[NVPTX] Add builtin support for 'nanosleep' PTX instrunction (#79888)

Summary: This patch adds a builtin for the `nanosleep` PTX function. It takes either an immediate or a register and sleeps for [0, 2t] nanoseconds given t. More information at the documentation: https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#miscellaneous-instructions-nanosleep
llvm · Jan 29, 2024 · 5f12cc9 · Artem-B · Feb 9, 2024 · 5f12cc9
1 parent d492faa
commit 5f12cc9
Show file tree

Hide file tree

Showing 5 changed files with 42 additions and 0 deletions.
diff --git a/clang/include/clang/Basic/BuiltinsNVPTX.def b/clang/include/clang/Basic/BuiltinsNVPTX.def
@@ -157,6 +157,7 @@ BUILTIN(__nvvm_read_ptx_sreg_pm3, "i", "n")
 // MISC
 
 BUILTIN(__nvvm_prmt, "UiUiUiUi", "")
+TARGET_BUILTIN(__nvvm_nanosleep, "vi", "n", AND(SM_70, PTX63))
 
 // Min Max
 

diff --git a/clang/test/CodeGen/builtins-nvptx.c b/clang/test/CodeGen/builtins-nvptx.c
@@ -818,6 +818,17 @@ __device__ void nvvm_vote(int pred) {
   // CHECK: ret void
 }
 
+// CHECK-LABEL: nvvm_nanosleep
+__device__ void nvvm_nanosleep(int d) {
+#if __CUDA_ARCH__ >= 700
+  // CHECK_PTX70_SM80: call void @llvm.nvvm.nanosleep
+  __nvvm_nanosleep(d);
+
+  // CHECK_PTX70_SM80: call void @llvm.nvvm.nanosleep
+  __nvvm_nanosleep(1);
+#endif
+}
+
 // CHECK-LABEL: nvvm_mbarrier
 __device__ void nvvm_mbarrier(long long* addr, __attribute__((address_space(3))) long long* sharedAddr, int count, long long state) {
   #if __CUDA_ARCH__ >= 800

diff --git a/llvm/include/llvm/IR/IntrinsicsNVVM.td b/llvm/include/llvm/IR/IntrinsicsNVVM.td
@@ -557,6 +557,10 @@ let TargetPrefix = "nvvm" in {
       DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty],
         [IntrNoMem, IntrSpeculatable]>;
 
+  def int_nvvm_nanosleep : ClangBuiltin<"__nvvm_nanosleep">,
+      DefaultAttrsIntrinsic<[], [llvm_i32_ty],
+                            [IntrConvergent, IntrNoMem, IntrHasSideEffects]>;
+
 //
 // Min Max
 //

diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -640,6 +640,12 @@ class F_MATH_3<string OpcStr, NVPTXRegClass t_regclass,
 def INT_NVVM_PRMT : F_MATH_3<"prmt.b32 \t$dst, $src0, $src1, $src2;", Int32Regs,
   Int32Regs, Int32Regs, Int32Regs, int_nvvm_prmt>;
 
+def INT_NVVM_NANOSLEEP_I : NVPTXInst<(outs), (ins i32imm:$i), "nanosleep.u32 \t$i;",
+                             [(int_nvvm_nanosleep imm:$i)]>,
+        Requires<[hasPTX<63>, hasSM<70>]>;
+def INT_NVVM_NANOSLEEP_R : NVPTXInst<(outs), (ins Int32Regs:$i), "nanosleep.u32 \t$i;",
+                             [(int_nvvm_nanosleep Int32Regs:$i)]>,
+        Requires<[hasPTX<63>, hasSM<70>]>;
 //
 // Min Max
 //

diff --git a/llvm/test/CodeGen/NVPTX/nanosleep.ll b/llvm/test/CodeGen/NVPTX/nanosleep.ll
@@ -0,0 +1,20 @@
+; RUN: llc < %s -march=nvptx64 -O2 -mcpu=sm_70 -mattr=+ptx63 | FileCheck %s
+; RUN: %if ptxas %{ llc < %s -march=nvptx64 -mcpu=sm_70 -mattr=+ptx63 | %ptxas-verify %}
+
+declare void @llvm.nvvm.nanosleep(i32)
+
+; CHECK-LABEL: test_nanosleep_r
+define void @test_nanosleep_r(i32 noundef %d) {
+entry:
+; CHECK: nanosleep.u32   %[[REG:.+]];
+  call void @llvm.nvvm.nanosleep(i32 %d)
+  ret void
+}
+
+; CHECK-LABEL: test_nanosleep_i
+define void @test_nanosleep_i() {
+entry:
+; CHECK: nanosleep.u32   42;
+  call void @llvm.nvvm.nanosleep(i32 42)
+  ret void
+}