qbit/go
1
0
Fork 0
mirror of https://github.com/golang/go synced 2024-10-05 11:31:22 -06:00
Code Releases Activity
90c0fefe51
go/src/cmd/5g/gsubr.go

1212 lines
25 KiB
Go
Raw Normal View History

[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
// Derived from Inferno utils/5c/txt.c
// http://code.google.com/p/inferno-os/source/browse/utils/5c/txt.c
//
// Copyright © 1994-1999 Lucent Technologies Inc. All rights reserved.
// Portions Copyright © 1995-1997 C H Forsyth (forsyth@terzarima.net)
// Portions Copyright © 1997-1999 Vita Nuova Limited
// Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com)
// Portions Copyright © 2004,2006 Bruce Ellis
// Portions Copyright © 2005-2007 C H Forsyth (forsyth@terzarima.net)
// Revisions Copyright © 2000-2007 Lucent Technologies Inc. and others
// Portions Copyright © 2009 The Go Authors. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.
package main
import (
cmd/5g etc: tweak import blocks Remove blank lines and merge lone imports into blocks. Change-Id: Ib46dad584456909969f1ba3a2c7f5667abf336ae Reviewed-on: https://go-review.googlesource.com/6792 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org> Reviewed-by: Rob Pike <r@golang.org>
2015-03-03 20:20:44 -07:00
"cmd/internal/gc"
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
"cmd/internal/obj"
"cmd/internal/obj/arm"
"fmt"
)
// TODO(rsc): Can make this bigger if we move
// the text segment up higher in 5l for all GOOS.
// At the same time, can raise StackBig in ../../runtime/stack.h.
var unmappedzero int = 4096
var resvd = []int{
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
arm.REG_R9, // formerly reserved for m; might be okay to reuse now; not sure about NaCl
arm.REG_R10, // reserved for g
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
/*
* return constant i node.
* overwritten by next call, but useful in calls to gins.
*/
var ncon_n gc.Node
func ncon(i uint32) *gc.Node {
if ncon_n.Type == nil {
gc.Nodconst(&ncon_n, gc.Types[gc.TUINT32], 0)
}
gc.Mpmovecfix(ncon_n.Val.U.Xval, int64(i))
return &ncon_n
}
var sclean [10]gc.Node
var nsclean int
/*
* n is a 64-bit value. fill in lo and hi to refer to its 32-bit halves.
*/
func split64(n *gc.Node, lo *gc.Node, hi *gc.Node) {
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Is64(n.Type) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gc.Fatal("split64 %v", gc.Tconv(n.Type, 0))
}
if nsclean >= len(sclean) {
gc.Fatal("split64 clean")
}
sclean[nsclean].Op = gc.OEMPTY
nsclean++
switch n.Op {
default:
switch n.Op {
default:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var n1 gc.Node
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !dotaddable(n, &n1) {
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Igen(n, &n1, nil)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
sclean[nsclean-1] = n1
}
n = &n1
case gc.ONAME:
if n.Class == gc.PPARAMREF {
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var n1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Cgen(n.Heapaddr, &n1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
sclean[nsclean-1] = n1
n = &n1
}
// nothing
case gc.OINDREG:
break
}
*lo = *n
*hi = *n
lo.Type = gc.Types[gc.TUINT32]
if n.Type.Etype == gc.TINT64 {
hi.Type = gc.Types[gc.TINT32]
} else {
hi.Type = gc.Types[gc.TUINT32]
}
hi.Xoffset += 4
case gc.OLITERAL:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var n1 gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gc.Convconst(&n1, n.Type, &n.Val)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
i := gc.Mpgetfix(n1.Val.U.Xval)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gc.Nodconst(lo, gc.Types[gc.TUINT32], int64(uint32(i)))
i >>= 32
if n.Type.Etype == gc.TINT64 {
gc.Nodconst(hi, gc.Types[gc.TINT32], int64(int32(i)))
} else {
gc.Nodconst(hi, gc.Types[gc.TUINT32], int64(uint32(i)))
}
}
}
func splitclean() {
if nsclean <= 0 {
gc.Fatal("splitclean")
}
nsclean--
if sclean[nsclean].Op != gc.OEMPTY {
