github/containerd
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases 2 Wiki Activity

go.mod: github.com/minio/sha256-simd v1.0.1

Browse Source 
https://github.com/minio/sha256-simd/compare/v1.0.0...v1.0.1

Signed-off-by: Akihiro Suda <akihiro.suda.cz@hco.ntt.co.jp>
This commit is contained in:
Akihiro Suda
2023-07-23 02:48:30 +09:00
parent 1f2216cc78
commit 68abb525a5
30 changed files with 1429 additions and 330 deletions
Download Patch File Download Diff File Expand all files Collapse all files

506
vendor/github.com/klauspost/cpuid/v2/cpuid.go generated vendored
View File

@@ -14,7 +14,9 @@ import (
"flag"
"fmt"
"math"
"math/bits"
"os"
"runtime"
"strings"
)
@@ -71,6 +73,7 @@ const (
AMD3DNOW // AMD 3DNOW
AMD3DNOWEXT // AMD 3DNowExt
AMXBF16 // Tile computational operations on BFLOAT16 numbers
AMXFP16 // Tile computational operations on FP16 numbers
AMXINT8 // Tile computational operations on 8-bit integers
AMXTILE // Tile architecture
AVX // AVX functions
@@ -82,6 +85,7 @@ const (
AVX512DQ // AVX-512 Doubleword and Quadword Instructions
AVX512ER // AVX-512 Exponential and Reciprocal Instructions
AVX512F // AVX-512 Foundation
AVX512FP16 // AVX-512 FP16 Instructions
AVX512IFMA // AVX-512 Integer Fused Multiply-Add Instructions
AVX512PF // AVX-512 Prefetch Instructions
AVX512VBMI // AVX-512 Vector Bit Manipulation Instructions
@@ -90,23 +94,51 @@ const (
AVX512VNNI // AVX-512 Vector Neural Network Instructions
AVX512VP2INTERSECT // AVX-512 Intersect for D/Q
AVX512VPOPCNTDQ // AVX-512 Vector Population Count Doubleword and Quadword
AVXSLOW // Indicates the CPU performs 2 128 bit operations instead of one.
AVXIFMA // AVX-IFMA instructions
AVXNECONVERT // AVX-NE-CONVERT instructions
AVXSLOW // Indicates the CPU performs 2 128 bit operations instead of one
AVXVNNI // AVX (VEX encoded) VNNI neural network instructions
AVXVNNIINT8 // AVX-VNNI-INT8 instructions
BHI_CTRL // Branch History Injection and Intra-mode Branch Target Injection / CVE-2022-0001, CVE-2022-0002 / INTEL-SA-00598
BMI1 // Bit Manipulation Instruction Set 1
BMI2 // Bit Manipulation Instruction Set 2
CETIBT // Intel CET Indirect Branch Tracking
CETSS // Intel CET Shadow Stack
CLDEMOTE // Cache Line Demote
CLMUL // Carry-less Multiplication
CLZERO // CLZERO instruction supported
CMOV // i686 CMOV
CMPCCXADD // CMPCCXADD instructions
CMPSB_SCADBS_SHORT // Fast short CMPSB and SCASB
CMPXCHG8 // CMPXCHG8 instruction
CPBOOST // Core Performance Boost
CPPC // AMD: Collaborative Processor Performance Control
CX16 // CMPXCHG16B Instruction
EFER_LMSLE_UNS // AMD: =Core::X86::Msr::EFER[LMSLE] is not supported, and MBZ
ENQCMD // Enqueue Command
ERMS // Enhanced REP MOVSB/STOSB
F16C // Half-precision floating-point conversion
FLUSH_L1D // Flush L1D cache
FMA3 // Intel FMA 3. Does not imply AVX.
FMA4 // Bulldozer FMA4 functions
GFNI // Galois Field New Instructions
FP128 // AMD: When set, the internal FP/SIMD execution datapath is no more than 128-bits wide
FP256 // AMD: When set, the internal FP/SIMD execution datapath is no more than 256-bits wide
FSRM // Fast Short Rep Mov
FXSR // FXSAVE, FXRESTOR instructions, CR4 bit 9
FXSROPT // FXSAVE/FXRSTOR optimizations
GFNI // Galois Field New Instructions. May require other features (AVX, AVX512VL,AVX512F) based on usage.
HLE // Hardware Lock Elision
HRESET // If set CPU supports history reset and the IA32_HRESET_ENABLE MSR
HTT // Hyperthreading (enabled)
HWA // Hardware assert supported. Indicates support for MSRC001_10
HYBRID_CPU // This part has CPUs of more than one type.
