cloudwego/frugal

A very fast dynamic Thrift serializer & deserializer.

frugal's Issues

[GLCC 赛题] SSA 后端 Auto-vectorization Pass 的实现

🚩GLCC 高校编程夏令营赛题
👏欢迎大家参与～

GLCC 是由**计算机学会举办的高校开源编程夏令营活动，活动为项目准备了开源奖学金。请在以下链接中报名参与⬇️
报名链接： https://www.gitlink.org.cn/glcc/subjects/detail/161
关于issue相关问题、解题思路的答疑将在群内进行

1、题目描述
Frugal 是字节自研的一款高性能 Thrift 库，旨在使用 JIT 技术来加速 Thrift 的编解码过程。由于 Thrift 采用了大端序，而 x86_64 等常见的处理器架构均为小端序，因此对于 list 这一类的数据类型需要一个循环来实现小端到大端的转换。而这个过程如果能够利用 SIMD 指令的话，可以极大的提高性能。

2、编码任务

• 实现一个 SSA 后端的 Auto-vectorization Pass
• 输入原始的 CFG，输出 vectorize 之后的 CFG
• IR 表示形式使用 Frugal SSA IR (internal/atm/ssa 目录)
• 该 Pass 的运行顺序在 Fusion 之后，在 Final TDCE 之前
• CPU 架构为 x86_64，SIMD 指令使用 AVX / AVX-2 指令集

3、技能要求和编程语言

• 编译原理
• x86_64 指令集知识
• 实现语言为 Go

4、预期完成结果

• 能够识别并且将 CFG 中的 load-swap-store 循环替换为正确的 SIMD 指令

readme 有疑问

无需生成代码的意思是什么
我理解直接通过idl就可以调用接口，可示例让我感觉很不清楚
可以增加一些示例么
无需生成 kitex_gen 么 我看示例也有引用呀
还是 将 kitex_gen换成 thriftgo生成的代码呢
可以给一些清楚的示例 仅转换了一个结构体 我没看懂 如何 无需生成代码

Does frugal not support ARM (Apple M1)?

Describe the bug

Can't use frugal to encode/decode on my MacBook Pro (M1)

To Reproduce

Steps to reproduce the behavior:

A local Docker is need for 'testcontainers'.

package main

import (
	"context"
	"fmt"
	"time"

	"github.com/cloudwego/frugal"
	"github.com/google/uuid"
	"github.com/nats-io/nats.go"
	"github.com/testcontainers/testcontainers-go"
	natscontainer "github.com/testcontainers/testcontainers-go/modules/nats"
)

const (
	SubjectEventUserRegister = "event.user.register"
)

type EventUserRegister struct {
	UserID       string `frugal:"1,required"`
	RegisterTime int64  `frugal:"2,required"`
}

func Encode[T any](data *T) []byte {
	bytes := make([]byte, frugal.EncodedSize(data))
	_ = must(frugal.EncodeObject(bytes, nil, data))
	return bytes
}

func Decode[T any](bytes []byte) *T {
	var data T
	_ = must(frugal.DecodeObject(bytes, &data))
	return &data
}

func main() {
	ctx := context.Background()

	natsContainer := must(natscontainer.RunContainer(ctx,
		testcontainers.WithImage("nats:2.9"),
	))
	should(natsContainer.IsRunning())

	nc := must(nats.Connect(must(natsContainer.ConnectionString(ctx))))
	should(nc.IsConnected())

	_ = must(nc.Subscribe(SubjectEventUserRegister, func(msg *nats.Msg) {
		event := Decode[EventUserRegister](msg.Data)
		fmt.Printf("%#v\n", event)
	}))

	must0(nc.Publish(SubjectEventUserRegister, Encode[EventUserRegister](&EventUserRegister{
		UserID:       uuid.New().String(),
		RegisterTime: time.Now().UnixMilli(),
	})))

	select {}
}

func must[T any](t T, err error) T {
	if err != nil {
		panic(err)
	}
	return t
}

func must0(err error) {
	if err != nil {
		panic(err)
	}
}

func should(b bool) {
	if !b {
		panic(b)
	}
}

I got this when I go run main.go:

# github.com/cloudwego/frugal/internal/atm/abi
../../../../go/go1.20/pkg/mod/github.com/cloudwego/[email protected]/internal/atm/abi/abi.go:31:11: undefined: ArchCreateABI

Expected behavior

print the value of EventUserRegister

Kitex version:

no kitex

Environment:

