internal/bytealg: add wasm architecture

This commit adds the wasm architecture to the internal/bytealg package.

Some parts of the assembly code have been extracted from WebAssembly
bytecode generated with Emscripten (which uses musl libc).

Updates #18892

Change-Id: Iba7f7158356b816c9ad03ca9223903a41a024da6
Reviewed-on: https://go-review.googlesource.com/103915Reviewed-by: Keith Randall <khr@golang.org>

src/internal/bytealg/compare_generic.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !386,!amd64,!amd64p32,!s390x,!arm,!arm64,!ppc64,!ppc64le,!mips,!mipsle
+// +build !386,!amd64,!amd64p32,!s390x,!arm,!arm64,!ppc64,!ppc64le,!mips,!mipsle,!wasm
 
 package bytealg
 

src/internal/bytealg/compare_native.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build 386 amd64 amd64p32 s390x arm arm64 ppc64 ppc64le mips mipsle
+// +build 386 amd64 amd64p32 s390x arm arm64 ppc64 ppc64le mips mipsle wasm
 
 package bytealg
 

src/internal/bytealg/compare_wasm.s

+// Copyright 2018 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.
+
+#include "go_asm.h"
+#include "textflag.h"
+
+TEXT ·Compare(SB), NOSPLIT, $0-56
+	Get SP
+	I64Load s1_base+0(FP)
+	I64Load s1_len+8(FP)
+	I64Load s2_base+24(FP)
+	I64Load s2_len+32(FP)
+	Call cmpbody<>(SB)
+	I64Store ret+48(FP)
+	RET
+
+TEXT bytes·Compare(SB), NOSPLIT, $0-56
+	Get SP
+	I64Load s1_base+0(FP)
+	I64Load s1_len+8(FP)
+	I64Load s2_base+24(FP)
+	I64Load s2_len+32(FP)
+	Call cmpbody<>(SB)
+	I64Store ret+48(FP)
+	RET
+
+TEXT runtime·cmpstring(SB), NOSPLIT, $0-40
+	Get SP
+	I64Load s1_base+0(FP)
+	I64Load s1_len+8(FP)
+	I64Load s2_base+16(FP)
+	I64Load s2_len+24(FP)
+	Call cmpbody<>(SB)
+	I64Store ret+32(FP)
+	RET
+
+// params: a, alen, b, blen
+// ret: -1/0/1
+TEXT cmpbody<>(SB), NOSPLIT, $0-0
+	// len = min(alen, blen)
+	Get R1
+	Get R3
+	Get R1
+	Get R3
+	I64LtU
+	Select
+	Set R4
+
+	Get R0
+	I32WrapI64
+	Get R2
+	I32WrapI64
+	Get R4
+	I32WrapI64
+	Call memcmp<>(SB)
+	I64ExtendSI32
+	Set R5
+
+	Get R5
+	I64Eqz
+	If
+		// check length
+		Get R1
+		Get R3
+		I64Sub
+		Set R5
+	End
+
+	I64Const $0
+	I64Const $-1
+	I64Const $1
+	Get R5
+	I64Const $0
+	I64LtS
+	Select
+	Get R5
+	I64Eqz
+	Select
+	Return
+
+// compiled with emscripten
+// params: a, b, len
+// ret: <0/0/>0
+TEXT memcmp<>(SB), NOSPLIT, $0-0
+	Get R2
+	If $1
+	Loop
+	Get R0
+	I32Load8S $0
+	Tee R3
+	Get R1
+	I32Load8S $0
+	Tee R4
+	I32Eq
+	If
+	Get R0
+	I32Const $1
+	I32Add
+	Set R0
+	Get R1
+	I32Const $1
+	I32Add
+	Set R1
+	I32Const $0
+	Get R2
+	I32Const $-1
+	I32Add
+	Tee R2
+	I32Eqz
+	BrIf $3
+	Drop
+	Br $1
+	End
+	End
+	Get R3
+	I32Const $255
+	I32And
+	Get R4
+	I32Const $255
+	I32And
+	I32Sub
+	Else
+	I32Const $0
+	End
+	Return