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&sclean[nsclean])
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
}
func gmove(f *gc.Node, t *gc.Node) {
if gc.Debug['M'] != 0 {
fmt.Printf("gmove %v -> %v\n", gc.Nconv(f, 0), gc.Nconv(t, 0))
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
ft := gc.Simsimtype(f.Type)
tt := gc.Simsimtype(t.Type)
cvt := t.Type
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/{5,6,8,9}g, cmd/internal/gc: use bools for is* and okfor* No functional changes. This diff was generated as follows: * Manually edit cmd/internal/gc/go.go to update types and group variables. * Manually edit initialization in cmd/internal/gc/align.go--localized s/1/true. * Manually fix the handling of sign in cmd/internal/gc/walk.go in func bounded (near line 4000). * Manually update go.y and regenerate y.go. * Run gofmt -r many times to do the rest, using https://gist.github.com/josharian/0f61dbb2dff81f938e70. toolstash -cmp on the stdlib comes back green. Change-Id: I19766ed551714e51b325133e7138818d117b3a9a Reviewed-on: https://go-review.googlesource.com/6530 Reviewed-by: Russ Cox <rsc@golang.org>
2015-03-01 00:54:01 -07:00
if gc.Iscomplex[ft] || gc.Iscomplex[tt] {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gc.Complexmove(f, t)
return
}
// cannot have two memory operands;
// except 64-bit, which always copies via registers anyway.
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var a int
var r1 gc.Node
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Is64(f.Type) && !gc.Is64(t.Type) && gc.Ismem(f) && gc.Ismem(t) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
goto hard
}
// convert constant to desired type
if f.Op == gc.OLITERAL {
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var con gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
switch tt {
default:
gc.Convconst(&con, t.Type, &f.Val)
case gc.TINT16,
gc.TINT8:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var con gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gc.Convconst(&con, gc.Types[gc.TINT32], &f.Val)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, con.Type, t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVW, &con, &r1)
gmove(&r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
case gc.TUINT16,
gc.TUINT8:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var con gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gc.Convconst(&con, gc.Types[gc.TUINT32], &f.Val)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, con.Type, t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVW, &con, &r1)
gmove(&r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
}
f = &con
ft = gc.Simsimtype(con.Type)
// constants can't move directly to memory
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if gc.Ismem(t) && !gc.Is64(t.Type) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
goto hard
}
}
// value -> value copy, only one memory operand.
// figure out the instruction to use.
// break out of switch for one-instruction gins.
// goto rdst for "destination must be register".
// goto hard for "convert to cvt type first".
// otherwise handle and return.
switch uint32(ft)<<16 | uint32(tt) {
default:
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
// should not happen
gc.Fatal("gmove %v -> %v", gc.Nconv(f, 0), gc.Nconv(t, 0))
return
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
/*
* integer copy and truncate
*/
case gc.TINT8<<16 | gc.TINT8: // same size
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Ismem(f) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
a = arm.AMOVB
break
}
fallthrough
case gc.TUINT8<<16 | gc.TINT8,
gc.TINT16<<16 | gc.TINT8, // truncate
gc.TUINT16<<16 | gc.TINT8,
gc.TINT32<<16 | gc.TINT8,
gc.TUINT32<<16 | gc.TINT8:
a = arm.AMOVBS
case gc.TUINT8<<16 | gc.TUINT8:
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Ismem(f) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
a = arm.AMOVB
break
}
fallthrough
case gc.TINT8<<16 | gc.TUINT8,
gc.TINT16<<16 | gc.TUINT8,
