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
|
#include "gpu.h"
#include "common/apple/cf_helpers.h"
#include "common/apple/smc_temps.h"
#include <IOKit/graphics/IOGraphicsLib.h>
const char* ffGpuDetectMetal(FFlist* gpus);
const char* ffGpuDetectDriverVersion(FFlist* gpus);
static double detectGpuTemp(const FFstrbuf* gpuName) {
double result = 0;
const char* error = NULL;
if (ffStrbufStartsWithS(gpuName, "Apple M")) {
switch (strtol(gpuName->chars + strlen("Apple M"), NULL, 10)) {
case 0:
error = "Invalid Apple Silicon GPU";
break;
case 1:
error = ffDetectSmcTemps(FF_TEMP_GPU_M1X, &result);
break;
case 2:
error = ffDetectSmcTemps(FF_TEMP_GPU_M2X, &result);
break;
case 3:
error = ffDetectSmcTemps(FF_TEMP_GPU_M3X, &result);
break;
case 4:
error = ffDetectSmcTemps(FF_TEMP_GPU_M4X, &result);
break;
default:
error = "Unsupported Apple Silicon GPU";
break;
}
} else if (ffStrbufStartsWithS(gpuName, "Intel")) {
error = ffDetectSmcTemps(FF_TEMP_GPU_INTEL, &result);
} else if (ffStrbufStartsWithS(gpuName, "Radeon") || ffStrbufStartsWithS(gpuName, "AMD")) {
error = ffDetectSmcTemps(FF_TEMP_GPU_AMD, &result);
} else {
error = ffDetectSmcTemps(FF_TEMP_GPU_UNKNOWN, &result);
}
if (error) {
return FF_GPU_TEMP_UNSET;
}
return result;
}
#ifdef __aarch64__
#include "common/apple/cf_helpers.h"
#include <IOKit/IOKitLib.h>
static const char* detectFrequency(FFGPUResult* gpu) {
// https://github.com/giampaolo/psutil/pull/2222/files
FF_IOOBJECT_AUTO_RELEASE io_registry_entry_t entryDevice = IOServiceGetMatchingService(MACH_PORT_NULL, IOServiceNameMatching("pmgr"));
if (!entryDevice) {
return "IOServiceGetMatchingServices() failed";
}
if (!IOObjectConformsTo(entryDevice, "AppleARMIODevice")) {
return "\"pmgr\" should conform to \"AppleARMIODevice\"";
}
FF_CFTYPE_AUTO_RELEASE CFDataRef freqProperty = (CFDataRef) IORegistryEntryCreateCFProperty(entryDevice, CFSTR("voltage-states9-sram"), kCFAllocatorDefault, kNilOptions);
if (!freqProperty || CFGetTypeID(freqProperty) != CFDataGetTypeID()) {
return "\"voltage-states9-sram\" in \"pmgr\" is not found";
}
// voltage-states9-sram stores supported <frequency / voltage> pairs of gpu from the lowest to the highest
CFIndex propLength = CFDataGetLength(freqProperty);
if (propLength == 0 || propLength % (CFIndex) sizeof(uint32_t) * 2 != 0) {
return "Invalid \"voltage-states9-sram\" length";
}
uint32_t* pStart = (uint32_t*) CFDataGetBytePtr(freqProperty);
uint32_t pMax = *pStart;
for (CFIndex i = 2; i < propLength / (CFIndex) sizeof(uint32_t) && pStart[i] > 0; i += 2 /* skip voltage */) {
pMax = pMax > pStart[i] ? pMax : pStart[i];
}
if (pMax > 0) {
// While this is not necessary for now (seems), we add this logic just in case. See cpu_apple.c
if (pMax > 100000000) { // Assume that pMax is in Hz
gpu->frequency = pMax / 1000 / 1000;
} else { // Assume that pMax is in kHz
gpu->frequency = pMax / 1000;
}
}
return NULL;
}
#endif
const char* ffDetectGPUImpl(const FFGPUOptions* options, FFlist* gpus) {
FF_IOOBJECT_AUTO_RELEASE io_iterator_t iterator = IO_OBJECT_NULL;
{
CFMutableDictionaryRef matches = IOServiceMatching(kIOAcceleratorClassName);
CFDictionaryAddValue(matches, CFSTR("IOMatchCategory"), CFSTR(kIOAcceleratorClassName));
if (IOServiceGetMatchingServices(MACH_PORT_NULL, matches, &iterator) != kIOReturnSuccess) {
return "IOServiceGetMatchingServices() failed";
}
}
io_registry_entry_t registryEntry;
while ((registryEntry = IOIteratorNext(iterator)) != IO_OBJECT_NULL) {
CFMutableDictionaryRef properties;
