CLRX  1
An unofficial OpenCL extensions designed for Radeon GPUs
GPUId.h
Go to the documentation of this file.
1 /*
2  * CLRadeonExtender - Unofficial OpenCL Radeon Extensions Library
3  * Copyright (C) 2014-2018 Mateusz Szpakowski
4  *
5  * This library is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU Lesser General Public
7  * License as published by the Free Software Foundation; either
8  * version 2.1 of the License, or (at your option) any later version.
9  *
10  * This library is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13  * Lesser General Public License for more details.
14  *
15  * You should have received a copy of the GNU Lesser General Public
16  * License along with this library; if not, write to the Free Software
17  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18  */
23 #ifndef __CLRX_GPUID_H__
24 #define __CLRX_GPUID_H__
25 
26 #include <CLRX/Config.h>
27 #include <CLRX/utils/Utilities.h>
28 #include <string>
29 
31 namespace CLRX
32 {
33 
36 {
37 public:
39  GPUIdException() = default;
41  explicit GPUIdException(const std::string& message);
43  virtual ~GPUIdException() noexcept = default;
44 };
45 
46 /*
47  * GPU identification utilities
48  */
49 
51 enum class GPUDeviceType: cxbyte
52 {
53  CAPE_VERDE = 0,
54  PITCAIRN,
55  TAHITI,
56  OLAND,
57  BONAIRE,
58  SPECTRE,
59  SPOOKY,
60  KALINDI,
61  HAINAN,
62  HAWAII,
63  ICELAND,
64  TONGA,
65  MULLINS,
66  FIJI,
67  CARRIZO,
68  DUMMY,
69  GOOSE,
70  HORSE,
71  STONEY,
72  ELLESMERE,
73  BAFFIN,
74  GFX804,
75  GFX900,
76  GFX901,
77  GFX902,
78  GFX903,
79  GFX904,
80  GFX905,
81  GPUDEVICE_MAX = GFX905,
82 
88  RADEON_R9_290 = HAWAII
89 };
90 
93 {
94  GCN1_0 = 0,
95  GCN1_1,
96  GCN1_2,
97  GCN1_4,
98  GPUARCH_MAX = GCN1_4
99 };
100 
102 extern GPUDeviceType getGPUDeviceTypeFromName(const char* name);
103 
105 extern const char* getGPUDeviceTypeName(GPUDeviceType deviceType);
106 
108 extern GPUArchitecture getGPUArchitectureFromName(const char* name);
109 
112 
115 
117 extern const char* getGPUArchitectureName(GPUArchitecture architecture);
118 
119 enum: Flags {
120  REGCOUNT_NO_VCC = 1,
121  REGCOUNT_NO_FLAT = 2,
122  REGCOUNT_NO_XNACK = 4,
123  REGCOUNT_NO_EXTRA = 0xffff
124 };
125 
126 enum: cxuint {
127  REGTYPE_SGPR = 0,
128  REGTYPE_VGPR
129 };
130 
131 enum : Flags
132 {
133  GCN_VCC = 1,
134  GCN_FLAT = 2,
135  GCN_XNACK = 4
136 };
137 
138 enum: Flags {
139  GPUSETUP_TGSIZE_EN = 1,
140  GPUSETUP_SCRATCH_EN = 2
141 };
142 
143 enum: cxuint {
144  MAX_REGTYPES_NUM = 4
145 };
146 
148 extern cxuint getGPUMaxRegistersNum(GPUArchitecture architecture, cxuint regType,
149  Flags flags = 0);
150 
152 extern cxuint getGPUMaxRegsNumByArchMask(uint16_t archMask, cxuint regType);
153 
155 extern void getGPUSetupMinRegistersNum(GPUArchitecture architecture, cxuint dimMask,
156  cxuint userDataNum, Flags flags, cxuint* gprsOut);
157 
159 extern size_t getGPUMaxLocalSize(GPUArchitecture architecture);
160 
162 extern size_t getGPUMaxGDSSize(GPUArchitecture architecture);
163 
165 extern cxuint getGPUExtraRegsNum(GPUArchitecture architecture, cxuint regType,
166  Flags flags);
167 
170 {
171  uint32_t major;
172  uint32_t minor;
173  uint32_t stepping;
174 };
175 
177 extern uint32_t calculatePgmRSrc1(GPUArchitecture arch, cxuint vgprsNum, cxuint sgprsNum,
178  cxuint priority, cxuint floatMode, bool privMode, bool dx10clamp,
179  bool debugMode, bool ieeeMode);
180 
182 extern uint32_t calculatePgmRSrc2(GPUArchitecture arch, bool scratchEn, cxuint userDataNum,
183  bool trapPresent, cxuint dimMask, cxuint defDimValues, bool tgSizeEn,
184  cxuint ldsSize, cxuint exceptions);
185 
186 
189 {
190  AMDCL2 = 0,
191  OPENSOURCE,
192  ROCM
193 };
194 
197  GPUArchVersionTable table);
198 
199 // get GPU device type from architecture version
200 extern GPUDeviceType getGPUDeviceTypeFromArchVersion(cxuint archMajor, cxuint archMinor,
201  cxuint archStepping);
202 
203 };
204 
205 #endif
uint32_t Flags
type for declaring various flags
Definition: Utilities.h:97
first iteration (Radeon HD7000 series)
virtual ~GPUIdException() noexcept=default
destructor
structure helper for AMDGPU architecture version
Definition: GPUId.h:169
AMD OpenCL 2.0 format.
