[dev.ssa] cmd/compile: support NaCl in SSA for ARM

NaCl code runs in sandbox and there are restrictions for its
instruction uses
(https://developer.chrome.com/native-client/reference/sandbox_internals/arm-32-bit-sandbox).

Like the legacy backend, on NaCl,
- don't use R9, which is used as NaCl's "thread pointer".
- don't use Duff's device.
- don't use indexed load/stores.
- the assembler rewrites DIV/MOD to runtime calls, which on NaCl
  clobbers R12, so R12 is marked as clobbered for DIV/MOD.
- other restrictions are satisfied by the assembler.

Enable SSA specific tests on nacl/arm, and disable non-SSA ones.

Updates #15365.

Change-Id: I9262693ec6756b89ca29d3ae4e52a96fe5403b02
Reviewed-on: https://go-review.googlesource.com/24859
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>

src/cmd/compile/internal/gc/ssa.go

@@ -37,12 +37,7 @@ func shouldssa(fn *Node) bool {
 		if os.Getenv("SSATEST") == "" {
 			return false
 		}
-	case "arm":
+	case "amd64", "arm":
-		// nacl/arm doesn't work yet
-		if obj.Getgoos() == "nacl" && os.Getenv("SSATEST") == "" {
-			return false
-		}
-	case "amd64":
 		// Generally available.
 	}
 	if !ssaEnabled {

src/cmd/compile/internal/ssa/config.go

@@ -30,6 +30,7 @@ type Config struct {
 	ctxt            *obj.Link                  // Generic arch information
 	optimize        bool                       // Do optimization
 	noDuffDevice    bool                       // Don't use Duff's device
+	nacl            bool                       // GOOS=nacl
 	sparsePhiCutoff uint64                     // Sparse phi location algorithm used above this #blocks*#variables score
 	curFunc         *Func
 
@@ -175,13 +176,25 @@ func NewConfig(arch string, fe Frontend, ctxt *obj.Link, optimize bool) *Config
 	}
 	c.ctxt = ctxt
 	c.optimize = optimize
+	c.nacl = obj.Getgoos() == "nacl"
 
-	// Don't use Duff's device on Plan 9, because floating
+	// Don't use Duff's device on Plan 9 AMD64, because floating
 	// point operations are not allowed in note handler.
-	if obj.Getgoos() == "plan9" {
+	if obj.Getgoos() == "plan9" && arch == "amd64" {
 		c.noDuffDevice = true
 	}
 
+	if c.nacl {
+		c.noDuffDevice = true // Don't use Duff's device on NaCl
+
+		// ARM assembler rewrites DIV/MOD to runtime calls, which
+		// clobber R12 on nacl
+		opcodeTable[OpARMDIV].reg.clobbers |= 1 << 12  // R12
+		opcodeTable[OpARMDIVU].reg.clobbers |= 1 << 12 // R12
+		opcodeTable[OpARMMOD].reg.clobbers |= 1 << 12  // R12
+		opcodeTable[OpARMMODU].reg.clobbers |= 1 << 12 // R12
+	}
+
 	// Assign IDs to preallocated values/blocks.
 	for i := range c.values {
 		c.values[i].ID = ID(i)

src/cmd/compile/internal/ssa/gen/ARM.rules

@@ -298,12 +298,12 @@
 // 4 and 128 are magic constants, see runtime/mkduff.go
 (Zero [s] ptr mem)
 	&& SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512
-	&& SizeAndAlign(s).Align()%4 == 0 ->
+	&& SizeAndAlign(s).Align()%4 == 0 && !config.noDuffDevice ->
 	(DUFFZERO [4 * (128 - int64(SizeAndAlign(s).Size()/4))] ptr (MOVWconst [0]) mem)
 
 // Large zeroing uses a loop
 (Zero [s] ptr mem)
-	&& SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 512
+	&& SizeAndAlign(s).Size()%4 == 0 && (SizeAndAlign(s).Size() > 512 || config.noDuffDevice)
 	&& SizeAndAlign(s).Align()%4 == 0 ->
 	(LoweredZero ptr (ADDconst <ptr.Type> ptr [SizeAndAlign(s).Size()]) (MOVWconst [0]) mem)
 
