1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
| //===-- CaymanInstructions.td - CM Instruction defs -------*- tablegen -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// TableGen definitions for instructions which are available only on Cayman
// family GPUs.
//
//===----------------------------------------------------------------------===//
def isCayman : Predicate<"Subtarget->hasCaymanISA()">;
//===----------------------------------------------------------------------===//
// Cayman Instructions
//===----------------------------------------------------------------------===//
let SubtargetPredicate = isCayman in {
def MULADD_INT24_cm : R600_3OP <0x08, "MULADD_INT24",
[(set i32:$dst, (AMDGPUmad_i24 i32:$src0, i32:$src1, i32:$src2))], VecALU
>;
def MUL_INT24_cm : R600_2OP <0x5B, "MUL_INT24",
[(set i32:$dst, (AMDGPUmul_i24 i32:$src0, i32:$src1))], VecALU
>;
def : IMad24Pat<MULADD_INT24_cm>;
let isVector = 1 in {
def RECIP_IEEE_cm : RECIP_IEEE_Common<0x86>;
def MULLO_INT_cm : MULLO_INT_Common<0x8F>;
def MULHI_INT_cm : MULHI_INT_Common<0x90>;
def MULLO_UINT_cm : MULLO_UINT_Common<0x91>;
def MULHI_UINT_cm : MULHI_UINT_Common<0x92>;
def MULHI_INT_cm24 : MULHI_INT24_Common<0x5c>;
def MULHI_UINT_cm24 : MULHI_UINT24_Common<0xb2>;
def RECIPSQRT_CLAMPED_cm : RECIPSQRT_CLAMPED_Common<0x87>;
def EXP_IEEE_cm : EXP_IEEE_Common<0x81>;
def LOG_IEEE_cm : LOG_IEEE_Common<0x83>;
def RECIP_CLAMPED_cm : RECIP_CLAMPED_Common<0x84>;
def RECIPSQRT_IEEE_cm : RECIPSQRT_IEEE_Common<0x89>;
def SIN_cm : SIN_Common<0x8D>;
def COS_cm : COS_Common<0x8E>;
} // End isVector = 1
def : RsqPat<RECIPSQRT_IEEE_cm, f32>;
def : POW_Common <LOG_IEEE_cm, EXP_IEEE_cm, MUL>;
defm DIV_cm : DIV_Common<RECIP_IEEE_cm>;
// RECIP_UINT emulation for Cayman
// The multiplication scales from [0,1] to the unsigned integer range
def : R600Pat <
(AMDGPUurecip i32:$src0),
(FLT_TO_UINT_eg (MUL_IEEE (RECIP_IEEE_cm (UINT_TO_FLT_eg $src0)),
(MOV_IMM_I32 CONST.FP_UINT_MAX_PLUS_1)))
>;
def CF_END_CM : CF_CLAUSE_EG<32, (ins), "CF_END"> {
let ADDR = 0;
let POP_COUNT = 0;
let COUNT = 0;
}
def : R600Pat<(fsqrt f32:$src), (MUL R600_Reg32:$src, (RECIPSQRT_CLAMPED_cm $src))>;
class RAT_STORE_DWORD <RegisterClass rc, ValueType vt, bits<4> mask> :
CF_MEM_RAT_CACHELESS <0x14, 0, mask,
(ins rc:$rw_gpr, R600_TReg32_X:$index_gpr),
"STORE_DWORD $rw_gpr, $index_gpr",
[(store_global vt:$rw_gpr, i32:$index_gpr)]> {
let eop = 0; // This bit is not used on Cayman.
}
def RAT_STORE_DWORD32 : RAT_STORE_DWORD <R600_TReg32_X, i32, 0x1>;
def RAT_STORE_DWORD64 : RAT_STORE_DWORD <R600_Reg64, v2i32, 0x3>;
def RAT_STORE_DWORD128 : RAT_STORE_DWORD <R600_Reg128, v4i32, 0xf>;
def RAT_STORE_TYPED_cm: CF_MEM_RAT_STORE_TYPED<0> {
let eop = 0; // This bit is not used on Cayman.
}
class VTX_READ_cm <string name, dag outs>
: VTX_WORD0_cm, VTX_READ<name, outs, []> {
// Static fields
let VC_INST = 0;
let FETCH_TYPE = 2;
let FETCH_WHOLE_QUAD = 0;
let SRC_REL = 0;
// XXX: We can infer this field based on the SRC_GPR. This would allow us
// to store vertex addresses in any channel, not just X.
