[llvm] [NVPTX] Implement `__builtin_readcyclecounter` on NVPTX (PR #81344)

[Joseph Huber via llvm-commits]

https://github.com/jhuber6 created https://github.com/llvm/llvm-project/pull/81344

Summary:
This patch simply states that `__builtin_readcyclecounter` is legal on
NVPTX and makes it  return the value from the `clock64` sreg. The timer
intrinsics are marked as having side effects, which is desireable for
timing primitives and required to pattern match the instrinic DAG.


>From 492cfa40c60111176bcbc7b758254b3a9eafe6d0 Mon Sep 17 00:00:00 2001
From: Joseph Huber <huberjn at outlook.com>
Date: Fri, 9 Feb 2024 19:27:15 -0600
Subject: [PATCH] [NVPTX] Implement `__builtin_readcyclecounter` on NVPTX

Summary:
This patch simply states that `__builtin_readcyclecounter` is legal on
NVPTX and makes it  return the value from the `clock64` sreg. The timer
intrinsics are marked as having side effects, which is desireable for
timing primitives and required to pattern match the instrinic DAG.
---
 llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp |  2 ++
 llvm/lib/Target/NVPTX/NVPTXIntrinsics.td    |  4 ++++
 llvm/test/CodeGen/NVPTX/intrinsics.ll       | 12 ++++++++++++
 3 files changed, 18 insertions(+)

diff --git a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
index c7bc623a88e1b9..603fbe09d545b0 100644
--- a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
+++ b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
@@ -489,6 +489,8 @@ NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM,
   setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v2f16, Expand);
   setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v2f16, Expand);
 
+  setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, Legal);
+
   setFP16OperationAction(ISD::SETCC, MVT::f16, Legal, Promote);
   setFP16OperationAction(ISD::SETCC, MVT::v2f16, Legal, Expand);
 
diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
index 2330d7213c26dc..133e2827008a87 100644
--- a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -6372,12 +6372,16 @@ def INT_PTX_SREG_LANEMASK_GE :
 def INT_PTX_SREG_LANEMASK_GT :
     PTX_READ_SREG_R32<"lanemask_gt", int_nvvm_read_ptx_sreg_lanemask_gt>;
 
+let hasSideEffects = 1 in {
 def INT_PTX_SREG_CLOCK :
     PTX_READ_SREG_R32<"clock", int_nvvm_read_ptx_sreg_clock>;
 def INT_PTX_SREG_CLOCK64 :
     PTX_READ_SREG_R64<"clock64", int_nvvm_read_ptx_sreg_clock64>;
 def INT_PTX_SREG_GLOBALTIMER :
     PTX_READ_SREG_R64<"globaltimer", int_nvvm_read_ptx_sreg_globaltimer>;
+}
+
+def: Pat <(i64 (readcyclecounter)), (INT_PTX_SREG_CLOCK64)>;
 
 def INT_PTX_SREG_PM0 : PTX_READ_SREG_R32<"pm0", int_nvvm_read_ptx_sreg_pm0>;
 def INT_PTX_SREG_PM1 : PTX_READ_SREG_R32<"pm1", int_nvvm_read_ptx_sreg_pm1>;
diff --git a/llvm/test/CodeGen/NVPTX/intrinsics.ll b/llvm/test/CodeGen/NVPTX/intrinsics.ll
index 2994f600d45c71..d84ee6754281e5 100644
--- a/llvm/test/CodeGen/NVPTX/intrinsics.ll
+++ b/llvm/test/CodeGen/NVPTX/intrinsics.ll
@@ -151,6 +151,17 @@ define i64 @test_globaltimer() {
   ret i64 %ret
 }
 
+; CHECK-LABEL: test_cyclecounter
+define i64 @test_cyclecounter() {
+; CHECK: mov.u64         %r{{.*}}, %clock64;
+  %a = tail call i64 @llvm.readcyclecounter()
+; CHECK: mov.u64         %r{{.*}}, %clock64;
+  %b = tail call i64 @llvm.readcyclecounter()
+  %ret = add i64 %a, %b
+; CHECK: ret
+  ret i64 %ret
+}
+
 declare float @llvm.fabs.f32(float)
 declare double @llvm.fabs.f64(double)
 declare float @llvm.nvvm.sqrt.f(float)
@@ -166,3 +177,4 @@ declare i32 @llvm.nvvm.read.ptx.sreg.clock()
 declare i64 @llvm.nvvm.read.ptx.sreg.clock64()
 declare void @llvm.nvvm.exit()
 declare i64 @llvm.nvvm.read.ptx.sreg.globaltimer()
+declare i64 @llvm.readcyclecounter()