HYPERVISOR // This bit has been reserved by Intel & AMD for use by hypervisors
IA32_ARCH_CAP // IA32_ARCH_CAPABILITIES MSR (Intel)
IA32_CORE_CAP // IA32_CORE_CAPABILITIES MSR
IBPB // Indirect Branch Restricted Speculation (IBRS) and Indirect Branch Predictor Barrier (IBPB)
IBRS // AMD: Indirect Branch Restricted Speculation
IBRS_PREFERRED // AMD: IBRS is preferred over software solution
IBRS_PROVIDES_SMP // AMD: IBRS provides Same Mode Protection
IBS // Instruction Based Sampling (AMD)
IBSBRNTRGT // Instruction Based Sampling Feature (AMD)
IBSFETCHSAM // Instruction Based Sampling Feature (AMD)
@@ -116,22 +148,63 @@ const (
IBSOPSAM // Instruction Based Sampling Feature (AMD)
IBSRDWROPCNT // Instruction Based Sampling Feature (AMD)
IBSRIPINVALIDCHK // Instruction Based Sampling Feature (AMD)
IBS_FETCH_CTLX // AMD: IBS fetch control extended MSR supported
IBS_OPDATA4 // AMD: IBS op data 4 MSR supported
IBS_OPFUSE // AMD: Indicates support for IbsOpFuse
IBS_PREVENTHOST // Disallowing IBS use by the host supported
IBS_ZEN4 // AMD: Fetch and Op IBS support IBS extensions added with Zen4
IDPRED_CTRL // IPRED_DIS
INT_WBINVD // WBINVD/WBNOINVD are interruptible.
INVLPGB // NVLPGB and TLBSYNC instruction supported
LAHF // LAHF/SAHF in long mode
LAM // If set, CPU supports Linear Address Masking
LBRVIRT // LBR virtualization
LZCNT // LZCNT instruction
MCAOVERFLOW // MCA overflow recovery support.
MCDT_NO // Processor do not exhibit MXCSR Configuration Dependent Timing behavior and do not need to mitigate it.
MCOMMIT // MCOMMIT instruction supported
MD_CLEAR // VERW clears CPU buffers
MMX // standard MMX
MMXEXT // SSE integer functions or AMD MMX ext
MOVBE // MOVBE instruction (big-endian)
MOVDIR64B // Move 64 Bytes as Direct Store
MOVDIRI // Move Doubleword as Direct Store
MOVSB_ZL // Fast Zero-Length MOVSB
MOVU // AMD: MOVU SSE instructions are more efficient and should be preferred to SSE MOVL/MOVH. MOVUPS is more efficient than MOVLPS/MOVHPS. MOVUPD is more efficient than MOVLPD/MOVHPD
MPX // Intel MPX (Memory Protection Extensions)
MSRIRC // Instruction Retired Counter MSR available
MSRLIST // Read/Write List of Model Specific Registers
MSR_PAGEFLUSH // Page Flush MSR available
NRIPS // Indicates support for NRIP save on VMEXIT
NX // NX (No-Execute) bit
OSXSAVE // XSAVE enabled by OS
PCONFIG // PCONFIG for Intel Multi-Key Total Memory Encryption
POPCNT // POPCNT instruction
PPIN // AMD: Protected Processor Inventory Number support. Indicates that Protected Processor Inventory Number (PPIN) capability can be enabled
PREFETCHI // PREFETCHIT0/1 instructions
PSFD // Predictive Store Forward Disable
RDPRU // RDPRU instruction supported
RDRAND // RDRAND instruction is available
RDSEED // RDSEED instruction is available
RDTSCP // RDTSCP Instruction
RRSBA_CTRL // Restricted RSB Alternate
RTM // Restricted Transactional Memory
RTM_ALWAYS_ABORT // Indicates that the loaded microcode is forcing RTM abort.
SERIALIZE // Serialize Instruction Execution
SEV // AMD Secure Encrypted Virtualization supported
SEV_64BIT // AMD SEV guest execution only allowed from a 64-bit host
SEV_ALTERNATIVE // AMD SEV Alternate Injection supported
SEV_DEBUGSWAP // Full debug state swap supported for SEV-ES guests
SEV_ES // AMD SEV Encrypted State supported
SEV_RESTRICTED // AMD SEV Restricted Injection supported
SEV_SNP // AMD SEV Secure Nested Paging supported
SGX // Software Guard Extensions
SGXLC // Software Guard Extensions Launch Control
SHA // Intel SHA Extensions
SME // AMD Secure Memory Encryption supported
SME_COHERENT // AMD Hardware cache coherency across encryption domains enforced
SPEC_CTRL_SSBD // Speculative Store Bypass Disable
SRBDS_CTRL // SRBDS mitigation MSR available
SSE // SSE functions
SSE2 // P4 SSE functions
SSE3 // Prescott SSE3 functions
@@ -140,14 +213,42 @@ const (
SSE4A // AMD Barcelona microarchitecture SSE4a instructions
SSSE3 // Conroe SSSE3 functions
STIBP // Single Thread Indirect Branch Predictors
STIBP_ALWAYSON // AMD: Single Thread Indirect Branch Prediction Mode has Enhanced Performance and may be left Always On
STOSB_SHORT // Fast short STOSB
SUCCOR // Software uncorrectable error containment and recovery capability.
SVM // AMD Secure Virtual Machine
SVMDA // Indicates support for the SVM decode assists.
SVMFBASID // SVM, Indicates that TLB flush events, including CR3 writes and CR4.PGE toggles, flush only the current ASID's TLB entries. Also indicates support for the extended VMCBTLB_Control
SVML // AMD SVM lock. Indicates support for SVM-Lock.
SVMNP // AMD SVM nested paging
SVMPF // SVM pause intercept filter. Indicates support for the pause intercept filter
SVMPFT // SVM PAUSE filter threshold. Indicates support for the PAUSE filter cycle count threshold
SYSCALL // System-Call Extension (SCE): SYSCALL and SYSRET instructions.
