Skip to content

Commit

Permalink
Add Intel apx, avx10, keylocker (#134)
Browse files Browse the repository at this point in the history 
* Add Intel APX/AVX10/Keylocker
* Bump CI
* Add to command as well.

Only KEYLOCKER has been released in the wild, AFAIK.
  • Loading branch information
@klauspost
klauspost committed Jul 25, 2023
1 parent 1af2d99 commit 666b9f5
Show file tree
Hide file tree
Showing 7 changed files with 259 additions and 206 deletions.
4 changes: 2 additions & 2 deletions .github/workflows/go.yml
View file
Expand Up @@ -10,7 +10,7 @@ jobs:
build:
strategy:
matrix:
go-version: [1.17.x, 1.18.x, 1.19.x]
go-version: [1.18.x, 1.19.x, 1.20.x]
os: [ubuntu-latest, macos-latest, windows-latest]
runs-on: ubuntu-latest
steps:
Expand Down Expand Up @@ -38,7 +38,7 @@ jobs:
- name: Set up Go
uses: actions/setup-go@v2
with:
go-version: 1.19.x
go-version: 1.20.x

- name: Checkout code
uses: actions/checkout@v2
Expand Down
9 changes: 8 additions & 1 deletion README.md
View file
Expand Up @@ -285,7 +285,12 @@ Exit Code 1
| AMXINT8 | Tile computational operations on 8-bit integers |
| AMXFP16 | Tile computational operations on FP16 numbers |
| AMXTILE | Tile architecture |
| APX_F | Intel APX |
| AVX | AVX functions |
| AVX10 | If set the Intel AVX10 Converged Vector ISA is supported |
| AVX10_128 | If set indicates that AVX10 128-bit vector support is present |
| AVX10_256 | If set indicates that AVX10 256-bit vector support is present |
| AVX10_512 | If set indicates that AVX10 512-bit vector support is present |
| AVX2 | AVX2 functions |
| AVX512BF16 | AVX-512 BFLOAT16 Instructions |
| AVX512BITALG | AVX-512 Bit Algorithms |
Expand Down Expand Up @@ -365,6 +370,8 @@ Exit Code 1
| IDPRED_CTRL | IPRED_DIS |
| INT_WBINVD | WBINVD/WBNOINVD are interruptible. |
| INVLPGB | NVLPGB and TLBSYNC instruction supported |
| KEYLOCKER | Key locker |
| KEYLOCKERW | Key locker wide |
| LAHF | LAHF/SAHF in long mode |
| LAM | If set, CPU supports Linear Address Masking |
| LBRVIRT | LBR virtualization |
Expand All @@ -380,7 +387,7 @@ Exit Code 1
| MOVDIRI | Move Doubleword as Direct Store |
| MOVSB_ZL | Fast Zero-Length MOVSB |
| MPX | Intel MPX (Memory Protection Extensions) |
| MOVU | 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 |
| MOVU | 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 |
| MSRIRC | Instruction Retired Counter MSR available |
| MSRLIST | Read/Write List of Model Specific Registers |
| MSR_PAGEFLUSH | Page Flush MSR available |
Expand Down
3 changes: 3 additions & 0 deletions cmd/cpuid/main.go
View file
Expand Up @@ -65,6 +65,9 @@ func main() {
fmt.Println("CPU Family", cpuid.CPU.Family, "Model:", cpuid.CPU.Model, "Stepping:", cpuid.CPU.Stepping)
fmt.Println("Features:", strings.Join(cpuid.CPU.FeatureSet(), ","))
fmt.Println("Microarchitecture level:", cpuid.CPU.X64Level())
if cpuid.CPU.AVX10Level > 0 {
fmt.Println("AVX10 level:", cpuid.CPU.AVX10Level)
}
fmt.Println("Cacheline bytes:", cpuid.CPU.CacheLine)
fmt.Println("L1 Instruction Cache:", cpuid.CPU.Cache.L1I, "bytes")
fmt.Println("L1 Data Cache:", cpuid.CPU.Cache.L1D, "bytes")
Expand Down
38 changes: 35 additions & 3 deletions cpuid.go
View file
Expand Up @@ -76,7 +76,12 @@ const (
AMXFP16 // Tile computational operations on FP16 numbers
AMXINT8 // Tile computational operations on 8-bit integers
AMXTILE // Tile architecture
APX_F // Intel APX
AVX // AVX functions
AVX10 // If set the Intel AVX10 Converged Vector ISA is supported
AVX10_128 // If set indicates that AVX10 128-bit vector support is present
AVX10_256 // If set indicates that AVX10 256-bit vector support is present
AVX10_512 // If set indicates that AVX10 512-bit vector support is present
AVX2 // AVX2 functions
AVX512BF16 // AVX-512 BFLOAT16 Instructions
AVX512BITALG // AVX-512 Bit Algorithms
Expand Down Expand Up @@ -156,6 +161,8 @@ const (
IDPRED_CTRL // IPRED_DIS
INT_WBINVD // WBINVD/WBNOINVD are interruptible.
INVLPGB // NVLPGB and TLBSYNC instruction supported
KEYLOCKER // Key locker
KEYLOCKERW // Key locker wide
LAHF // LAHF/SAHF in long mode
LAM // If set, CPU supports Linear Address Masking
LBRVIRT // LBR virtualization
Expand Down Expand Up @@ -302,9 +309,10 @@ type CPUInfo struct {
L2 int // L2 Cache (per core or shared). Will be -1 if undetected
L3 int // L3 Cache (per core, per ccx or shared). Will be -1 if undetected
}
SGX SGXSupport
maxFunc uint32
maxExFunc uint32
SGX SGXSupport
AVX10Level uint8
maxFunc uint32
maxExFunc uint32
}

