// Copyright 2017 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

//go:build 386 || amd64

package cpu

const CacheLinePadSize = 64

// cpuid is implemented in cpu_x86.s.
func cpuid(,  uint32) (, , ,  uint32)

// xgetbv with ecx = 0 is implemented in cpu_x86.s.
func xgetbv() (,  uint32)

// getGOAMD64level is implemented in cpu_x86.s. Returns number in [1,4].
func getGOAMD64level() int32

const (
	// ecx bits
	cpuid_SSE3      = 1 << 0
	cpuid_PCLMULQDQ = 1 << 1
	cpuid_SSSE3     = 1 << 9
	cpuid_FMA       = 1 << 12
	cpuid_SSE41     = 1 << 19
	cpuid_SSE42     = 1 << 20
	cpuid_POPCNT    = 1 << 23
	cpuid_AES       = 1 << 25
	cpuid_OSXSAVE   = 1 << 27
	cpuid_AVX       = 1 << 28

	// ebx bits
	cpuid_BMI1     = 1 << 3
	cpuid_AVX2     = 1 << 5
	cpuid_BMI2     = 1 << 8
	cpuid_ERMS     = 1 << 9
	cpuid_AVX512F  = 1 << 16
	cpuid_ADX      = 1 << 19
	cpuid_SHA      = 1 << 29
	cpuid_AVX512BW = 1 << 30
	cpuid_AVX512VL = 1 << 31

	// edx bits for CPUID 0x80000001
	cpuid_RDTSCP = 1 << 27
)

var maxExtendedFunctionInformation uint32

func doinit() {
	options = []option{
		{Name: "adx", Feature: &X86.HasADX},
		{Name: "aes", Feature: &X86.HasAES},
		{Name: "erms", Feature: &X86.HasERMS},
		{Name: "pclmulqdq", Feature: &X86.HasPCLMULQDQ},
		{Name: "rdtscp", Feature: &X86.HasRDTSCP},
		{Name: "sha", Feature: &X86.HasSHA},
	}
	 := getGOAMD64level()
	if  < 2 {
		// These options are required at level 2. At lower levels
		// they can be turned off.
		options = append(options,
			option{Name: "popcnt", Feature: &X86.HasPOPCNT},
			option{Name: "sse3", Feature: &X86.HasSSE3},
			option{Name: "sse41", Feature: &X86.HasSSE41},
			option{Name: "sse42", Feature: &X86.HasSSE42},
			option{Name: "ssse3", Feature: &X86.HasSSSE3})
	}
	if  < 3 {
		// These options are required at level 3. At lower levels
		// they can be turned off.
		options = append(options,
			option{Name: "avx", Feature: &X86.HasAVX},
			option{Name: "avx2", Feature: &X86.HasAVX2},
			option{Name: "bmi1", Feature: &X86.HasBMI1},
			option{Name: "bmi2", Feature: &X86.HasBMI2},
			option{Name: "fma", Feature: &X86.HasFMA})
	}
	if  < 4 {
		// These options are required at level 4. At lower levels
		// they can be turned off.
		options = append(options,
			option{Name: "avx512f", Feature: &X86.HasAVX512F},
			option{Name: "avx512bw", Feature: &X86.HasAVX512BW},
			option{Name: "avx512vl", Feature: &X86.HasAVX512VL},
		)
	}

	, , ,  := cpuid(0, 0)

	if  < 1 {
		return
	}

	maxExtendedFunctionInformation, _, _, _ = cpuid(0x80000000, 0)

	, , ,  := cpuid(1, 0)

	X86.HasSSE3 = isSet(, cpuid_SSE3)
	X86.HasPCLMULQDQ = isSet(, cpuid_PCLMULQDQ)
	X86.HasSSSE3 = isSet(, cpuid_SSSE3)
	X86.HasSSE41 = isSet(, cpuid_SSE41)
	X86.HasSSE42 = isSet(, cpuid_SSE42)
	X86.HasPOPCNT = isSet(, cpuid_POPCNT)
	X86.HasAES = isSet(, cpuid_AES)

	// OSXSAVE can be false when using older Operating Systems
	// or when explicitly disabled on newer Operating Systems by
	// e.g. setting the xsavedisable boot option on Windows 10.
	X86.HasOSXSAVE = isSet(, cpuid_OSXSAVE)

	// The FMA instruction set extension only has VEX prefixed instructions.
	// VEX prefixed instructions require OSXSAVE to be enabled.
	// See Intel 64 and IA-32 Architecture Software Developer’s Manual Volume 2
	// Section 2.4 "AVX and SSE Instruction Exception Specification"
	X86.HasFMA = isSet(, cpuid_FMA) && X86.HasOSXSAVE

	 := false
	 := false
	// For XGETBV, OSXSAVE bit is required and sufficient.
	if X86.HasOSXSAVE {
		,  := xgetbv()
		// Check if XMM and YMM registers have OS support.
		 = isSet(, 1<<1) && isSet(, 1<<2)

		// AVX512 detection does not work on Darwin,
		// see https://github.com/golang/go/issues/49233
		//
		// Check if opmask, ZMMhi256 and Hi16_ZMM have OS support.
		 =  && isSet(, 1<<5) && isSet(, 1<<6) && isSet(, 1<<7)
	}

	X86.HasAVX = isSet(, cpuid_AVX) && 

	if  < 7 {
		return
	}

	, , ,  := cpuid(7, 0)
	X86.HasBMI1 = isSet(, cpuid_BMI1)
	X86.HasAVX2 = isSet(, cpuid_AVX2) && 
	X86.HasBMI2 = isSet(, cpuid_BMI2)
	X86.HasERMS = isSet(, cpuid_ERMS)
	X86.HasADX = isSet(, cpuid_ADX)
	X86.HasSHA = isSet(, cpuid_SHA)

	X86.HasAVX512F = isSet(, cpuid_AVX512F) && 
	if X86.HasAVX512F {
		X86.HasAVX512BW = isSet(, cpuid_AVX512BW)
		X86.HasAVX512VL = isSet(, cpuid_AVX512VL)
	}

	var  uint32
	, _, _, _ = cpuid(0x80000000, 0)

	if  < 0x80000001 {
		return
	}

	, , ,  := cpuid(0x80000001, 0)
	X86.HasRDTSCP = isSet(, cpuid_RDTSCP)
}

func isSet( uint32,  uint32) bool {
	return & != 0
}

// Name returns the CPU name given by the vendor.
// If the CPU name can not be determined an
// empty string is returned.
func () string {
	if maxExtendedFunctionInformation < 0x80000004 {
		return ""
	}

	 := make([]byte, 0, 3*4*4)

	var , , ,  uint32
	, , ,  = cpuid(0x80000002, 0)
	 = appendBytes(, , , , )
	, , ,  = cpuid(0x80000003, 0)
	 = appendBytes(, , , , )
	, , ,  = cpuid(0x80000004, 0)
	 = appendBytes(, , , , )

	// Trim leading spaces.
	for len() > 0 && [0] == ' ' {
		 = [1:]
	}

	// Trim tail after and including the first null byte.
	for ,  := range  {
		if  == '\x00' {
			 = [:]
			break
		}
	}

	return string()
}

func appendBytes( []byte,  ...uint32) []byte {
	for ,  := range  {
		 = append(,
			byte(( >> 0)),
			byte(( >> 8)),
			byte(( >> 16)),
			byte(( >> 24)))
	}
	return 
}