GPUArchitecture getGPUArchitectureFromDeviceType(GPUDeviceType deviceType)
get GPUArchitecture from GPU device type
GPUArchitecture getGPUArchitectureFromName(const char *name)
get GPU architecture from name
GPUArchitecture
GPU architecture.
Definition: GPUId.h:92
GPUId exception class.
Definition: GPUId.h:35
GPUDeviceType getLowestGPUDeviceTypeFromArchitecture(GPUArchitecture arch)
get lowest GPU device for architecture
GPUArchVersionTable
ADMGPUArchValues table type.
Definition: GPUId.h:188
Configuration header.
void getGPUSetupMinRegistersNum(GPUArchitecture architecture, cxuint dimMask, cxuint userDataNum, Flags flags, cxuint *gprsOut)
get minimal number of required registers
uint32_t calculatePgmRSrc2(GPUArchitecture arch, bool scratchEn, cxuint userDataNum, bool trapPresent, cxuint dimMask, cxuint defDimValues, bool tgSizeEn, cxuint ldsSize, cxuint exceptions)
calculate PGMRSRC2 register value
const char * getGPUDeviceTypeName(GPUDeviceType deviceType)
get GPU device type name
size_t getGPUMaxLocalSize(GPUArchitecture architecture)
get maximum local size for GPU architecture
cxuint getGPUMaxRegistersNum(GPUArchitecture architecture, cxuint regType, Flags flags=0)
get maximum available registers for GPU (type: 0 - scalar, 1 - vector)
uint32_t major
arch major number
Definition: GPUId.h:171
uint32_t stepping
arch stepping number
Definition: GPUId.h:173
unsigned char cxbyte
unsigned byte
Definition: Config.h:215
uint32_t calculatePgmRSrc1(GPUArchitecture arch, cxuint vgprsNum, cxuint sgprsNum, cxuint priority, cxuint floatMode, bool privMode, bool dx10clamp, bool debugMode, bool ieeeMode)
calculate PGMRSRC1 register value
main namespace
Definition: AsmDefs.h:38
GFX9 architecture (AMD RX VEGA)
unsigned int cxuint
unsigned int
Definition: Config.h:223
size_t getGPUMaxGDSSize(GPUArchitecture architecture)
get maximum GDS size for GPU architecture
cxuint getGPUMaxRegsNumByArchMask(uint16_t archMask, cxuint regType)
get maximum available registers for GPU (type: 0 - scalar, 1 - vector)
uint32_t minor
arch minor number
Definition: GPUId.h:172
GPUDeviceType
type of GPU device
Definition: GPUId.h:51
third iteration (Radeon Rx 300 series and Tonga)
ROCm (RadeonOpenCompute) format.
GPUIdException()=default
empty constructor
utilities for other libraries and programs
const char * getGPUArchitectureName(GPUArchitecture architecture)
get GPU architecture name
std::string message
message
Definition: Utilities.h:61
GPUDeviceType getGPUDeviceTypeFromName(const char *name)
get GPU device type from name
exception class
Definition: Utilities.h:58
AMDGPUArchVersion getGPUArchVersion(GPUDeviceType deviceType, GPUArchVersionTable table)
get AMD GPU architecture version for specific device type and driver
cxuint getGPUExtraRegsNum(GPUArchitecture architecture, cxuint regType, Flags flags)
get extra registers (like VCC,FLAT_SCRATCH)
second iteration (Radeon Rx 200 series)