AMDGPU: Select branch on undef to uniform scc branch

llvm-svn: 289877

llvm/lib/Target/AMDGPU/AMDGPUISelDAGToDAG.cpp

@@ -1458,6 +1458,12 @@ bool AMDGPUDAGToDAGISel::isCBranchSCC(const SDNode *N) const {
 void AMDGPUDAGToDAGISel::SelectBRCOND(SDNode *N) {
   SDValue Cond = N->getOperand(1);
 
+  if (Cond.isUndef()) {
+    CurDAG->SelectNodeTo(N, AMDGPU::SI_BR_UNDEF, MVT::Other,
+                         N->getOperand(2), N->getOperand(0));
+    return;
+  }
+
   if (isCBranchSCC(N)) {
     // This brcond will use S_CBRANCH_SCC*, so let tablegen handle it.
     SelectCode(N);

llvm/lib/Target/AMDGPU/SIISelLowering.cpp

@@ -1762,6 +1762,15 @@ MachineBasicBlock *SITargetLowering::EmitInstrWithCustomInserter(
     MI.eraseFromParent();
     return BB;
   }
+  case AMDGPU::SI_BR_UNDEF: {
+    const SIInstrInfo *TII = getSubtarget()->getInstrInfo();
+    const DebugLoc &DL = MI.getDebugLoc();
+    MachineInstr *Br = BuildMI(*BB, MI, DL, TII->get(AMDGPU::S_CBRANCH_SCC1))
+      .addOperand(MI.getOperand(0));
+    Br->getOperand(1).setIsUndef(true); // read undef SCC
+    MI.eraseFromParent();
+    return BB;
+  }
   default:
     return AMDGPUTargetLowering::EmitInstrWithCustomInserter(MI, BB);
   }

llvm/lib/Target/AMDGPU/SIInstructions.td

@@ -246,6 +246,12 @@ def SI_KILL_TERMINATOR : SPseudoInstSI <
 
 } // End Uses = [EXEC], Defs = [EXEC,VCC]
 
+// Branch on undef scc. Used to avoid intermediate copy from
+// IMPLICIT_DEF to SCC.
+def SI_BR_UNDEF : SPseudoInstSI <(outs), (ins sopp_brtarget:$simm16)> {
+  let isTerminator = 1;
+  let usesCustomInserter = 1;
+}
 
 def SI_PS_LIVE : PseudoInstSI <
   (outs SReg_64:$dst), (ins),

llvm/test/CodeGen/AMDGPU/cgp-bitfield-extract.ll

@@ -27,7 +27,7 @@
 
 ; GCN-LABEL: {{^}}sink_ubfe_i32:
 ; GCN-NOT: lshr
-; GCN: s_cbranch_vccnz
+; GCN: s_cbranch_scc1
 
 ; GCN: s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80008
 ; GCN: BB0_2:
@@ -121,7 +121,7 @@ ret:
 ; GCN-LABEL: {{^}}sink_ubfe_i16:
 ; GCN-NOT: lshr
 ; VI: s_bfe_u32 s0, s0, 0xc0004
-; GCN: s_cbranch_vccnz
+; GCN: s_cbranch_scc1
 
 ; SI: s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80004
 ; VI: s_and_b32 s0, s0, 0xff
@@ -175,12 +175,13 @@ ret:
 ; OPT: ret
 
 ; GCN-LABEL: {{^}}sink_ubfe_i64_span_midpoint:
-; GCN: s_lshr_b64 s{{\[}}[[LO:[0-9]+]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, 30
-; GCN: s_cbranch_vccnz BB3_2
+; GCN: s_cbranch_scc1 BB3_2
 
+; GCN: s_lshr_b64 s{{\[}}[[LO:[0-9]+]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, 30
 ; GCN: s_and_b32 s{{[0-9]+}}, s[[LO]], 0xff
 
 ; GCN: BB3_2:
+; GCN: s_lshr_b64 s{{\[}}[[LO:[0-9]+]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, 30
 ; GCN: s_and_b32 s{{[0-9]+}}, s[[LO]], 0x7f
 
 ; GCN: BB3_3:
@@ -225,7 +226,7 @@ ret:
 
