(&SIRemoveShortExecBranchesID);
   addPass(&SIInsertSkipsPassID);
   addPass(&BranchRelaxationPassID);
 }

llvm/lib/Target/AMDGPU/CMakeLists.txt

@@ -116,7 +116,6 @@ add_llvm_target(AMDGPUCodeGen
   SIOptimizeExecMaskingPreRA.cpp
   SIPeepholeSDWA.cpp
   SIRegisterInfo.cpp
-  SIRemoveShortExecBranches.cpp
   SIShrinkInstructions.cpp
   SIWholeQuadMode.cpp
   GCNILPSched.cpp

llvm/lib/Target/AMDGPU/SIInsertSkips.cpp

@@ -41,7 +41,7 @@ using namespace llvm;
 #define DEBUG_TYPE "si-insert-skips"
 
 static cl::opt<unsigned> SkipThresholdFlag(
-  "amdgpu-skip-threshold-legacy",
+  "amdgpu-skip-threshold",
   cl::desc("Number of instructions before jumping over divergent control flow"),
   cl::init(12), cl::Hidden);
 
@@ -466,9 +466,6 @@ bool SIInsertSkips::runOnMachineFunction(MachineFunction &MF) {
       MachineInstr &MI = *I;
 
       switch (MI.getOpcode()) {
-      case AMDGPU::S_CBRANCH_EXECZ:
-        ExecBranchStack.push_back(MI.getOperand(0).getMBB());
-        break;
       case AMDGPU::SI_MASK_BRANCH:
         ExecBranchStack.push_back(MI.getOperand(0).getMBB());
         MadeChange |= skipMaskBranch(MI, MBB);

llvm/lib/Target/AMDGPU/SILowerControlFlow.cpp

@@ -244,9 +244,9 @@ void SILowerControlFlow::emitIf(MachineInstr &MI) {
     BuildMI(MBB, I, DL, TII->get(MovTermOpc), Exec)
     .addReg(Tmp, RegState::Kill);
 
-  // Insert the S_CBRANCH_EXECZ instruction which will be optimized later
-  // during SIRemoveShortExecBranches.
-  MachineInstr *NewBr = BuildMI(MBB, I, DL, TII->get(AMDGPU::S_CBRANCH_EXECZ))
+  // Insert a pseudo terminator to help keep the verifier happy. This will also
+  // be used later when inserting skips.
+  MachineInstr *NewBr = BuildMI(MBB, I, DL, TII->get(AMDGPU::SI_MASK_BRANCH))
                             .add(MI.getOperand(2));
 
   if (!LIS) {
@@ -323,8 +323,8 @@ void SILowerControlFlow::emitElse(MachineInstr &MI) {
     .addReg(DstReg);
 
   MachineInstr *Branch =
-      BuildMI(MBB, ElsePt, DL, TII->get(AMDGPU::S_CBRANCH_EXECZ))
-          .addMBB(DestBB);
+    BuildMI(MBB, ElsePt, DL, TII->get(AMDGPU::SI_MASK_BRANCH))
+    .addMBB(DestBB);
 
   if (!LIS) {
     MI.eraseFromParent();

llvm/lib/Target/AMDGPU/SIRemoveShortExecBranches.cpp

@@ -10,8 +10,9 @@ define i32 @divergent_if_swap_brtarget_order0(i32 %value) {
 ; CHECK-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; CHECK-NEXT:    ; implicit-def: $vgpr0
 ; CHECK-NEXT:    s_and_saveexec_b64 s[4:5], vcc
+; CHECK-NEXT:    ; mask branch BB0_2
 ; CHECK-NEXT:    s_cbranch_execz BB0_2
-; CHECK-NEXT:  ; %bb.1: ; %if.true
+; CHECK-NEXT:  BB0_1: ; %if.true
 ; CHECK-NEXT:    global_load_dword v0, v[0:1], off
 ; CHECK-NEXT:  BB0_2: ; %endif
 ; CHECK-NEXT:    s_or_b64 exec, exec, s[4:5]
@@ -37,10 +38,12 @@ define i32 @divergent_if_swap_brtarget_order1(i32 %value) {
 ; CHECK-NEXT:    v_cmp_ne_u32_e32 vcc, 0, v0
 ; CHECK-NEXT:    ; implicit-def: $vgpr0
 ; CHECK-NEXT:    s_and_saveexec_b64 s[4:5], vcc
-; CHECK-NEXT:    s_cbranch_execnz BB1_2
-; CHECK-NEXT:  ; %bb.1: ; %if.true
+; CHECK-NEXT:    ; mask branch BB1_2
+; CHECK-NEXT:  BB1_1: ; %endif
+; CHECK-NEXT:    s_or_b64 exec, exec, s[4:5]
+; CHECK-NEXT:    s_setpc_b64 s[30:31]
+; CHECK-NEXT:  BB1_2: ; %if.true
 ; CHECK-NEXT:    global_load_dword v0, v[0:1], off
-; CHECK-NEXT:  BB1_2: ; %endif
 ; CHECK-NEXT:    s_or_b64 exec, exec, s[4:5]
 ; CHECK-NEXT:    s_waitcnt vmcnt(0)
 ; CHECK-NEXT:    s_setpc_b64 s[30:31]