huff0: decompress directly into output

huff0/_generate/gen.go

@@ -0,0 +1,310 @@
+package main
+
+//go:generate go run gen.go -out ../decompress_amd64.s -pkg=huff0
+
+import (
+	"flag"
+	"fmt"
+	"strconv"
+
+	_ "github.com/klauspost/compress"
+
+	. "github.com/mmcloughlin/avo/build"
+	. "github.com/mmcloughlin/avo/operand"
+	"github.com/mmcloughlin/avo/reg"
+)
+
+func main() {
+	flag.Parse()
+
+	ConstraintExpr("amd64,!appengine,!noasm,gc")
+
+	decompress := decompress4x{}
+	decompress.generateProcedure("decompress4x_main_loop_amd64")
+	decompress.generateProcedure4x8bit("decompress4x_8b_main_loop_amd64")
+
+	Generate()
+}
+
+type decompress4x struct {
+}
+
+func (d decompress4x) generateProcedure(name string) {
+	Package("github.com/klauspost/compress/huff0")
+	TEXT(name, 0, "func(ctx* decompress4xContext)")
+	Doc(name+" is an x86 assembler implementation of Decompress4X when tablelog > 8.decodes a sequence", "")
+	Pragma("noescape")
+
+	exhausted := GP64()
+	XORQ(exhausted.As64(), exhausted.As64()) // exhausted = false
+
+	limitPtr := AllocLocal(8)
+
+	bufferOrigin := GP64()
+	peekBits := GP64()
+	buffer := GP64()
+	dstEvery := GP64()
+	table := GP64()
+
+	br0 := GP64()
+	br1 := GP64()
+	br2 := GP64()
+	br3 := GP64()
+
+	Comment("Preload values")
+	{
+		ctx := Dereference(Param("ctx"))
+		Load(ctx.Field("peekBits"), peekBits)
+		Load(ctx.Field("out"), buffer)
+		MOVQ(buffer, bufferOrigin)
+		limit := Load(ctx.Field("limit"), GP64())
+		MOVQ(limit, limitPtr)
+		Load(ctx.Field("dstEvery"), dstEvery)
+		Load(ctx.Field("tbl"), table)
+		Load(ctx.Field("pbr0"), br0)
+		Load(ctx.Field("pbr1"), br1)
+		Load(ctx.Field("pbr2"), br2)
+		Load(ctx.Field("pbr3"), br3)
+	}
+
+	Comment("Main loop")
+	Label("main_loop")
+
+	MOVQ(bufferOrigin, buffer)
+	// Check if we have space
+	CMPQ(buffer, limitPtr)
+	SETGE(exhausted.As8())
+	d.decodeTwoValues(0, br0, peekBits, table, buffer, exhausted)
+	ADDQ(dstEvery, buffer)
+	d.decodeTwoValues(1, br1, peekBits, table, buffer, exhausted)
+	ADDQ(dstEvery, buffer)
+	d.decodeTwoValues(2, br2, peekBits, table, buffer, exhausted)
+	ADDQ(dstEvery, buffer)
+	d.decodeTwoValues(3, br3, peekBits, table, buffer, exhausted)
+
+	ADDQ(U8(2), bufferOrigin) // off += 2
+
+	TESTB(exhausted.As8(), exhausted.As8()) // any br[i].ofs < 4?
+	JZ(LabelRef("main_loop"))
+
+	{
+		ctx := Dereference(Param("ctx"))
+		tmp := Load(ctx.Field("out"), GP64())
+		decoded := GP64()
+		MOVQ(bufferOrigin, decoded)
+		SUBQ(tmp, decoded)
+		SHLQ(U8(2), decoded) // decoded *= 4
+
+		Store(decoded, ctx.Field("decoded"))
+	}
+
+	RET()
+}
+
+// TODO [wmu]: I believe it's doable in avo, but can't figure out how to deal
+//             with arbitrary pointers to a given type
+const bitReader_in = 0
+const bitReader_off = bitReader_in + 3*8 // {ptr, len, cap}
+const bitReader_value = bitReader_off + 8
+const bitReader_bitsRead = bitReader_value + 8
+
+func (d decompress4x) decodeTwoValues(id int, br, peekBits, table, buffer, exhausted reg.GPVirtual) {
+	brValue, brBitsRead := d.fillFast32(id, 32, br, exhausted)
+
+	val := GP64()