 ; GCN-LABEL: {{^}}sink_ubfe_i64_low32:
 
-; GCN: s_cbranch_vccnz BB4_2
+; GCN: s_cbranch_scc1 BB4_2
 
 ; GCN: s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x8000f
 
@@ -273,7 +274,7 @@ ret:
 ; OPT: ret
 
 ; GCN-LABEL: {{^}}sink_ubfe_i64_high32:
-; GCN: s_cbranch_vccnz BB5_2
+; GCN: s_cbranch_scc1 BB5_2
 ; GCN: s_bfe_u32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80003
 
 ; GCN: BB5_2:

llvm/test/CodeGen/AMDGPU/i1-copy-implicit-def.ll

@@ -4,7 +4,7 @@
 ; SILowerI1Copies was not handling IMPLICIT_DEF
 ; SI-LABEL: {{^}}br_implicit_def:
 ; SI: BB#0:
-; SI-NEXT: s_cbranch_vccnz
+; SI-NEXT: s_cbranch_scc1
 define void @br_implicit_def(i32 addrspace(1)* %out, i32 %arg) #0 {
 bb:
   br i1 undef, label %bb1, label %bb2

llvm/test/CodeGen/AMDGPU/ret_jump.ll

@@ -5,7 +5,7 @@
 ; Mask should be in original state after executed unreachable block
 
 ; GCN-LABEL: {{^}}main:
-; GCN: s_cbranch_vccnz [[RET_BB:BB[0-9]+_[0-9]+]]
+; GCN: s_cbranch_scc1 [[RET_BB:BB[0-9]+_[0-9]+]]
 
 ; GCN: s_and_saveexec_b64 [[SAVE_EXEC:s\[[0-9]+:[0-9]+\]]], vcc
 ; GCN-NEXT: s_xor_b64 [[XOR_EXEC:s\[[0-9]+:[0-9]+\]]], exec, [[SAVE_EXEC]]

llvm/test/CodeGen/AMDGPU/si-annotate-cf-noloop.ll

@@ -37,7 +37,7 @@ bb5:                                              ; preds = %bb3, %bb1
 ; OPT-NOT: call i1 @llvm.amdgcn.loop
 
 ; GCN-LABEL: {{^}}annotate_ret_noloop:
-; GCN: s_cbranch_vccnz
+; GCN: s_cbranch_scc1
 ; GCN: s_endpgm
 ; GCN: .Lfunc_end1
 define void @annotate_ret_noloop(<4 x float> addrspace(1)* noalias nocapture readonly %arg) #0 {

llvm/test/CodeGen/AMDGPU/skip-if-dead.ll

@@ -263,7 +263,7 @@ exit:
 ; CHECK-NEXT: s_endpgm
 
 ; CHECK: [[KILLBB:BB[0-9]+_[0-9]+]]:
-; CHECK-NEXT: s_cbranch_vccz [[PHIBB:BB[0-9]+_[0-9]+]]
+; CHECK-NEXT: s_cbranch_scc0 [[PHIBB:BB[0-9]+_[0-9]+]]
 
 ; CHECK: [[PHIBB]]:
 ; CHECK: v_cmp_eq_f32_e32 vcc, 0, [[PHIREG]]

llvm/test/CodeGen/AMDGPU/unhandled-loop-condition-assertion.ll

@@ -40,7 +40,7 @@ for.end:                                          ; preds = %for.body, %entry
 
 ; COMMON-LABEL: {{^}}branch_false:
 ; SI: s_cbranch_vccnz
-; SI: s_cbranch_vccnz
+; SI: s_cbranch_scc1
 ; SI: s_endpgm
 define void @branch_false(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
 entry:
@@ -76,8 +76,8 @@ for.end:                                          ; preds = %for.body, %entry
 }
 
 ; COMMON-LABEL: {{^}}branch_undef:
-; SI: s_cbranch_vccnz
-; SI: s_cbranch_vccnz
+; SI: s_cbranch_scc1
+; SI: s_cbranch_scc1
 ; SI: s_endpgm
 define void @branch_undef(i8 addrspace(1)* nocapture %main, i32 %main_stride) #0 {
 entry: