AMDGPU: Force skip over SMRD, VMEM and s_waitcnt instructions

Summary: This fixes a large Dawn of War 3 performance regression with RADV from Mesa 19.0 to master which was caused by creating less code in some branches.

Reviewers: arsen, nhaehnle

Reviewed By: nhaehnle

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

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D60824

llvm-svn: 358592

llvm/lib/Target/AMDGPU/SIInsertSkips.cpp

@@ -137,6 +137,10 @@ bool SIInsertSkips::shouldSkip(const MachineBasicBlock &From,
       if (TII->hasUnwantedEffectsWhenEXECEmpty(*I))
         return true;
 
+      // These instructions are potentially expensive even if EXEC = 0.
+      if (TII->isSMRD(*I) || TII->isVMEM(*I) || I->getOpcode() == AMDGPU::S_WAITCNT)
+        return true;
+
       ++NumInstr;
       if (NumInstr >= SkipThreshold)
         return true;

llvm/test/CodeGen/AMDGPU/collapse-endcf.ll

@@ -7,6 +7,7 @@
 ; GCN-NEXT: s_cbranch_execz [[ENDIF]]
 ; GCN:      s_and_b64 exec, exec, vcc
 ; GCN-NEXT: ; mask branch [[ENDIF]]
+; GCN-NEXT: s_cbranch_execz [[ENDIF]]
 ; GCN-NEXT: {{^BB[0-9_]+}}:
 ; GCN:      store_dword
 ; GCN-NEXT: {{^}}[[ENDIF]]:
@@ -46,6 +47,7 @@ bb.outer.end:                                     ; preds = %bb.outer.then, %bb.
 ; GCN-NEXT: s_cbranch_execz [[ENDIF_OUTER]]
 ; GCN:      s_and_saveexec_b64 [[SAVEEXEC_INNER:s\[[0-9:]+\]]]
 ; GCN-NEXT: ; mask branch [[ENDIF_INNER:BB[0-9_]+]]
+; GCN-NEXT: s_cbranch_execz [[ENDIF_INNER]]
 ; GCN-NEXT: {{^BB[0-9_]+}}:
 ; GCN:      store_dword
 ; GCN-NEXT: {{^}}[[ENDIF_INNER]]:
@@ -91,6 +93,7 @@ bb.outer.end:                                     ; preds = %bb.inner.then, %bb
 ; GCN:      s_and_saveexec_b64 [[SAVEEXEC_INNER:s\[[0-9:]+\]]]
 ; GCN-NEXT: s_xor_b64 [[SAVEEXEC_INNER2:s\[[0-9:]+\]]], exec, [[SAVEEXEC_INNER]]
 ; GCN-NEXT: ; mask branch [[THEN_INNER:BB[0-9_]+]]
+; GCN-NEXT: s_cbranch_execz [[THEN_INNER]]
 ; GCN-NEXT: {{^BB[0-9_]+}}:
 ; GCN:      store_dword
 ; GCN-NEXT: {{^}}[[THEN_INNER]]:
@@ -140,6 +143,7 @@ bb.outer.end:                                        ; preds = %bb, %bb.then, %b
 ; GCN:      store_dword
 ; GCN-NEXT: s_and_saveexec_b64 [[SAVEEXEC_INNER_IF_OUTER_ELSE:s\[[0-9:]+\]]]
 ; GCN-NEXT: ; mask branch [[THEN_OUTER_FLOW:BB[0-9_]+]]
+; GCN-NEXT: s_cbranch_execz [[THEN_OUTER_FLOW]]
 ; GCN-NEXT: {{^BB[0-9_]+}}:
 ; GCN:      store_dword
 ; GCN-NEXT: {{^}}[[THEN_OUTER_FLOW]]:
@@ -153,6 +157,7 @@ bb.outer.end:                                        ; preds = %bb, %bb.then, %b
 ; GCN:      store_dword
 ; GCN-NEXT: s_and_saveexec_b64 [[SAVEEXEC_INNER_IF_OUTER_THEN:s\[[0-9:]+\]]]
 ; GCN-NEXT: ; mask branch [[FLOW1:BB[0-9_]+]]
+; GCN-NEXT: s_cbranch_execz [[FLOW1]]
 ; GCN-NEXT: {{^BB[0-9_]+}}:
 ; GCN:      store_dword
 ; GCN-NEXT: [[FLOW1]]:
@@ -199,6 +204,7 @@ bb.outer.end:
 ; ALL-LABEL: {{^}}s_endpgm_unsafe_barrier:
 ; GCN:      s_and_saveexec_b64 [[SAVEEXEC:s\[[0-9:]+\]]]
 ; GCN-NEXT: ; mask branch [[ENDIF:BB[0-9_]+]]
+; GCN-NEXT: s_cbranch_execz [[ENDIF]]
 ; GCN-NEXT: {{^BB[0-9_]+}}:
 ; GCN:      store_dword
 ; GCN-NEXT: {{^}}[[ENDIF]]:

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

@@ -367,6 +367,7 @@ bb7:                                              ; preds = %bb4
 ; CHECK: v_cmp_neq_f32_e32 vcc, 0,
 ; CHECK: s_and_saveexec_b64 s{{\[[0-9]+:[0-9]+\]}}, vcc
 ; CHECK: mask branch [[END:BB[0-9]+_[0-9]+]]
+; CHECK-NEXT: s_cbranch_execz [[END]]
 ; CHECK-NOT: branch
 
 ; CHECK: BB{{[0-9]+_[0-9]+}}: ; %bb8

llvm/test/CodeGen/AMDGPU/valu-i1.ll

@@ -66,6 +66,7 @@ end:
 ; SI: v_cmp_ne_u32_e32 vcc, 0, v{{[0-9]+}}
 ; SI: s_and_saveexec_b64 [[BR_SREG:s\[[0-9]+:[0-9]+\]]], vcc
 ; SI-NEXT: ; mask branch [[EXIT:BB[0-9]+_[0-9]+]]
+; SI-NEXT: s_cbranch_execz [[EXIT]]
 
 ; SI-NEXT: BB{{[0-9]+_[0-9]+}}:
 ; SI: buffer_store_dword
@@ -92,6 +93,7 @@ exit:
 ; SI: v_cmp_ne_u32_e32 vcc, 0, v{{[0-9]+}}
 ; SI: s_and_saveexec_b64 [[BR_SREG:s\[[0-9]+:[0-9]+\]]], vcc
 ; SI-NEXT: ; mask branch [[EXIT:BB[0-9]+_[0-9]+]]
+; SI-NEXT: s_cbranch_execz [[EXIT]]
 
 ; SI-NEXT: BB{{[0-9]+_[0-9]+}}:
 ; SI: buffer_store_dword
@@ -129,6 +131,7 @@ exit:
 ; SI-NEXT: s_or_saveexec_b64
 ; SI-NEXT: s_xor_b64 exec, exec
 ; SI-NEXT: ; mask branch [[UNIFIED_RETURN:BB[0-9]+_[0-9]+]]
+; SI-NEXT: s_cbranch_execz [[UNIFIED_RETURN]]
 
 ; SI-NEXT: {{^BB[0-9]+_[0-9]+}}: ; %then
 ; SI: s_waitcnt

llvm/test/CodeGen/AMDGPU/wqm.ll

@@ -423,6 +423,7 @@ END:
 ;CHECK-NEXT: s_and_b64 [[SAVED]], exec, [[SAVED]]
 ;CHECK-NEXT: s_xor_b64 exec, exec, [[SAVED]]
 ;CHECK-NEXT: mask branch [[END_BB:BB[0-9]+_[0-9]+]]
+;CHECK-NEXT: s_cbranch_execz [[END_BB]]
 ;CHECK-NEXT: BB{{[0-9]+_[0-9]+}}: ; %ELSE
 ;CHECK: store_dword
 ;CHECK: [[END_BB]]: ; %END