Skip to content

G
go
Commit a9ba3e30 authored by Richard Musiol's avatar Richard Musiol Committed by Brad Fitzpatrick
Browse files
Options

cmd/compile: add SSA config options useAvg and useHmul 

This commit allows architectures to disable optimizations that need the
Avg* and Hmul* operations.

WebAssembly has no such operations, so using them as an optimization
but then having to emulate them with multiple instructions makes no
sense, especially since the WebAssembly compiler may do the same
optimizations internally.

Updates #18892

Change-Id: If57b59e3235482a9e0ec334a7312b3e3b5fc2b61
Reviewed-on: https://go-review.googlesource.com/103256
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: default avatarJosh Bleecher Snyder <josharian@gmail.com>
parent 80e69220
Hide whitespace changes
Inline Side-by-side
Showing with 42 additions and 36 deletions
src/cmd/compile/internal/ssa/config.go
View file @ a9ba3e30
...@@ -34,6 +34,8 @@ type Config struct { ...@@ -34,6 +34,8 @@ type Config struct {
optimize bool // Do optimization optimize bool // Do optimization
noDuffDevice bool // Don't use Duff's device noDuffDevice bool // Don't use Duff's device
useSSE bool // Use SSE for non-float operations useSSE bool // Use SSE for non-float operations
useAvg bool // Use optimizations that need Avg* operations
useHmul bool // Use optimizations that need Hmul* operations
nacl bool // GOOS=nacl nacl bool // GOOS=nacl
use387 bool // GO386=387 use387 bool // GO386=387
SoftFloat bool // SoftFloat bool //
...@@ -190,6 +192,8 @@ const ( ...@@ -190,6 +192,8 @@ const (
// NewConfig returns a new configuration object for the given architecture. // NewConfig returns a new configuration object for the given architecture.
func NewConfig(arch string, types Types, ctxt *obj.Link, optimize bool) *Config { func NewConfig(arch string, types Types, ctxt *obj.Link, optimize bool) *Config {
c := &Config{arch: arch, Types: types} c := &Config{arch: arch, Types: types}
c.useAvg = true
c.useHmul = true
switch arch { switch arch {
case "amd64": case "amd64":
c.PtrSize = 8 c.PtrSize = 8
......
src/cmd/compile/internal/ssa/gen/generic.rules
View file @ a9ba3e30
...@@ -845,7 +845,7 @@ ...@@ -845,7 +845,7 @@
(Const32 <typ.UInt32> [int64(1<<15+(umagic(16,c).m+1)/2)]) (Const32 <typ.UInt32> [int64(1<<15+(umagic(16,c).m+1)/2)])
(Rsh32Ux64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [1]))) (Rsh32Ux64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [1])))
(Const64 <typ.UInt64> [16+umagic(16,c).s-2]))) (Const64 <typ.UInt64> [16+umagic(16,c).s-2])))
(Div16u x (Const16 [c])) && umagicOK(16, c) && config.RegSize == 4 -> (Div16u x (Const16 [c])) && umagicOK(16, c) && config.RegSize == 4 && config.useAvg ->
(Trunc32to16 (Trunc32to16
(Rsh32Ux64 <typ.UInt32> (Rsh32Ux64 <typ.UInt32>
(Avg32u (Avg32u
...@@ -856,19 +856,19 @@ ...@@ -856,19 +856,19 @@
(Const64 <typ.UInt64> [16+umagic(16,c).s-1]))) (Const64 <typ.UInt64> [16+umagic(16,c).s-1])))
// For 32-bit divides on 32-bit machines // For 32-bit divides on 32-bit machines
(Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 4 && umagic(32,c).m&1 == 0 -> (Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 4 && umagic(32,c).m&1 == 0 && config.useHmul ->
(Rsh32Ux64 <typ.UInt32> (Rsh32Ux64 <typ.UInt32>
