[NVPTX] Add cta_group support to TMA G2S intrinsics

llvm/docs/NVPTXUsage.rst

@@ -1016,7 +1016,7 @@ Syntax:
 
 .. code-block:: llvm
 
-  declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.tile.1d(ptr addrspace(7) %dst, ptr addrspace(3) %bar, ptr %tensor_map, i32 %d0, i16 %mc, i64 %ch, i1 %flag_mc, i1 %flag_ch)
+  declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.tile.1d(ptr addrspace(7) %dst, ptr addrspace(3) %bar, ptr %tensor_map, i32 %d0, i16 %mc, i64 %ch, i1 %flag_mc, i1 %flag_ch, i32 %flag_cta_group)
   declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.tile.2d(..., i32 %d0, i32 %d1, ...)
   declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.tile.3d(..., i32 %d0, i32 %d1, i32 %d2, ...)
   declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.tile.4d(..., i32 %d0, i32 %d1, i32 %d2, i32 %d3, ...)
@@ -1034,18 +1034,26 @@ source tensor is preserved at the destination. The dimension of the
 tensor data ranges from 1d to 5d with the coordinates specified
 by the ``i32 %d0 ... i32 %d4`` arguments.
 
-* The last two arguments to these intrinsics are boolean flags
-  indicating support for cache_hint and/or multicast modifiers.
-  These flag arguments must be compile-time constants. The backend
-  looks through these flags and lowers the intrinsics appropriately.
+* The last three arguments to these intrinsics are flags
+  indicating support for multicast, cache_hint and cta_group::1/2
+  modifiers. These flag arguments must be compile-time constants.
+  The backend looks through these flags and lowers the intrinsics
+  appropriately.
 
-* The Nth argument (denoted by ``i1 flag_ch``) when set, indicates
+* The argument denoted by ``i1 %flag_ch`` when set, indicates
   a valid cache_hint (``i64 %ch``) and generates the ``.L2::cache_hint``
   variant of the PTX instruction.
 
-* The [N-1]th argument (denoted by ``i1 flag_mc``) when set, indicates
-  the presence of a multicast mask (``i16 %mc``) and generates the PTX
-  instruction with the ``.multicast::cluster`` modifier.
+* The argument denoted by ``i1 %flag_mc`` when set, indicates
+  the presence of a multicast mask (``i16 %mc``) and generates
+  the PTX instruction with the ``.multicast::cluster`` modifier.
+
+* The argument denoted by ``i32 %flag_cta_group`` takes values within
+  the range [0, 3) i.e. {0,1,2}. When the value of ``%flag_cta_group``
+  is not within the range, it may raise an error from the Verifier.
+  The default value is '0' with no cta_group modifier in the
+  instruction. The values of '1' and '2' lower to ``cta_group::1``
+  and ``cta_group::2`` variants of the PTX instruction respectively.
 
 For more information, refer PTX ISA
 `<https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#data-movement-and-conversion-instructions-cp-async-bulk-tensor>`_.
@@ -1058,7 +1066,7 @@ Syntax:
 
 .. code-block:: llvm
 
-  declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.im2col.3d(ptr addrspace(3) %dst, ptr addrspace(3) %bar, ptr %tensor_map, i32 %d0, i32 %d1, i32 %d2, i16 %im2col0, i16 %mc, i64 %ch, i1 %flag_mc, i1 %flag_ch)
+  declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.im2col.3d(ptr addrspace(3) %dst, ptr addrspace(3) %bar, ptr %tensor_map, i32 %d0, i32 %d1, i32 %d2, i16 %im2col0, i16 %mc, i64 %ch, i1 %flag_mc, i1 %flag_ch, i32 %flag_cta_group)
   declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.im2col.4d(..., i32 %d0, i32 %d1, i32 %d2, i32 %d3, i16 %im2col0, i16 %im2col1, ...)
   declare void @llvm.nvvm.cp.async.bulk.tensor.g2s.im2col.5d(..., i32 %d0, i32 %d1, i32 %d2, i32 %d3, i32 %d4, i16 %im2col0, i16 %im2col1, i16 %im2col2, ...)
 
