Revert "runtime: add wasm bulk memory operations" by Frank

ethstorage · Sep 11, 2023 · 85bdb1e · 85bdb1e
1 parent 91c77a5
commit 85bdb1e
Show file tree

Hide file tree

Showing 15 changed files with 355 additions and 66 deletions.
diff --git a/src/cmd/compile/internal/ir/symtab.go b/src/cmd/compile/internal/ir/symtab.go
@@ -61,6 +61,10 @@ var Syms struct {
 	// Wasm
 	WasmDiv *obj.LSym
 	// Wasm
+	WasmMove *obj.LSym
+	// Wasm
+	WasmZero *obj.LSym
+	// Wasm
 	WasmTruncS *obj.LSym
 	// Wasm
 	WasmTruncU *obj.LSym

diff --git a/src/cmd/compile/internal/ssa/_gen/Wasm.rules b/src/cmd/compile/internal/ssa/_gen/Wasm.rules
@@ -234,9 +234,24 @@
 	(I64Store [s-8] dst (I64Load [s-8] src mem)
 		(I64Store dst (I64Load src mem) mem))
 
+// Adjust moves to be a multiple of 16 bytes.
+(Move [s] dst src mem)
+	&& s > 16 && s%16 != 0 && s%16 <= 8 =>
+	(Move [s-s%16]
+		(OffPtr <dst.Type> dst [s%16])
+		(OffPtr <src.Type> src [s%16])
+		(I64Store dst (I64Load src mem) mem))
+(Move [s] dst src mem)
+	&& s > 16 && s%16 != 0 && s%16 > 8 =>
+	(Move [s-s%16]
+		(OffPtr <dst.Type> dst [s%16])
+		(OffPtr <src.Type> src [s%16])
+		(I64Store [8] dst (I64Load [8] src mem)
+			(I64Store dst (I64Load src mem) mem)))
+
 // Large copying uses helper.
-(Move [s] dst src mem) && logLargeCopy(v, s) =>
-	(LoweredMove [s] dst src mem)
+(Move [s] dst src mem) && s%8 == 0 && logLargeCopy(v, s) =>
+	(LoweredMove [s/8] dst src mem)
 
 // Lowering Zero instructions
 (Zero [0] _ mem) => mem
@@ -259,7 +274,7 @@
 		(I64Store32 destptr (I64Const [0]) mem))
 
 // Strip off any fractional word zeroing.
-(Zero [s] destptr mem) && s%8 != 0 && s > 8 && s < 32 =>
+(Zero [s] destptr mem) && s%8 != 0 && s > 8 =>
 	(Zero [s-s%8] (OffPtr <destptr.Type> destptr [s%8])
 		(I64Store destptr (I64Const [0]) mem))
 
@@ -278,8 +293,8 @@
 				(I64Store destptr (I64Const [0]) mem))))
 
 // Large zeroing uses helper.
-(Zero [s] destptr mem) =>
-	(LoweredZero [s] destptr mem)
+(Zero [s] destptr mem) && s%8 == 0 && s > 32 =>
+	(LoweredZero [s/8] destptr mem)
 
 // Lowering constants
 (Const64 ...) => (I64Const ...)

diff --git a/src/cmd/compile/internal/ssa/_gen/WasmOps.go b/src/cmd/compile/internal/ssa/_gen/WasmOps.go
@@ -126,8 +126,8 @@ func init() {
 		{name: "LoweredInterCall", argLength: 2, reg: regInfo{inputs: []regMask{gp}, clobbers: callerSave}, aux: "CallOff", call: true},          // call fn by pointer. arg0=codeptr, arg1=mem, auxint=argsize, returns mem
 
 		{name: "LoweredAddr", argLength: 1, reg: gp11, aux: "SymOff", rematerializeable: true, symEffect: "Addr"}, // returns base+aux+auxint, arg0=base
-		{name: "LoweredMove", argLength: 3, reg: regInfo{inputs: []regMask{gp, gp}}, aux: "Int64"},                // large move. arg0=dst, arg1=src, arg2=mem, auxint=len, returns mem
-		{name: "LoweredZero", argLength: 2, reg: regInfo{inputs: []regMask{gp}}, aux: "Int64"},                    // large zeroing. arg0=start, arg1=mem, auxint=len, returns mem
+		{name: "LoweredMove", argLength: 3, reg: regInfo{inputs: []regMask{gp, gp}}, aux: "Int64"},                // large move. arg0=dst, arg1=src, arg2=mem, auxint=len/8, returns mem
+		{name: "LoweredZero", argLength: 2, reg: regInfo{inputs: []regMask{gp}}, aux: "Int64"},                    // large zeroing. arg0=start, arg1=mem, auxint=len/8, returns mem                 // large zeroing. arg0=start, arg1=mem, auxint=len, returns mem
 
