Run DCE after a LoopFlatten test to reduce spurious output [nfc]
[llvm-project.git] / llvm / lib / Target / NVPTX / NVPTXSubtarget.cpp
blob7fa64af196b936be37898742e4dce0f824a6eed5
1 //===- NVPTXSubtarget.cpp - NVPTX Subtarget Information -------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements the NVPTX specific subclass of TargetSubtarget.
11 //===----------------------------------------------------------------------===//
13 #include "NVPTXSubtarget.h"
14 #include "NVPTXTargetMachine.h"
16 using namespace llvm;
18 #define DEBUG_TYPE "nvptx-subtarget"
20 #define GET_SUBTARGETINFO_ENUM
21 #define GET_SUBTARGETINFO_TARGET_DESC
22 #define GET_SUBTARGETINFO_CTOR
23 #include "NVPTXGenSubtargetInfo.inc"
25 static cl::opt<bool>
26 NoF16Math("nvptx-no-f16-math", cl::Hidden,
27 cl::desc("NVPTX Specific: Disable generation of f16 math ops."),
28 cl::init(false));
29 // Pin the vtable to this file.
30 void NVPTXSubtarget::anchor() {}
32 NVPTXSubtarget &NVPTXSubtarget::initializeSubtargetDependencies(StringRef CPU,
33 StringRef FS) {
34 // Provide the default CPU if we don't have one.
35 TargetName = std::string(CPU.empty() ? "sm_30" : CPU);
37 ParseSubtargetFeatures(TargetName, /*TuneCPU*/ TargetName, FS);
39 // Set default to PTX 6.0 (CUDA 9.0)
40 if (PTXVersion == 0) {
41 PTXVersion = 60;
44 return *this;
47 NVPTXSubtarget::NVPTXSubtarget(const Triple &TT, const std::string &CPU,
48 const std::string &FS,
49 const NVPTXTargetMachine &TM)
50 : NVPTXGenSubtargetInfo(TT, CPU, /*TuneCPU*/ CPU, FS), PTXVersion(0),
51 SmVersion(20), TM(TM),
52 TLInfo(TM, initializeSubtargetDependencies(CPU, FS)) {}
54 bool NVPTXSubtarget::hasImageHandles() const {
55 // Enable handles for Kepler+, where CUDA supports indirect surfaces and
56 // textures
57 if (TM.getDrvInterface() == NVPTX::CUDA)
58 return (SmVersion >= 30);
60 // Disabled, otherwise
61 return false;
64 bool NVPTXSubtarget::allowFP16Math() const {
65 return hasFP16Math() && NoF16Math == false;