@@ -1074,8 +1082,8 @@ are unrolled into a single dimensional column at the destination. In this
 mode, the tensor has to be at least three-dimensional. Along with the tensor
 coordinates, im2col offsets are also specified (denoted by
 ``i16 im2col0...i16 %im2col2``). The number of im2col offsets is two less
-than the number of dimensions of the tensor operation. The last two arguments
-to these intrinsics are boolean flags, with the same functionality as described
+than the number of dimensions of the tensor operation. The last three arguments
+to these intrinsics are flags, with the same functionality as described
 in the ``tile`` mode intrinsics above.
 
 For more information, refer PTX ISA

llvm/include/llvm/IR/IntrinsicsNVVM.td

@@ -2020,20 +2020,26 @@ foreach dim = 1...5 in {
     defvar num_im2col_offsets = !if(is_im2col, !add(dim, -2), 0);
     defvar im2col_offsets_args = !listsplat(llvm_i16_ty, num_im2col_offsets);
 
+    defvar g2s_params = !listconcat(
+                          [llvm_shared_cluster_ptr_ty, // dst_ptr
+                           llvm_shared_ptr_ty,  // mbarrier_ptr
+                           llvm_ptr_ty],        // tensormap_ptr
+                          tensor_dim_args,      // actual tensor dims
+                          im2col_offsets_args,  // im2col offsets
+                          [llvm_i16_ty,         // cta_mask
+                           llvm_i64_ty]);       // cache_hint
+    defvar g2s_flags = [llvm_i1_ty,             // Flag for cta_mask
+                        llvm_i1_ty,             // Flag for cache_hint
+                        llvm_i32_ty];           // Flag for cta_group
+    defvar cta_group_idx = !add(
+                             !size(g2s_params),
+                             !sub(!size(g2s_flags), 1));
+    defvar g2s_props = [IntrConvergent,
+                        WriteOnly<ArgIndex<0>>, ReadOnly<ArgIndex<2>>,
+                        // Allowed values for cta_group are {0,1,2} i.e [0, 3).
+                        Range<ArgIndex<cta_group_idx>, 0, 3>];
     def int_nvvm_cp_async_bulk_tensor_g2s_ # mode # _ # dim # d :
-      DefaultAttrsIntrinsicFlags<[],
-          !listconcat([llvm_shared_cluster_ptr_ty,  // dst_shared_cluster_ptr
-                       llvm_shared_ptr_ty,          // mbarrier_smem_ptr
-                       llvm_ptr_ty],                // tensormap_ptr
-                      tensor_dim_args,              // actual tensor dims
-                      im2col_offsets_args,          // im2col offsets
-                      [llvm_i16_ty,                 // cta_mask
-                       llvm_i64_ty]),               // cache_hint
-          [llvm_i1_ty,                              // Flag for cta_mask
-           llvm_i1_ty],                             // Flag for cache_hint
-          [IntrConvergent,
-           WriteOnly<ArgIndex<0>>, ReadOnly<ArgIndex<2>>,
-           NoCapture<ArgIndex<0>>, NoCapture<ArgIndex<1>>, NoCapture<ArgIndex<2>>]>;
+      DefaultAttrsIntrinsicFlags<[], g2s_params, g2s_flags, g2s_props>;
 
     def int_nvvm_cp_async_bulk_tensor_s2g_ # mode # _ # dim # d :
       DefaultAttrsIntrinsicFlags<[],

llvm/include/llvm/IR/NVVMIntrinsicUtils.h

@@ -38,6 +38,15 @@ enum class TMAReductionOp : uint8_t {
   XOR = 7,
 };
 
