more review feedback

Author: modiking

llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp

@@ -3050,13 +3050,14 @@ SDValue NVPTXTargetLowering::LowerADDRSPACECAST(SDValue Op,
          DestAS == llvm::ADDRESS_SPACE_SHARED_CLUSTER) ||
         (SrcAS == llvm::ADDRESS_SPACE_SHARED_CLUSTER &&
          DestAS == llvm::ADDRESS_SPACE_SHARED)) {
+      SDLoc DL(Op.getNode());
       const MVT GenerictVT =
           getPointerTy(DAG.getDataLayout(), ADDRESS_SPACE_GENERIC);
-      const MVT OutputVT = getPointerTy(DAG.getDataLayout(), DestAS);
       SDValue GenericConversion = DAG.getAddrSpaceCast(
-          SDLoc(), GenerictVT, Op.getOperand(0), SrcAS, ADDRESS_SPACE_GENERIC);
-      SDValue SharedClusterConversion = DAG.getAddrSpaceCast(
-          SDLoc(), OutputVT, GenericConversion, ADDRESS_SPACE_GENERIC, DestAS);
+          DL, GenerictVT, Op.getOperand(0), SrcAS, ADDRESS_SPACE_GENERIC);
+      SDValue SharedClusterConversion =
+          DAG.getAddrSpaceCast(DL, Op.getValueType(), GenericConversion,
+                               ADDRESS_SPACE_GENERIC, DestAS);
       return SharedClusterConversion;
     }
 

llvm/lib/Target/NVPTX/NVPTXIntrinsics.td

@@ -2416,10 +2416,8 @@ foreach space = ["local", "shared", "global", "const", "param"] in {
   defm cvta_to_#space : G_TO_NG<space>;
 }
 
-foreach space = ["shared::cluster"] in {
-  defm cvta_#space : NG_TO_G<space, false, [hasclusters]>;
-  defm cvta_to_#space : G_TO_NG<space, false, [hasclusters]>;
-}
+defm cvta_shared_cluster : NG_TO_G<"shared::cluster", false, [hasClusters]>;
+defm cvta_to_shared_cluster : G_TO_NG<"shared::cluster", false, [hasClusters]>;
 
 def : Pat<(int_nvvm_ptr_param_to_gen i32:$src),
           (cvta_param $src)>;

llvm/lib/Target/NVPTX/NVPTXTargetMachine.cpp

@@ -282,8 +282,10 @@ NVPTXTargetMachine::getPredicatedAddrSpace(const Value *V) const {
     case Intrinsic::nvvm_isspacep_local:
       return std::make_pair(II->getArgOperand(0), llvm::ADDRESS_SPACE_LOCAL);
     case Intrinsic::nvvm_isspacep_shared:
-    case Intrinsic::nvvm_isspacep_shared_cluster:
       return std::make_pair(II->getArgOperand(0), llvm::ADDRESS_SPACE_SHARED);
+    case Intrinsic::nvvm_isspacep_shared_cluster:
+      return std::make_pair(II->getArgOperand(0),
+                            llvm::ADDRESS_SPACE_SHARED_CLUSTER);
     default:
       break;
     }

llvm/lib/Target/NVPTX/NVPTXTargetTransformInfo.cpp

@@ -424,9 +424,13 @@ static std::optional<bool> evaluateIsSpace(Intrinsic::ID IID, unsigned AS) {
   case Intrinsic::nvvm_isspacep_local:
     return AS == NVPTXAS::ADDRESS_SPACE_LOCAL;
   case Intrinsic::nvvm_isspacep_shared:
+    // If shared cluster this can't be evaluated at compile time.
+    if (AS == NVPTXAS::ADDRESS_SPACE_SHARED_CLUSTER)
+      return std::nullopt;
     return AS == NVPTXAS::ADDRESS_SPACE_SHARED;
   case Intrinsic::nvvm_isspacep_shared_cluster:
-    return AS == NVPTXAS::ADDRESS_SPACE_SHARED_CLUSTER;
+    return AS == NVPTXAS::ADDRESS_SPACE_SHARED_CLUSTER ||
+           AS == NVPTXAS::ADDRESS_SPACE_SHARED;
   case Intrinsic::nvvm_isspacep_const:
     return AS == NVPTXAS::ADDRESS_SPACE_CONST;
   default:

llvm/test/Transforms/InferAddressSpaces/NVPTX/isspacep.ll

@@ -114,8 +114,8 @@ entry:
   ret i1 %val
 }
 
-define i1 @test_isspacep_cluster_shared_true(ptr addrspace(7) %addr) {
-; CHECK-LABEL: define i1 @test_isspacep_cluster_shared_true(
+define i1 @test_isspacep_shared_cluster_true(ptr addrspace(7) %addr) {
+; CHECK-LABEL: define i1 @test_isspacep_shared_cluster_true(
 ; CHECK-SAME: ptr addrspace(7) [[ADDR:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    ret i1 true
@@ -127,8 +127,8 @@ entry:
   ret i1 %val
 }
 
-define i1 @test_isspacep_cluster_shared_false(ptr addrspace(1) %addr) {
-; CHECK-LABEL: define i1 @test_isspacep_cluster_shared_false(
+define i1 @test_isspacep_shared_cluster_false(ptr addrspace(1) %addr) {
+; CHECK-LABEL: define i1 @test_isspacep_shared_cluster_false(
 ; CHECK-SAME: ptr addrspace(1) [[ADDR:%.*]]) {
 ; CHECK-NEXT:  [[ENTRY:.*:]]
 ; CHECK-NEXT:    ret i1 false
@@ -139,3 +139,34 @@ entry:
   %val = call i1 @llvm.nvvm.isspacep.shared.cluster(ptr %addr1)
   ret i1 %val
 }
+
+; isspacep_shared_cluster returns true for shared
+define i1 @test_isspacep_cluster_shared_shared(ptr addrspace(3) %addr) {
+; CHECK-LABEL: define i1 @test_isspacep_cluster_shared_shared(
+; CHECK-SAME: ptr addrspace(3) [[ADDR:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    ret i1 true
+;
+entry:
+  %addr0 = addrspacecast ptr addrspace(3) %addr to ptr
+  %addr1 = getelementptr i8, ptr %addr0, i32 10
+  %val = call i1 @llvm.nvvm.isspacep.shared.cluster(ptr %addr1)
+  ret i1 %val
+}
+
+; shared cluster cannot be evaluated to shared at compile time
+define i1 @test_isspacep_shared_shared_cluster(ptr addrspace(7) %addr) {
+; CHECK-LABEL: define i1 @test_isspacep_shared_shared_cluster(
+; CHECK-SAME: ptr addrspace(7) [[ADDR:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[ADDR2:%.*]] = getelementptr i8, ptr addrspace(7) [[ADDR]], i32 10
+; CHECK-NEXT:    [[ADDR1:%.*]] = addrspacecast ptr addrspace(7) [[ADDR2]] to ptr
+; CHECK-NEXT:    [[VAL:%.*]] = call i1 @llvm.nvvm.isspacep.shared(ptr [[ADDR1]])
+; CHECK-NEXT:    ret i1 [[VAL]]
+;
+entry:
+  %addr0 = addrspacecast ptr addrspace(7) %addr to ptr
+  %addr1 = getelementptr i8, ptr %addr0, i32 10
+  %val = call i1 @llvm.nvvm.isspacep.shared(ptr %addr1)
+  ret i1 %val
+}