1 //===-- SIRegisterInfo.td - SI Register defs ---------------*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 //===----------------------------------------------------------------------===//
11 // Declarations that describe the SI registers
12 //===----------------------------------------------------------------------===//
14 class SIReg <string n, bits<16> encoding = 0> : Register<n> {
15 let Namespace = "AMDGPU";
16 let HWEncoding = encoding;
20 def VCC : SIReg<"VCC", 106>;
21 def EXEC : SIReg<"EXEC", 126>;
22 def SCC : SIReg<"SCC", 253>;
23 def M0 : SIReg <"M0", 124>;
26 foreach Index = 0-101 in {
27 def SGPR#Index : SIReg <"SGPR"#Index, Index>;
31 foreach Index = 0-255 in {
32 def VGPR#Index : SIReg <"VGPR"#Index, Index> {
33 let HWEncoding{8} = 1;
37 // virtual Interpolation registers
38 def PERSP_SAMPLE_I : SIReg <"PERSP_SAMPLE_I">;
39 def PERSP_SAMPLE_J : SIReg <"PERSP_SAMPLE_J">;
40 def PERSP_CENTER_I : SIReg <"PERSP_CENTER_I">;
41 def PERSP_CENTER_J : SIReg <"PERSP_CENTER_J">;
42 def PERSP_CENTROID_I : SIReg <"PERSP_CENTROID_I">;
43 def PERSP_CENTROID_J : SIReg <"PERP_CENTROID_J">;
44 def PERSP_I_W : SIReg <"PERSP_I_W">;
45 def PERSP_J_W : SIReg <"PERSP_J_W">;
46 def PERSP_1_W : SIReg <"PERSP_1_W">;
47 def LINEAR_SAMPLE_I : SIReg <"LINEAR_SAMPLE_I">;
48 def LINEAR_SAMPLE_J : SIReg <"LINEAR_SAMPLE_J">;
49 def LINEAR_CENTER_I : SIReg <"LINEAR_CENTER_I">;
50 def LINEAR_CENTER_J : SIReg <"LINEAR_CENTER_J">;
51 def LINEAR_CENTROID_I : SIReg <"LINEAR_CENTROID_I">;
52 def LINEAR_CENTROID_J : SIReg <"LINEAR_CENTROID_J">;
53 def LINE_STIPPLE_TEX_COORD : SIReg <"LINE_STIPPLE_TEX_COORD">;
54 def POS_X_FLOAT : SIReg <"POS_X_FLOAT">;
55 def POS_Y_FLOAT : SIReg <"POS_Y_FLOAT">;
56 def POS_Z_FLOAT : SIReg <"POS_Z_FLOAT">;
57 def POS_W_FLOAT : SIReg <"POS_W_FLOAT">;
58 def FRONT_FACE : SIReg <"FRONT_FACE">;
59 def ANCILLARY : SIReg <"ANCILLARY">;
60 def SAMPLE_COVERAGE : SIReg <"SAMPLE_COVERAGE">;
61 def POS_FIXED_PT : SIReg <"POS_FIXED_PT">;
63 //===----------------------------------------------------------------------===//
64 // Groupings using register classes and tuples
65 //===----------------------------------------------------------------------===//
67 // SGPR 32-bit registers
68 def SGPR_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
69 (add (sequence "SGPR%u", 0, 101))>;
71 // SGPR 64-bit registers
72 def SGPR_64 : RegisterTuples<[sub0, sub1],
73 [(add (decimate (trunc SGPR_32, 101), 2)),
74 (add (decimate (shl SGPR_32, 1), 2))]>;
76 // SGPR 128-bit registers
77 def SGPR_128 : RegisterTuples<[sub0, sub1, sub2, sub3],
78 [(add (decimate (trunc SGPR_32, 99), 4)),
79 (add (decimate (shl SGPR_32, 1), 4)),
80 (add (decimate (shl SGPR_32, 2), 4)),
81 (add (decimate (shl SGPR_32, 3), 4))]>;
83 // SGPR 256-bit registers
84 def SGPR_256 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7],
85 [(add (decimate (trunc SGPR_32, 95), 4)),
86 (add (decimate (shl SGPR_32, 1), 4)),
87 (add (decimate (shl SGPR_32, 2), 4)),
88 (add (decimate (shl SGPR_32, 3), 4)),
89 (add (decimate (shl SGPR_32, 4), 4)),
90 (add (decimate (shl SGPR_32, 5), 4)),
91 (add (decimate (shl SGPR_32, 6), 4)),
92 (add (decimate (shl SGPR_32, 7), 4))]>;
94 // SGPR 512-bit registers
95 def SGPR_512 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7,
96 sub8, sub9, sub10, sub11, sub12, sub13, sub14, sub15],
97 [(add (decimate (trunc SGPR_32, 87), 4)),
98 (add (decimate (shl SGPR_32, 1), 4)),
99 (add (decimate (shl SGPR_32, 2), 4)),
100 (add (decimate (shl SGPR_32, 3), 4)),
101 (add (decimate (shl SGPR_32, 4), 4)),
102 (add (decimate (shl SGPR_32, 5), 4)),
103 (add (decimate (shl SGPR_32, 6), 4)),
104 (add (decimate (shl SGPR_32, 7), 4)),
105 (add (decimate (shl SGPR_32, 8), 4)),
106 (add (decimate (shl SGPR_32, 9), 4)),
107 (add (decimate (shl SGPR_32, 10), 4)),
108 (add (decimate (shl SGPR_32, 11), 4)),
109 (add (decimate (shl SGPR_32, 12), 4)),
