Stowage/go
2
0
Fork 0
mirror of https://github.com/golang/go.git synced 2025-12-08 06:10:04 +00:00
Code Issues Projects Releases Packages Wiki Activity

cmd/compile: rework handling of udiv on ARM

Browse source 
Instead of populating the aux symbol
of CALLudiv during rewrite rules,
populate it during genssa.

This simplifies the rewrite rules.
It also removes all remaining calls
to ctxt.Lookup from any rewrite rules.
This is a first step towards removing
ctxt from ssa.Cache entirely,
and also a first step towards converting
the obj.LSym.Version field into a boolean.
It should also speed up compilation.

Also, move func udiv into package runtime.
That's where it is anyway,
and it lets udiv look and act like the rest of
the runtime support functions.

Change-Id: I41462a632c14fdc41f61b08049ec13cd80a87bfe
Reviewed-on: https://go-review.googlesource.com/41191
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
This commit is contained in:
Josh Bleecher Snyder 2017-04-20 07:50:17 -07:00
parent 6e97c71cb7
commit 01b1a34aac
9 changed files with 27 additions and 38 deletions
Download patch file Download diff file Expand all files Collapse all files

5
src/cmd/compile/internal/arm/ssa.go
View file

@ -622,7 +622,10 @@ func ssaGenValue(s *gc.SSAGenState, v *ssa.Value) {
p.From.Offset = v.AuxInt
p.To.Type = obj.TYPE_REG
p.To.Reg = v.Reg()
case ssa.OpARMCALLstatic, ssa.OpARMCALLclosure, ssa.OpARMCALLinter, ssa.OpARMCALLudiv:
case ssa.OpARMCALLstatic, ssa.OpARMCALLclosure, ssa.OpARMCALLinter:
s.Call(v)
case ssa.OpARMCALLudiv:
v.Aux = gc.Udiv
s.Call(v)
case ssa.OpARMDUFFZERO:
p := s.Prog(obj.ADUFFZERO)

3
src/cmd/compile/internal/gc/go.go
View file

@ -270,5 +270,6 @@ var (
writeBarrier,
writebarrierptr,
typedmemmove,
typedmemclr *obj.LSym
typedmemclr,
Udiv *obj.LSym
)

1
src/cmd/compile/internal/gc/ssa.go
View file

@ -91,6 +91,7 @@ func initssaconfig() {
writebarrierptr = Sysfunc("writebarrierptr")
typedmemmove = Sysfunc("typedmemmove")
typedmemclr = Sysfunc("typedmemclr")
Udiv = Sysfunc("udiv")
}
// buildssa builds an SSA function.

8
src/cmd/compile/internal/ssa/gen/ARM.rules
View file

@ -35,11 +35,11 @@
(Div32 x y) ->
(SUB (XOR <types.UInt32> // negate the result if one operand is negative
(Select0 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)}
(Select0 <types.UInt32> (CALLudiv
(SUB <types.UInt32> (XOR x <types.UInt32> (Signmask x)) (Signmask x)) // negate x if negative
(SUB <types.UInt32> (XOR y <types.UInt32> (Signmask y)) (Signmask y)))) // negate y if negative
(Signmask (XOR <types.UInt32> x y))) (Signmask (XOR <types.UInt32> x y)))
(Div32u x y) -> (Select0 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)} x y))
(Div32u x y) -> (Select0 <types.UInt32> (CALLudiv x y))
(Div16 x y) -> (Div32 (SignExt16to32 x) (SignExt16to32 y))
(Div16u x y) -> (Div32u (ZeroExt16to32 x) (ZeroExt16to32 y))
(Div8 x y) -> (Div32 (SignExt8to32 x) (SignExt8to32 y))
@ -49,11 +49,11 @@
(Mod32 x y) ->
(SUB (XOR <types.UInt32> // negate the result if x is negative
(Select1 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)}
(Select1 <types.UInt32> (CALLudiv
(SUB <types.UInt32> (XOR <types.UInt32> x (Signmask x)) (Signmask x)) // negate x if negative
(SUB <types.UInt32> (XOR <types.UInt32> y (Signmask y)) (Signmask y)))) // negate y if negative
(Signmask x)) (Signmask x))
(Mod32u x y) -> (Select1 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)} x y))
(Mod32u x y) -> (Select1 <types.UInt32> (CALLudiv x y))
(Mod16 x y) -> (Mod32 (SignExt16to32 x) (SignExt16to32 y))
(Mod16u x y) -> (Mod32u (ZeroExt16to32 x) (ZeroExt16to32 y))
(Mod8 x y) -> (Mod32 (SignExt8to32 x) (SignExt8to32 y))