SYSEE // SYSENTER and SYSEXIT instructions
TBM // AMD Trailing Bit Manipulation
TDX_GUEST // Intel Trust Domain Extensions Guest
TLB_FLUSH_NESTED // AMD: Flushing includes all the nested translations for guest translations
TME // Intel Total Memory Encryption. The following MSRs are supported: IA32_TME_CAPABILITY, IA32_TME_ACTIVATE, IA32_TME_EXCLUDE_MASK, and IA32_TME_EXCLUDE_BASE.
TOPEXT // TopologyExtensions: topology extensions support. Indicates support for CPUID Fn8000_001D_EAX_x[N:0]-CPUID Fn8000_001E_EDX.
TSCRATEMSR // MSR based TSC rate control. Indicates support for MSR TSC ratio MSRC000_0104
TSXLDTRK // Intel TSX Suspend Load Address Tracking
VAES // Vector AES
VAES // Vector AES. AVX(512) versions requires additional checks.
VMCBCLEAN // VMCB clean bits. Indicates support for VMCB clean bits.
VMPL // AMD VM Permission Levels supported
VMSA_REGPROT // AMD VMSA Register Protection supported
VMX // Virtual Machine Extensions
VPCLMULQDQ // Carry-Less Multiplication Quadword
VPCLMULQDQ // Carry-Less Multiplication Quadword. Requires AVX for 3 register versions.
VTE // AMD Virtual Transparent Encryption supported
WAITPKG // TPAUSE, UMONITOR, UMWAIT
WBNOINVD // Write Back and Do Not Invalidate Cache
WRMSRNS // Non-Serializing Write to Model Specific Register
X87 // FPU
XGETBV1 // Supports XGETBV with ECX = 1
XOP // Bulldozer XOP functions
XSAVE // XSAVE, XRESTOR, XSETBV, XGETBV
XSAVEC // Supports XSAVEC and the compacted form of XRSTOR.
XSAVEOPT // XSAVEOPT available
XSAVES // Supports XSAVES/XRSTORS and IA32_XSS
// ARM features:
AESARM // AES instructions
@@ -174,7 +275,6 @@ const (
SM3 // SM3 instructions
SM4 // SM4 instructions
SVE // Scalable Vector Extension
// Keep it last. It automatically defines the size of []flagSet
lastID
@@ -192,8 +292,10 @@ type CPUInfo struct {
LogicalCores int // Number of physical cores times threads that can run on each core through the use of hyperthreading. Will be 0 if undetectable.
Family int // CPU family number
Model int // CPU model number
Stepping int // CPU stepping info
CacheLine int // Cache line size in bytes. Will be 0 if undetectable.
Hz int64 // Clock speed, if known, 0 otherwise
Hz int64 // Clock speed, if known, 0 otherwise. Will attempt to contain base clock speed.
BoostFreq int64 // Max clock speed, if known, 0 otherwise
Cache struct {
L1I int // L1 Instruction Cache (per core or shared). Will be -1 if undetected
L1D int // L1 Data Cache (per core or shared). Will be -1 if undetected
@@ -209,6 +311,7 @@ var cpuid func(op uint32) (eax, ebx, ecx, edx uint32)
var cpuidex func(op, op2 uint32) (eax, ebx, ecx, edx uint32)
var xgetbv func(index uint32) (eax, edx uint32)
var rdtscpAsm func() (eax, ebx, ecx, edx uint32)
var darwinHasAVX512 = func() bool { return false }
// CPU contains information about the CPU as detected on startup,
// or when Detect last was called.
@@ -292,10 +395,66 @@ func (c CPUInfo) Supports(ids ...FeatureID) bool {
// Has allows for checking a single feature.
// Should be inlined by the compiler.
func (c CPUInfo) Has(id FeatureID) bool {
func (c *CPUInfo) Has(id FeatureID) bool {
return c.featureSet.inSet(id)
}
// AnyOf returns whether the CPU supports one or more of the requested features.
func (c CPUInfo) AnyOf(ids ...FeatureID) bool {
for _, id := range ids {
if c.featureSet.inSet(id) {
return true
}
}
return false
}
// Features contains several features combined for a fast check using
// CpuInfo.HasAll
type Features *flagSet
// CombineFeatures allows to combine several features for a close to constant time lookup.
func CombineFeatures(ids ...FeatureID) Features {
var v flagSet
for _, id := range ids {
v.set(id)
}
return &v
}
func (c *CPUInfo) HasAll(f Features) bool {
return c.featureSet.hasSetP(f)
}
// https://en.wikipedia.org/wiki/X86-64#Microarchitecture_levels
var oneOfLevel = CombineFeatures(SYSEE, SYSCALL)
var level1Features = CombineFeatures(CMOV, CMPXCHG8, X87, FXSR, MMX, SSE, SSE2)
var level2Features = CombineFeatures(CMOV, CMPXCHG8, X87, FXSR, MMX, SSE, SSE2, CX16, LAHF, POPCNT, SSE3, SSE4, SSE42, SSSE3)
var level3Features = CombineFeatures(CMOV, CMPXCHG8, X87, FXSR, MMX, SSE, SSE2, CX16, LAHF, POPCNT, SSE3, SSE4, SSE42, SSSE3, AVX, AVX2, BMI1, BMI2, F16C, FMA3, LZCNT, MOVBE, OSXSAVE)
var level4Features = CombineFeatures(CMOV, CMPXCHG8, X87, FXSR, MMX, SSE, SSE2, CX16, LAHF, POPCNT, SSE3, SSE4, SSE42, SSSE3, AVX, AVX2, BMI1, BMI2, F16C, FMA3, LZCNT, MOVBE, OSXSAVE, AVX512F, AVX512BW, AVX512CD, AVX512DQ, AVX512VL)
// X64Level returns the microarchitecture level detected on the CPU.