+// Enum to represent the cta_group::1 and
+// cta_group::2 variants in TMA/TCGEN05 family of
+// PTX instructions.
+enum class CTAGroupKind : uint8_t {
+  CG_NONE = 0, // default with no cta_group modifier
+  CG_1 = 1,    // cta_group::1 modifier
+  CG_2 = 2,    // cta_group::2 modifier
+};
+
 inline bool FPToIntegerIntrinsicShouldFTZ(Intrinsic::ID IntrinsicID) {
   switch (IntrinsicID) {
   case Intrinsic::nvvm_f2i_rm_ftz:

llvm/lib/IR/AutoUpgrade.cpp

@@ -945,6 +945,53 @@ static bool upgradeArmOrAarch64IntrinsicFunction(bool IsArm, Function *F,
   return false; // No other 'arm.*', 'aarch64.*'.
 }
 
+static Intrinsic::ID shouldUpgradeNVPTXTMAG2SIntrinsics(Function *F,
+                                                        StringRef Name) {
+  if (Name.consume_front("cp.async.bulk.tensor.g2s.")) {
+    Intrinsic::ID ID =
+        StringSwitch<Intrinsic::ID>(Name)
+            .Case("im2col.3d",
+                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_3d)
+            .Case("im2col.4d",
+                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_4d)
+            .Case("im2col.5d",
+                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_5d)
+            .Case("tile.1d", Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_1d)
+            .Case("tile.2d", Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_2d)
+            .Case("tile.3d", Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_3d)
+            .Case("tile.4d", Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_4d)
+            .Case("tile.5d", Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_5d)
+            .Default(Intrinsic::not_intrinsic);
+
+    if (ID == Intrinsic::not_intrinsic)
+      return ID;
+
+    // These intrinsics may need upgrade for two reasons:
+    // (1) When the address-space of the first argument is shared[AS=3]
+    //     (and we upgrade it to use shared_cluster address-space[AS=7])
+    if (F->getArg(0)->getType()->getPointerAddressSpace() ==
+        NVPTXAS::ADDRESS_SPACE_SHARED)
+      return ID;
+
+    // (2) When there are only two boolean flag arguments at the end:
+    //
+    // The last three parameters of the older version of these
+    // intrinsics are: arg1, arg2, .. i64 ch, i1 mc_flag, i1 ch_flag
+    //
+    // The newer version reads as:
+    // arg1, arg2, .. i64 ch, i1 mc_flag, i1 ch_flag, i32 cta_group_flag
+    //
+    // So, when the type of the [N-3]rd argument is "not i1", then
+    // it is the older version and we need to upgrade.
+    size_t FlagStartIndex = F->getFunctionType()->getNumParams() - 3;
+    Type *ArgType = F->getFunctionType()->getParamType(FlagStartIndex);
+    if (!ArgType->isIntegerTy(1))
+      return ID;
+  }
+
+  return Intrinsic::not_intrinsic;
+}
+
 static Intrinsic::ID shouldUpgradeNVPTXSharedClusterIntrinsic(Function *F,
                                                               StringRef Name) {
   if (Name.consume_front("mapa.shared.cluster"))
@@ -959,22 +1006,6 @@ static Intrinsic::ID shouldUpgradeNVPTXSharedClusterIntrinsic(Function *F,
                   Intrinsic::nvvm_cp_async_bulk_global_to_shared_cluster)
             .Case("shared.cta.to.cluster",
                   Intrinsic::nvvm_cp_async_bulk_shared_cta_to_cluster)
-            .Case("tensor.g2s.im2col.3d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_3d)
-            .Case("tensor.g2s.im2col.4d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_4d)
-            .Case("tensor.g2s.im2col.5d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_5d)
-            .Case("tensor.g2s.tile.1d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_1d)
-            .Case("tensor.g2s.tile.2d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_2d)
-            .Case("tensor.g2s.tile.3d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_3d)
-            .Case("tensor.g2s.tile.4d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_4d)
-            .Case("tensor.g2s.tile.5d",
-                  Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_5d)
             .Default(Intrinsic::not_intrinsic);
 
     if (ID != Intrinsic::not_intrinsic)
@@ -1339,6 +1370,14 @@ static bool upgradeIntrinsicFunction1(Function *F, Function *&NewFn,
         return true;
       }
 