(Hmul32u <typ.UInt32> (Hmul32u <typ.UInt32>
(Const32 <typ.UInt32> [int64(int32(1<<31+umagic(32,c).m/2))]) (Const32 <typ.UInt32> [int64(int32(1<<31+umagic(32,c).m/2))])
x) x)
(Const64 <typ.UInt64> [umagic(32,c).s-1])) (Const64 <typ.UInt64> [umagic(32,c).s-1]))
(Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 4 && c&1 == 0 -> (Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 4 && c&1 == 0 && config.useHmul ->
(Rsh32Ux64 <typ.UInt32> (Rsh32Ux64 <typ.UInt32>
(Hmul32u <typ.UInt32> (Hmul32u <typ.UInt32>
(Const32 <typ.UInt32> [int64(int32(1<<31+(umagic(32,c).m+1)/2))]) (Const32 <typ.UInt32> [int64(int32(1<<31+(umagic(32,c).m+1)/2))])
(Rsh32Ux64 <typ.UInt32> x (Const64 <typ.UInt64> [1]))) (Rsh32Ux64 <typ.UInt32> x (Const64 <typ.UInt64> [1])))
(Const64 <typ.UInt64> [umagic(32,c).s-2])) (Const64 <typ.UInt64> [umagic(32,c).s-2]))
(Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 4 -> (Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 4 && config.useAvg && config.useHmul ->
(Rsh32Ux64 <typ.UInt32> (Rsh32Ux64 <typ.UInt32>
(Avg32u (Avg32u
x x
...@@ -893,7 +893,7 @@ ...@@ -893,7 +893,7 @@
(Const64 <typ.UInt64> [int64(1<<31+(umagic(32,c).m+1)/2)]) (Const64 <typ.UInt64> [int64(1<<31+(umagic(32,c).m+1)/2)])
(Rsh64Ux64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [1]))) (Rsh64Ux64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [1])))
(Const64 <typ.UInt64> [32+umagic(32,c).s-2]))) (Const64 <typ.UInt64> [32+umagic(32,c).s-2])))
(Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 8 -> (Div32u x (Const32 [c])) && umagicOK(32, c) && config.RegSize == 8 && config.useAvg ->
(Trunc64to32 (Trunc64to32
(Rsh64Ux64 <typ.UInt64> (Rsh64Ux64 <typ.UInt64>
(Avg64u (Avg64u
...@@ -905,19 +905,19 @@ ...@@ -905,19 +905,19 @@
// For 64-bit divides on 64-bit machines // For 64-bit divides on 64-bit machines
// (64-bit divides on 32-bit machines are lowered to a runtime call by the walk pass.) // (64-bit divides on 32-bit machines are lowered to a runtime call by the walk pass.)
(Div64u x (Const64 [c])) && umagicOK(64, c) && config.RegSize == 8 && umagic(64,c).m&1 == 0 -> (Div64u x (Const64 [c])) && umagicOK(64, c) && config.RegSize == 8 && umagic(64,c).m&1 == 0 && config.useHmul ->
(Rsh64Ux64 <typ.UInt64> (Rsh64Ux64 <typ.UInt64>
(Hmul64u <typ.UInt64> (Hmul64u <typ.UInt64>
(Const64 <typ.UInt64> [int64(1<<63+umagic(64,c).m/2)]) (Const64 <typ.UInt64> [int64(1<<63+umagic(64,c).m/2)])
x) x)
(Const64 <typ.UInt64> [umagic(64,c).s-1])) (Const64 <typ.UInt64> [umagic(64,c).s-1]))
(Div64u x (Const64 [c])) && umagicOK(64, c) && config.RegSize == 8 && c&1 == 0 -> (Div64u x (Const64 [c])) && umagicOK(64, c) && config.RegSize == 8 && c&1 == 0 && config.useHmul ->
(Rsh64Ux64 <typ.UInt64> (Rsh64Ux64 <typ.UInt64>
(Hmul64u <typ.UInt64> (Hmul64u <typ.UInt64>
(Const64 <typ.UInt64> [int64(1<<63+(umagic(64,c).m+1)/2)]) (Const64 <typ.UInt64> [int64(1<<63+(umagic(64,c).m+1)/2)])
(Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [1]))) (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [1])))
(Const64 <typ.UInt64> [umagic(64,c).s-2])) (Const64 <typ.UInt64> [umagic(64,c).s-2]))
(Div64u x (Const64 [c])) && umagicOK(64, c) && config.RegSize == 8 -> (Div64u x (Const64 [c])) && umagicOK(64, c) && config.RegSize == 8 && config.useAvg && config.useHmul ->
