Skip to content

Fix the heuristics for L3 cache size for Arm64 #71029

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
kunalspathak merged 13 commits into from
Jun 22, 2022
Merged

Fix the heuristics for L3 cache size for Arm64 #71029

Show file tree
Hide file tree
Changes from 11 commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
f6310e4
Get cacheLevel for unix
kunalspathak Jun 20, 2022
45c4138
Get cacheLevel for windows
kunalspathak Jun 20, 2022
8b6862c
Get cacheLevel for os
kunalspathak Jun 20, 2022
363490c
add the macro CHECK_CACHE_SIZE
kunalspathak Jun 20, 2022
29ed6db
fix multiplier
kunalspathak Jun 21, 2022
468dc4a
Review comments
kunalspathak Jun 21, 2022
b09e495
Fix linux arm64 build
kunalspathak Jun 21, 2022
c7c905a
Change from std:string to char[]
kunalspathak Jun 21, 2022
0bf5ec1
minor nit
kunalspathak Jun 21, 2022
1f7ce1c
opps should be outside the loop
kunalspathak Jun 21, 2022
cc72139
Replace assert with _ASSERTE
kunalspathak Jun 21, 2022
5671866
CHECK_CACHE_SIZE -> UPDATE_CACHE_SIZE_AND_LEVEL
kunalspathak Jun 21, 2022
aaab0bf
Use GetTotalProcessorCount for windows
kunalspathak Jun 22, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
97 changes: 73 additions & 24 deletions src/coreclr/gc/unix/gcenv.unix.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -876,21 +876,29 @@ bool ReadMemoryValueFromFile(const char* filename, uint64_t* val)
return result;
}

#define CHECK_CACHE_SIZE(CACHE_LEVEL) if (size > cacheSize) { cacheSize = size; cacheLevel = CACHE_LEVEL; }

static size_t GetLogicalProcessorCacheSizeFromOS()
{
size_t cacheLevel = 0;
size_t cacheSize = 0;
size_t size;

#ifdef _SC_LEVEL1_DCACHE_SIZE
cacheSize = std::max(cacheSize, ( size_t) sysconf(_SC_LEVEL1_DCACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL1_DCACHE_SIZE);
Copy link
Member

@Maoni0 Maoni0 Jun 21, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Q (also applies to the old src) - what does sysconf return if there's no cache of that level? I presume it returns 0, not -1 (the doc says it returns -1 if there's an error).

CHECK_CACHE_SIZE(1)
#endif
#ifdef _SC_LEVEL2_CACHE_SIZE
cacheSize = std::max(cacheSize, ( size_t) sysconf(_SC_LEVEL2_CACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL2_CACHE_SIZE);
CHECK_CACHE_SIZE(2)
#endif
#ifdef _SC_LEVEL3_CACHE_SIZE
cacheSize = std::max(cacheSize, ( size_t) sysconf(_SC_LEVEL3_CACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL3_CACHE_SIZE);
CHECK_CACHE_SIZE(3)
#endif
#ifdef _SC_LEVEL4_CACHE_SIZE
cacheSize = std::max(cacheSize, ( size_t) sysconf(_SC_LEVEL4_CACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL4_CACHE_SIZE);
CHECK_CACHE_SIZE(4)
#endif

#if defined(TARGET_LINUX) && !defined(HOST_ARM) && !defined(HOST_X86)
Expand All @@ -901,25 +909,39 @@ static size_t GetLogicalProcessorCacheSizeFromOS()
// for the platform. Currently musl and arm64 should be only cases to use
// this method to determine cache size.
//
size_t size;

if (ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index0/size", &size))
cacheSize = std::max(cacheSize, size);
if (ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index1/size", &size))
cacheSize = std::max(cacheSize, size);
if (ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index2/size", &size))
cacheSize = std::max(cacheSize, size);
if (ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index3/size", &size))
cacheSize = std::max(cacheSize, size);
if (ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index4/size", &size))
cacheSize = std::max(cacheSize, size);
size_t level;
char path_to_size_file[] = "/sys/devices/system/cpu/cpu0/cache/index-/size";
char path_to_level_file[] = "/sys/devices/system/cpu/cpu0/cache/index-/level";
int index = 40;
assert(path_to_size_file[index] == '-');
assert(path_to_level_file[index] == '-');

for (int i = 0; i < 5; i++)
{
path_to_size_file[index] = (char)(48 + i);

if (ReadMemoryValueFromFile(path_to_size_file, &size))
{
path_to_level_file[index] = (char)(48 + i);

if (ReadMemoryValueFromFile(path_to_level_file, &level))
{
CHECK_CACHE_SIZE(level)
}
else
{
cacheSize = std::max(cacheSize, size);
}
}
}
}
#endif

#if defined(HOST_ARM64) && !defined(TARGET_OSX)
#if (defined(HOST_ARM64) || defined(HOST_LOONGARCH64)) && !defined(TARGET_OSX)
if (cacheSize == 0)
{
// It is currently expected to be missing cache size info
// We expect to get the L3 cache size for Arm64 but currently expected to be missing that info
// from most of the machines.
//
// _SC_LEVEL*_*CACHE_SIZE is not yet present. Work is in progress to enable this for arm64
//
Expand Down Expand Up @@ -964,6 +986,38 @@ static size_t GetLogicalProcessorCacheSizeFromOS()
}
#endif