gc.TUINT16<<16 | gc.TUINT8,
gc.TINT32<<16 | gc.TUINT8,
gc.TUINT32<<16 | gc.TUINT8:
a = arm.AMOVBU
case gc.TINT64<<16 | gc.TINT8, // truncate low word
gc.TUINT64<<16 | gc.TINT8:
a = arm.AMOVBS
goto trunc64
case gc.TINT64<<16 | gc.TUINT8,
gc.TUINT64<<16 | gc.TUINT8:
a = arm.AMOVBU
goto trunc64
case gc.TINT16<<16 | gc.TINT16: // same size
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Ismem(f) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
a = arm.AMOVH
break
}
fallthrough
case gc.TUINT16<<16 | gc.TINT16,
gc.TINT32<<16 | gc.TINT16, // truncate
gc.TUINT32<<16 | gc.TINT16:
a = arm.AMOVHS
case gc.TUINT16<<16 | gc.TUINT16:
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Ismem(f) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
a = arm.AMOVH
break
}
fallthrough
case gc.TINT16<<16 | gc.TUINT16,
gc.TINT32<<16 | gc.TUINT16,
gc.TUINT32<<16 | gc.TUINT16:
a = arm.AMOVHU
case gc.TINT64<<16 | gc.TINT16, // truncate low word
gc.TUINT64<<16 | gc.TINT16:
a = arm.AMOVHS
goto trunc64
case gc.TINT64<<16 | gc.TUINT16,
gc.TUINT64<<16 | gc.TUINT16:
a = arm.AMOVHU
goto trunc64
case gc.TINT32<<16 | gc.TINT32, // same size
gc.TINT32<<16 | gc.TUINT32,
gc.TUINT32<<16 | gc.TINT32,
gc.TUINT32<<16 | gc.TUINT32:
a = arm.AMOVW
case gc.TINT64<<16 | gc.TINT32, // truncate
gc.TUINT64<<16 | gc.TINT32,
gc.TINT64<<16 | gc.TUINT32,
gc.TUINT64<<16 | gc.TUINT32:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var flo gc.Node
var fhi gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
split64(f, &flo, &fhi)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, t.Type, nil)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVW, &flo, &r1)
gins(arm.AMOVW, &r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
splitclean()
return
case gc.TINT64<<16 | gc.TINT64, // same size
gc.TINT64<<16 | gc.TUINT64,
gc.TUINT64<<16 | gc.TINT64,
gc.TUINT64<<16 | gc.TUINT64:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var fhi gc.Node
var flo gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
split64(f, &flo, &fhi)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var tlo gc.Node
var thi gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
split64(t, &tlo, &thi)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, flo.Type, nil)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r2 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r2, fhi.Type, nil)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVW, &flo, &r1)
gins(arm.AMOVW, &fhi, &r2)
gins(arm.AMOVW, &r1, &tlo)
gins(arm.AMOVW, &r2, &thi)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
gc.Regfree(&r2)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
splitclean()
splitclean()
return
/*
* integer up-conversions
*/
case gc.TINT8<<16 | gc.TINT16, // sign extend int8
gc.TINT8<<16 | gc.TUINT16,
gc.TINT8<<16 | gc.TINT32,
gc.TINT8<<16 | gc.TUINT32:
a = arm.AMOVBS
goto rdst
case gc.TINT8<<16 | gc.TINT64, // convert via int32
gc.TINT8<<16 | gc.TUINT64:
cvt = gc.Types[gc.TINT32]
goto hard
case gc.TUINT8<<16 | gc.TINT16, // zero extend uint8
gc.TUINT8<<16 | gc.TUINT16,
gc.TUINT8<<16 | gc.TINT32,
gc.TUINT8<<16 | gc.TUINT32:
a = arm.AMOVBU
goto rdst
case gc.TUINT8<<16 | gc.TINT64, // convert via uint32
gc.TUINT8<<16 | gc.TUINT64:
cvt = gc.Types[gc.TUINT32]
goto hard
case gc.TINT16<<16 | gc.TINT32, // sign extend int16
gc.TINT16<<16 | gc.TUINT32:
a = arm.AMOVHS
goto rdst
case gc.TINT16<<16 | gc.TINT64, // convert via int32
gc.TINT16<<16 | gc.TUINT64:
cvt = gc.Types[gc.TINT32]
goto hard
case gc.TUINT16<<16 | gc.TINT32, // zero extend uint16
gc.TUINT16<<16 | gc.TUINT32:
a = arm.AMOVHU
goto rdst
case gc.TUINT16<<16 | gc.TINT64, // convert via uint32
gc.TUINT16<<16 | gc.TUINT64:
cvt = gc.Types[gc.TUINT32]
goto hard
case gc.TINT32<<16 | gc.TINT64, // sign extend int32
gc.TINT32<<16 | gc.TUINT64:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var tlo gc.Node
var thi gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