@@ -339,12 +339,12 @@
 // 8 and 128 are magic constants, see runtime/mkduff.go
 (Move [s] dst src mem)
 	&& SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512
-	&& SizeAndAlign(s).Align()%4 == 0 ->
+	&& SizeAndAlign(s).Align()%4 == 0 && !config.noDuffDevice ->
 	(DUFFCOPY [8 * (128 - int64(SizeAndAlign(s).Size()/4))] dst src mem)
 
 // Large move uses a loop
 (Move [s] dst src mem)
-	&& SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 512
+	&& SizeAndAlign(s).Size()%4 == 0 && (SizeAndAlign(s).Size() > 512 || config.noDuffDevice)
 	&& SizeAndAlign(s).Align()%4 == 0 ->
 	(LoweredMove dst src (ADDconst <src.Type> src [SizeAndAlign(s).Size()]) mem)
 
@@ -1128,14 +1128,14 @@
 (CMPshiftRAreg x y (MOVWconst [c])) -> (CMPshiftRA x y [c])
 
 // use indexed loads and stores
-(MOVWload [0] {sym} (ADD ptr idx) mem) && sym == nil -> (MOVWloadidx ptr idx mem)
+(MOVWload [0] {sym} (ADD ptr idx) mem) && sym == nil && !config.nacl -> (MOVWloadidx ptr idx mem)
-(MOVWstore [0] {sym} (ADD ptr idx) val mem) && sym == nil -> (MOVWstoreidx ptr idx val mem)
+(MOVWstore [0] {sym} (ADD ptr idx) val mem) && sym == nil && !config.nacl -> (MOVWstoreidx ptr idx val mem)
-(MOVWload [0] {sym} (ADDshiftLL ptr idx [c]) mem) && sym == nil -> (MOVWloadshiftLL ptr idx [c] mem)
+(MOVWload [0] {sym} (ADDshiftLL ptr idx [c]) mem) && sym == nil && !config.nacl -> (MOVWloadshiftLL ptr idx [c] mem)
-(MOVWload [0] {sym} (ADDshiftRL ptr idx [c]) mem) && sym == nil -> (MOVWloadshiftRL ptr idx [c] mem)
+(MOVWload [0] {sym} (ADDshiftRL ptr idx [c]) mem) && sym == nil && !config.nacl -> (MOVWloadshiftRL ptr idx [c] mem)
-(MOVWload [0] {sym} (ADDshiftRA ptr idx [c]) mem) && sym == nil -> (MOVWloadshiftRA ptr idx [c] mem)
+(MOVWload [0] {sym} (ADDshiftRA ptr idx [c]) mem) && sym == nil && !config.nacl -> (MOVWloadshiftRA ptr idx [c] mem)
-(MOVWstore [0] {sym} (ADDshiftLL ptr idx [c]) val mem) && sym == nil -> (MOVWstoreshiftLL ptr idx [c] val mem)
+(MOVWstore [0] {sym} (ADDshiftLL ptr idx [c]) val mem) && sym == nil && !config.nacl -> (MOVWstoreshiftLL ptr idx [c] val mem)
-(MOVWstore [0] {sym} (ADDshiftRL ptr idx [c]) val mem) && sym == nil -> (MOVWstoreshiftRL ptr idx [c] val mem)
+(MOVWstore [0] {sym} (ADDshiftRL ptr idx [c]) val mem) && sym == nil && !config.nacl -> (MOVWstoreshiftRL ptr idx [c] val mem)
-(MOVWstore [0] {sym} (ADDshiftRA ptr idx [c]) val mem) && sym == nil -> (MOVWstoreshiftRA ptr idx [c] val mem)
+(MOVWstore [0] {sym} (ADDshiftRA ptr idx [c]) val mem) && sym == nil && !config.nacl -> (MOVWstoreshiftRA ptr idx [c] val mem)
 