GO111MODULE='on'
GOARCH='arm64'
GOBIN=''
GOCACHE='/Users/tzq0301/Library/Caches/go-build'
GOENV='/Users/tzq0301/Library/Application Support/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFLAGS=''
GOHOSTARCH='arm64'
GOHOSTOS='darwin'
GOINSECURE=''
GOMODCACHE='/Users/tzq0301/go/go1.20/pkg/mod'
GONOPROXY=''
GONOSUMDB=''
GOOS='darwin'
GOPATH='/Users/tzq0301/go/go1.20:/Users/tzq0301/go/go1.20/go'
GOPRIVATE=''
GOPROXY='https://goproxy.cn,direct'
GOROOT='/Users/tzq0301/go/go1.21.4'
GOSUMDB='sum.golang.org'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/Users/tzq0301/go/go1.21.4/pkg/tool/darwin_arm64'
GOVCS=''
GOVERSION='go1.21.4'
GCCGO='gccgo'
AR='ar'
CC='clang'
CXX='clang++'
CGO_ENABLED='1'
GOMOD='/Users/tzq0301/Code/Go/nats-demo/frugal/go.mod'
GOWORK=''
CGO_CFLAGS='-O2 -g'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-O2 -g'
CGO_FFLAGS='-O2 -g'
CGO_LDFLAGS='-O2 -g'
PKG_CONFIG='pkg-config'
GOGCCFLAGS='-fPIC -arch arm64 -pthread -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -ffile-prefix-map=/var/folders/sk/nrp4br617kscldkf4jsr0dsh0000gn/T/go-build3032409702=/tmp/go-build -gno-record-gcc-switches -fno-common'

add support for go1.20

When use the kitex with go1.20 I encountered the following problems , I hope there will be support for go 1.20 soon

# github.com/cloudwego/frugal/internal/loader
../../../../go/pkg/mod/github.com/cloudwego/[email protected]/internal/loader/funcdata.go:32:22: undefined: _ModuleData
../../../../go/pkg/mod/github.com/cloudwego/[email protected]/internal/loader/funcdata.go:35:27: undefined: _ModuleData
../../../../go/pkg/mod/github.com/cloudwego/[email protected]/internal/loader/funcdata.go:39:16: undefined: _ModuleData
../../../../go/pkg/mod/github.com/cloudwego/[email protected]/internal/loader/funcdata.go:75:26: undefined: _ModuleData
../../../../go/pkg/mod/github.com/cloudwego/[email protected]/internal/loader/funcdata_invalid.go:27:9: panic("Unsupported Go version. Supported versions are: 1.15, 1.16, 1.17, 1.18, 1.19") (no value) used as value

Proposal: fallback to thrift-iterator when frugal is not available

Frugal is only available on AMD64 arch machines, while some user have both AMD64 & ARM64 machines, which makes it difficult to keep the same codebase.

frugal field type not provided will cause panic

Produce

If I simply run the examples in the Serialization and deserialization on a customized Go struct section in README.md, it will get an error message:

panic: frugal: cannot measure encoded size: invalid tag for field main.MyStruct.Msg

goroutine 1 [running]:
github.com/cloudwego/frugal/internal/binary/encoder.EncodedSize({0xf714e0, 0xc00010f620})
        /home/yunwei/coding/frugal/internal/binary/encoder/encoder.go:69 +0x8e
github.com/cloudwego/frugal.EncodedSize(...)
        /home/yunwei/coding/frugal/frugal.go:27
main.main()
        /home/yunwei/coding/frugal/cmd/main.go:20 +0xa9
exit status 2

the test code is like this, which is copied from README:

package main

import (
    "github.com/cloudwego/frugal"
)

type MyStruct struct {
    Msg     string  `frugal:"1,default"`
    Code    int64   `frugal:"2,default"`
    Numbers []int64 `frugal:"3,default,list<i64>"`
}

func main() {
    ms := &MyStruct{
        Msg: "my message",
        Code: 1024,
        Numbers: []int64{0, 1, 2, 3, 4},
    }

    buf := make([]byte, frugal.EncodedSize(ms))
    frugal.EncodeObject(buf, nil, ms)

    got := &MyStruct{}
    frugal.DecodeObject(buf, got)
}

However, if I change the struct to:

type MyStruct struct {
    Msg     string  `frugal:"1,default,string"`
    Code    int64   `frugal:"2,default,i64"`
    Numbers []int64 `frugal:"3,default,list<i64>"`
}

Then the example can be run successfully.