split64(t, &tlo, &thi)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, tlo.Type, nil)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r2 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r2, thi.Type, nil)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gmove(f, &r1)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
p1 := gins(arm.AMOVW, &r1, &r2)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
p1.From.Type = obj.TYPE_SHIFT
p1.From.Offset = 2<<5 | 31<<7 | int64(r1.Val.U.Reg)&15 // r1->31
p1.From.Reg = 0
//print("gmove: %P\n", p1);
gins(arm.AMOVW, &r1, &tlo)
gins(arm.AMOVW, &r2, &thi)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
gc.Regfree(&r2)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
splitclean()
return
case gc.TUINT32<<16 | gc.TINT64, // zero extend uint32
gc.TUINT32<<16 | gc.TUINT64:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var thi gc.Node
var tlo gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
split64(t, &tlo, &thi)
gmove(f, &tlo)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, thi.Type, nil)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVW, ncon(0), &r1)
gins(arm.AMOVW, &r1, &thi)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
splitclean()
return
// case CASE(TFLOAT64, TUINT64):
/*
* float to integer
*/
case gc.TFLOAT32<<16 | gc.TINT8,
gc.TFLOAT32<<16 | gc.TUINT8,
gc.TFLOAT32<<16 | gc.TINT16,
gc.TFLOAT32<<16 | gc.TUINT16,
gc.TFLOAT32<<16 | gc.TINT32,
gc.TFLOAT32<<16 | gc.TUINT32,
// case CASE(TFLOAT32, TUINT64):
gc.TFLOAT64<<16 | gc.TINT8,
gc.TFLOAT64<<16 | gc.TUINT8,
gc.TFLOAT64<<16 | gc.TINT16,
gc.TFLOAT64<<16 | gc.TUINT16,
gc.TFLOAT64<<16 | gc.TINT32,
gc.TFLOAT64<<16 | gc.TUINT32:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
fa := arm.AMOVF
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
a := arm.AMOVFW
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if ft == gc.TFLOAT64 {
fa = arm.AMOVD
a = arm.AMOVDW
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
ta := arm.AMOVW
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
switch tt {
case gc.TINT8:
ta = arm.AMOVBS
case gc.TUINT8:
ta = arm.AMOVBU
case gc.TINT16:
ta = arm.AMOVHS
case gc.TUINT16:
ta = arm.AMOVHU
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, gc.Types[ft], f)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r2 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r2, gc.Types[tt], t)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
gins(fa, f, &r1) // load to fpu
p1 := gins(a, &r1, &r1) // convert to w
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
switch tt {
case gc.TUINT8,
gc.TUINT16,
gc.TUINT32:
p1.Scond |= arm.C_UBIT
}
gins(arm.AMOVW, &r1, &r2) // copy to cpu
gins(ta, &r2, t) // store
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
gc.Regfree(&r2)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
/*
* integer to float
*/
case gc.TINT8<<16 | gc.TFLOAT32,
gc.TUINT8<<16 | gc.TFLOAT32,
gc.TINT16<<16 | gc.TFLOAT32,
gc.TUINT16<<16 | gc.TFLOAT32,
gc.TINT32<<16 | gc.TFLOAT32,
gc.TUINT32<<16 | gc.TFLOAT32,
gc.TINT8<<16 | gc.TFLOAT64,
gc.TUINT8<<16 | gc.TFLOAT64,
gc.TINT16<<16 | gc.TFLOAT64,
gc.TUINT16<<16 | gc.TFLOAT64,
gc.TINT32<<16 | gc.TFLOAT64,
gc.TUINT32<<16 | gc.TFLOAT64:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
fa := arm.AMOVW
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
switch ft {
case gc.TINT8:
fa = arm.AMOVBS
case gc.TUINT8:
fa = arm.AMOVBU
case gc.TINT16:
fa = arm.AMOVHS
case gc.TUINT16:
fa = arm.AMOVHU
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
a := arm.AMOVWF
ta := arm.AMOVF
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if tt == gc.TFLOAT64 {
a = arm.AMOVWD
ta = arm.AMOVD
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, gc.Types[ft], f)
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r2 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r2, gc.Types[tt], t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(fa, f, &r1) // load to cpu
gins(arm.AMOVW, &r1, &r2) // copy to fpu