src/internal/bytealg/equal_wasm.s

+// Copyright 2018 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.
+
+#include "go_asm.h"
+#include "textflag.h"
+
+TEXT ·Equal(SB), NOSPLIT, $0-49
+	MOVD a_len+8(FP), R0
+	MOVD b_len+32(FP), R1
+	Get R0
+	Get R1
+	I64Eq
+	If
+		Get SP
+		I64Load a+0(FP)
+		I64Load b+24(FP)
+		Get R0
+		Call memeqbody<>(SB)
+		I64Store8 ret+48(FP)
+	Else
+		Get SP
+		I64Const $0
+		I64Store8 ret+48(FP)
+	End
+	RET
+
+TEXT bytes·Equal(SB), NOSPLIT, $0-49
+	MOVD a_len+8(FP), R0
+	MOVD b_len+32(FP), R1
+	Get R0
+	Get R1
+	I64Eq
+	If
+		Get SP
+		I64Load a+0(FP)
+		I64Load b+24(FP)
+		Get R0
+		Call memeqbody<>(SB)
+		I64Store8 ret+48(FP)
+	Else
+		Get SP
+		I64Const $0
+		I64Store8 ret+48(FP)
+	End
+	RET
+
+// memequal(p, q unsafe.Pointer, size uintptr) bool
+TEXT runtime·memequal(SB), NOSPLIT, $0-25
+	Get SP
+	I64Load a+0(FP)
+	I64Load b+8(FP)
+	I64Load size+16(FP)
+	Call memeqbody<>(SB)
+	I64Store8 ret+24(FP)
+	RET
+
+// memequal_varlen(a, b unsafe.Pointer) bool
+TEXT runtime·memequal_varlen(SB), NOSPLIT, $0-17
+	Get SP
+	I64Load a+0(FP)
+	I64Load b+8(FP)
+	I64Load 8(CTXT) // compiler stores size at offset 8 in the closure
+	Call memeqbody<>(SB)
+	I64Store8 ret+16(FP)
+	RET
+
+// params: a, b, len
+// ret: 0/1
+TEXT memeqbody<>(SB), NOSPLIT, $0-0
+	Get R0
+	Get R1
+	I64Eq
+	If
+		I64Const $1
+		Return
+	End
+
+loop:
+	Loop
+		Get R2
+		I64Eqz
+		If
+			I64Const $1
+			Return
+		End
+
+		Get R0
+		I32WrapI64
+		I64Load8U $0
+		Get R1
+		I32WrapI64
+		I64Load8U $0
+		I64Ne
+		If
+			I64Const $0
+			Return
+		End
+
+		Get R0
+		I64Const $1
+		I64Add
+		Set R0
+
+		Get R1
+		I64Const $1
+		I64Add
+		Set R1
+
+		Get R2
+		I64Const $1
+		I64Sub
+		Set R2
+
+		Br loop
+	End
+	UNDEF

src/internal/bytealg/indexbyte_generic.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build !386,!amd64,!amd64p32,!s390x,!arm,!arm64,!ppc64,!ppc64le,!mips,!mipsle,!mips64,!mips64le
+// +build !386,!amd64,!amd64p32,!s390x,!arm,!arm64,!ppc64,!ppc64le,!mips,!mipsle,!mips64,!mips64le,!wasm
 
 package bytealg
 

src/internal/bytealg/indexbyte_native.go

@@ -2,7 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
-// +build 386 amd64 amd64p32 s390x arm arm64 ppc64 ppc64le mips mipsle mips64 mips64le
+// +build 386 amd64 amd64p32 s390x arm arm64 ppc64 ppc64le mips mipsle mips64 mips64le wasm
 