(Rsh64Ux64 <typ.UInt64> (Rsh64Ux64 <typ.UInt64>
(Avg64u (Avg64u
x x
...@@ -992,7 +992,7 @@ ...@@ -992,7 +992,7 @@
(Rsh64x64 <t> (Rsh64x64 <t>
(SignExt32to64 x) (SignExt32to64 x)
(Const64 <typ.UInt64> [63]))) (Const64 <typ.UInt64> [63])))
(Div32 <t> x (Const32 [c])) && smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 == 0 -> (Div32 <t> x (Const32 [c])) && smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 == 0 && config.useHmul ->
(Sub32 <t> (Sub32 <t>
(Rsh32x64 <t> (Rsh32x64 <t>
(Hmul32 <t> (Hmul32 <t>
...@@ -1002,7 +1002,7 @@ ...@@ -1002,7 +1002,7 @@
(Rsh32x64 <t> (Rsh32x64 <t>
x x
(Const64 <typ.UInt64> [31]))) (Const64 <typ.UInt64> [31])))
(Div32 <t> x (Const32 [c])) && smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 != 0 -> (Div32 <t> x (Const32 [c])) && smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 != 0 && config.useHmul ->
(Sub32 <t> (Sub32 <t>
(Rsh32x64 <t> (Rsh32x64 <t>
(Add32 <t> (Add32 <t>
...@@ -1014,7 +1014,7 @@ ...@@ -1014,7 +1014,7 @@
(Rsh32x64 <t> (Rsh32x64 <t>
x x
(Const64 <typ.UInt64> [31]))) (Const64 <typ.UInt64> [31])))
(Div64 <t> x (Const64 [c])) && smagicOK(64,c) && smagic(64,c).m&1 == 0 -> (Div64 <t> x (Const64 [c])) && smagicOK(64,c) && smagic(64,c).m&1 == 0 && config.useHmul ->
(Sub64 <t> (Sub64 <t>
(Rsh64x64 <t> (Rsh64x64 <t>
(Hmul64 <t> (Hmul64 <t>
...@@ -1024,7 +1024,7 @@ ...@@ -1024,7 +1024,7 @@
(Rsh64x64 <t> (Rsh64x64 <t>
x x
(Const64 <typ.UInt64> [63]))) (Const64 <typ.UInt64> [63])))
(Div64 <t> x (Const64 [c])) && smagicOK(64,c) && smagic(64,c).m&1 != 0 -> (Div64 <t> x (Const64 [c])) && smagicOK(64,c) && smagic(64,c).m&1 != 0 && config.useHmul ->
(Sub64 <t> (Sub64 <t>
(Rsh64x64 <t> (Rsh64x64 <t>
(Add64 <t> (Add64 <t>
......
src/cmd/compile/internal/ssa/rewritegeneric.go
View file @ a9ba3e30
...@@ -7990,7 +7990,7 @@ func rewriteValuegeneric_OpDiv16u_0(v *Value) bool { ...@@ -7990,7 +7990,7 @@ func rewriteValuegeneric_OpDiv16u_0(v *Value) bool {
return true return true
} }
// match: (Div16u x (Const16 [c])) // match: (Div16u x (Const16 [c]))
// cond: umagicOK(16, c) && config.RegSize == 4 // cond: umagicOK(16, c) && config.RegSize == 4 && config.useAvg
// result: (Trunc32to16 (Rsh32Ux64 <typ.UInt32> (Avg32u (Lsh32x64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [16])) (Mul32 <typ.UInt32> (Const32 <typ.UInt32> [int64(umagic(16,c).m)]) (ZeroExt16to32 x))) (Const64 <typ.UInt64> [16+umagic(16,c).s-1]))) // result: (Trunc32to16 (Rsh32Ux64 <typ.UInt32> (Avg32u (Lsh32x64 <typ.UInt32> (ZeroExt16to32 x) (Const64 <typ.UInt64> [16])) (Mul32 <typ.UInt32> (Const32 <typ.UInt32> [int64(umagic(16,c).m)]) (ZeroExt16to32 x))) (Const64 <typ.UInt64> [16+umagic(16,c).s-1])))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8000,7 +8000,7 @@ func rewriteValuegeneric_OpDiv16u_0(v *Value) bool { ...@@ -8000,7 +8000,7 @@ func rewriteValuegeneric_OpDiv16u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(16, c) && config.RegSize == 4) { if !(umagicOK(16, c) && config.RegSize == 4 && config.useAvg) {
break break
} }
v.reset(OpTrunc32to16) v.reset(OpTrunc32to16)
...@@ -8207,7 +8207,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool { ...@@ -8207,7 +8207,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool {
return true return true
} }
// match: (Div32 <t> x (Const32 [c])) // match: (Div32 <t> x (Const32 [c]))