+      // Upgrade TMA copy G2S Intrinsics
+      IID = shouldUpgradeNVPTXTMAG2SIntrinsics(F, Name);
+      if (IID != Intrinsic::not_intrinsic) {
+        rename(F);
+        NewFn = Intrinsic::getOrInsertDeclaration(F->getParent(), IID);
+        return true;
+      }
+
       // The following nvvm intrinsics correspond exactly to an LLVM idiom, but
       // not to an intrinsic alone.  We expand them in UpgradeIntrinsicCall.
       //
@@ -4831,7 +4870,18 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
     return;
   }
   case Intrinsic::nvvm_cp_async_bulk_global_to_shared_cluster:
-  case Intrinsic::nvvm_cp_async_bulk_shared_cta_to_cluster:
+  case Intrinsic::nvvm_cp_async_bulk_shared_cta_to_cluster: {
+    // Create a new call with the correct address space.
+    SmallVector<Value *, 4> Args(CI->args());
+    Args[0] = Builder.CreateAddrSpaceCast(
+        Args[0], Builder.getPtrTy(NVPTXAS::ADDRESS_SPACE_SHARED_CLUSTER));
+
+    NewCall = Builder.CreateCall(NewFn, Args);
+    NewCall->takeName(CI);
+    CI->replaceAllUsesWith(NewCall);
+    CI->eraseFromParent();
+    return;
+  }
   case Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_3d:
   case Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_4d:
   case Intrinsic::nvvm_cp_async_bulk_tensor_g2s_im2col_5d:
@@ -4840,10 +4890,22 @@ void llvm::UpgradeIntrinsicCall(CallBase *CI, Function *NewFn) {
   case Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_3d:
   case Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_4d:
   case Intrinsic::nvvm_cp_async_bulk_tensor_g2s_tile_5d: {
-    // Create a new call with the correct address space.
-    SmallVector<Value *, 4> Args(CI->args());
-    Args[0] = Builder.CreateAddrSpaceCast(
-        Args[0], Builder.getPtrTy(NVPTXAS::ADDRESS_SPACE_SHARED_CLUSTER));
+    SmallVector<Value *, 16> Args(CI->args());
+
+    // Create AddrSpaceCast to shared_cluster if needed.
+    // This handles case (1) in shouldUpgradeNVPTXTMAG2SIntrinsics().
+    unsigned AS = CI->getArgOperand(0)->getType()->getPointerAddressSpace();
+    if (AS == NVPTXAS::ADDRESS_SPACE_SHARED)
+      Args[0] = Builder.CreateAddrSpaceCast(
+          Args[0], Builder.getPtrTy(NVPTXAS::ADDRESS_SPACE_SHARED_CLUSTER));
+
+    // Attach the flag argument for cta_group, with a
+    // default value of 0. This handles case (2) in
+    // shouldUpgradeNVPTXTMAG2SIntrinsics().
+    size_t NumArgs = CI->arg_size();
+    Value *FlagArg = CI->getArgOperand(NumArgs - 3);
+    if (!FlagArg->getType()->isIntegerTy(1))
+      Args.push_back(ConstantInt::get(Builder.getInt32Ty(), 0));
 