 // constant folding in indexed loads and stores
 (MOVWloadidx ptr (MOVWconst [c]) mem) -> (MOVWload [c] ptr mem)

src/cmd/compile/internal/ssa/regalloc.go

@@ -490,6 +490,9 @@ func (s *regAllocState) init(f *Func) {
 			s.f.Config.fe.Unimplementedf(0, "arch %s not implemented", s.f.Config.arch)
 		}
 	}
+	if s.f.Config.nacl && s.f.Config.arch == "arm" {
+		s.allocatable &^= 1 << 9 // R9 is "thread pointer" on nacl/arm
+	}
 
 	s.regs = make([]regState, s.numRegs)
 	s.values = make([]valState, f.NumValues())

src/cmd/compile/internal/ssa/rewriteARM.go

@@ -8382,7 +8382,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWload [0] {sym} (ADD ptr idx) mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWloadidx ptr idx mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8396,7 +8396,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		ptr := v_0.Args[0]
 		idx := v_0.Args[1]
 		mem := v.Args[1]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWloadidx)
@@ -8406,7 +8406,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWload [0] {sym} (ADDshiftLL ptr idx [c]) mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWloadshiftLL ptr idx [c] mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8421,7 +8421,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		idx := v_0.Args[1]
 		c := v_0.AuxInt
 		mem := v.Args[1]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWloadshiftLL)
@@ -8432,7 +8432,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWload [0] {sym} (ADDshiftRL ptr idx [c]) mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWloadshiftRL ptr idx [c] mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8447,7 +8447,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		idx := v_0.Args[1]
 		c := v_0.AuxInt
 		mem := v.Args[1]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWloadshiftRL)
@@ -8458,7 +8458,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWload [0] {sym} (ADDshiftRA ptr idx [c]) mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWloadshiftRA ptr idx [c] mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8473,7 +8473,7 @@ func rewriteValueARM_OpARMMOVWload(v *Value, config *Config) bool {
 		idx := v_0.Args[1]
 		c := v_0.AuxInt
 		mem := v.Args[1]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWloadshiftRA)
@@ -8875,7 +8875,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWstore [0] {sym} (ADD ptr idx) val mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWstoreidx ptr idx val mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8890,7 +8890,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		idx := v_0.Args[1]
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWstoreidx)
@@ -8901,7 +8901,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWstore [0] {sym} (ADDshiftLL ptr idx [c]) val mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWstoreshiftLL ptr idx [c] val mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8917,7 +8917,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		c := v_0.AuxInt
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWstoreshiftLL)
@@ -8929,7 +8929,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWstore [0] {sym} (ADDshiftRL ptr idx [c]) val mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWstoreshiftRL ptr idx [c] val mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8945,7 +8945,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		c := v_0.AuxInt
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWstoreshiftRL)
@@ -8957,7 +8957,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (MOVWstore [0] {sym} (ADDshiftRA ptr idx [c]) val mem)
-	// cond: sym == nil
+	// cond: sym == nil && !config.nacl
 	// result: (MOVWstoreshiftRA ptr idx [c] val mem)
 	for {
 		if v.AuxInt != 0 {
@@ -8973,7 +8973,7 @@ func rewriteValueARM_OpARMMOVWstore(v *Value, config *Config) bool {
 		c := v_0.AuxInt
 		val := v.Args[1]
 		mem := v.Args[2]
-		if !(sym == nil) {
+		if !(sym == nil && !config.nacl) {
 			break
 		}
 		v.reset(OpARMMOVWstoreshiftRA)
@@ -10670,14 +10670,14 @@ func rewriteValueARM_OpMove(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Move [s] dst src mem)
-	// cond: SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 	&& SizeAndAlign(s).Align()%4 == 0
+	// cond: SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 	&& SizeAndAlign(s).Align()%4 == 0 && !config.noDuffDevice
 	// result: (DUFFCOPY [8 * (128 - int64(SizeAndAlign(s).Size()/4))] dst src mem)
 	for {
 		s := v.AuxInt
 		dst := v.Args[0]
 		src := v.Args[1]
 		mem := v.Args[2]
-		if !(SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 && SizeAndAlign(s).Align()%4 == 0) {
+		if !(SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 && SizeAndAlign(s).Align()%4 == 0 && !config.noDuffDevice) {
 			break
 		}
 		v.reset(OpARMDUFFCOPY)
@@ -10688,14 +10688,14 @@ func rewriteValueARM_OpMove(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Move [s] dst src mem)
-	// cond: SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 512 	&& SizeAndAlign(s).Align()%4 == 0
+	// cond: SizeAndAlign(s).Size()%4 == 0 && (SizeAndAlign(s).Size() > 512 || config.noDuffDevice) 	&& SizeAndAlign(s).Align()%4 == 0
 	// result: (LoweredMove dst src (ADDconst <src.Type> src [SizeAndAlign(s).Size()]) mem)
 	for {
 		s := v.AuxInt
 		dst := v.Args[0]
 		src := v.Args[1]
 		mem := v.Args[2]
-		if !(SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 512 && SizeAndAlign(s).Align()%4 == 0) {
+		if !(SizeAndAlign(s).Size()%4 == 0 && (SizeAndAlign(s).Size() > 512 || config.noDuffDevice) && SizeAndAlign(s).Align()%4 == 0) {
 			break
 		}
 		v.reset(OpARMLoweredMove)
@@ -16786,13 +16786,13 @@ func rewriteValueARM_OpZero(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Zero [s] ptr mem)
-	// cond: SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 	&& SizeAndAlign(s).Align()%4 == 0
+	// cond: SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 	&& SizeAndAlign(s).Align()%4 == 0 && !config.noDuffDevice
 	// result: (DUFFZERO [4 * (128 - int64(SizeAndAlign(s).Size()/4))] ptr (MOVWconst [0]) mem)
 	for {
 		s := v.AuxInt
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 && SizeAndAlign(s).Align()%4 == 0) {
+		if !(SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 4 && SizeAndAlign(s).Size() <= 512 && SizeAndAlign(s).Align()%4 == 0 && !config.noDuffDevice) {
 			break
 		}
 		v.reset(OpARMDUFFZERO)
@@ -16805,13 +16805,13 @@ func rewriteValueARM_OpZero(v *Value, config *Config) bool {
 		return true
 	}
 	// match: (Zero [s] ptr mem)
-	// cond: SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 512 	&& SizeAndAlign(s).Align()%4 == 0
+	// cond: SizeAndAlign(s).Size()%4 == 0 && (SizeAndAlign(s).Size() > 512 || config.noDuffDevice) 	&& SizeAndAlign(s).Align()%4 == 0
 	// result: (LoweredZero ptr (ADDconst <ptr.Type> ptr [SizeAndAlign(s).Size()]) (MOVWconst [0]) mem)
 	for {
 		s := v.AuxInt
 		ptr := v.Args[0]
 		mem := v.Args[1]
-		if !(SizeAndAlign(s).Size()%4 == 0 && SizeAndAlign(s).Size() > 512 && SizeAndAlign(s).Align()%4 == 0) {
+		if !(SizeAndAlign(s).Size()%4 == 0 && (SizeAndAlign(s).Size() > 512 || config.noDuffDevice) && SizeAndAlign(s).Align()%4 == 0) {
 			break
 		}
 		v.reset(OpARMLoweredZero)