// cond: smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 == 0 // cond: smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 == 0 && config.useHmul
// result: (Sub32 <t> (Rsh32x64 <t> (Hmul32 <t> (Const32 <typ.UInt32> [int64(int32(smagic(32,c).m/2))]) x) (Const64 <typ.UInt64> [smagic(32,c).s-1])) (Rsh32x64 <t> x (Const64 <typ.UInt64> [31]))) // result: (Sub32 <t> (Rsh32x64 <t> (Hmul32 <t> (Const32 <typ.UInt32> [int64(int32(smagic(32,c).m/2))]) x) (Const64 <typ.UInt64> [smagic(32,c).s-1])) (Rsh32x64 <t> x (Const64 <typ.UInt64> [31])))
for { for {
t := v.Type t := v.Type
...@@ -8218,7 +8218,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool { ...@@ -8218,7 +8218,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(smagicOK(32, c) && config.RegSize == 4 && smagic(32, c).m&1 == 0) { if !(smagicOK(32, c) && config.RegSize == 4 && smagic(32, c).m&1 == 0 && config.useHmul) {
break break
} }
v.reset(OpSub32) v.reset(OpSub32)
...@@ -8243,7 +8243,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool { ...@@ -8243,7 +8243,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool {
return true return true
} }
// match: (Div32 <t> x (Const32 [c])) // match: (Div32 <t> x (Const32 [c]))
// cond: smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 != 0 // cond: smagicOK(32,c) && config.RegSize == 4 && smagic(32,c).m&1 != 0 && config.useHmul
// result: (Sub32 <t> (Rsh32x64 <t> (Add32 <t> (Hmul32 <t> (Const32 <typ.UInt32> [int64(int32(smagic(32,c).m))]) x) x) (Const64 <typ.UInt64> [smagic(32,c).s])) (Rsh32x64 <t> x (Const64 <typ.UInt64> [31]))) // result: (Sub32 <t> (Rsh32x64 <t> (Add32 <t> (Hmul32 <t> (Const32 <typ.UInt32> [int64(int32(smagic(32,c).m))]) x) x) (Const64 <typ.UInt64> [smagic(32,c).s])) (Rsh32x64 <t> x (Const64 <typ.UInt64> [31])))
for { for {
t := v.Type t := v.Type
...@@ -8254,7 +8254,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool { ...@@ -8254,7 +8254,7 @@ func rewriteValuegeneric_OpDiv32_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(smagicOK(32, c) && config.RegSize == 4 && smagic(32, c).m&1 != 0) { if !(smagicOK(32, c) && config.RegSize == 4 && smagic(32, c).m&1 != 0 && config.useHmul) {
break break
} }
v.reset(OpSub32) v.reset(OpSub32)
...@@ -8380,7 +8380,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8380,7 +8380,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
return true return true
} }
// match: (Div32u x (Const32 [c])) // match: (Div32u x (Const32 [c]))
// cond: umagicOK(32, c) && config.RegSize == 4 && umagic(32,c).m&1 == 0 // cond: umagicOK(32, c) && config.RegSize == 4 && umagic(32,c).m&1 == 0 && config.useHmul
// result: (Rsh32Ux64 <typ.UInt32> (Hmul32u <typ.UInt32> (Const32 <typ.UInt32> [int64(int32(1<<31+umagic(32,c).m/2))]) x) (Const64 <typ.UInt64> [umagic(32,c).s-1])) // result: (Rsh32Ux64 <typ.UInt32> (Hmul32u <typ.UInt32> (Const32 <typ.UInt32> [int64(int32(1<<31+umagic(32,c).m/2))]) x) (Const64 <typ.UInt64> [umagic(32,c).s-1]))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8390,7 +8390,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8390,7 +8390,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(32, c) && config.RegSize == 4 && umagic(32, c).m&1 == 0) { if !(umagicOK(32, c) && config.RegSize == 4 && umagic(32, c).m&1 == 0 && config.useHmul) {