// If features are lacking or non x64 mode, 0 is returned.
// See https://en.wikipedia.org/wiki/X86-64#Microarchitecture_levels
func (c CPUInfo) X64Level() int {
if !c.featureSet.hasOneOf(oneOfLevel) {
return 0
}
if c.featureSet.hasSetP(level4Features) {
return 4
}
if c.featureSet.hasSetP(level3Features) {
return 3
}
if c.featureSet.hasSetP(level2Features) {
return 2
}
if c.featureSet.hasSetP(level1Features) {
return 1
}
return 0
}
// Disable will disable one or several features.
func (c *CPUInfo) Disable(ids ...FeatureID) bool {
for _, id := range ids {
@@ -318,11 +477,10 @@ func (c CPUInfo) IsVendor(v Vendor) bool {
return c.VendorID == v
}
// FeatureSet returns all available features as strings.
func (c CPUInfo) FeatureSet() []string {
s := make([]string, 0)
for _, f := range c.featureSet.Strings() {
s = append(s, f)
}
s := make([]string, 0, c.featureSet.nEnabled())
s = append(s, c.featureSet.Strings()...)
return s
}
@@ -361,25 +519,42 @@ func (c CPUInfo) LogicalCPU() int {
return int(ebx >> 24)
}
// hertz tries to compute the clock speed of the CPU. If leaf 15 is
// frequencies tries to compute the clock speed of the CPU. If leaf 15 is
// supported, use it, otherwise parse the brand string. Yes, really.
func hertz(model string) int64 {
func (c *CPUInfo) frequencies() {
c.Hz, c.BoostFreq = 0, 0
mfi := maxFunctionID()
if mfi >= 0x15 {
eax, ebx, ecx, _ := cpuid(0x15)
if eax != 0 && ebx != 0 && ecx != 0 {
return int64((int64(ecx) * int64(ebx)) / int64(eax))
c.Hz = (int64(ecx) * int64(ebx)) / int64(eax)
}
}
if mfi >= 0x16 {
a, b, _, _ := cpuid(0x16)
// Base...
if a&0xffff > 0 {
c.Hz = int64(a&0xffff) * 1_000_000
}
// Boost...
if b&0xffff > 0 {
c.BoostFreq = int64(b&0xffff) * 1_000_000
}
}
if c.Hz > 0 {
return
}
// computeHz determines the official rated speed of a CPU from its brand
// string. This insanity is *actually the official documented way to do
// this according to Intel*, prior to leaf 0x15 existing. The official
// documentation only shows this working for exactly `x.xx` or `xxxx`
// cases, e.g., `2.50GHz` or `1300MHz`; this parser will accept other
// sizes.
model := c.BrandName
hz := strings.LastIndex(model, "Hz")
if hz < 3 {
return 0
return
}
var multiplier int64
switch model[hz-1] {
@@ -391,7 +566,7 @@ func hertz(model string) int64 {
multiplier = 1000 * 1000 * 1000 * 1000
}
if multiplier == 0 {
return 0
return
}
freq := int64(0)
divisor := int64(0)
@@ -403,21 +578,22 @@ func hertz(model string) int64 {
decimalShift *= 10
} else if model[i] == '.' {
if divisor != 0 {
return 0
return
}
divisor = decimalShift
} else {
return 0
return
}
}
// we didn't find a space
if i < 0 {
return 0
return
}
if divisor != 0 {
return (freq * multiplier) / divisor
c.Hz = (freq * multiplier) / divisor
return
}
return freq * multiplier
c.Hz = freq * multiplier
}
// VM Will return true if the cpu id indicates we are in
@@ -437,7 +613,7 @@ const flagMask = flagBits - 1
// flagSet contains detected cpu features and characteristics in an array of flags
type flagSet [(lastID + flagMask) / flagBits]flags
func (s flagSet) inSet(feat FeatureID) bool {
func (s *flagSet) inSet(feat FeatureID) bool {
return s[feat>>flagBitsLog2]&(1<<(feat&flagMask)) != 0
}
@@ -466,6 +642,52 @@ func (s *flagSet) or(other flagSet) {
}
}
// hasSet returns whether all features are present.
func (s *flagSet) hasSet(other flagSet) bool {
for i, v := range other[:] {
if s[i]&v != v {
return false
}
}
return true
}
// hasSet returns whether all features are present.
func (s *flagSet) hasSetP(other *flagSet) bool {
for i, v := range other[:] {
if s[i]&v != v {
return false
}
}
return true
}
// hasOneOf returns whether one or more features are present.
func (s *flagSet) hasOneOf(other *flagSet) bool {
for i, v := range other[:] {
if s[i]&v != 0 {
return true
}
}
return false
}
// nEnabled will return the number of enabled flags.
func (s *flagSet) nEnabled() (n int) {
for _, v := range s[:] {
n += bits.OnesCount64(uint64(v))
}
return n
}
func flagSetWith(feat ...FeatureID) flagSet {
var res flagSet
for _, f := range feat {
res.set(f)
}
return res
}
// ParseFeature will parse the string and return the ID of the matching feature.