     NewCall = Builder.CreateCall(NewFn, Args);
     NewCall->takeName(CI);

llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXInstPrinter.cpp

@@ -437,3 +437,22 @@ void NVPTXInstPrinter::printTmaReductionMode(const MCInst *MI, int OpNum,
   llvm_unreachable(
       "Invalid Reduction Op in printCpAsyncBulkTensorReductionMode");
 }
+
+void NVPTXInstPrinter::printCTAGroup(const MCInst *MI, int OpNum,
+                                     raw_ostream &O) {
+  const MCOperand &MO = MI->getOperand(OpNum);
+  using CGTy = nvvm::CTAGroupKind;
+
+  switch (static_cast<CGTy>(MO.getImm())) {
+  case CGTy::CG_NONE:
+    O << "";
+    return;
+  case CGTy::CG_1:
+    O << ".cta_group::1";
+    return;
+  case CGTy::CG_2:
+    O << ".cta_group::2";
+    return;
+  }
+  llvm_unreachable("Invalid cta_group in printCTAGroup");
+}

llvm/lib/Target/NVPTX/MCTargetDesc/NVPTXInstPrinter.h

@@ -51,6 +51,7 @@ class NVPTXInstPrinter : public MCInstPrinter {
   void printProtoIdent(const MCInst *MI, int OpNum, raw_ostream &O);
   void printPrmtMode(const MCInst *MI, int OpNum, raw_ostream &O);
   void printTmaReductionMode(const MCInst *MI, int OpNum, raw_ostream &O);
+  void printCTAGroup(const MCInst *MI, int OpNum, raw_ostream &O);
 };
 
 }

llvm/lib/Target/NVPTX/NVPTXISelDAGToDAG.cpp

@@ -2556,19 +2556,25 @@ void NVPTXDAGToDAGISel::SelectCpAsyncBulkTensorG2SCommon(SDNode *N,
   // We have {Chain, Intrinsic-ID} followed by the actual intrisic args:
   // {dst, mbar, src, dims{d0...dN}, im2col_offsets{dims-2}
   // multicast, cache_hint,
-  // multicast_flag, cache_hint_flag}
+  // multicast_flag, cache_hint_flag, cta_group_flag}
   // NumOperands = {Chain, IID} + {Actual intrinsic args}
-  //             = {2}          + {7 + dims + im2col_offsets}
+  //             = {2}          + {8 + dims + im2col_offsets}
   size_t NumOps = N->getNumOperands();
   size_t NumDims = IsIm2Col ? GetDimsFromIntrinsic(N->getConstantOperandVal(1))
-                            : (NumOps - 9);
+                            : (NumOps - 10);
   // Offsets is always 'NumDims - 2' and only for im2col mode
   size_t NumOffsets = IsIm2Col ? (NumDims - 2) : 0;
-  bool IsCacheHint = N->getConstantOperandVal(NumOps - 1) == 1;
-  bool IsMultiCast = N->getConstantOperandVal(NumOps - 2) == 1;
+  bool IsCacheHint = N->getConstantOperandVal(NumOps - 2) == 1;
+  bool IsMultiCast = N->getConstantOperandVal(NumOps - 3) == 1;
   size_t NumBaseArgs = NumDims + NumOffsets + 3; // for {dst, mbar, src}
   size_t MultiCastIdx = NumBaseArgs + 2;         // for Chain and IID
 
+  unsigned CTAGroupVal = N->getConstantOperandVal(NumOps - 1);
+  if ((CTAGroupVal > 0) && !Subtarget->hasCpAsyncBulkTensorCTAGroupSupport())
+    report_fatal_error(
+        formatv("CpAsyncBulkTensorG2S cta_group::1/2 is not supported on sm_{}",
+                Subtarget->getSmVersion()));
+
   SDLoc DL(N);
   SmallVector<SDValue, 8> Ops(N->ops().slice(2, NumBaseArgs));
 
@@ -2580,6 +2586,9 @@ void NVPTXDAGToDAGISel::SelectCpAsyncBulkTensorG2SCommon(SDNode *N,
   if (IsCacheHint)
     Ops.push_back(N->getOperand(MultiCastIdx + 1));
 
+  // Flag for CTA Group
+  Ops.push_back(getI32Imm(CTAGroupVal, DL));
+
   // Finally, the chain operand
   Ops.push_back(N->getOperand(0));
 

llvm/lib/Target/NVPTX/NVPTXIntrinsics.td

@@ -578,10 +578,14 @@ class G2S_STRINGS<int dim, string mode, bit mc, bit ch, bit is_shared32 = 0> {
                      # !if(!eq(mode, "tile"), "_TILE", "_IM2COL");
 }
 