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
p1 := gins(a, &r2, &r2) // convert
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
switch ft {
case gc.TUINT8,
gc.TUINT16,
gc.TUINT32:
p1.Scond |= arm.C_UBIT
}
gins(ta, &r2, t) // store
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
gc.Regfree(&r2)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
case gc.TUINT64<<16 | gc.TFLOAT32,
gc.TUINT64<<16 | gc.TFLOAT64:
gc.Fatal("gmove UINT64, TFLOAT not implemented")
return
/*
* float to float
*/
case gc.TFLOAT32<<16 | gc.TFLOAT32:
a = arm.AMOVF
case gc.TFLOAT64<<16 | gc.TFLOAT64:
a = arm.AMOVD
case gc.TFLOAT32<<16 | gc.TFLOAT64:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, gc.Types[gc.TFLOAT64], t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVF, f, &r1)
gins(arm.AMOVFD, &r1, &r1)
gins(arm.AMOVD, &r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
case gc.TFLOAT64<<16 | gc.TFLOAT32:
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var r1 gc.Node
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, gc.Types[gc.TFLOAT64], t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(arm.AMOVD, f, &r1)
gins(arm.AMOVDF, &r1, &r1)
gins(arm.AMOVF, &r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
}
gins(a, f, t)
return
// TODO(kaib): we almost always require a register dest anyway, this can probably be
// removed.
// requires register destination
rdst:
cmd/internal/gc: manual goto removal + grind to move var decls Also change gc.Naddr to return the Addr instead of filling it in. Change-Id: I98a86705d23bee49626a12a042a4d51cabe290ea Reviewed-on: https://go-review.googlesource.com/6601 Reviewed-by: Rob Pike <r@golang.org>
2015-03-02 18:34:22 -07:00
{
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, t.Type, t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/internal/gc: manual goto removal + grind to move var decls Also change gc.Naddr to return the Addr instead of filling it in. Change-Id: I98a86705d23bee49626a12a042a4d51cabe290ea Reviewed-on: https://go-review.googlesource.com/6601 Reviewed-by: Rob Pike <r@golang.org>
2015-03-02 18:34:22 -07:00
gins(a, f, &r1)
gmove(&r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
cmd/internal/gc: manual goto removal + grind to move var decls Also change gc.Naddr to return the Addr instead of filling it in. Change-Id: I98a86705d23bee49626a12a042a4d51cabe290ea Reviewed-on: https://go-review.googlesource.com/6601 Reviewed-by: Rob Pike <r@golang.org>
2015-03-02 18:34:22 -07:00
return
}
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
// requires register intermediate
hard:
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, cvt, t)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gmove(f, &r1)
gmove(&r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
return
// truncate 64 bit integer
trunc64:
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
var fhi gc.Node
var flo gc.Node
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
split64(f, &flo, &fhi)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(&r1, t.Type, nil)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
gins(a, &flo, &r1)
gins(a, &r1, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&r1)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
splitclean()
return
}
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
func samaddr(f *gc.Node, t *gc.Node) bool {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if f.Op != t.Op {
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
switch f.Op {
case gc.OREGISTER:
if f.Val.U.Reg != t.Val.U.Reg {
break
}
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return true
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
/*
* generate one instruction:
* as f, t
*/
func gins(as int, f *gc.Node, t *gc.Node) *obj.Prog {
// Node nod;
// int32 v;
if f != nil && f.Op == gc.OINDEX {
gc.Fatal("gins OINDEX not implemented")
}
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
// gc.Regalloc(&nod, &regnode, Z);