// Will return UNKNOWN if not found.
func ParseFeature(s string) FeatureID {
@@ -546,7 +768,7 @@ func threadsPerCore() int {
if vend == AMD {
// Workaround for AMD returning 0, assume 2 if >= Zen 2
// It will be more correct than not.
fam, _ := familyModel()
fam, _, _ := familyModel()
_, _, _, d := cpuid(1)
if (d&(1<<28)) != 0 && fam >= 23 {
return 2
@@ -584,14 +806,27 @@ func logicalCores() int {
}
}
func familyModel() (int, int) {
func familyModel() (family, model, stepping int) {
if maxFunctionID() < 0x1 {
return 0, 0
return 0, 0, 0
}
eax, _, _, _ := cpuid(1)
family := ((eax >> 8) & 0xf) + ((eax >> 20) & 0xff)
model := ((eax >> 4) & 0xf) + ((eax >> 12) & 0xf0)
return int(family), int(model)
// If BaseFamily[3:0] is less than Fh then ExtendedFamily[7:0] is reserved and Family is equal to BaseFamily[3:0].
family = int((eax >> 8) & 0xf)
extFam := family == 0x6 // Intel is 0x6, needs extended model.
if family == 0xf {
// Add ExtFamily
family += int((eax >> 20) & 0xff)
extFam = true
}
// If BaseFamily[3:0] is less than 0Fh then ExtendedModel[3:0] is reserved and Model is equal to BaseModel[3:0].
model = int((eax >> 4) & 0xf)
if extFam {
// Add ExtModel
model += int((eax >> 12) & 0xf0)
}
stepping = int(eax & 0xf)
return family, model, stepping
}
func physicalCores() int {
@@ -675,6 +910,7 @@ func (c *CPUInfo) cacheSize() {
if maxFunctionID() < 4 {
return
}
c.Cache.L1I, c.Cache.L1D, c.Cache.L2, c.Cache.L3 = 0, 0, 0, 0
for i := uint32(0); ; i++ {
eax, ebx, ecx, _ := cpuidex(4, i)
cacheType := eax & 15
@@ -725,9 +961,14 @@ func (c *CPUInfo) cacheSize() {
c.Cache.L2 = int(((ecx >> 16) & 0xFFFF) * 1024)
// CPUID Fn8000_001D_EAX_x[N:0] Cache Properties
if maxExtendedFunction() < 0x8000001D {
if maxExtendedFunction() < 0x8000001D || !c.Has(TOPEXT) {
return
}
// Xen Hypervisor is buggy and returns the same entry no matter ECX value.
// Hack: When we encounter the same entry 100 times we break.
nSame := 0
var last uint32
for i := uint32(0); i < math.MaxUint32; i++ {
eax, ebx, ecx, _ := cpuidex(0x8000001D, i)
@@ -743,6 +984,16 @@ func (c *CPUInfo) cacheSize() {
return
}
// Check for the same value repeated.
comb := eax ^ ebx ^ ecx
if comb == last {
nSame++
if nSame == 100 {
return
}
}
last = comb
switch level {
case 1:
switch typ {
@@ -767,8 +1018,6 @@ func (c *CPUInfo) cacheSize() {
}
}
}
return
}
type SGXEPCSection struct {
@@ -829,21 +1078,26 @@ func support() flagSet {
if mfi < 0x1 {
return fs
}
family, model := familyModel()
family, model, _ := familyModel()
_, _, c, d := cpuid(1)
fs.setIf((d&(1<<0)) != 0, X87)
fs.setIf((d&(1<<8)) != 0, CMPXCHG8)
fs.setIf((d&(1<<11)) != 0, SYSEE)
fs.setIf((d&(1<<15)) != 0, CMOV)
fs.setIf((d&(1<<23)) != 0, MMX)
fs.setIf((d&(1<<25)) != 0, MMXEXT)
fs.setIf((d&(1<<24)) != 0, FXSR)
fs.setIf((d&(1<<25)) != 0, FXSROPT)
fs.setIf((d&(1<<25)) != 0, SSE)
fs.setIf((d&(1<<26)) != 0, SSE2)
fs.setIf((c&1) != 0, SSE3)
fs.setIf((c&(1<<5)) != 0, VMX)
fs.setIf((c&0x00000200) != 0, SSSE3)
fs.setIf((c&0x00080000) != 0, SSE4)
fs.setIf((c&0x00100000) != 0, SSE42)
fs.setIf((c&(1<<9)) != 0, SSSE3)
fs.setIf((c&(1<<19)) != 0, SSE4)
fs.setIf((c&(1<<20)) != 0, SSE42)
fs.setIf((c&(1<<25)) != 0, AESNI)
fs.setIf((c&(1<<1)) != 0, CLMUL)
fs.setIf(c&(1<<22) != 0, MOVBE)
fs.setIf(c&(1<<23) != 0, POPCNT)
fs.setIf(c&(1<<30) != 0, RDRAND)
@@ -859,6 +1113,8 @@ func support() flagSet {
if vend == AMD && (d&(1<<28)) != 0 && mfi >= 4 {
fs.setIf(threadsPerCore() > 1, HTT)
}
fs.setIf(c&1<<26 != 0, XSAVE)
fs.setIf(c&1<<27 != 0, OSXSAVE)
// Check XGETBV/XSAVE (26), OXSAVE (27) and AVX (28) bits
const avxCheck = 1<<26 | 1<<27 | 1<<28
if c&avxCheck == avxCheck {
@@ -884,7 +1140,6 @@ func support() flagSet {
// Check AVX2, AVX2 requires OS support, but BMI1/2 don't.