#if (defined(HOST_ARM64) || defined(HOST_LOONGARCH64)) && !defined(TARGET_OSX)
if (cacheLevel != 3)
{
// We expect to get the L3 cache size for Arm64 but currently expected to be missing that info
// from most of the machines.
// Hence, just use the following heuristics at best depending on the CPU count
// 1 ~ 4 : 4 MB
// 5 ~ 16 : 8 MB
// 17 ~ 64 : 16 MB
// 65+ : 32 MB
DWORD logicalCPUs = g_totalCpuCount;
if (logicalCPUs < 5)
{
cacheSize = 4;
}
else if (logicalCPUs < 17)
{
cacheSize = 8;
}
else if (logicalCPUs < 65)
{
cacheSize = 16;
}
else
{
cacheSize = 32;
}

cacheSize *= (1024 * 1024);
}
#endif

return cacheSize;
}

Expand Down Expand Up @@ -1037,15 +1091,10 @@ size_t GCToOSInterface::GetCacheSizePerLogicalCpu(bool trueSize)
size_t maxSize, maxTrueSize;
maxSize = maxTrueSize = GetLogicalProcessorCacheSizeFromOS(); // Returns the size of the highest level processor cache

#if defined(HOST_ARM64)
// Bigger gen0 size helps arm64 targets
maxSize = maxTrueSize * 3;
#endif

s_maxSize = maxSize;
s_maxTrueSize = maxTrueSize;

// printf("GetCacheSizePerLogicalCpu returns %d, adjusted size %d\n", maxSize, maxTrueSize);
// printf("GetCacheSizePerLogicalCpu returns %zu, adjusted size %zu\n", maxSize, maxTrueSize);
return trueSize ? maxTrueSize : maxSize;
}

Expand Down
51 changes: 44 additions & 7 deletions src/coreclr/gc/windows/gcenv.windows.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -402,6 +402,9 @@ SYSTEM_LOGICAL_PROCESSOR_INFORMATION *GetLPI(PDWORD nEntries)
size_t GetLogicalProcessorCacheSizeFromOS()
{
size_t cache_size = 0;
size_t cache_level = 0;
uint32_t totalCPUCount = 0;

DWORD nEntries = 0;

// Try to use GetLogicalProcessorInformation API and get a valid pointer to the SLPI array if successful. Returns NULL
Expand All @@ -424,7 +427,15 @@ size_t GetLogicalProcessorCacheSizeFromOS()
{
if (pslpi[i].Relationship == RelationCache)
{
last_cache_size = max(last_cache_size, pslpi[i].Cache.Size);
if (last_cache_size < pslpi[i].Cache.Size)
{
last_cache_size = pslpi[i].Cache.Size;
cache_level = pslpi[i].Cache.Level;
}
}
else if (pslpi[i].Relationship == RelationProcessorCore)
{
totalCPUCount++;
}
}
cache_size = last_cache_size;
Expand All @@ -434,6 +445,37 @@ size_t GetLogicalProcessorCacheSizeFromOS()
if(pslpi)
delete[] pslpi; // release the memory allocated for the SLPI array.

#if defined(TARGET_ARM64)
if (cache_level != 3)
{
// We expect to get the L3 cache size for Arm64 but currently expected to be missing that info
// from most of the machines.
// Hence, just use the following heuristics at best depending on the CPU count
// 1 ~ 4 : 4 MB
// 5 ~ 16 : 8 MB
// 17 ~ 64 : 16 MB
// 65+ : 32 MB
if (totalCPUCount < 5)
{
cache_size = 4;
}
else if (totalCPUCount < 17)
{
cache_size = 8;
}
else if (totalCPUCount < 65)
{
cache_size = 16;
}
else
{
cache_size = 32;
}

cache_size *= (1024 * 1024);
}
#endif // TARGET_ARM64

return cache_size;
}

Expand Down Expand Up @@ -836,15 +878,10 @@ size_t GCToOSInterface::GetCacheSizePerLogicalCpu(bool trueSize)

maxSize = maxTrueSize = GetLogicalProcessorCacheSizeFromOS() ; // Returns the size of the highest level processor cache

#if defined(TARGET_ARM64)
// Bigger gen0 size helps arm64 targets
maxSize = maxTrueSize * 3;
#endif

s_maxSize = maxSize;
s_maxTrueSize = maxTrueSize;

// printf("GetCacheSizePerLogicalCpu returns %d, adjusted size %d\n", maxSize, maxTrueSize);
// printf("GetCacheSizePerLogicalCpu returns %zu, adjusted size %zu\n", maxSize, maxTrueSize);
return trueSize ? maxTrueSize : maxSize;
}

Expand Down
91 changes: 72 additions & 19 deletions src/coreclr/pal/src/misc/sysinfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -539,23 +539,31 @@ ReadMemoryValueFromFile(const char* filename, uint64_t* val)
return result;
}