5
src/cmd/compile/internal/ssa/gen/ARMOps.go
View file

@ -152,10 +152,7 @@ func init() {
},
clobberFlags: true,
typ: "(UInt32,UInt32)",
aux: "SymOff",
// TODO(mdempsky): Should this be true?
call: false,
symEffect: "None",
call: false, // TODO(mdempsky): Should this be true?
},
{name: "ADDS", argLength: 2, reg: gp21carry, asm: "ADD", commutative: true}, // arg0 + arg1, set carry flag

2
src/cmd/compile/internal/ssa/opGen.go
View file

@ -8151,10 +8151,8 @@ var opcodeTable = [...]opInfo{
},
{
name: "CALLudiv",
auxType: auxSymOff,
argLen: 2,
clobberFlags: true,
symEffect: SymNone,
reg: regInfo{
inputs: []inputInfo{
{0, 2}, // R1

20
src/cmd/compile/internal/ssa/rewriteARM.go
View file

@ -13582,13 +13582,11 @@ func rewriteValueARM_OpDiv16u(v *Value) bool {
func rewriteValueARM_OpDiv32(v *Value) bool {
b := v.Block
_ = b
config := b.Func.Config
_ = config
types := &b.Func.Config.Types
_ = types
// match: (Div32 x y)
// cond:
// result: (SUB (XOR <types.UInt32> (Select0 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)} (SUB <types.UInt32> (XOR x <types.UInt32> (Signmask x)) (Signmask x)) (SUB <types.UInt32> (XOR y <types.UInt32> (Signmask y)) (Signmask y)))) (Signmask (XOR <types.UInt32> x y))) (Signmask (XOR <types.UInt32> x y)))
// result: (SUB (XOR <types.UInt32> (Select0 <types.UInt32> (CALLudiv (SUB <types.UInt32> (XOR x <types.UInt32> (Signmask x)) (Signmask x)) (SUB <types.UInt32> (XOR y <types.UInt32> (Signmask y)) (Signmask y)))) (Signmask (XOR <types.UInt32> x y))) (Signmask (XOR <types.UInt32> x y)))
for {
x := v.Args[0]
y := v.Args[1]
@ -13596,7 +13594,6 @@ func rewriteValueARM_OpDiv32(v *Value) bool {
v0 := b.NewValue0(v.Pos, OpARMXOR, types.UInt32)
v1 := b.NewValue0(v.Pos, OpSelect0, types.UInt32)
v2 := b.NewValue0(v.Pos, OpARMCALLudiv, MakeTuple(types.UInt32, types.UInt32))
v2.Aux = config.ctxt.Lookup("udiv", 0)
v3 := b.NewValue0(v.Pos, OpARMSUB, types.UInt32)
v4 := b.NewValue0(v.Pos, OpARMXOR, types.UInt32)
v4.AddArg(x)
@ -13653,20 +13650,17 @@ func rewriteValueARM_OpDiv32F(v *Value) bool {
func rewriteValueARM_OpDiv32u(v *Value) bool {
b := v.Block
_ = b
config := b.Func.Config
_ = config
types := &b.Func.Config.Types
_ = types
// match: (Div32u x y)
// cond:
// result: (Select0 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)} x y))
// result: (Select0 <types.UInt32> (CALLudiv x y))
for {
x := v.Args[0]
y := v.Args[1]
v.reset(OpSelect0)
v.Type = types.UInt32
v0 := b.NewValue0(v.Pos, OpARMCALLudiv, MakeTuple(types.UInt32, types.UInt32))
v0.Aux = config.ctxt.Lookup("udiv", 0)
v0.AddArg(x)
v0.AddArg(y)
v.AddArg(v0)
@ -15088,13 +15082,11 @@ func rewriteValueARM_OpMod16u(v *Value) bool {
func rewriteValueARM_OpMod32(v *Value) bool {
b := v.Block
_ = b
config := b.Func.Config
_ = config
types := &b.Func.Config.Types
_ = types
// match: (Mod32 x y)
// cond:
// result: (SUB (XOR <types.UInt32> (Select1 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)} (SUB <types.UInt32> (XOR <types.UInt32> x (Signmask x)) (Signmask x)) (SUB <types.UInt32> (XOR <types.UInt32> y (Signmask y)) (Signmask y)))) (Signmask x)) (Signmask x))
// result: (SUB (XOR <types.UInt32> (Select1 <types.UInt32> (CALLudiv (SUB <types.UInt32> (XOR <types.UInt32> x (Signmask x)) (Signmask x)) (SUB <types.UInt32> (XOR <types.UInt32> y (Signmask y)) (Signmask y)))) (Signmask x)) (Signmask x))
for {
x := v.Args[0]
y := v.Args[1]
@ -15102,7 +15094,6 @@ func rewriteValueARM_OpMod32(v *Value) bool {
v0 := b.NewValue0(v.Pos, OpARMXOR, types.UInt32)
v1 := b.NewValue0(v.Pos, OpSelect1, types.UInt32)
v2 := b.NewValue0(v.Pos, OpARMCALLudiv, MakeTuple(types.UInt32, types.UInt32))
v2.Aux = config.ctxt.Lookup("udiv", 0)
v3 := b.NewValue0(v.Pos, OpARMSUB, types.UInt32)
v4 := b.NewValue0(v.Pos, OpARMXOR, types.UInt32)
v4.AddArg(x)
@ -15140,20 +15131,17 @@ func rewriteValueARM_OpMod32(v *Value) bool {
func rewriteValueARM_OpMod32u(v *Value) bool {
b := v.Block
_ = b
config := b.Func.Config
_ = config
types := &b.Func.Config.Types
_ = types
// match: (Mod32u x y)
// cond:
// result: (Select1 <types.UInt32> (CALLudiv {config.ctxt.Lookup("udiv", 0)} x y))
// result: (Select1 <types.UInt32> (CALLudiv x y))
for {
x := v.Args[0]
y := v.Args[1]
v.reset(OpSelect1)
v.Type = types.UInt32
v0 := b.NewValue0(v.Pos, OpARMCALLudiv, MakeTuple(types.UInt32, types.UInt32))
v0.Aux = config.ctxt.Lookup("udiv", 0)
v0.AddArg(x)
v0.AddArg(y)
v.AddArg(v0)