Question

The README said:

Frugal tag is like frugal:"1,default,string", 1 is field ID, default is field requiredness, string is field type. Field ID and requiredness is always required, but field type is only required for list, set and enum.

while the code in the internal/binary/defs/resolver.go gives that:

        /* must have 3 fields: ID, Requiredness, Type */
        if ft = strings.Split(tv, ","); len(ft) != 3 {
            return nil, fmt.Errorf("invalid tag for field %s.%s", vt, sf.Name)
        }

Is the field type optional tag not support currently, or the tag should be provided each time in Frugal?

By the way, another minor fix for README example is in PR #7 .

Unlicensed dependency: `github.com/chenzhuoyu/iasm`

Describe the bug

Your dependency github.com/chenzhuoyu/iasm is unlicensed. As a result cloudwego/frugal cannot be used in MIT / Apache 2 licensed projects.

Please update github.com/chenzhuoyu/iasm to v0.9.0 which added a license. Then create a new release.

Please also update cloudwego/kitex and other projects to use the new cloudwego/frugal release with the updated dependency.

谢谢
Bernd, Maintainer of Dapr (dapr.io) -- github.com/dapr

在Kitex中无法使用Frugal功能

Env

• Kitex：0.4.2

利用kitex根据IDL生成代码

kitex -module "xxx" -thrift frugal_tag,template=slim -service xxx idl/xxx.thrift

这里使用的是不需要编解码代码，生成了代码

但在初始化 client 和 server 时使用 WithPayloadCodec(thrift.NewThriftFrugalCodec())，发现无法引入FrugalWrite、FrugalRead

`reflect.Pointer` is not compatible with go lower than 1.18

Describe the bug

I cannot compile my project with go 1.16.

Error message:

15# github.com/cloudwego/frugal/internal/rt
16../compile_path/pkg/mod/github.com/cloudwego/[email protected]/internal/rt/runtime.go:120:14: undefined: reflect.Pointer

The reflect.Pointer is added to go in 1.18 golang/go@17910ed . But this package's mod version is 1.15, which has caused some misunderstandings.

Please use reflect.Ptr to ensure compatibility with lower versions.

To Reproduce

Build this project with a version of go lower than 1.18.

一些AMD64后端优化建议

frugal/internal/atm/ssa/pass_fusion_amd64.go

Line 111 in 167cbd2

* leaq {off1}(%r0), %r1 ; leaq {off2}(%r1,%r2), %r3 --> leaq {off1+off2}(%r0,%r2), %r3 */

在现代处理器上三操作数LEA比两操作数LEA慢。

Intel® 64 and IA-32 Architectures Optimization Reference Manual
3.5.1.3 Using LEA

For LEA instructions with three source operands and some specific situations, instruction latency has increased to 3 cycles, and must dispatch via port 1:
— LEA that has all three source operands: base, index, and offset.
— LEA that uses base and index registers where the base is EBP, RBP, or R13.
...

see https://reviews.llvm.org/D32277

---

frugal/internal/atm/ssa/pass_fusion_amd64.go

Lines 160 to 161 in 167cbd2

	* movx {mem}, %r0; btsq %r2, %r0, %r1; movx %r1, {mem} --> btsq %r2, {mem}
	* movx {mem}, %r0; btsq {imm}, %r0, %r1; movx %r1, {mem} --> btsq {imm}, {mem} */

bts指令对于内存目标很慢.

see https://go-review.googlesource.com/c/go/+/318149

---

frugal/internal/atm/ssa/pass_fusion_amd64.go

Lines 138 to 144 in 167cbd2

	/* movx {mem}, %r0; bswapx %r0, %r1 --> movbex {mem}, %r1 */
	case *IrAMD64_BSWAP: {
	if ins, ok := defs[p.V].(*IrAMD64_MOV_load); ok && ins.N != 1 {
	done = false
	bb.Ins[i] = &IrAMD64_MOV_load_be { R: p.R, M: ins.M, N: ins.N }
	}
	}

1. movw时选用的指令为带零扩展的指令(movzx)， 但是movbew不具备零拓展
2. movbe是x86_64v3的一部分, 使用movbe指令需要先判断本机是否支持movbe

Support go1.19

Is your feature request related to a problem? Please describe.

Compiling using go1.19 will failed with the following messages:
../../../../go/pkg/mod/github.com/cloudwego/[email protected]/internal/loader/funcdata_invalid.go:27:9: panic("Unsupported Go version. Supported versions are: 1.15, 1.16, 1.17, 1.18") (no value) used as value

Describe the solution you'd like
support compile using go1.19

Describe alternatives you've considered

Additional context

go 1.21 support

why not support it in rc cycle? or add a default slow support for outter version