var cpuid func(op uint32) (eax, ebx, ecx, edx uint32)
Expand Down Expand Up @@ -1165,6 +1173,7 @@ func support() flagSet {
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<<23) != 0, KEYLOCKER)
fs.setIf(ecx&(1<<27) != 0, MOVDIRI)
fs.setIf(ecx&(1<<28) != 0, MOVDIR64B)
fs.setIf(ecx&(1<<29) != 0, ENQCMD)
Expand Down Expand Up @@ -1202,6 +1211,8 @@ func support() flagSet {
fs.setIf(edx1&(1<<4) != 0, AVXVNNIINT8)
fs.setIf(edx1&(1<<5) != 0, AVXNECONVERT)
fs.setIf(edx1&(1<<14) != 0, PREFETCHI)
fs.setIf(edx1&(1<<19) != 0, AVX10)
fs.setIf(edx1&(1<<21) != 0, APX_F)

// Only detect AVX-512 features if XGETBV is supported
if c&((1<<26)|(1<<27)) == (1<<26)|(1<<27) {
Expand Down Expand Up @@ -1252,6 +1263,19 @@ func support() flagSet {
fs.setIf(edx&(1<<4) != 0, BHI_CTRL)
fs.setIf(edx&(1<<5) != 0, MCDT_NO)

// Add keylocker features.
if fs.inSet(KEYLOCKER) && mfi >= 0x19 {
_, ebx, _, _ := cpuidex(0x19, 0)
fs.setIf(ebx&5 == 5, KEYLOCKERW) // Bit 0 and 2 (1+4)
}

// Add AVX10 features.
if fs.inSet(AVX10) && mfi >= 0x24 {
_, ebx, _, _ := cpuidex(0x24, 0)
fs.setIf(ebx&(1<<16) != 0, AVX10_128)
fs.setIf(ebx&(1<<17) != 0, AVX10_256)
fs.setIf(ebx&(1<<18) != 0, AVX10_512)
}
}

// Processor Extended State Enumeration Sub-leaf (EAX = 0DH, ECX = 1)
Expand Down Expand Up @@ -1404,6 +1428,14 @@ func support() flagSet {
return fs
}

func (c *CPUInfo) supportAVX10() uint8 {
if c.maxFunc >= 0x24 && c.featureSet.inSet(AVX10) {
_, ebx, _, _ := cpuidex(0x24, 0)
return uint8(ebx)
}
return 0
}

func valAsString(values ...uint32) []byte {
r := make([]byte, 4*len(values))
for i, v := range values {
Expand Down
1 change: 1 addition & 0 deletions detect_x86.go
View file
Expand Up @@ -31,6 +31,7 @@ func addInfo(c *CPUInfo, safe bool) {
c.LogicalCores = logicalCores()
c.PhysicalCores = physicalCores()
c.VendorID, c.VendorString = vendorID()
c.AVX10Level = c.supportAVX10()
c.cacheSize()
c.frequencies()
}

0 comments on commit 666b9f5

Please sign in to comment.