if mfi >= 7 {
_, ebx, ecx, edx := cpuidex(7, 0)
eax1, _, _, _ := cpuidex(7, 1)
if fs.inSet(AVX) && (ebx&0x00000020) != 0 {
fs.set(AVX2)
}
@@ -901,18 +1156,52 @@ func support() flagSet {
fs.setIf(ebx&(1<<18) != 0, RDSEED)
fs.setIf(ebx&(1<<19) != 0, ADX)
fs.setIf(ebx&(1<<29) != 0, SHA)
// CPUID.(EAX=7, ECX=0).ECX
fs.setIf(ecx&(1<<5) != 0, WAITPKG)
fs.setIf(ecx&(1<<7) != 0, CETSS)
fs.setIf(ecx&(1<<8) != 0, GFNI)
fs.setIf(ecx&(1<<9) != 0, VAES)
fs.setIf(ecx&(1<<10) != 0, VPCLMULQDQ)
fs.setIf(ecx&(1<<13) != 0, TME)
fs.setIf(ecx&(1<<25) != 0, CLDEMOTE)
fs.setIf(ecx&(1<<27) != 0, MOVDIRI)
fs.setIf(ecx&(1<<28) != 0, MOVDIR64B)
fs.setIf(ecx&(1<<29) != 0, ENQCMD)
fs.setIf(ecx&(1<<30) != 0, SGXLC)
// CPUID.(EAX=7, ECX=0).EDX
fs.setIf(edx&(1<<4) != 0, FSRM)
fs.setIf(edx&(1<<9) != 0, SRBDS_CTRL)
fs.setIf(edx&(1<<10) != 0, MD_CLEAR)
fs.setIf(edx&(1<<11) != 0, RTM_ALWAYS_ABORT)
fs.setIf(edx&(1<<14) != 0, SERIALIZE)
fs.setIf(edx&(1<<15) != 0, HYBRID_CPU)
fs.setIf(edx&(1<<16) != 0, TSXLDTRK)
fs.setIf(edx&(1<<18) != 0, PCONFIG)
fs.setIf(edx&(1<<20) != 0, CETIBT)
fs.setIf(edx&(1<<26) != 0, IBPB)
fs.setIf(edx&(1<<27) != 0, STIBP)
fs.setIf(edx&(1<<28) != 0, FLUSH_L1D)
fs.setIf(edx&(1<<29) != 0, IA32_ARCH_CAP)
fs.setIf(edx&(1<<30) != 0, IA32_CORE_CAP)
fs.setIf(edx&(1<<31) != 0, SPEC_CTRL_SSBD)
// CPUID.(EAX=7, ECX=1).EAX
eax1, _, _, edx1 := cpuidex(7, 1)
fs.setIf(fs.inSet(AVX) && eax1&(1<<4) != 0, AVXVNNI)
fs.setIf(eax1&(1<<7) != 0, CMPCCXADD)
fs.setIf(eax1&(1<<10) != 0, MOVSB_ZL)
fs.setIf(eax1&(1<<11) != 0, STOSB_SHORT)
fs.setIf(eax1&(1<<12) != 0, CMPSB_SCADBS_SHORT)
fs.setIf(eax1&(1<<22) != 0, HRESET)
fs.setIf(eax1&(1<<23) != 0, AVXIFMA)
fs.setIf(eax1&(1<<26) != 0, LAM)
// CPUID.(EAX=7, ECX=1).EDX
fs.setIf(edx1&(1<<4) != 0, AVXVNNIINT8)
fs.setIf(edx1&(1<<5) != 0, AVXNECONVERT)
fs.setIf(edx1&(1<<14) != 0, PREFETCHI)
// Only detect AVX-512 features if XGETBV is supported
if c&((1<<26)|(1<<27)) == (1<<26)|(1<<27) {
@@ -922,7 +1211,11 @@ func support() flagSet {
// Verify that XCR0[7:5] = ‘111b’ (OPMASK state, upper 256-bit of ZMM0-ZMM15 and
// ZMM16-ZMM31 state are enabled by OS)
/// and that XCR0[2:1] = ‘11b’ (XMM state and YMM state are enabled by OS).