 		{name: "LoweredGetClosurePtr", reg: gp01},                                                                          // returns wasm.REG_CTXT, the closure pointer
 		{name: "LoweredGetCallerPC", reg: gp01, rematerializeable: true},                                                   // returns the PC of the caller of the current function

diff --git a/src/cmd/compile/internal/ssa/rewriteWasm.go b/src/cmd/compile/internal/ssa/rewriteWasm.go
diff --git a/src/cmd/compile/internal/ssagen/ssa.go b/src/cmd/compile/internal/ssagen/ssa.go
@@ -202,6 +202,8 @@ func InitConfig() {
 	}
 
 	// Wasm (all asm funcs with special ABIs)
+	ir.Syms.WasmMove = typecheck.LookupRuntimeVar("wasmMove")
+	ir.Syms.WasmZero = typecheck.LookupRuntimeVar("wasmZero")
 	ir.Syms.WasmDiv = typecheck.LookupRuntimeVar("wasmDiv")
 	ir.Syms.WasmTruncS = typecheck.LookupRuntimeVar("wasmTruncS")
 	ir.Syms.WasmTruncU = typecheck.LookupRuntimeVar("wasmTruncU")

diff --git a/src/cmd/compile/internal/wasm/ssa.go b/src/cmd/compile/internal/wasm/ssa.go
@@ -262,13 +262,14 @@ func ssaGenValue(s *ssagen.State, v *ssa.Value) {
 		getValue32(s, v.Args[0])
 		getValue32(s, v.Args[1])
 		i32Const(s, int32(v.AuxInt))
-		s.Prog(wasm.AMemoryCopy)
+		p := s.Prog(wasm.ACall)
+		p.To = obj.Addr{Type: obj.TYPE_MEM, Name: obj.NAME_EXTERN, Sym: ir.Syms.WasmMove}
 
 	case ssa.OpWasmLoweredZero:
 		getValue32(s, v.Args[0])
-		i32Const(s, 0)
 		i32Const(s, int32(v.AuxInt))
-		s.Prog(wasm.AMemoryFill)
+		p := s.Prog(wasm.ACall)
+		p.To = obj.Addr{Type: obj.TYPE_MEM, Name: obj.NAME_EXTERN, Sym: ir.Syms.WasmZero}
 
 	case ssa.OpWasmLoweredNilCheck:
 		getValue64(s, v.Args[0])

diff --git a/src/cmd/internal/obj/wasm/a.out.go b/src/cmd/internal/obj/wasm/a.out.go
@@ -230,17 +230,6 @@ const (
 	AI64TruncSatF64S
 	AI64TruncSatF64U
 
-	AMemoryInit
-	ADataDrop
-	AMemoryCopy
-	AMemoryFill
-	ATableInit
-	AElemDrop
-	ATableCopy
-	ATableGrow
-	ATableSize
-	ATableFill
-
 	ALast // Sentinel: End of low-level WebAssembly instructions.
 
 	ARESUMEPOINT

diff --git a/src/cmd/internal/obj/wasm/anames.go b/src/cmd/internal/obj/wasm/anames.go
diff --git a/src/cmd/internal/obj/wasm/wasmobj.go b/src/cmd/internal/obj/wasm/wasmobj.go
@@ -920,6 +920,8 @@ var notUsePC_B = map[string]bool{
 	"runtime.gcWriteBarrier6": true,
 	"runtime.gcWriteBarrier7": true,
 	"runtime.gcWriteBarrier8": true,
+	"runtime.wasmMove":        true,
+	"runtime.wasmZero":        true,
 	"runtime.wasmDiv":         true,
 	"runtime.wasmTruncS":      true,
 	"runtime.wasmTruncU":      true,
@@ -962,7 +964,8 @@ func assemble(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) {
 	// Some functions use a special calling convention.
 	switch s.Name {
 	case "_rt0_wasm_js", "_rt0_wasm_wasip1", "wasm_export_run", "wasm_export_resume", "wasm_export_getsp",
-		"wasm_pc_f_loop", "runtime.wasmDiv", "runtime.wasmTruncS", "runtime.wasmTruncU", "memeqbody":
+		"wasm_pc_f_loop", "runtime.wasmMove", "runtime.wasmZero", "runtime.wasmDiv", "runtime.wasmTruncS",
+		"runtime.wasmTruncU", "memeqbody":
 		varDecls = []*varDecl{}
 		useAssemblyRegMap()
 	case "memchr", "memcmp":
@@ -1216,11 +1219,7 @@ func assemble(ctxt *obj.Link, s *obj.LSym, newprog obj.ProgAlloc) {
 			writeUleb128(w, align(p.As))
 			writeUleb128(w, uint64(p.To.Offset))
 
