[LLVM][NVPTX] Add cp.async.bulk.commit/wait intrinsics (#78698)

This patch adds NVVM intrinsics and NVPTX codegen for the bulk variants
of the async-copy commit/wait instructions.
lit tests are added to verify the generated PTX.

PTX Doc link:

https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#data-movement-and-conversion-instructions-cp-async-bulk-commit-group

Signed-off-by: Durgadoss R <durgadossr@nvidia.com>

llvm/include/llvm/IR/IntrinsicsNVVM.td

@@ -1454,6 +1454,16 @@ def int_nvvm_cp_async_wait_all :
     ClangBuiltin<"__nvvm_cp_async_wait_all">,
     Intrinsic<[],[],[]>;
 
+// cp.async.bulk variants of the commit/wait group
+def int_nvvm_cp_async_bulk_commit_group :
+    Intrinsic<[],[],[]>;
+
+def int_nvvm_cp_async_bulk_wait_group :
+    Intrinsic<[],[llvm_i32_ty],[ImmArg<ArgIndex<0>>]>;
+
+def int_nvvm_cp_async_bulk_wait_group_read :
+    Intrinsic<[],[llvm_i32_ty],[ImmArg<ArgIndex<0>>]>;
+
 // mbarrier
 def int_nvvm_mbarrier_init : ClangBuiltin<"__nvvm_mbarrier_init">,
     Intrinsic<[],[llvm_ptr_ty, llvm_i32_ty],[IntrConvergent, IntrNoCallback]>;

llvm/lib/Target/NVPTX/NVPTXIntrinsics.td

@@ -411,6 +411,22 @@ def CP_ASYNC_WAIT_ALL :
   [(int_nvvm_cp_async_wait_all)]>,
   Requires<[hasPTX<70>, hasSM<80>]>;
 
+// cp.async.bulk variants of the commit/wait group
+def CP_ASYNC_BULK_COMMIT_GROUP :
+  NVPTXInst<(outs), (ins), "cp.async.bulk.commit_group;",
+  [(int_nvvm_cp_async_bulk_commit_group)]>,
+  Requires<[hasPTX<80>, hasSM<90>]>;
+
+def CP_ASYNC_BULK_WAIT_GROUP :
+  NVPTXInst<(outs), (ins i32imm:$n), "cp.async.bulk.wait_group $n;",
+  [(int_nvvm_cp_async_bulk_wait_group (i32 timm:$n))]>,
+  Requires<[hasPTX<80>, hasSM<90>]>;
+
+def CP_ASYNC_BULK_WAIT_GROUP_READ :
+  NVPTXInst<(outs), (ins i32imm:$n), "cp.async.bulk.wait_group.read $n;",
+  [(int_nvvm_cp_async_bulk_wait_group_read (i32 timm:$n))]>,
+  Requires<[hasPTX<80>, hasSM<90>]>;
+
 //-----------------------------------
 // MBarrier Functions
 //-----------------------------------

llvm/test/CodeGen/NVPTX/intrinsics-sm90.ll

@@ -144,6 +144,31 @@ define void @test_barrier_cluster_aligned() {
        ret void
 }
 
+; CHECK-LABEL: test_cp_async_bulk_commit_group(
+define void @test_cp_async_bulk_commit_group() {
+; CHECK: cp.async.bulk.commit_group;
+       call void @llvm.nvvm.cp.async.bulk.commit.group()
+       ret void
+}
+
+; CHECK-LABEL: test_cp_async_bulk_wait_group(
+define void @test_cp_async_bulk_wait_group() {
+; CHECK: cp.async.bulk.wait_group 8;
+       call void @llvm.nvvm.cp.async.bulk.wait.group(i32 8)
+; CHECK: cp.async.bulk.wait_group 0;
+       call void @llvm.nvvm.cp.async.bulk.wait.group(i32 0)
+       ret void
+}
+
+; CHECK-LABEL: test_cp_async_bulk_wait_group_read(
+define void @test_cp_async_bulk_wait_group_read() {
+; CHECK: cp.async.bulk.wait_group.read 8;
+       call void @llvm.nvvm.cp.async.bulk.wait.group.read(i32 8)
+; CHECK: cp.async.bulk.wait_group.read 0;
+       call void @llvm.nvvm.cp.async.bulk.wait.group.read(i32 0)
+       ret void
+}
+
 declare i1 @llvm.nvvm.isspacep.shared.cluster(ptr %p);
 declare ptr @llvm.nvvm.mapa(ptr %p, i32 %r);
 declare ptr addrspace(3) @llvm.nvvm.mapa.shared.cluster(ptr addrspace(3) %p, i32 %r);
@@ -167,3 +192,6 @@ declare void @llvm.nvvm.barrier.cluster.arrive.aligned()
 declare void @llvm.nvvm.barrier.cluster.arrive.relaxed.aligned()
 declare void @llvm.nvvm.barrier.cluster.wait.aligned()
 declare void @llvm.nvvm.fence.sc.cluster()
+declare void @llvm.nvvm.cp.async.bulk.commit.group()
+declare void @llvm.nvvm.cp.async.bulk.wait.group(i32)
+declare void @llvm.nvvm.cp.async.bulk.wait.group.read(i32)