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
// v = constnode.vconst;
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
// gc.Cgen(f->right, &nod);
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
// constnode.vconst = v;
// idx.reg = nod.reg;
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
// gc.Regfree(&nod);
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if t != nil && t.Op == gc.OINDEX {
gc.Fatal("gins OINDEX not implemented")
}
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
// gc.Regalloc(&nod, &regnode, Z);
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
// v = constnode.vconst;
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
// gc.Cgen(t->right, &nod);
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
// constnode.vconst = v;
// idx.reg = nod.reg;
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
// gc.Regfree(&nod);
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
p := gc.Prog(as)
cmd/internal/gc: change Naddr to take a *Addr to fill in This allows gins to let Naddr fill in p.From and p.To directly, avoiding the zeroing and copying of a temporary. Change-Id: I96d120afe266e68f94d5e82b00886bf6bd458f85 Reviewed-on: https://go-review.googlesource.com/7742 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-03-16 13:27:19 -06:00
gc.Naddr(&p.From, f)
gc.Naddr(&p.To, t)
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
switch as {
case arm.ABL:
if p.To.Type == obj.TYPE_REG {
p.To.Type = obj.TYPE_MEM
}
case arm.ACMP, arm.ACMPF, arm.ACMPD:
if t != nil {
if f.Op != gc.OREGISTER {
/* generate a comparison
TODO(kaib): one of the args can actually be a small constant. relax the constraint and fix call sites.
*/
gc.Fatal("bad operands to gcmp")
}
p.From = p.To
p.To = obj.Addr{}
raddr(f, p)
}
case arm.AMULU:
if f != nil && f.Op != gc.OREGISTER {
gc.Fatal("bad operands to mul")
}
case arm.AMOVW:
if (p.From.Type == obj.TYPE_MEM || p.From.Type == obj.TYPE_ADDR || p.From.Type == obj.TYPE_CONST) && (p.To.Type == obj.TYPE_MEM || p.To.Type == obj.TYPE_ADDR) {
gc.Fatal("gins double memory")
}
case arm.AADD:
if p.To.Type == obj.TYPE_MEM {
gc.Fatal("gins arith to mem")
}
case arm.ARSB:
if p.From.Type == obj.TYPE_NONE {
gc.Fatal("rsb with no from")
}
}
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if gc.Debug['g'] != 0 {
fmt.Printf("%v\n", p)
}
return p
}
/*
* insert n into reg slot of p
*/
func raddr(n *gc.Node, p *obj.Prog) {
var a obj.Addr
cmd/internal/gc: change Naddr to take a *Addr to fill in This allows gins to let Naddr fill in p.From and p.To directly, avoiding the zeroing and copying of a temporary. Change-Id: I96d120afe266e68f94d5e82b00886bf6bd458f85 Reviewed-on: https://go-review.googlesource.com/7742 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-03-16 13:27:19 -06:00
gc.Naddr(&a, n)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if a.Type != obj.TYPE_REG {
if n != nil {
gc.Fatal("bad in raddr: %v", gc.Oconv(int(n.Op), 0))
} else {
gc.Fatal("bad in raddr: <null>")
}
p.Reg = 0
} else {
p.Reg = a.Reg
}
}
/* generate a constant shift
* arm encodes a shift by 32 as 0, thus asking for 0 shift is illegal.
*/
func gshift(as int, lhs *gc.Node, stype int32, sval int32, rhs *gc.Node) *obj.Prog {
if sval <= 0 || sval > 32 {
gc.Fatal("bad shift value: %d", sval)
}
sval = sval & 0x1f
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
p := gins(as, nil, rhs)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
p.From.Type = obj.TYPE_SHIFT
p.From.Offset = int64(stype) | int64(sval)<<7 | int64(lhs.Val.U.Reg)&15
return p
}
/* generate a register shift
*/
func gregshift(as int, lhs *gc.Node, stype int32, reg *gc.Node, rhs *gc.Node) *obj.Prog {
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
p := gins(as, nil, rhs)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
p.From.Type = obj.TYPE_SHIFT
p.From.Offset = int64(stype) | (int64(reg.Val.U.Reg)&15)<<8 | 1<<4 | int64(lhs.Val.U.Reg)&15
return p
}
/*
* return Axxx for Oxxx on type t.