 package bytealg
 

src/internal/bytealg/indexbyte_wasm.s

+// Copyright 2018 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.
+
+#include "go_asm.h"
+#include "textflag.h"
+
+TEXT ·IndexByte(SB), NOSPLIT, $0-40
+	I64Load s+0(FP)
+	I32WrapI64
+	I32Load8U c+24(FP)
+	I64Load s_len+8(FP)
+	I32WrapI64
+	Call memchr<>(SB)
+	I64ExtendSI32
+	Set R0
+
+	Get SP
+	I64Const $-1
+	Get R0
+	I64Load s+0(FP)
+	I64Sub
+	Get R0
+	I64Eqz $0
+	Select
+	I64Store ret+32(FP)
+
+	RET
+
+TEXT ·IndexByteString(SB), NOSPLIT, $0-32
+	Get SP
+	I64Load s+0(FP)
+	I32WrapI64
+	I32Load8U c+16(FP)
+	I64Load s_len+8(FP)
+	I32WrapI64
+	Call memchr<>(SB)
+	I64ExtendSI32
+	Set R0
+
+	I64Const $-1
+	Get R0
+	I64Load s+0(FP)
+	I64Sub
+	Get R0
+	I64Eqz $0
+	Select
+	I64Store ret+24(FP)
+
+	RET
+
+TEXT bytes·IndexByte(SB), NOSPLIT, $0-40
+	Get SP
+	I64Load s+0(FP)
+	I32WrapI64
+	I32Load8U c+24(FP)
+	I64Load s_len+8(FP)
+	I32WrapI64
+	Call memchr<>(SB)
+	I64ExtendSI32
+	Set R0
+
+	I64Const $-1
+	Get R0
+	I64Load s+0(FP)
+	I64Sub
+	Get R0
+	I64Eqz $0
+	Select
+	I64Store ret+32(FP)
+
+	RET
+
+TEXT strings·IndexByte(SB), NOSPLIT, $0-32
+	Get SP
+	I64Load s+0(FP)
+	I32WrapI64
+	I32Load8U c+16(FP)
+	I64Load s_len+8(FP)
+	I32WrapI64
+	Call memchr<>(SB)
+	I64ExtendSI32
+	Set R0
+
+	I64Const $-1
+	Get R0
+	I64Load s+0(FP)
+	I64Sub
+	Get R0
+	I64Eqz $0
+	Select
+	I64Store ret+24(FP)
+	RET
+
+// compiled with emscripten
+// params: s, c, len
+// ret: index
+TEXT memchr<>(SB), NOSPLIT, $0
+	Get R1
+	I32Const $255
+	I32And
+	Set R4
+	Block
+	Block
+	Get R2
+	I32Const $0
+	I32Ne
+	Tee R3
+	Get R0
+	I32Const $3
+	I32And
+	I32Const $0
+	I32Ne
+	I32And
+	If
+	Get R1
+	I32Const $255
+	I32And
+	Set R5
+	Loop
+	Get R0
+	I32Load8U $0
+	Get R5
+	I32Eq
+	BrIf $2
+	Get R2
+	I32Const $-1
+	I32Add
+	Tee R2
+	I32Const $0
+	I32Ne
+	Tee R3
+	Get R0
+	I32Const $1
+	I32Add
+	Tee R0
+	I32Const $3
+	I32And
+	I32Const $0
+	I32Ne
+	I32And
+	BrIf $0
+	End
+	End
+	Get R3
+	BrIf $0
+	I32Const $0
+	Set R1
+	Br $1
+	End
+	Get R0
+	I32Load8U $0
+	Get R1
+	I32Const $255
+	I32And
+	Tee R3
+	I32Eq
+	If
+	Get R2
+	Set R1
+	Else
+	Get R4
+	I32Const $16843009
+	I32Mul
+	Set R4
+	Block
+	Block
+	Get R2
+	I32Const $3
+	I32GtU
+	If
+	Get R2
+	Set R1
+	Loop
+	Get R0
+	I32Load $0
+	Get R4
+	I32Xor
+	Tee R2
+	I32Const $-2139062144
+	I32And
+	I32Const $-2139062144
+	I32Xor
+	Get R2
+	I32Const $-16843009
+	I32Add
+	I32And
+	I32Eqz
+	If
+	Get R0
+	I32Const $4
+	I32Add
+	Set R0
+	Get R1
+	I32Const $-4
+	I32Add
+	Tee R1
+	I32Const $3
+	I32GtU
+	BrIf $1
+	Br $3
+	End
+	End
+	Else
+	Get R2
+	Set R1
+	Br $1
+	End
+	Br $1
+	End
+	Get R1
+	I32Eqz
+	If
+	I32Const $0
+	Set R1
+	Br $3
+	End
+	End
+	Loop
+	Get R0
+	I32Load8U $0
+	Get R3
+	I32Eq
+	BrIf $2
+	Get R0
+	I32Const $1
+	I32Add
+	Set R0
+	Get R1
+	I32Const $-1
+	I32Add
+	Tee R1
+	BrIf $0
+	I32Const $0
+	Set R1
+	End
+	End
+	End
+	Get R0
+	I32Const $0
+	Get R1
+	Select
+	Return