2 let Namespace = "AMDGPU" in {
4 def high : SubRegIndex;
6 def sub0 : SubRegIndex;
7 def sub1 : SubRegIndex;
8 def sub2 : SubRegIndex;
9 def sub3 : SubRegIndex;
10 def sub4 : SubRegIndex;
11 def sub5 : SubRegIndex;
12 def sub6 : SubRegIndex;
13 def sub7 : SubRegIndex;
16 class SIReg <string n, bits<16> encoding = 0> : Register<n> {
17 let Namespace = "AMDGPU";
18 let HWEncoding = encoding;
21 class SI_64 <string n, list<Register> subregs, bits<16> encoding> : RegisterWithSubRegs<n, subregs> {
22 let Namespace = "AMDGPU";
23 let SubRegIndices = [low, high];
24 let HWEncoding = encoding;
27 class SGPR_32 <bits<16> num, string name> : SIReg<name, num>;
29 class VGPR_32 <bits<16> num, string name> : SIReg<name, num>;
32 def VCC : SIReg<"VCC", 106>;
33 def EXEC_LO : SIReg <"EXEC LO", 126>;
34 def EXEC_HI : SIReg <"EXEC HI", 127>;
35 def EXEC : SI_64<"EXEC", [EXEC_LO, EXEC_HI], 126>;
36 def SCC : SIReg<"SCC", 253>;
37 def SREG_LIT_0 : SIReg <"S LIT 0", 128>;
38 def SI_LITERAL_CONSTANT : SIReg<"LITERAL CONSTANT", 255>;
39 def M0 : SIReg <"M0", 124>;
41 //Interpolation registers
42 def PERSP_SAMPLE_I : SIReg <"PERSP_SAMPLE_I">;
43 def PERSP_SAMPLE_J : SIReg <"PERSP_SAMPLE_J">;
44 def PERSP_CENTER_I : SIReg <"PERSP_CENTER_I">;
45 def PERSP_CENTER_J : SIReg <"PERSP_CENTER_J">;
46 def PERSP_CENTROID_I : SIReg <"PERSP_CENTROID_I">;
47 def PERSP_CENTROID_J : SIReg <"PERP_CENTROID_J">;
48 def PERSP_I_W : SIReg <"PERSP_I_W">;
49 def PERSP_J_W : SIReg <"PERSP_J_W">;
50 def PERSP_1_W : SIReg <"PERSP_1_W">;
51 def LINEAR_SAMPLE_I : SIReg <"LINEAR_SAMPLE_I">;
52 def LINEAR_SAMPLE_J : SIReg <"LINEAR_SAMPLE_J">;
53 def LINEAR_CENTER_I : SIReg <"LINEAR_CENTER_I">;
54 def LINEAR_CENTER_J : SIReg <"LINEAR_CENTER_J">;
55 def LINEAR_CENTROID_I : SIReg <"LINEAR_CENTROID_I">;
56 def LINEAR_CENTROID_J : SIReg <"LINEAR_CENTROID_J">;
57 def LINE_STIPPLE_TEX_COORD : SIReg <"LINE_STIPPLE_TEX_COORD">;
58 def POS_X_FLOAT : SIReg <"POS_X_FLOAT">;
59 def POS_Y_FLOAT : SIReg <"POS_Y_FLOAT">;
60 def POS_Z_FLOAT : SIReg <"POS_Z_FLOAT">;
61 def POS_W_FLOAT : SIReg <"POS_W_FLOAT">;
62 def FRONT_FACE : SIReg <"FRONT_FACE">;
63 def ANCILLARY : SIReg <"ANCILLARY">;
64 def SAMPLE_COVERAGE : SIReg <"SAMPLE_COVERAGE">;
65 def POS_FIXED_PT : SIReg <"POS_FIXED_PT">;
67 // SGPR 32-bit registers
68 foreach Index = 0-101 in {
69 def SGPR#Index : SGPR_32 <Index, "SGPR"#Index>;
72 def SGPR_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
73 (add (sequence "SGPR%u", 0, 101))>;
75 // SGPR 64-bit registers
76 def SGPR_64 : RegisterTuples<[low, high],
77 [(add (decimate SGPR_32, 2)),
78 (add(decimate (rotl SGPR_32, 1), 2))]>;
80 // SGPR 128-bit registers
81 def SGPR_128 : RegisterTuples<[sel_x, sel_y, sel_z, sel_w],
82 [(add (decimate SGPR_32, 4)),