test/live.go

@@ -1,4 +1,4 @@
-// +build !amd64,!arm nacl,arm
+// +build !amd64,!arm
 // errorcheck -0 -l -live -wb=0
 
 // Copyright 2014 The Go Authors. All rights reserved.

test/live_ssa.go

@@ -1,4 +1,4 @@
-// +build amd64 arm,!nacl
+// +build amd64 arm
 // errorcheck -0 -l -live -wb=0
 
 // Copyright 2014 The Go Authors. All rights reserved.

test/nilptr3.go

@@ -2,7 +2,7 @@
 // Fails on ppc64x because of incomplete optimization.
 // See issues 9058.
 // Same reason for mips64x and s390x.
-// +build !ppc64,!ppc64le,!mips64,!mips64le,!amd64,!s390x,!arm nacl,arm
+// +build !ppc64,!ppc64le,!mips64,!mips64le,!amd64,!s390x,!arm
 
 // Copyright 2013 The Go Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style

test/nilptr3_ssa.go

@@ -1,5 +1,5 @@
 // errorcheck -0 -d=nil
-// +build amd64 arm,!nacl
+// +build amd64 arm
 
 // Copyright 2013 The Go Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style

test/sliceopt.go

@@ -1,4 +1,4 @@
-// +build !amd64,!arm nacl,arm
+// +build !amd64,!arm
 // errorcheck -0 -d=append,slice
 
 // Copyright 2015 The Go Authors. All rights reserved.