110 (add (decimate (shl SGPR_32, 13), 4)),
111 (add (decimate (shl SGPR_32, 14), 4)),
112 (add (decimate (shl SGPR_32, 15), 4))]>;
114 // VGPR 32-bit registers
115 def VGPR_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
116 (add (sequence "VGPR%u", 0, 255))>;
118 // VGPR 64-bit registers
119 def VGPR_64 : RegisterTuples<[sub0, sub1],
120 [(add (trunc VGPR_32, 255)),
121 (add (shl VGPR_32, 1))]>;
123 // VGPR 128-bit registers
124 def VGPR_128 : RegisterTuples<[sub0, sub1, sub2, sub3],
125 [(add (trunc VGPR_32, 253)),
126 (add (shl VGPR_32, 1)),
127 (add (shl VGPR_32, 2)),
128 (add (shl VGPR_32, 3))]>;
130 // VGPR 256-bit registers
131 def VGPR_256 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7],
132 [(add (trunc VGPR_32, 249)),
133 (add (shl VGPR_32, 1)),
134 (add (shl VGPR_32, 2)),
135 (add (shl VGPR_32, 3)),
136 (add (shl VGPR_32, 4)),
137 (add (shl VGPR_32, 5)),
138 (add (shl VGPR_32, 6)),
139 (add (shl VGPR_32, 7))]>;
141 // VGPR 512-bit registers
142 def VGPR_512 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7,
143 sub8, sub9, sub10, sub11, sub12, sub13, sub14, sub15],
144 [(add (trunc VGPR_32, 241)),
145 (add (shl VGPR_32, 1)),
146 (add (shl VGPR_32, 2)),
147 (add (shl VGPR_32, 3)),
148 (add (shl VGPR_32, 4)),
149 (add (shl VGPR_32, 5)),
150 (add (shl VGPR_32, 6)),
151 (add (shl VGPR_32, 7)),
152 (add (shl VGPR_32, 8)),
153 (add (shl VGPR_32, 9)),
154 (add (shl VGPR_32, 10)),
155 (add (shl VGPR_32, 11)),
156 (add (shl VGPR_32, 12)),
157 (add (shl VGPR_32, 13)),
158 (add (shl VGPR_32, 14)),
159 (add (shl VGPR_32, 15))]>;
161 //===----------------------------------------------------------------------===//
162 // Register classes used as source and destination
163 //===----------------------------------------------------------------------===//
165 // Special register classes for predicates and the M0 register
166 def SCCReg : RegisterClass<"AMDGPU", [i32, i1], 32, (add SCC)>;
167 def VCCReg : RegisterClass<"AMDGPU", [i64, i1], 64, (add VCC)>;
168 def EXECReg : RegisterClass<"AMDGPU", [i64, i1], 64, (add EXEC)>;
169 def M0Reg : RegisterClass<"AMDGPU", [i32], 32, (add M0)>;
171 // Register class for all scalar registers (SGPRs + Special Registers)
172 def SReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
176 def SReg_64 : RegisterClass<"AMDGPU", [i64, i1], 64,
177 (add SGPR_64, VCCReg, EXECReg)
180 def SReg_128 : RegisterClass<"AMDGPU", [v4f32, v4i32], 128, (add SGPR_128)>;
182 def SReg_256 : RegisterClass<"AMDGPU", [v8i32], 256, (add SGPR_256)>;
184 def SReg_512 : RegisterClass<"AMDGPU", [v16i32], 512, (add SGPR_512)>;
186 // Register class for all vector registers (VGPRs + Interploation Registers)
187 def VReg_32 : RegisterClass<"AMDGPU", [f32, i32, v1i32], 32, (add VGPR_32)>;
189 def VReg_64 : RegisterClass<"AMDGPU", [i64, v2i32], 64, (add VGPR_64)>;
191 def VReg_128 : RegisterClass<"AMDGPU", [v4f32, v4i32], 128, (add VGPR_128)>;
193 def VReg_256 : RegisterClass<"AMDGPU", [v8i32], 256, (add VGPR_256)>;
195 def VReg_512 : RegisterClass<"AMDGPU", [v16i32], 512, (add VGPR_512)>;
197 //===----------------------------------------------------------------------===//
198 // [SV]Src_* register classes, can have either an immediate or an register
199 //===----------------------------------------------------------------------===//
201 def SSrc_32 : RegisterClass<"AMDGPU", [i32, f32], 32, (add SReg_32)>;
203 def SSrc_64 : RegisterClass<"AMDGPU", [i64, i1], 64, (add SReg_64)>;
205 def VSrc_32 : RegisterClass<"AMDGPU", [i32, f32], 32,
206 (add VReg_32, SReg_32,
207 PERSP_SAMPLE_I, PERSP_SAMPLE_J,
208 PERSP_CENTER_I, PERSP_CENTER_J,
209 PERSP_CENTROID_I, PERSP_CENTROID_J,
210 PERSP_I_W, PERSP_J_W, PERSP_1_W,
211 LINEAR_SAMPLE_I, LINEAR_SAMPLE_J,
212 LINEAR_CENTER_I, LINEAR_CENTER_J,
213 LINEAR_CENTROID_I, LINEAR_CENTROID_J,
214 LINE_STIPPLE_TEX_COORD,
226 def VSrc_64 : RegisterClass<"AMDGPU", [i64], 64, (add VReg_64, SReg_64)>;