83 (add (decimate (rotl SGPR_32, 1), 4)),
84 (add (decimate (rotl SGPR_32, 2), 4)),
85 (add (decimate (rotl SGPR_32, 3), 4))]>;
87 // SGPR 256-bit registers
88 def SGPR_256 : RegisterTuples<[sub0, sub1, sub2, sub3, sub4, sub5, sub6, sub7],
89 [(add (decimate SGPR_32, 8)),
90 (add (decimate (rotl SGPR_32, 1), 8)),
91 (add (decimate (rotl SGPR_32, 2), 8)),
92 (add (decimate (rotl SGPR_32, 3), 8)),
93 (add (decimate (rotl SGPR_32, 4), 8)),
94 (add (decimate (rotl SGPR_32, 5), 8)),
95 (add (decimate (rotl SGPR_32, 6), 8)),
96 (add (decimate (rotl SGPR_32, 7), 8))]>;
98 // VGPR 32-bit registers
99 foreach Index = 0-255 in {
100 def VGPR#Index : VGPR_32 <Index, "VGPR"#Index>;
103 def VGPR_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
104 (add (sequence "VGPR%u", 0, 255))>;
106 // VGPR 64-bit registers
107 def VGPR_64 : RegisterTuples<[low, high],
109 (add (rotl VGPR_32, 1))]>;
111 // VGPR 128-bit registers
112 def VGPR_128 : RegisterTuples<[sel_x, sel_y, sel_z, sel_w],
114 (add (rotl VGPR_32, 1)),
115 (add (rotl VGPR_32, 2)),
116 (add (rotl VGPR_32, 3))]>;
118 // Register class for all scalar registers (SGPRs + Special Registers)
119 def SReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
120 (add SGPR_32, SREG_LIT_0, M0, EXEC_LO, EXEC_HI)
123 def SReg_64 : RegisterClass<"AMDGPU", [i64], 64, (add SGPR_64, VCC, EXEC)>;
125 def SReg_1 : RegisterClass<"AMDGPU", [i1], 1, (add VCC, SGPR_64, EXEC)>;
127 def SReg_128 : RegisterClass<"AMDGPU", [v4f32, v4i32], 128, (add SGPR_128)>;
129 def SReg_256 : RegisterClass<"AMDGPU", [v8i32], 256, (add SGPR_256)>;
131 // Register class for all vector registers (VGPRs + Interploation Registers)
132 def VReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32,
134 PERSP_SAMPLE_I, PERSP_SAMPLE_J,
135 PERSP_CENTER_I, PERSP_CENTER_J,
136 PERSP_CENTROID_I, PERSP_CENTROID_J,
137 PERSP_I_W, PERSP_J_W, PERSP_1_W,
138 LINEAR_SAMPLE_I, LINEAR_SAMPLE_J,
139 LINEAR_CENTER_I, LINEAR_CENTER_J,
140 LINEAR_CENTROID_I, LINEAR_CENTROID_J,
141 LINE_STIPPLE_TEX_COORD,
153 def VReg_64 : RegisterClass<"AMDGPU", [i64], 64, (add VGPR_64)>;
155 def VReg_128 : RegisterClass<"AMDGPU", [v4f32], 128, (add VGPR_128)>;
157 // AllReg_* - A set of all scalar and vector registers of a given width.
158 def AllReg_32 : RegisterClass<"AMDGPU", [f32, i32], 32, (add VReg_32, SReg_32)>;
160 def AllReg_64 : RegisterClass<"AMDGPU", [f64, i64], 64, (add SReg_64, VReg_64)>;
162 // Special register classes for predicates and the M0 register
163 def SCCReg : RegisterClass<"AMDGPU", [i1], 1, (add SCC)>;
164 def VCCReg : RegisterClass<"AMDGPU", [i1], 1, (add VCC)>;
165 def EXECReg : RegisterClass<"AMDGPU", [i1], 1, (add EXEC)>;
166 def M0Reg : RegisterClass<"AMDGPU", [i32], 32, (add M0)>;