if (eax>>5)&7 == 7 && (eax>>1)&3 == 3 {
hasAVX512 := (eax>>5)&7 == 7 && (eax>>1)&3 == 3
if runtime.GOOS == "darwin" {
hasAVX512 = fs.inSet(AVX) && darwinHasAVX512()
}
if hasAVX512 {
fs.setIf(ebx&(1<<16) != 0, AVX512F)
fs.setIf(ebx&(1<<17) != 0, AVX512DQ)
fs.setIf(ebx&(1<<21) != 0, AVX512IFMA)
@@ -934,49 +1227,137 @@ func support() flagSet {
// ecx
fs.setIf(ecx&(1<<1) != 0, AVX512VBMI)
fs.setIf(ecx&(1<<6) != 0, AVX512VBMI2)
fs.setIf(ecx&(1<<8) != 0, GFNI)
fs.setIf(ecx&(1<<9) != 0, VAES)
fs.setIf(ecx&(1<<10) != 0, VPCLMULQDQ)
fs.setIf(ecx&(1<<11) != 0, AVX512VNNI)
fs.setIf(ecx&(1<<12) != 0, AVX512BITALG)
fs.setIf(ecx&(1<<14) != 0, AVX512VPOPCNTDQ)
// edx
fs.setIf(edx&(1<<8) != 0, AVX512VP2INTERSECT)
fs.setIf(edx&(1<<22) != 0, AMXBF16)
fs.setIf(edx&(1<<23) != 0, AVX512FP16)
fs.setIf(edx&(1<<24) != 0, AMXTILE)
fs.setIf(edx&(1<<25) != 0, AMXINT8)
// eax1 = CPUID.(EAX=7, ECX=1).EAX
fs.setIf(eax1&(1<<5) != 0, AVX512BF16)
fs.setIf(eax1&(1<<19) != 0, WRMSRNS)
fs.setIf(eax1&(1<<21) != 0, AMXFP16)
fs.setIf(eax1&(1<<27) != 0, MSRLIST)
}
}
// CPUID.(EAX=7, ECX=2)
_, _, _, edx = cpuidex(7, 2)
fs.setIf(edx&(1<<0) != 0, PSFD)
fs.setIf(edx&(1<<1) != 0, IDPRED_CTRL)
fs.setIf(edx&(1<<2) != 0, RRSBA_CTRL)
fs.setIf(edx&(1<<4) != 0, BHI_CTRL)
fs.setIf(edx&(1<<5) != 0, MCDT_NO)
}
// Processor Extended State Enumeration Sub-leaf (EAX = 0DH, ECX = 1)
// EAX
// Bit 00: XSAVEOPT is available.
// Bit 01: Supports XSAVEC and the compacted form of XRSTOR if set.
// Bit 02: Supports XGETBV with ECX = 1 if set.
// Bit 03: Supports XSAVES/XRSTORS and IA32_XSS if set.
// Bits 31 - 04: Reserved.
// EBX
// Bits 31 - 00: The size in bytes of the XSAVE area containing all states enabled by XCRO | IA32_XSS.
// ECX
// Bits 31 - 00: Reports the supported bits of the lower 32 bits of the IA32_XSS MSR. IA32_XSS[n] can be set to 1 only if ECX[n] is 1.
// EDX?
// Bits 07 - 00: Used for XCR0. Bit 08: PT state. Bit 09: Used for XCR0. Bits 12 - 10: Reserved. Bit 13: HWP state. Bits 31 - 14: Reserved.
if mfi >= 0xd {
if fs.inSet(XSAVE) {
eax, _, _, _ := cpuidex(0xd, 1)
fs.setIf(eax&(1<<0) != 0, XSAVEOPT)
fs.setIf(eax&(1<<1) != 0, XSAVEC)
fs.setIf(eax&(1<<2) != 0, XGETBV1)
fs.setIf(eax&(1<<3) != 0, XSAVES)
}
}
if maxExtendedFunction() >= 0x80000001 {
_, _, c, d := cpuid(0x80000001)
if (c & (1 << 5)) != 0 {
fs.set(LZCNT)
fs.set(POPCNT)
}
fs.setIf((c&(1<<10)) != 0, IBS)
fs.setIf((d&(1<<31)) != 0, AMD3DNOW)
fs.setIf((d&(1<<30)) != 0, AMD3DNOWEXT)
fs.setIf((d&(1<<23)) != 0, MMX)
fs.setIf((d&(1<<22)) != 0, MMXEXT)
// ECX
fs.setIf((c&(1<<0)) != 0, LAHF)
fs.setIf((c&(1<<2)) != 0, SVM)
fs.setIf((c&(1<<6)) != 0, SSE4A)
fs.setIf((c&(1<<10)) != 0, IBS)
fs.setIf((c&(1<<22)) != 0, TOPEXT)
// EDX
fs.setIf(d&(1<<11) != 0, SYSCALL)
fs.setIf(d&(1<<20) != 0, NX)
fs.setIf(d&(1<<22) != 0, MMXEXT)
fs.setIf(d&(1<<23) != 0, MMX)
fs.setIf(d&(1<<24) != 0, FXSR)
fs.setIf(d&(1<<25) != 0, FXSROPT)
fs.setIf(d&(1<<27) != 0, RDTSCP)
fs.setIf(d&(1<<30) != 0, AMD3DNOWEXT)
fs.setIf(d&(1<<31) != 0, AMD3DNOW)
/* XOP and FMA4 use the AVX instruction coding scheme, so they can't be
