R600/SI: Add VI versions of MUBUF loads and stores
[oota-llvm.git] / lib / Target / R600 / AMDGPU.td
1 //===-- AMDIL.td - AMDIL Tablegen files --*- tablegen -*-------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //==-----------------------------------------------------------------------===//
9
10 include "llvm/Target/Target.td"
11
12 //===----------------------------------------------------------------------===//
13 // Subtarget Features
14 //===----------------------------------------------------------------------===//
15
16 // Debugging Features
17
18 def FeatureDumpCode : SubtargetFeature <"DumpCode",
19         "DumpCode",
20         "true",
21         "Dump MachineInstrs in the CodeEmitter">;
22
23 def FeatureIRStructurizer : SubtargetFeature <"disable-irstructurizer",
24         "EnableIRStructurizer",
25         "false",
26         "Disable IR Structurizer">;
27
28 def FeaturePromoteAlloca : SubtargetFeature <"promote-alloca",
29         "EnablePromoteAlloca",
30         "true",
31         "Enable promote alloca pass">;
32
33 // Target features
34
35 def FeatureIfCvt : SubtargetFeature <"disable-ifcvt",
36         "EnableIfCvt",
37         "false",
38         "Disable the if conversion pass">;
39
40 def FeatureFP64 : SubtargetFeature<"fp64",
41         "FP64",
42         "true",
43         "Enable double precision operations">;
44
45 def FeatureFP64Denormals : SubtargetFeature<"fp64-denormals",
46         "FP64Denormals",
47         "true",
48         "Enable double precision denormal handling",
49         [FeatureFP64]>;
50
51 // Some instructions do not support denormals despite this flag. Using
52 // fp32 denormals also causes instructions to run at the double
53 // precision rate for the device.
54 def FeatureFP32Denormals : SubtargetFeature<"fp32-denormals",
55         "FP32Denormals",
56         "true",
57         "Enable single precision denormal handling">;
58
59 def Feature64BitPtr : SubtargetFeature<"64BitPtr",
60         "Is64bit",
61         "true",
62         "Specify if 64-bit addressing should be used">;
63
64 def FeatureR600ALUInst : SubtargetFeature<"R600ALUInst",
65         "R600ALUInst",
66         "false",
67         "Older version of ALU instructions encoding">;
68
69 def FeatureVertexCache : SubtargetFeature<"HasVertexCache",
70         "HasVertexCache",
71         "true",
72         "Specify use of dedicated vertex cache">;
73
74 def FeatureCaymanISA : SubtargetFeature<"caymanISA",
75         "CaymanISA",
76         "true",
77         "Use Cayman ISA">;
78
79 def FeatureCFALUBug : SubtargetFeature<"cfalubug",
80         "CFALUBug",
81         "true",
82         "GPU has CF_ALU bug">;
83
84 // XXX - This should probably be removed once enabled by default
85 def FeatureEnableLoadStoreOpt : SubtargetFeature <"load-store-opt",
86         "EnableLoadStoreOpt",
87         "true",
88         "Enable SI load/store optimizer pass">;
89
90 def FeatureFlatAddressSpace : SubtargetFeature<"flat-address-space",
91         "FlatAddressSpace",
92         "true",
93         "Support flat address space">;
94
95 def FeatureVGPRSpilling : SubtargetFeature<"vgpr-spilling",
96         "EnableVGPRSpilling",
97         "true",
98         "Enable spilling of VGPRs to scratch memory">;
99
100 class SubtargetFeatureFetchLimit <string Value> :
101                           SubtargetFeature <"fetch"#Value,
102         "TexVTXClauseSize",
103         Value,
104         "Limit the maximum number of fetches in a clause to "#Value>;
105
106 def FeatureFetchLimit8 : SubtargetFeatureFetchLimit <"8">;
107 def FeatureFetchLimit16 : SubtargetFeatureFetchLimit <"16">;
108
109 class SubtargetFeatureWavefrontSize <int Value> : SubtargetFeature<
110         "wavefrontsize"#Value,