+	Commentf("val0 := br%d.peekTopBits(peekBits)", id)
+	CX := reg.CL
+	MOVQ(brValue, val.As64())
+	MOVQ(peekBits, CX.As64())
+	SHRQ(CX, val.As64()) // val = (value >> peek_bits) & mask
+
+	Comment("v0 := table[val0&mask]")
+	MOVW(Mem{Base: table, Index: val.As64(), Scale: 2}, CX.As16())
+
+	Commentf("br%d.advance(uint8(v0.entry)", id)
+	out := reg.RAX             // Fixed since we need 8H
+	MOVB(CX.As8H(), out.As8()) // AL = uint8(v0.entry >> 8)
+
+	SHLQ(CX, brValue)                // value <<= n
+	ADDB(CX.As8(), brBitsRead.As8()) // bits_read += n
+
+	Commentf("val1 := br%d.peekTopBits(peekBits)", id)
+	MOVQ(peekBits, CX.As64())
+	MOVQ(brValue, val.As64())
+	SHRQ(CX, val.As64()) // val = (value >> peek_bits) & mask
+
+	Comment("v1 := table[val1&mask]")
+	MOVW(Mem{Base: table, Index: val.As64(), Scale: 2}, CX.As16()) // tmp - v1
+
+	Commentf("br%d.advance(uint8(v1.entry))", id)
+	MOVB(CX.As8H(), out.As8H())      // AH = uint8(v0.entry >> 8)
+	SHLQ(CX, brValue)                // value <<= n
+	ADDB(CX.As8(), brBitsRead.As8()) // bits_read += n
+
+	Comment("these two writes get coalesced")
+	Comment("out[id * dstEvery + 0] = uint8(v0.entry >> 8)")
+	Comment("out[id * dstEvery + 1] = uint8(v1.entry >> 8)")
+	MOVW(out.As16(), Mem{Base: buffer})
+
+	Comment("update the bitrader reader structure")
+	MOVQ(brValue, Mem{Base: br, Disp: bitReader_value})
+	MOVB(brBitsRead.As8(), Mem{Base: br, Disp: bitReader_bitsRead})
+}
+
+func (d decompress4x) generateProcedure4x8bit(name string) {
+	Package("github.com/klauspost/compress/huff0")
+	TEXT(name, 0, "func(ctx* decompress4xContext)")
+	Doc(name+" is an x86 assembler implementation of Decompress4X when tablelog > 8.decodes a sequence", "")
+	Pragma("noescape")
+
+	exhausted := GP64()                      // Fixed since we need 8H
+	XORQ(exhausted.As64(), exhausted.As64()) // exhausted = false
+
+	bufferOrigin := AllocLocal(8)
+	limitPtr := AllocLocal(8)
+
+	peekBits := GP64()
+	buffer := GP64()
+	dstEvery := GP64()
+	table := GP64()
+
+	br0 := GP64()
+	br1 := GP64()
+	br2 := GP64()
+	br3 := GP64()
+
+	Comment("Preload values")
+	{
+		ctx := Dereference(Param("ctx"))
+		Load(ctx.Field("peekBits"), peekBits)
+		Load(ctx.Field("out"), buffer)
+		MOVQ(buffer, bufferOrigin)
+		limit := Load(ctx.Field("limit"), GP64())
+		MOVQ(limit, limitPtr)
+		Load(ctx.Field("dstEvery"), dstEvery)
+		Load(ctx.Field("tbl"), table)
+		Load(ctx.Field("pbr0"), br0)
+		Load(ctx.Field("pbr1"), br1)
+		Load(ctx.Field("pbr2"), br2)
+		Load(ctx.Field("pbr3"), br3)
+	}
+
+	Comment("Main loop")
+	Label("main_loop")
+
+	MOVQ(bufferOrigin, buffer)
+	// Check if we have space
+	CMPQ(buffer, limitPtr)
+	SETGE(exhausted.As8())
+	d.decodeFourValues(0, br0, peekBits, table, buffer, exhausted)
+	ADDQ(dstEvery, buffer)
+	d.decodeFourValues(1, br1, peekBits, table, buffer, exhausted)
+	ADDQ(dstEvery, buffer)
+	d.decodeFourValues(2, br2, peekBits, table, buffer, exhausted)
+	ADDQ(dstEvery, buffer)
+	d.decodeFourValues(3, br3, peekBits, table, buffer, exhausted)
+
+	ADDQ(U8(4), bufferOrigin) // off += 4
+
+	TESTB(exhausted.As8(), exhausted.As8()) // any br[i].ofs < 4?