let SRC_SEL_X = 0;
let SRC_SEL_Y = 0;
let STRUCTURED_READ = 0;
let LDS_REQ = 0;
let COALESCED_READ = 0;
let Inst{31-0} = Word0;
}
def VTX_READ_8_cm
: VTX_READ_cm <"VTX_READ_8 $dst_gpr, $src_gpr",
(outs R600_TReg32_X:$dst_gpr)> {
let DST_SEL_X = 0;
let DST_SEL_Y = 7; // Masked
let DST_SEL_Z = 7; // Masked
let DST_SEL_W = 7; // Masked
let DATA_FORMAT = 1; // FMT_8
}
def VTX_READ_16_cm
: VTX_READ_cm <"VTX_READ_16 $dst_gpr, $src_gpr",
(outs R600_TReg32_X:$dst_gpr)> {
let DST_SEL_X = 0;
let DST_SEL_Y = 7; // Masked
let DST_SEL_Z = 7; // Masked
let DST_SEL_W = 7; // Masked
let DATA_FORMAT = 5; // FMT_16
}
def VTX_READ_32_cm
: VTX_READ_cm <"VTX_READ_32 $dst_gpr, $src_gpr",
(outs R600_TReg32_X:$dst_gpr)> {
let DST_SEL_X = 0;
let DST_SEL_Y = 7; // Masked
let DST_SEL_Z = 7; // Masked
let DST_SEL_W = 7; // Masked
let DATA_FORMAT = 0xD; // COLOR_32
// This is not really necessary, but there were some GPU hangs that appeared
// to be caused by ALU instructions in the next instruction group that wrote
// to the $src_gpr registers of the VTX_READ.
// e.g.
// %t3_x = VTX_READ_PARAM_32_eg killed %t2_x, 24
// %t2_x = MOV %zero
//Adding this constraint prevents this from happening.
let Constraints = "$src_gpr.ptr = $dst_gpr";
}
def VTX_READ_64_cm
: VTX_READ_cm <"VTX_READ_64 $dst_gpr.XY, $src_gpr",
(outs R600_Reg64:$dst_gpr)> {
let DST_SEL_X = 0;
let DST_SEL_Y = 1;
let DST_SEL_Z = 7;
let DST_SEL_W = 7;
let DATA_FORMAT = 0x1D; // COLOR_32_32
}
def VTX_READ_128_cm
: VTX_READ_cm <"VTX_READ_128 $dst_gpr.XYZW, $src_gpr",
(outs R600_Reg128:$dst_gpr)> {
let DST_SEL_X = 0;
let DST_SEL_Y = 1;
let DST_SEL_Z = 2;
let DST_SEL_W = 3;
let DATA_FORMAT = 0x22; // COLOR_32_32_32_32
// XXX: Need to force VTX_READ_128 instructions to write to the same register
// that holds its buffer address to avoid potential hangs. We can't use
// the same constraint as VTX_READ_32_eg, because the $src_gpr.ptr and $dst
// registers are different sizes.
}
//===----------------------------------------------------------------------===//
// VTX Read from parameter memory space
//===----------------------------------------------------------------------===//
def : R600Pat<(i32:$dst_gpr (vtx_id3_az_extloadi8 ADDRVTX_READ:$src_gpr)),
(VTX_READ_8_cm MEMxi:$src_gpr, 3)>;
def : R600Pat<(i32:$dst_gpr (vtx_id3_az_extloadi16 ADDRVTX_READ:$src_gpr)),
(VTX_READ_16_cm MEMxi:$src_gpr, 3)>;
def : R600Pat<(i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_32_cm MEMxi:$src_gpr, 3)>;
def : R600Pat<(v2i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_64_cm MEMxi:$src_gpr, 3)>;
def : R600Pat<(v4i32:$dst_gpr (vtx_id3_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_128_cm MEMxi:$src_gpr, 3)>;
//===----------------------------------------------------------------------===//
// VTX Read from constant memory space
//===----------------------------------------------------------------------===//
def : R600Pat<(i32:$dst_gpr (vtx_id2_az_extloadi8 ADDRVTX_READ:$src_gpr)),
(VTX_READ_8_cm MEMxi:$src_gpr, 2)>;
def : R600Pat<(i32:$dst_gpr (vtx_id2_az_extloadi16 ADDRVTX_READ:$src_gpr)),
(VTX_READ_16_cm MEMxi:$src_gpr, 2)>;
def : R600Pat<(i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_32_cm MEMxi:$src_gpr, 2)>;
def : R600Pat<(v2i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_64_cm MEMxi:$src_gpr, 2)>;
def : R600Pat<(v4i32:$dst_gpr (vtx_id2_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_128_cm MEMxi:$src_gpr, 2)>;
//===----------------------------------------------------------------------===//
// VTX Read from global memory space
//===----------------------------------------------------------------------===//
def : R600Pat<(i32:$dst_gpr (vtx_id1_az_extloadi8 ADDRVTX_READ:$src_gpr)),
(VTX_READ_8_cm MEMxi:$src_gpr, 1)>;
def : R600Pat<(i32:$dst_gpr (vtx_id1_az_extloadi16 ADDRVTX_READ:$src_gpr)),
(VTX_READ_16_cm MEMxi:$src_gpr, 1)>;
def : R600Pat<(i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_32_cm MEMxi:$src_gpr, 1)>;
def : R600Pat<(v2i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_64_cm MEMxi:$src_gpr, 1)>;
def : R600Pat<(v4i32:$dst_gpr (vtx_id1_load ADDRVTX_READ:$src_gpr)),
(VTX_READ_128_cm MEMxi:$src_gpr, 1)>;
} // End let SubtargetPredicate = isCayman
|