#define CHECK_CACHE_SIZE(CACHE_LEVEL) if (size > cacheSize) { cacheSize = size; cacheLevel = CACHE_LEVEL; }

size_t
PALAPI
PAL_GetLogicalProcessorCacheSizeFromOS()
{
size_t cacheLevel = 0;
size_t cacheSize = 0;
size_t size;

#ifdef _SC_LEVEL1_DCACHE_SIZE
cacheSize = std::max(cacheSize, (size_t)sysconf(_SC_LEVEL1_DCACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL1_DCACHE_SIZE);
CHECK_CACHE_SIZE(1)
#endif
#ifdef _SC_LEVEL2_CACHE_SIZE
cacheSize = std::max(cacheSize, (size_t)sysconf(_SC_LEVEL2_CACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL2_CACHE_SIZE);
CHECK_CACHE_SIZE(2)
#endif
#ifdef _SC_LEVEL3_CACHE_SIZE
cacheSize = std::max(cacheSize, (size_t)sysconf(_SC_LEVEL3_CACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL3_CACHE_SIZE);
CHECK_CACHE_SIZE(3)
#endif
#ifdef _SC_LEVEL4_CACHE_SIZE
cacheSize = std::max(cacheSize, (size_t)sysconf(_SC_LEVEL4_CACHE_SIZE));
size = ( size_t) sysconf(_SC_LEVEL4_CACHE_SIZE);
CHECK_CACHE_SIZE(4)
#endif

#if defined(TARGET_LINUX) && !defined(HOST_ARM) && !defined(HOST_X86)
Expand All @@ -566,25 +574,39 @@ PAL_GetLogicalProcessorCacheSizeFromOS()
// for the platform. Currently musl and arm64 should be only cases to use
// this method to determine cache size.
//
size_t size;

if(ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index0/size", &size))
cacheSize = std::max(cacheSize, size);
if(ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index1/size", &size))
cacheSize = std::max(cacheSize, size);
if(ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index2/size", &size))
cacheSize = std::max(cacheSize, size);
if(ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index3/size", &size))
cacheSize = std::max(cacheSize, size);
if(ReadMemoryValueFromFile("/sys/devices/system/cpu/cpu0/cache/index4/size", &size))
cacheSize = std::max(cacheSize, size);
size_t level;
char path_to_size_file[] = "/sys/devices/system/cpu/cpu0/cache/index-/size";
char path_to_level_file[] = "/sys/devices/system/cpu/cpu0/cache/index-/level";
int index = 40;
_ASSERTE(path_to_size_file[index] == '-');
_ASSERTE(path_to_level_file[index] == '-');

for (int i = 0; i < 5; i++)
{
path_to_size_file[index] = (char)(48 + i);

if (ReadMemoryValueFromFile(path_to_size_file, &size))
{
path_to_level_file[index] = (char)(48 + i);

if (ReadMemoryValueFromFile(path_to_level_file, &level))
{
CHECK_CACHE_SIZE(level)
}
else
{
cacheSize = std::max(cacheSize, size);
}
}
}
}
#endif

#if (defined(HOST_ARM64) || defined(HOST_LOONGARCH64)) && !defined(TARGET_OSX)
if (cacheSize == 0)
{
// It is currently expected to be missing cache size info
// We expect to get the L3 cache size for Arm64 but currently expected to be missing that info
// from most of the machines with an exceptions on some machines.
//
// _SC_LEVEL*_*CACHE_SIZE is not yet present. Work is in progress to enable this for arm64
//
Expand Down Expand Up @@ -621,12 +643,43 @@ PAL_GetLogicalProcessorCacheSizeFromOS()
|| sysctlbyname("hw.l3cachesize", &cacheSizeFromSysctl, &sz, nullptr, 0) == 0
|| sysctlbyname("hw.l2cachesize", &cacheSizeFromSysctl, &sz, nullptr, 0) == 0
|| sysctlbyname("hw.l1dcachesize", &cacheSizeFromSysctl, &sz, nullptr, 0) == 0;

if (success)
{
_ASSERTE(cacheSizeFromSysctl > 0);
cacheSize = (size_t) cacheSizeFromSysctl;
cacheSize = ( size_t) cacheSizeFromSysctl;
}
}
#endif

#if (defined(HOST_ARM64) || defined(HOST_LOONGARCH64)) && !defined(TARGET_OSX)
if (cacheLevel != 3)
{
// We expect to get the L3 cache size for Arm64 but currently expected to be missing that info
// from most of the machines.
// Hence, just use the following heuristics at best depending on the CPU count
// 1 ~ 4 : 4 MB
// 5 ~ 16 : 8 MB
// 17 ~ 64 : 16 MB
// 65+ : 32 MB
DWORD logicalCPUs = PAL_GetLogicalCpuCountFromOS();
if (logicalCPUs < 5)
{
cacheSize = 4;
}
else if (logicalCPUs < 17)
{
cacheSize = 8;
}
else if (logicalCPUs < 65)
{
cacheSize = 16;
}
else
{
cacheSize = 32;
}

cacheSize *= (1024 * 1024);
}
#endif

Expand Down
Loading