+	JZ(LabelRef("main_loop"))
+
+	{
+		ctx := Dereference(Param("ctx"))
+		tmp := Load(ctx.Field("out"), GP64())
+		decoded := GP64()
+		MOVQ(bufferOrigin, decoded)
+		SUBQ(tmp, decoded)
+		SHLQ(U8(2), decoded) // decoded *= 4
+
+		Store(decoded, ctx.Field("decoded"))
+	}
+	RET()
+}
+
+func (d decompress4x) decodeFourValues(id int, br, peekBits, table, buffer, exhausted reg.GPVirtual) {
+	brValue, brBitsRead := d.fillFast32(id+1000, 32, br, exhausted)
+
+	decompress := func(valID int, outByte reg.Register) {
+		CX := reg.CL
+		val := GP64()
+		Commentf("val%d := br%d.peekTopBits(peekBits)", valID, id)
+		MOVQ(brValue, val.As64())
+		MOVQ(peekBits, CX.As64())
+		SHRQ(CX, val.As64()) // val = (value >> peek_bits) & mask
+
+		Commentf("v%d := table[val0&mask]", valID)
+		MOVW(Mem{Base: table, Index: val.As64(), Scale: 2}, CX.As16())
+
+		Commentf("br%d.advance(uint8(v%d.entry)", id, valID)
+		MOVB(CX.As8H(), outByte) // outByte = uint8(v0.entry >> 8)
+
+		SHLQ(CX, brValue)          // value <<= n
+		ADDB(CX, brBitsRead.As8()) // bits_read += n
+	}
+
+	out := reg.RAX // Fixed since we need 8H
+	decompress(0, out.As8L())
+	decompress(1, out.As8H())
+	BSWAPL(out.As32())
+	decompress(2, out.As8H())
+	decompress(3, out.As8L())
+	BSWAPL(out.As32())
+
+	Comment("these four writes get coalesced")
+	Comment("out[id * dstEvery + 0] = uint8(v0.entry >> 8)")
+	Comment("out[id * dstEvery + 1] = uint8(v1.entry >> 8)")
+	Comment("out[id * dstEvery + 3] = uint8(v2.entry >> 8)")
+	Comment("out[id * dstEvery + 4] = uint8(v3.entry >> 8)")
+	MOVL(out.As32(), Mem{Base: buffer})
+
+	Comment("update the bitreader reader structure")
+	MOVQ(brValue, Mem{Base: br, Disp: bitReader_value})
+	MOVB(brBitsRead.As8(), Mem{Base: br, Disp: bitReader_bitsRead})
+}
+
+func (d decompress4x) fillFast32(id, atLeast int, br, exhausted reg.GPVirtual) (brValue, brBitsRead reg.GPVirtual) {
+	if atLeast > 32 {
+		panic(fmt.Sprintf("at least (%d) cannot be >32", atLeast))
+	}
+	Commentf("br%d.fillFast32()", id)
+	brValue = GP64()
+	brBitsRead = GP64()
+	MOVQ(Mem{Base: br, Disp: bitReader_value}, brValue)
+	MOVBQZX(Mem{Base: br, Disp: bitReader_bitsRead}, brBitsRead)
+
+	// We must have at least 2 * max tablelog left
+	CMPQ(brBitsRead, U8(64-atLeast))
+	JBE(LabelRef("skip_fill" + strconv.Itoa(id)))
+	brOffset := GP64()
+	MOVQ(Mem{Base: br, Disp: bitReader_off}, brOffset)
+
+	SUBQ(U8(32), brBitsRead) // b.bitsRead -= 32
+	SUBQ(U8(4), brOffset)    // b.off -= 4
+
+	// v := b.in[b.off-4 : b.off]
+	// v = v[:4]
+	// low := (uint32(v[0])) | (uint32(v[1]) << 8) | (uint32(v[2]) << 16) | (uint32(v[3]) << 24)
+	tmp := GP64()
+	MOVQ(Mem{Base: br, Disp: bitReader_in}, tmp)
+
+	Comment("b.value |= uint64(low) << (b.bitsRead & 63)")
+	addr := Mem{Base: brOffset, Index: tmp.As64(), Scale: 1}
+	CX := reg.CL
+	MOVL(addr, tmp.As32()) // tmp = uint32(b.in[b.off:b.off+4])
+	MOVQ(brBitsRead, CX.As64())
+	SHLQ(CX, tmp.As64())
+
+	MOVQ(brOffset, Mem{Base: br, Disp: bitReader_off})
+	ORQ(tmp.As64(), brValue)
+	{
+		Commentf("exhausted = exhausted || (br%d.off < 4)", id)
+		CMPQ(brOffset, U8(4))
+		tmp = GP64()
+		SETLT(tmp.As8())
+		ORB(tmp.As8(), exhausted.As8())
+	}
+
+	Label("skip_fill" + strconv.Itoa(id))
+	return
+}

huff0/_generate/go.mod

@@ -0,0 +1,10 @@
+module github.com/klauspost/compress/s2/_generate
+
+go 1.15
+
+require (
+	github.com/klauspost/compress v1.15.1
+	github.com/mmcloughlin/avo v0.4.0
+)
+
+replace github.com/klauspost/compress => ../..