break break
} }
v.reset(OpRsh32Ux64) v.reset(OpRsh32Ux64)
...@@ -8407,7 +8407,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8407,7 +8407,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
return true return true
} }
// match: (Div32u x (Const32 [c])) // match: (Div32u x (Const32 [c]))
// cond: umagicOK(32, c) && config.RegSize == 4 && c&1 == 0 // cond: umagicOK(32, c) && config.RegSize == 4 && c&1 == 0 && config.useHmul
// result: (Rsh32Ux64 <typ.UInt32> (Hmul32u <typ.UInt32> (Const32 <typ.UInt32> [int64(int32(1<<31+(umagic(32,c).m+1)/2))]) (Rsh32Ux64 <typ.UInt32> x (Const64 <typ.UInt64> [1]))) (Const64 <typ.UInt64> [umagic(32,c).s-2])) // result: (Rsh32Ux64 <typ.UInt32> (Hmul32u <typ.UInt32> (Const32 <typ.UInt32> [int64(int32(1<<31+(umagic(32,c).m+1)/2))]) (Rsh32Ux64 <typ.UInt32> x (Const64 <typ.UInt64> [1]))) (Const64 <typ.UInt64> [umagic(32,c).s-2]))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8417,7 +8417,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8417,7 +8417,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(32, c) && config.RegSize == 4 && c&1 == 0) { if !(umagicOK(32, c) && config.RegSize == 4 && c&1 == 0 && config.useHmul) {
break break
} }
v.reset(OpRsh32Ux64) v.reset(OpRsh32Ux64)
...@@ -8439,7 +8439,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8439,7 +8439,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
return true return true
} }
// match: (Div32u x (Const32 [c])) // match: (Div32u x (Const32 [c]))
// cond: umagicOK(32, c) && config.RegSize == 4 // cond: umagicOK(32, c) && config.RegSize == 4 && config.useAvg && config.useHmul
// result: (Rsh32Ux64 <typ.UInt32> (Avg32u x (Hmul32u <typ.UInt32> (Const32 <typ.UInt32> [int64(int32(umagic(32,c).m))]) x)) (Const64 <typ.UInt64> [umagic(32,c).s-1])) // result: (Rsh32Ux64 <typ.UInt32> (Avg32u x (Hmul32u <typ.UInt32> (Const32 <typ.UInt32> [int64(int32(umagic(32,c).m))]) x)) (Const64 <typ.UInt64> [umagic(32,c).s-1]))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8449,7 +8449,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8449,7 +8449,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(32, c) && config.RegSize == 4) { if !(umagicOK(32, c) && config.RegSize == 4 && config.useAvg && config.useHmul) {
break break
} }
v.reset(OpRsh32Ux64) v.reset(OpRsh32Ux64)
...@@ -8534,7 +8534,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8534,7 +8534,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
return true return true
} }
// match: (Div32u x (Const32 [c])) // match: (Div32u x (Const32 [c]))
// cond: umagicOK(32, c) && config.RegSize == 8 // cond: umagicOK(32, c) && config.RegSize == 8 && config.useAvg
// result: (Trunc64to32 (Rsh64Ux64 <typ.UInt64> (Avg64u (Lsh64x64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [32])) (Mul64 <typ.UInt64> (Const64 <typ.UInt32> [int64(umagic(32,c).m)]) (ZeroExt32to64 x))) (Const64 <typ.UInt64> [32+umagic(32,c).s-1]))) // result: (Trunc64to32 (Rsh64Ux64 <typ.UInt64> (Avg64u (Lsh64x64 <typ.UInt64> (ZeroExt32to64 x) (Const64 <typ.UInt64> [32])) (Mul64 <typ.UInt64> (Const64 <typ.UInt32> [int64(umagic(32,c).m)]) (ZeroExt32to64 x))) (Const64 <typ.UInt64> [32+umagic(32,c).s-1])))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8544,7 +8544,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8544,7 +8544,7 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(32, c) && config.RegSize == 8) { if !(umagicOK(32, c) && config.RegSize == 8 && config.useAvg) {