1
src/cmd/vet/all/whitelist/arm.txt
View file

@ -19,6 +19,7 @@ runtime/duff_arm.s: [arm] duffcopy: function duffcopy missing Go declaration
runtime/tls_arm.s: [arm] save_g: function save_g missing Go declaration
runtime/tls_arm.s: [arm] load_g: function load_g missing Go declaration
runtime/tls_arm.s: [arm] _initcgo: function _initcgo missing Go declaration
runtime/vlop_arm.s: [arm] udiv: function udiv missing Go declaration
// Clearer using FP than SP, but that requires named offsets.
runtime/asm_arm.s: [arm] rt0_go: use of 4(R13) points beyond argument frame

16
src/runtime/vlop_arm.s
View file

@ -106,7 +106,7 @@ TEXT runtime·_sfloatpanic(SB),NOSPLIT,$-4
MOVW g_sigpc(g), LR
B runtime·sigpanic(SB)
// func udiv(n, d uint32) (q, r uint32)
// func runtime·udiv(n, d uint32) (q, r uint32)
// compiler knowns the register usage of this function
// Reference:
// Sloss, Andrew et. al; ARM System Developer's Guide: Designing and Optimizing System Software
@ -118,7 +118,7 @@ TEXT runtime·_sfloatpanic(SB),NOSPLIT,$-4
#define Ra R11
// Be careful: Ra == R11 will be used by the linker for synthesized instructions.
TEXT udiv(SB),NOSPLIT,$-4
TEXT runtime·udiv(SB),NOSPLIT,$-4
MOVBU runtime·hardDiv(SB), Ra
CMP $0, Ra
BNE udiv_hardware
@ -241,7 +241,7 @@ TEXT _divu(SB), NOSPLIT, $16-0
MOVW Rn, Rr /* numerator */
MOVW g_m(g), Rq
MOVW m_divmod(Rq), Rq /* denominator */
BL udiv(SB)
BL runtime·udiv(SB)
MOVW Rq, RTMP
MOVW 4(R13), Rq
MOVW 8(R13), Rr
@ -259,7 +259,7 @@ TEXT _modu(SB), NOSPLIT, $16-0
MOVW Rn, Rr /* numerator */
MOVW g_m(g), Rq
MOVW m_divmod(Rq), Rq /* denominator */
BL udiv(SB)
BL runtime·udiv(SB)
MOVW Rr, RTMP
MOVW 4(R13), Rq
MOVW 8(R13), Rr
@ -283,7 +283,7 @@ TEXT _div(SB),NOSPLIT,$16-0
BGE d2
RSB $0, Rq, Rq
d0:
BL udiv(SB) /* none/both neg */
BL runtime·udiv(SB) /* none/both neg */
MOVW Rq, RTMP
B out1
d1:
@ -291,7 +291,7 @@ d1:
BGE d0
RSB $0, Rq, Rq
d2:
BL udiv(SB) /* one neg */
BL runtime·udiv(SB) /* one neg */
RSB $0, Rq, RTMP
out1:
MOVW 4(R13), Rq
@ -314,11 +314,11 @@ TEXT _mod(SB),NOSPLIT,$16-0
CMP $0, Rr
BGE m1
RSB $0, Rr, Rr
BL udiv(SB) /* neg numerator */
BL runtime·udiv(SB) /* neg numerator */
RSB $0, Rr, RTMP
B out
m1:
BL udiv(SB) /* pos numerator */
BL runtime·udiv(SB) /* pos numerator */
MOVW Rr, RTMP
out:
MOVW 4(R13), Rq