huff0/_generate/go.sum

@@ -0,0 +1,32 @@
+github.com/mmcloughlin/avo v0.4.0 h1:jeHDRktVD+578ULxWpQHkilor6pkdLF7u7EiTzDbfcU=
+github.com/mmcloughlin/avo v0.4.0/go.mod h1:RW9BfYA3TgO9uCdNrKU2h6J8cPD8ZLznvfgHAeszb1s=
+github.com/yuin/goldmark v1.4.0/go.mod h1:mwnBkeHKe2W/ZEtQ+71ViKU8L12m81fl3OWwC1Zlc8k=
+golang.org/x/arch v0.0.0-20210923205945-b76863e36670/go.mod h1:5om86z9Hs0C8fWVUuoMHwpExlXzs5Tkyp9hOrfG7pp8=
+golang.org/x/crypto v0.0.0-20190308221718-c2843e01d9a2/go.mod h1:djNgcEr1/C05ACkg1iLfiJU5Ep61QUkGW8qpdssI0+w=
+golang.org/x/crypto v0.0.0-20191011191535-87dc89f01550/go.mod h1:yigFU9vqHzYiE8UmvKecakEJjdnWj3jj499lnFckfCI=
+golang.org/x/mod v0.4.2 h1:Gz96sIWK3OalVv/I/qNygP42zyoKp3xptRVCWRFEBvo=
+golang.org/x/mod v0.4.2/go.mod h1:s0Qsj1ACt9ePp/hMypM3fl4fZqREWJwdYDEqhRiZZUA=
+golang.org/x/net v0.0.0-20190404232315-eb5bcb51f2a3/go.mod h1:t9HGtf8HONx5eT2rtn7q6eTqICYqUVnKs3thJo3Qplg=
+golang.org/x/net v0.0.0-20190620200207-3b0461eec859/go.mod h1:z5CRVTTTmAJ677TzLLGU+0bjPO0LkuOLi4/5GtJWs/s=
+golang.org/x/net v0.0.0-20210805182204-aaa1db679c0d/go.mod h1:9nx3DQGgdP8bBQD5qxJ1jj9UTztislL4KSBs9R2vV5Y=
+golang.org/x/sync v0.0.0-20190423024810-112230192c58/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sync v0.0.0-20210220032951-036812b2e83c/go.mod h1:RxMgew5VJxzue5/jJTE5uejpjVlOe/izrB70Jof72aM=
+golang.org/x/sys v0.0.0-20190215142949-d0b11bdaac8a/go.mod h1:STP8DvDyc/dI5b8T5hshtkjS+E42TnysNCUPdjciGhY=
+golang.org/x/sys v0.0.0-20190412213103-97732733099d/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20201119102817-f84b799fce68/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210423082822-04245dca01da/go.mod h1:h1NjWce9XRLGQEsW7wpKNCjG9DtNlClVuFLEZdDNbEs=
+golang.org/x/sys v0.0.0-20210809222454-d867a43fc93e/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.0.0-20211030160813-b3129d9d1021 h1:giLT+HuUP/gXYrG2Plg9WTjj4qhfgaW424ZIFog3rlk=
+golang.org/x/sys v0.0.0-20211030160813-b3129d9d1021/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1/go.mod h1:bj7SfCRtBDWHUb9snDiAeCFNEtKQo2Wmx5Cou7ajbmo=
+golang.org/x/text v0.3.0/go.mod h1:NqM8EUOU14njkJ3fqMW+pc6Ldnwhi/IjpwHt7yyuwOQ=
+golang.org/x/text v0.3.6/go.mod h1:5Zoc/QRtKVWzQhOtBMvqHzDpF6irO9z98xDceosuGiQ=
+golang.org/x/tools v0.0.0-20180917221912-90fa682c2a6e/go.mod h1:n7NCudcB/nEzxVGmLbDWY5pfWTLqBcC2KZ6jyYvM4mQ=
+golang.org/x/tools v0.0.0-20191119224855-298f0cb1881e/go.mod h1:b+2E5dAYhXwXZwtnZ6UAqBI28+e2cm9otk0dWdXHAEo=
+golang.org/x/tools v0.1.7 h1:6j8CgantCy3yc8JGBqkDLMKWqZ0RDU2g1HVgacojGWQ=
+golang.org/x/tools v0.1.7/go.mod h1:LGqMHiF4EqQNHR1JncWGqT5BVaXmza+X+BDGol+dOxo=
+golang.org/x/xerrors v0.0.0-20190717185122-a985d3407aa7/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20191011141410-1b5146add898/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1 h1:go1bK/D/BFZV2I8cIQd1NKEZ+0owSTG1fDTci4IqFcE=
+golang.org/x/xerrors v0.0.0-20200804184101-5ec99f83aff1/go.mod h1:I/5z698sn9Ka8TeJc9MKroUUfqBBauWjQqLJ2OPfmY0=
+rsc.io/pdf v0.1.1/go.mod h1:n8OzWcQ6Sp37PL01nO98y4iUCRdTGarVfzxY20ICaU4=