break break
} }
v.reset(OpTrunc64to32) v.reset(OpTrunc64to32)
...@@ -8578,6 +8578,8 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool { ...@@ -8578,6 +8578,8 @@ func rewriteValuegeneric_OpDiv32u_0(v *Value) bool {
func rewriteValuegeneric_OpDiv64_0(v *Value) bool { func rewriteValuegeneric_OpDiv64_0(v *Value) bool {
b := v.Block b := v.Block
_ = b _ = b
config := b.Func.Config
_ = config
typ := &b.Func.Config.Types typ := &b.Func.Config.Types
_ = typ _ = typ
// match: (Div64 (Const64 [c]) (Const64 [d])) // match: (Div64 (Const64 [c]) (Const64 [d]))
...@@ -8729,7 +8731,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool { ...@@ -8729,7 +8731,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool {
return true return true
} }
// match: (Div64 <t> x (Const64 [c])) // match: (Div64 <t> x (Const64 [c]))
// cond: smagicOK(64,c) && smagic(64,c).m&1 == 0 // cond: smagicOK(64,c) && smagic(64,c).m&1 == 0 && config.useHmul
// result: (Sub64 <t> (Rsh64x64 <t> (Hmul64 <t> (Const64 <typ.UInt64> [int64(smagic(64,c).m/2)]) x) (Const64 <typ.UInt64> [smagic(64,c).s-1])) (Rsh64x64 <t> x (Const64 <typ.UInt64> [63]))) // result: (Sub64 <t> (Rsh64x64 <t> (Hmul64 <t> (Const64 <typ.UInt64> [int64(smagic(64,c).m/2)]) x) (Const64 <typ.UInt64> [smagic(64,c).s-1])) (Rsh64x64 <t> x (Const64 <typ.UInt64> [63])))
for { for {
t := v.Type t := v.Type
...@@ -8740,7 +8742,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool { ...@@ -8740,7 +8742,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(smagicOK(64, c) && smagic(64, c).m&1 == 0) { if !(smagicOK(64, c) && smagic(64, c).m&1 == 0 && config.useHmul) {
break break
} }
v.reset(OpSub64) v.reset(OpSub64)
...@@ -8765,7 +8767,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool { ...@@ -8765,7 +8767,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool {
return true return true
} }
// match: (Div64 <t> x (Const64 [c])) // match: (Div64 <t> x (Const64 [c]))
// cond: smagicOK(64,c) && smagic(64,c).m&1 != 0 // cond: smagicOK(64,c) && smagic(64,c).m&1 != 0 && config.useHmul
// result: (Sub64 <t> (Rsh64x64 <t> (Add64 <t> (Hmul64 <t> (Const64 <typ.UInt64> [int64(smagic(64,c).m)]) x) x) (Const64 <typ.UInt64> [smagic(64,c).s])) (Rsh64x64 <t> x (Const64 <typ.UInt64> [63]))) // result: (Sub64 <t> (Rsh64x64 <t> (Add64 <t> (Hmul64 <t> (Const64 <typ.UInt64> [int64(smagic(64,c).m)]) x) x) (Const64 <typ.UInt64> [smagic(64,c).s])) (Rsh64x64 <t> x (Const64 <typ.UInt64> [63])))
for { for {
t := v.Type t := v.Type
...@@ -8776,7 +8778,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool { ...@@ -8776,7 +8778,7 @@ func rewriteValuegeneric_OpDiv64_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(smagicOK(64, c) && smagic(64, c).m&1 != 0) { if !(smagicOK(64, c) && smagic(64, c).m&1 != 0 && config.useHmul) {
break break
} }
v.reset(OpSub64) v.reset(OpSub64)
...@@ -8922,7 +8924,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool { ...@@ -8922,7 +8924,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool {
return true return true
} }
// match: (Div64u x (Const64 [c])) // match: (Div64u x (Const64 [c]))
