[AMDGPU] Optimze old value of v_mov_b32_dpp

[rampitec]

Summary

We can eliminate old value if bound_ctrl = 1 and row_mask = bank_mask = 0xf.
This is alternative implementation working with the intrinsic in InstCombine.
Original review for past-ISel optimization: D46570.

[rampitec]

Ping.

[arsenm]

lib/Transforms/InstCombine/InstCombineCalls.cpp

3480–3482

I would prefer to use checked dyn_casts here rather than crash on invalid intrinsic uses

test/CodeGen/AMDGPU/llvm.amdgcn.update.dpp.ll

2–3 ↗

(On Diff #145754)

These should go in the test/InstCombine/AMDGPU/amdgcn-intrinsics.ll

[arsenm]

LGTM

lib/Transforms/InstCombine/InstCombineCalls.cpp

   case Intrinsic::amdgcn_kill: {
     const ConstantInt *C = dyn_cast<ConstantInt>(II->getArgOperand(0));
     if (!C || !C->getZExtValue())
       break;
 
     // amdgcn.kill(i1 1) is a no-op
     return eraseInstFromFunction(CI);
   }
+  case Intrinsic::amdgcn_update_dpp: {
+    Value *Old = II->getArgOperand(0);
+
+    auto BC = dyn_cast<ConstantInt>(II->getArgOperand(5));
+    auto RM = dyn_cast<ConstantInt>(II->getArgOperand(3));
+    auto BM = dyn_cast<ConstantInt>(II->getArgOperand(4));
+    if (!BC || !RM || !BM ||
+        BC->isZeroValue() ||
+        RM->getZExtValue() != 0xF ||
+        BM->getZExtValue() != 0xF ||
+        isa<UndefValue>(Old))
+      break;
+
+    // If bound_ctrl = 1, row mask = bank mask = 0xf we can omit old value.
+    II->setOperand(0, UndefValue::get(Old->getType()));
+    return II;
+  }
   case Intrinsic::stackrestore: {
     // If the save is right next to the restore, remove the restore.  This can
     // happen when variable allocas are DCE'd.
     if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
       if (SS->getIntrinsicID() == Intrinsic::stacksave) {
         if (&*++SS->getIterator() == II)
           return eraseInstFromFunction(CI);
       }

test/Transforms/InstCombine/AMDGPU/amdgcn-intrinsics.ll

 ; CHECK-LABEL: @kill_true() {
 ; CHECK-NEXT: ret void
 ; CHECK-NEXT: }
 define void @kill_true() {
   call void @llvm.amdgcn.kill(i1 true)
   ret void
 }
 
+; --------------------------------------------------------------------
+; llvm.amdgcn.update.dpp.i32
+; --------------------------------------------------------------------
+
+declare i32 @llvm.amdgcn.update.dpp.i32(i32, i32, i32, i32, i32, i1)
+
+; CHECK-LABEL: {{^}}define amdgpu_kernel void @update_dpp_no_combine(
+; CHECK: @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 false)
+define amdgpu_kernel void @update_dpp_no_combine(i32 addrspace(1)* %out, i32 %in1, i32 %in2) {
+  %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 1, i32 1, i32 1, i1 0)
+  store i32 %tmp0, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: {{^}}define amdgpu_kernel void @update_dpp_drop_old(
+; CHECK: @llvm.amdgcn.update.dpp.i32(i32 undef, i32 %in2, i32 3, i32 15, i32 15, i1 true)
+define amdgpu_kernel void @update_dpp_drop_old(i32 addrspace(1)* %out, i32 %in1, i32 %in2) {
+  %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 %in1, i32 %in2, i32 3, i32 15, i32 15, i1 1)
+  store i32 %tmp0, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK-LABEL: {{^}}define amdgpu_kernel void @update_dpp_undef_old(
+; CHECK: @llvm.amdgcn.update.dpp.i32(i32 undef, i32 %in1, i32 4, i32 15, i32 15, i1 true)
+define amdgpu_kernel void @update_dpp_undef_old(i32 addrspace(1)* %out, i32 %in1) {
+  %tmp0 = call i32 @llvm.amdgcn.update.dpp.i32(i32 undef, i32 %in1, i32 4, i32 15, i32 15, i1 1)
+  store i32 %tmp0, i32 addrspace(1)* %out
+  ret void
+}
 
 ; CHECK: attributes #5 = { convergent }