-		case ACurrentMemory, AGrowMemory, AMemoryFill:
-			w.WriteByte(0x00)
-
-		case AMemoryCopy:
-			w.WriteByte(0x00)
+		case ACurrentMemory, AGrowMemory:
 			w.WriteByte(0x00)
 
 		}

diff --git a/src/cmd/link/internal/wasm/asm.go b/src/cmd/link/internal/wasm/asm.go
@@ -113,13 +113,15 @@ func readWasmImport(ldr *loader.Loader, s loader.Sym) obj.WasmImport {
 }
 
 var wasmFuncTypes = map[string]*wasmFuncType{
-	"_rt0_wasm_js":            {Params: []byte{}},                                         //
-	"_rt0_wasm_wasip1":        {Params: []byte{}},                                         //
-	"wasm_export__start":      {},                                                         //
-	"wasm_export_run":         {Params: []byte{I32, I32}},                                 // argc, argv
-	"wasm_export_resume":      {Params: []byte{}},                                         //
-	"wasm_export_getsp":       {Results: []byte{I32}},                                     // sp
-	"wasm_pc_f_loop":          {Params: []byte{}},                                         //
+	"_rt0_wasm_js":            {Params: []byte{}},         //
+	"_rt0_wasm_wasip1":        {Params: []byte{}},         //
+	"wasm_export__start":      {},                         //
+	"wasm_export_run":         {Params: []byte{I32, I32}}, // argc, argv
+	"wasm_export_resume":      {Params: []byte{}},         //
+	"wasm_export_getsp":       {Results: []byte{I32}},     // sp
+	"wasm_pc_f_loop":          {Params: []byte{}},
+	"runtime.wasmMove":        {Params: []byte{I32, I32, I32}},                            // dst, src, len
+	"runtime.wasmZero":        {Params: []byte{I32, I32}},                                 // ptr, len                                    //
 	"runtime.wasmDiv":         {Params: []byte{I64, I64}, Results: []byte{I64}},           // x, y -> x/y
 	"runtime.wasmTruncS":      {Params: []byte{F64}, Results: []byte{I64}},                // x -> int(x)
 	"runtime.wasmTruncU":      {Params: []byte{F64}, Results: []byte{I64}},                // x -> uint(x)

diff --git a/src/runtime/asm_wasm.s b/src/runtime/asm_wasm.s
@@ -322,8 +322,10 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-48; \
 		I64Load stackArgs+16(FP); \
 		I32WrapI64; \
 		I64Load stackArgsSize+24(FP); \
+		I64Const $3; \
+		I64ShrU; \
 		I32WrapI64; \
-		MemoryCopy; \
+		Call runtime·wasmMove(SB); \
 	End; \
 	\
 	MOVD f+8(FP), CTXT; \

diff --git a/src/runtime/memclr_wasm.s b/src/runtime/memclr_wasm.s
@@ -11,10 +11,29 @@ TEXT runtime·memclrNoHeapPointers(SB), NOSPLIT, $0-16
 	MOVD ptr+0(FP), R0
 	MOVD n+8(FP), R1
 
-	Get R0
-	I32WrapI64
-	I32Const $0
-	Get R1
-	I32WrapI64
-	MemoryFill
-	RET
+loop:
+	Loop
+		Get R1
+		I64Eqz
+		If
+			RET
+		End
+
+		Get R0
+		I32WrapI64
+		I64Const $0
+		I64Store8 $0
+
+		Get R0
+		I64Const $1
+		I64Add
+		Set R0
+
+		Get R1
+		I64Const $1
+		I64Sub
+		Set R1
+
+		Br loop
+	End
+	UNDEF