111         "WavefrontSize",
112         !cast<string>(Value),
113         "The number of threads per wavefront">;
114
115 def FeatureWavefrontSize16 : SubtargetFeatureWavefrontSize<16>;
116 def FeatureWavefrontSize32 : SubtargetFeatureWavefrontSize<32>;
117 def FeatureWavefrontSize64 : SubtargetFeatureWavefrontSize<64>;
118
119 class SubtargetFeatureLocalMemorySize <int Value> : SubtargetFeature<
120         "localmemorysize"#Value,
121         "LocalMemorySize",
122         !cast<string>(Value),
123         "The size of local memory in bytes">;
124
125 class SubtargetFeatureGeneration <string Value,
126                                   list<SubtargetFeature> Implies> :
127         SubtargetFeature <Value, "Gen", "AMDGPUSubtarget::"#Value,
128                           Value#" GPU generation", Implies>;
129
130 def FeatureLocalMemorySize0 : SubtargetFeatureLocalMemorySize<0>;
131 def FeatureLocalMemorySize32768 : SubtargetFeatureLocalMemorySize<32768>;
132 def FeatureLocalMemorySize65536 : SubtargetFeatureLocalMemorySize<65536>;
133
134 def FeatureR600 : SubtargetFeatureGeneration<"R600",
135         [FeatureR600ALUInst, FeatureFetchLimit8, FeatureLocalMemorySize0]>;
136
137 def FeatureR700 : SubtargetFeatureGeneration<"R700",
138         [FeatureFetchLimit16, FeatureLocalMemorySize0]>;
139
140 def FeatureEvergreen : SubtargetFeatureGeneration<"EVERGREEN",
141         [FeatureFetchLimit16, FeatureLocalMemorySize32768]>;
142
143 def FeatureNorthernIslands : SubtargetFeatureGeneration<"NORTHERN_ISLANDS",
144         [FeatureFetchLimit16, FeatureWavefrontSize64,
145          FeatureLocalMemorySize32768]
146 >;
147
148 def FeatureSouthernIslands : SubtargetFeatureGeneration<"SOUTHERN_ISLANDS",
149         [Feature64BitPtr, FeatureFP64, FeatureLocalMemorySize32768,
150          FeatureWavefrontSize64]>;
151
152 def FeatureSeaIslands : SubtargetFeatureGeneration<"SEA_ISLANDS",
153         [Feature64BitPtr, FeatureFP64, FeatureLocalMemorySize65536,
154          FeatureWavefrontSize64, FeatureFlatAddressSpace]>;
155
156 def FeatureVolcanicIslands : SubtargetFeatureGeneration<"VOLCANIC_ISLANDS",
157         [Feature64BitPtr, FeatureFP64, FeatureLocalMemorySize65536,
158          FeatureWavefrontSize64, FeatureFlatAddressSpace]>;
159
160 //===----------------------------------------------------------------------===//
161
162 def AMDGPUInstrInfo : InstrInfo {
163   let guessInstructionProperties = 1;
164 }
165
166 def AMDGPUAsmParser : AsmParser {
167   // Some of the R600 registers have the same name, so this crashes.
168   // For example T0_XYZW and T0_XY both have the asm name T0.
169   let ShouldEmitMatchRegisterName = 0;
170 }
171
172 def AMDGPU : Target {
173   // Pull in Instruction Info:
174   let InstructionSet = AMDGPUInstrInfo;
175   let AssemblyParsers = [AMDGPUAsmParser];
176 }
177
178 // Dummy Instruction itineraries for pseudo instructions
179 def ALU_NULL : FuncUnit;
180 def NullALU : InstrItinClass;
181
182 //===----------------------------------------------------------------------===//
183 // Predicate helper class
184 //===----------------------------------------------------------------------===//
185
186 class PredicateControl {
187   Predicate SubtargetPredicate;
188   list<Predicate> OtherPredicates = [];
189   list<Predicate> Predicates = !listconcat([SubtargetPredicate],
190                                             OtherPredicates);
191 }
192
193 // Include AMDGPU TD files
194 include "R600Schedule.td"
195 include "SISchedule.td"
196 include "Processors.td"
197 include "AMDGPUInstrInfo.td"
198 include "AMDGPUIntrinsics.td"
199 include "AMDGPURegisterInfo.td"
200 include "AMDGPUInstructions.td"
201 include "AMDGPUCallingConv.td"