* used unless the OS has AVX support. */
if fs.inSet(AVX) {
fs.setIf((c&0x00000800) != 0, XOP)
fs.setIf((c&0x00010000) != 0, FMA4)
fs.setIf((c&(1<<11)) != 0, XOP)
fs.setIf((c&(1<<16)) != 0, FMA4)
}
}
if maxExtendedFunction() >= 0x80000007 {
_, b, _, d := cpuid(0x80000007)
fs.setIf((b&(1<<0)) != 0, MCAOVERFLOW)
fs.setIf((b&(1<<1)) != 0, SUCCOR)
fs.setIf((b&(1<<2)) != 0, HWA)
fs.setIf((d&(1<<9)) != 0, CPBOOST)
}
if maxExtendedFunction() >= 0x80000008 {
_, b, _, _ := cpuid(0x80000008)
fs.setIf(b&(1<<28) != 0, PSFD)
fs.setIf(b&(1<<27) != 0, CPPC)
fs.setIf(b&(1<<24) != 0, SPEC_CTRL_SSBD)
fs.setIf(b&(1<<23) != 0, PPIN)
fs.setIf(b&(1<<21) != 0, TLB_FLUSH_NESTED)
fs.setIf(b&(1<<20) != 0, EFER_LMSLE_UNS)
fs.setIf(b&(1<<19) != 0, IBRS_PROVIDES_SMP)
fs.setIf(b&(1<<18) != 0, IBRS_PREFERRED)
fs.setIf(b&(1<<17) != 0, STIBP_ALWAYSON)
fs.setIf(b&(1<<15) != 0, STIBP)
fs.setIf(b&(1<<14) != 0, IBRS)
fs.setIf((b&(1<<13)) != 0, INT_WBINVD)
fs.setIf(b&(1<<12) != 0, IBPB)
fs.setIf((b&(1<<9)) != 0, WBNOINVD)
fs.setIf((b&(1<<8)) != 0, MCOMMIT)
fs.setIf((b&(1<<4)) != 0, RDPRU)
fs.setIf((b&(1<<3)) != 0, INVLPGB)
fs.setIf((b&(1<<1)) != 0, MSRIRC)
fs.setIf((b&(1<<0)) != 0, CLZERO)
}
if fs.inSet(SVM) && maxExtendedFunction() >= 0x8000000A {
_, _, _, edx := cpuid(0x8000000A)
fs.setIf((edx>>0)&1 == 1, SVMNP)
fs.setIf((edx>>1)&1 == 1, LBRVIRT)
fs.setIf((edx>>2)&1 == 1, SVML)
fs.setIf((edx>>3)&1 == 1, NRIPS)
fs.setIf((edx>>4)&1 == 1, TSCRATEMSR)
fs.setIf((edx>>5)&1 == 1, VMCBCLEAN)
fs.setIf((edx>>6)&1 == 1, SVMFBASID)
fs.setIf((edx>>7)&1 == 1, SVMDA)
fs.setIf((edx>>10)&1 == 1, SVMPF)
fs.setIf((edx>>12)&1 == 1, SVMPFT)
}
if maxExtendedFunction() >= 0x8000001a {
eax, _, _, _ := cpuid(0x8000001a)
fs.setIf((eax>>0)&1 == 1, FP128)
fs.setIf((eax>>1)&1 == 1, MOVU)
fs.setIf((eax>>2)&1 == 1, FP256)
}
if maxExtendedFunction() >= 0x8000001b && fs.inSet(IBS) {
@@ -989,6 +1370,35 @@ func support() flagSet {
fs.setIf((eax>>5)&1 == 1, IBSBRNTRGT)
fs.setIf((eax>>6)&1 == 1, IBSOPCNTEXT)
fs.setIf((eax>>7)&1 == 1, IBSRIPINVALIDCHK)
fs.setIf((eax>>8)&1 == 1, IBS_OPFUSE)
fs.setIf((eax>>9)&1 == 1, IBS_FETCH_CTLX)
fs.setIf((eax>>10)&1 == 1, IBS_OPDATA4) // Doc says "Fixed,0. IBS op data 4 MSR supported", but assuming they mean 1.
fs.setIf((eax>>11)&1 == 1, IBS_ZEN4)
}
if maxExtendedFunction() >= 0x8000001f && vend == AMD {
a, _, _, _ := cpuid(0x8000001f)
fs.setIf((a>>0)&1 == 1, SME)
fs.setIf((a>>1)&1 == 1, SEV)
fs.setIf((a>>2)&1 == 1, MSR_PAGEFLUSH)
fs.setIf((a>>3)&1 == 1, SEV_ES)
fs.setIf((a>>4)&1 == 1, SEV_SNP)
fs.setIf((a>>5)&1 == 1, VMPL)
fs.setIf((a>>10)&1 == 1, SME_COHERENT)
fs.setIf((a>>11)&1 == 1, SEV_64BIT)
fs.setIf((a>>12)&1 == 1, SEV_RESTRICTED)
fs.setIf((a>>13)&1 == 1, SEV_ALTERNATIVE)
fs.setIf((a>>14)&1 == 1, SEV_DEBUGSWAP)
fs.setIf((a>>15)&1 == 1, IBS_PREVENTHOST)
fs.setIf((a>>16)&1 == 1, VTE)
fs.setIf((a>>24)&1 == 1, VMSA_REGPROT)
}
if mfi >= 0x21 {
// Intel Trusted Domain Extensions Guests have their own cpuid leaf (0x21).
_, ebx, ecx, edx := cpuid(0x21)
identity := string(valAsString(ebx, edx, ecx))
fs.setIf(identity == "IntelTDX ", TDX_GUEST)
}
return fs