// cond: umagicOK(64, c) && config.RegSize == 8 && umagic(64,c).m&1 == 0 // cond: umagicOK(64, c) && config.RegSize == 8 && umagic(64,c).m&1 == 0 && config.useHmul
// result: (Rsh64Ux64 <typ.UInt64> (Hmul64u <typ.UInt64> (Const64 <typ.UInt64> [int64(1<<63+umagic(64,c).m/2)]) x) (Const64 <typ.UInt64> [umagic(64,c).s-1])) // result: (Rsh64Ux64 <typ.UInt64> (Hmul64u <typ.UInt64> (Const64 <typ.UInt64> [int64(1<<63+umagic(64,c).m/2)]) x) (Const64 <typ.UInt64> [umagic(64,c).s-1]))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8932,7 +8934,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool { ...@@ -8932,7 +8934,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(64, c) && config.RegSize == 8 && umagic(64, c).m&1 == 0) { if !(umagicOK(64, c) && config.RegSize == 8 && umagic(64, c).m&1 == 0 && config.useHmul) {
break break
} }
v.reset(OpRsh64Ux64) v.reset(OpRsh64Ux64)
...@@ -8949,7 +8951,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool { ...@@ -8949,7 +8951,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool {
return true return true
} }
// match: (Div64u x (Const64 [c])) // match: (Div64u x (Const64 [c]))
// cond: umagicOK(64, c) && config.RegSize == 8 && c&1 == 0 // cond: umagicOK(64, c) && config.RegSize == 8 && c&1 == 0 && config.useHmul
// result: (Rsh64Ux64 <typ.UInt64> (Hmul64u <typ.UInt64> (Const64 <typ.UInt64> [int64(1<<63+(umagic(64,c).m+1)/2)]) (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [1]))) (Const64 <typ.UInt64> [umagic(64,c).s-2])) // result: (Rsh64Ux64 <typ.UInt64> (Hmul64u <typ.UInt64> (Const64 <typ.UInt64> [int64(1<<63+(umagic(64,c).m+1)/2)]) (Rsh64Ux64 <typ.UInt64> x (Const64 <typ.UInt64> [1]))) (Const64 <typ.UInt64> [umagic(64,c).s-2]))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8959,7 +8961,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool { ...@@ -8959,7 +8961,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(64, c) && config.RegSize == 8 && c&1 == 0) { if !(umagicOK(64, c) && config.RegSize == 8 && c&1 == 0 && config.useHmul) {
break break
} }
v.reset(OpRsh64Ux64) v.reset(OpRsh64Ux64)
...@@ -8981,7 +8983,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool { ...@@ -8981,7 +8983,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool {
return true return true
} }
// match: (Div64u x (Const64 [c])) // match: (Div64u x (Const64 [c]))
// cond: umagicOK(64, c) && config.RegSize == 8 // cond: umagicOK(64, c) && config.RegSize == 8 && config.useAvg && config.useHmul
// result: (Rsh64Ux64 <typ.UInt64> (Avg64u x (Hmul64u <typ.UInt64> (Const64 <typ.UInt64> [int64(umagic(64,c).m)]) x)) (Const64 <typ.UInt64> [umagic(64,c).s-1])) // result: (Rsh64Ux64 <typ.UInt64> (Avg64u x (Hmul64u <typ.UInt64> (Const64 <typ.UInt64> [int64(umagic(64,c).m)]) x)) (Const64 <typ.UInt64> [umagic(64,c).s-1]))
for { for {
_ = v.Args[1] _ = v.Args[1]
...@@ -8991,7 +8993,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool { ...@@ -8991,7 +8993,7 @@ func rewriteValuegeneric_OpDiv64u_0(v *Value) bool {
break break
} }
c := v_1.AuxInt c := v_1.AuxInt
if !(umagicOK(64, c) && config.RegSize == 8) { if !(umagicOK(64, c) && config.RegSize == 8 && config.useAvg && config.useHmul) {
break break
} }
v.reset(OpRsh64Ux64) v.reset(OpRsh64Ux64)
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7