*/
func optoas(op int, t *gc.Type) int {
if t == nil {
gc.Fatal("optoas: t is nil")
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
a := obj.AXXX
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
switch uint32(op)<<16 | uint32(gc.Simtype[t.Etype]) {
default:
gc.Fatal("optoas: no entry %v-%v etype %v simtype %v", gc.Oconv(int(op), 0), gc.Tconv(t, 0), gc.Tconv(gc.Types[t.Etype], 0), gc.Tconv(gc.Types[gc.Simtype[t.Etype]], 0))
/* case CASE(OADDR, TPTR32):
a = ALEAL;
break;
case CASE(OADDR, TPTR64):
a = ALEAQ;
break;
*/
// TODO(kaib): make sure the conditional branches work on all edge cases
case gc.OEQ<<16 | gc.TBOOL,
gc.OEQ<<16 | gc.TINT8,
gc.OEQ<<16 | gc.TUINT8,
gc.OEQ<<16 | gc.TINT16,
gc.OEQ<<16 | gc.TUINT16,
gc.OEQ<<16 | gc.TINT32,
gc.OEQ<<16 | gc.TUINT32,
gc.OEQ<<16 | gc.TINT64,
gc.OEQ<<16 | gc.TUINT64,
gc.OEQ<<16 | gc.TPTR32,
gc.OEQ<<16 | gc.TPTR64,
gc.OEQ<<16 | gc.TFLOAT32,
gc.OEQ<<16 | gc.TFLOAT64:
a = arm.ABEQ
case gc.ONE<<16 | gc.TBOOL,
gc.ONE<<16 | gc.TINT8,
gc.ONE<<16 | gc.TUINT8,
gc.ONE<<16 | gc.TINT16,
gc.ONE<<16 | gc.TUINT16,
gc.ONE<<16 | gc.TINT32,
gc.ONE<<16 | gc.TUINT32,
gc.ONE<<16 | gc.TINT64,
gc.ONE<<16 | gc.TUINT64,
gc.ONE<<16 | gc.TPTR32,
gc.ONE<<16 | gc.TPTR64,
gc.ONE<<16 | gc.TFLOAT32,
gc.ONE<<16 | gc.TFLOAT64:
a = arm.ABNE
case gc.OLT<<16 | gc.TINT8,
gc.OLT<<16 | gc.TINT16,
gc.OLT<<16 | gc.TINT32,
gc.OLT<<16 | gc.TINT64,
gc.OLT<<16 | gc.TFLOAT32,
gc.OLT<<16 | gc.TFLOAT64:
a = arm.ABLT
case gc.OLT<<16 | gc.TUINT8,
gc.OLT<<16 | gc.TUINT16,
gc.OLT<<16 | gc.TUINT32,
gc.OLT<<16 | gc.TUINT64:
a = arm.ABLO
case gc.OLE<<16 | gc.TINT8,
gc.OLE<<16 | gc.TINT16,
gc.OLE<<16 | gc.TINT32,
gc.OLE<<16 | gc.TINT64,
gc.OLE<<16 | gc.TFLOAT32,
gc.OLE<<16 | gc.TFLOAT64:
a = arm.ABLE
case gc.OLE<<16 | gc.TUINT8,
gc.OLE<<16 | gc.TUINT16,
gc.OLE<<16 | gc.TUINT32,
gc.OLE<<16 | gc.TUINT64:
a = arm.ABLS
case gc.OGT<<16 | gc.TINT8,
gc.OGT<<16 | gc.TINT16,
gc.OGT<<16 | gc.TINT32,
gc.OGT<<16 | gc.TINT64,
gc.OGT<<16 | gc.TFLOAT32,
gc.OGT<<16 | gc.TFLOAT64:
a = arm.ABGT
case gc.OGT<<16 | gc.TUINT8,
gc.OGT<<16 | gc.TUINT16,
gc.OGT<<16 | gc.TUINT32,
gc.OGT<<16 | gc.TUINT64:
a = arm.ABHI
case gc.OGE<<16 | gc.TINT8,
