xref: /freebsd/contrib/llvm-project/llvm/lib/Target/NVPTX/NVPTXSubtarget.cpp (revision 6966ac055c3b7a39266fb982493330df7a097997)
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.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "NVPTXSubtarget.h"
14 #include "NVPTXTargetMachine.h"
15 
16 using namespace llvm;
17 
18 #define DEBUG_TYPE "nvptx-subtarget"
19 
20 #define GET_SUBTARGETINFO_ENUM
21 #define GET_SUBTARGETINFO_TARGET_DESC
22 #define GET_SUBTARGETINFO_CTOR
23 #include "NVPTXGenSubtargetInfo.inc"
24 
25 static cl::opt<bool>
26     NoF16Math("nvptx-no-f16-math", cl::ZeroOrMore, cl::Hidden,
27               cl::desc("NVPTX Specific: Disable generation of f16 math ops."),
28               cl::init(false));
29 
30 // Pin the vtable to this file.
31 void NVPTXSubtarget::anchor() {}
32 
33 NVPTXSubtarget &NVPTXSubtarget::initializeSubtargetDependencies(StringRef CPU,
34                                                                 StringRef FS) {
35     // Provide the default CPU if we don't have one.
36   TargetName = CPU.empty() ? "sm_20" : CPU;
37 
38   ParseSubtargetFeatures(TargetName, FS);
39 
40   // Set default to PTX 3.2 (CUDA 5.5)
41   if (PTXVersion == 0) {
42     PTXVersion = 32;
43   }
44 
45   return *this;
46 }
47 
48 NVPTXSubtarget::NVPTXSubtarget(const Triple &TT, const std::string &CPU,
49                                const std::string &FS,
50                                const NVPTXTargetMachine &TM)
51     : NVPTXGenSubtargetInfo(TT, CPU, FS), PTXVersion(0), SmVersion(20), TM(TM),
52       InstrInfo(), TLInfo(TM, initializeSubtargetDependencies(CPU, FS)),
53       FrameLowering() {}
54 
55 bool NVPTXSubtarget::hasImageHandles() const {
56   // Enable handles for Kepler+, where CUDA supports indirect surfaces and
57   // textures
58   if (TM.getDrvInterface() == NVPTX::CUDA)
59     return (SmVersion >= 30);
60 
61   // Disabled, otherwise
62   return false;
63 }
64 
65 bool NVPTXSubtarget::allowFP16Math() const {
66   return hasFP16Math() && NoF16Math == false;
67 }
68