+def CTAGroupFlags : Operand<i32> {
+  let PrintMethod = "printCTAGroup";
+}
+
 multiclass CP_ASYNC_BULK_TENSOR_G2S_INTR<int dim, bit is_shared32, string mode> {
   defvar dims_dag = !dag(ins, !listsplat(Int32Regs, dim), !foreach(i, !range(dim), "d" # i));
   defvar dims_str = !interleave(!foreach(i, !range(dim), "$d" # i), ", ");
-  defvar asm_str_default = " [$dst], [$tmap, {{" # dims_str # "}}], [$mbar]";
+  defvar asm_str_default = "$cg [$dst], [$tmap, {{" # dims_str # "}}], [$mbar]";
   defvar rc = !if(is_shared32, Int32Regs, Int64Regs);
 
   defvar num_im2col = !if(!ge(dim, 3), !add(dim, -2), 0);
@@ -595,19 +599,22 @@ multiclass CP_ASYNC_BULK_TENSOR_G2S_INTR<int dim, bit is_shared32, string mode>
     !strconcat(asm_str_default, im2col_asm_str), asm_str_default);
 
   def "" : NVPTXInst<(outs),
-            !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag),
+            !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag, (ins CTAGroupFlags:$cg)),
             !strconcat(G2S_STRINGS<dim, mode, 0, 0>.inst_name, asm_str, ";"), []>,
             Requires<[hasPTX<80>, hasSM<90>]>;
   def _MC : NVPTXInst<(outs),
-                  !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag, (ins Int16Regs:$mc)),
+                  !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag,
+                       (ins Int16Regs:$mc, CTAGroupFlags:$cg)),
                   !strconcat(G2S_STRINGS<dim, mode, 1, 0>.inst_name, asm_str, ", $mc;"), []>,
                   Requires<[hasPTX<80>, hasSM<90>]>;
   def _CH : NVPTXInst<(outs),
-                  !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag, (ins Int64Regs:$ch)),
+                  !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag,
+                       (ins Int64Regs:$ch, CTAGroupFlags:$cg)),
                   !strconcat(G2S_STRINGS<dim, mode, 0, 1>.inst_name, asm_str, ", $ch;"), []>,
                   Requires<[hasPTX<80>, hasSM<90>]>;
   def _MC_CH : NVPTXInst<(outs),
-                     !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag, (ins Int16Regs:$mc, Int64Regs:$ch)),
+                     !con((ins rc:$dst, rc:$mbar, Int64Regs:$tmap), dims_dag, im2col_dag,
+                          (ins Int16Regs:$mc, Int64Regs:$ch, CTAGroupFlags:$cg)),
                      !strconcat(G2S_STRINGS<dim, mode, 1, 1>.inst_name, asm_str, ", $mc, $ch;"), []>,
                      Requires<[hasPTX<80>, hasSM<90>]>;
 }

llvm/lib/Target/NVPTX/NVPTXSubtarget.h

@@ -117,6 +117,14 @@ class NVPTXSubtarget : public NVPTXGenSubtargetInfo {
     return HasTcgen05 && PTXVersion >= 86;
   }
 
+  // TMA G2S copy with cta_group::1/2 support
+  bool hasCpAsyncBulkTensorCTAGroupSupport() const {
+    // TODO: Update/tidy-up after the family-conditional support arrives
+    return ((FullSmVersion == 1001 || FullSmVersion == 1011) &&
+            PTXVersion >= 86) ||
+           (FullSmVersion == 1031 && PTXVersion >= 88);
+  }
+
   // Prior to CUDA 12.3 ptxas did not recognize that the trap instruction
   // terminates a basic block. Instead, it would assume that control flow
   // continued to the next instruction. The next instruction could be in the