if (IORegistryEntryCreateCFProperties(registryEntry, &properties, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess) {
IOObjectRelease(registryEntry);
continue;
}
FFGPUResult* gpu = FF_LIST_ADD(FFGPUResult, *gpus);
gpu->index = FF_GPU_INDEX_UNSET;
ffStrbufInit(&gpu->memoryType);
gpu->dedicated.total = gpu->dedicated.used = gpu->shared.total = gpu->shared.used = FF_GPU_VMEM_SIZE_UNSET;
gpu->type = FF_GPU_TYPE_UNKNOWN;
gpu->frequency = FF_GPU_FREQUENCY_UNSET;
IORegistryEntryGetRegistryEntryID(registryEntry, &gpu->deviceId);
ffStrbufInitStatic(&gpu->platformApi, "IOKit");
ffStrbufInit(&gpu->driver); // Ok for both Apple and Intel
ffCfDictGetString(properties, CFSTR("CFBundleIdentifier"), &gpu->driver);
if (ffCfDictGetInt(properties, CFSTR("gpu-core-count"), &gpu->coreCount) != NULL) { // For Apple
gpu->coreCount = FF_GPU_CORE_COUNT_UNSET;
}
gpu->coreUsage = FF_GPU_CORE_USAGE_UNSET;
CFDictionaryRef perfStatistics = NULL;
uint64_t vramUsed = 0, vramTotal = 0;
if (ffCfDictGetDict(properties, CFSTR("PerformanceStatistics"), &perfStatistics) == NULL) {
int64_t utilization;
if (ffCfDictGetInt64(perfStatistics, CFSTR("Device Utilization %"), &utilization) == NULL) {
gpu->coreUsage = (double) utilization;
} else if (ffCfDictGetInt64(perfStatistics, CFSTR("GPU Core Utilization"), &utilization) == NULL) {
gpu->coreUsage = (double) utilization / 10000000.; // Nvidia?
}
if (ffCfDictGetInt64(perfStatistics, CFSTR("Alloc system memory"), (int64_t*) &vramTotal) == NULL) {
if (ffCfDictGetInt64(perfStatistics, CFSTR("In use system memory"), (int64_t*) &vramUsed) != NULL) {
vramTotal = 0;
}
} else if (ffCfDictGetInt64(perfStatistics, CFSTR("vramFreeBytes"), (int64_t*) &vramTotal) == NULL) {
if (ffCfDictGetInt64(perfStatistics, CFSTR("vramUsedBytes"), (int64_t*) &vramUsed) == NULL) {
vramTotal += vramUsed;
} else {
vramTotal = 0;
}
}
}
ffStrbufInit(&gpu->name);
// IOAccelerator returns model / vendor-id properties for Apple Silicon, but not for Intel Iris GPUs.
// Still needs testing for AMD's
if (ffCfDictGetString(properties, CFSTR("model"), &gpu->name) != NULL) {
CFRelease(properties);
properties = NULL;
FF_IOOBJECT_AUTO_RELEASE io_registry_entry_t parentEntry = 0;
if (IORegistryEntryGetParentEntry(registryEntry, kIOServicePlane, &parentEntry) != kIOReturnSuccess ||
IORegistryEntryCreateCFProperties(parentEntry, &properties, kCFAllocatorDefault, kNilOptions) != kIOReturnSuccess) {
IOObjectRelease(registryEntry);
continue;
}
ffCfDictGetString(properties, CFSTR("model"), &gpu->name);
}
ffStrbufInit(&gpu->vendor);
int vendorId;
if (ffCfDictGetInt(properties, CFSTR("vendor-id"), &vendorId) == NULL) {
const char* vendorStr = ffGPUGetVendorString((unsigned) vendorId);
ffStrbufAppendS(&gpu->vendor, vendorStr);
if (vendorStr == FF_GPU_VENDOR_NAME_APPLE || vendorStr == FF_GPU_VENDOR_NAME_INTEL) {
gpu->type = FF_GPU_TYPE_INTEGRATED;
} else if (vendorStr == FF_GPU_VENDOR_NAME_NVIDIA || vendorStr == FF_GPU_VENDOR_NAME_AMD) {
gpu->type = FF_GPU_TYPE_DISCRETE;
}
#ifdef __aarch64__
if (vendorStr == FF_GPU_VENDOR_NAME_APPLE) {
detectFrequency(gpu);
}
#endif
if (gpu->type == FF_GPU_TYPE_INTEGRATED) {
gpu->shared.total = vramTotal;
gpu->shared.used = vramUsed;
} else if (gpu->type == FF_GPU_TYPE_DISCRETE) {
gpu->dedicated.total = vramTotal;
gpu->dedicated.used = vramUsed;
}
}
gpu->temperature = options->temp ? detectGpuTemp(&gpu->name) : FF_GPU_TEMP_UNSET;
CFRelease(properties);
IOObjectRelease(registryEntry);
}
ffGpuDetectMetal(gpus);
if (instance.config.general.detectVersion) {
ffGpuDetectDriverVersion(gpus);
}
return NULL;
}
|