gc.OGE<<16 | gc.TINT16,
gc.OGE<<16 | gc.TINT32,
gc.OGE<<16 | gc.TINT64,
gc.OGE<<16 | gc.TFLOAT32,
gc.OGE<<16 | gc.TFLOAT64:
a = arm.ABGE
case gc.OGE<<16 | gc.TUINT8,
gc.OGE<<16 | gc.TUINT16,
gc.OGE<<16 | gc.TUINT32,
gc.OGE<<16 | gc.TUINT64:
a = arm.ABHS
case gc.OCMP<<16 | gc.TBOOL,
gc.OCMP<<16 | gc.TINT8,
gc.OCMP<<16 | gc.TUINT8,
gc.OCMP<<16 | gc.TINT16,
gc.OCMP<<16 | gc.TUINT16,
gc.OCMP<<16 | gc.TINT32,
gc.OCMP<<16 | gc.TUINT32,
gc.OCMP<<16 | gc.TPTR32:
a = arm.ACMP
case gc.OCMP<<16 | gc.TFLOAT32:
a = arm.ACMPF
case gc.OCMP<<16 | gc.TFLOAT64:
a = arm.ACMPD
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
case gc.OPS<<16 | gc.TFLOAT32,
gc.OPS<<16 | gc.TFLOAT64:
a = arm.ABVS
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
case gc.OAS<<16 | gc.TBOOL:
a = arm.AMOVB
case gc.OAS<<16 | gc.TINT8:
a = arm.AMOVBS
case gc.OAS<<16 | gc.TUINT8:
a = arm.AMOVBU
case gc.OAS<<16 | gc.TINT16:
a = arm.AMOVHS
case gc.OAS<<16 | gc.TUINT16:
a = arm.AMOVHU
case gc.OAS<<16 | gc.TINT32,
gc.OAS<<16 | gc.TUINT32,
gc.OAS<<16 | gc.TPTR32:
a = arm.AMOVW
case gc.OAS<<16 | gc.TFLOAT32:
a = arm.AMOVF
case gc.OAS<<16 | gc.TFLOAT64:
a = arm.AMOVD
case gc.OADD<<16 | gc.TINT8,
gc.OADD<<16 | gc.TUINT8,
gc.OADD<<16 | gc.TINT16,
gc.OADD<<16 | gc.TUINT16,
gc.OADD<<16 | gc.TINT32,
gc.OADD<<16 | gc.TUINT32,
gc.OADD<<16 | gc.TPTR32:
a = arm.AADD
case gc.OADD<<16 | gc.TFLOAT32:
a = arm.AADDF
case gc.OADD<<16 | gc.TFLOAT64:
a = arm.AADDD
case gc.OSUB<<16 | gc.TINT8,
gc.OSUB<<16 | gc.TUINT8,
gc.OSUB<<16 | gc.TINT16,
gc.OSUB<<16 | gc.TUINT16,
gc.OSUB<<16 | gc.TINT32,
gc.OSUB<<16 | gc.TUINT32,
gc.OSUB<<16 | gc.TPTR32:
a = arm.ASUB
case gc.OSUB<<16 | gc.TFLOAT32:
a = arm.ASUBF
case gc.OSUB<<16 | gc.TFLOAT64:
a = arm.ASUBD
case gc.OMINUS<<16 | gc.TINT8,
gc.OMINUS<<16 | gc.TUINT8,
gc.OMINUS<<16 | gc.TINT16,
gc.OMINUS<<16 | gc.TUINT16,
gc.OMINUS<<16 | gc.TINT32,
gc.OMINUS<<16 | gc.TUINT32,
gc.OMINUS<<16 | gc.TPTR32:
a = arm.ARSB
case gc.OAND<<16 | gc.TINT8,
gc.OAND<<16 | gc.TUINT8,
gc.OAND<<16 | gc.TINT16,
gc.OAND<<16 | gc.TUINT16,
gc.OAND<<16 | gc.TINT32,
gc.OAND<<16 | gc.TUINT32,
gc.OAND<<16 | gc.TPTR32:
a = arm.AAND
case gc.OOR<<16 | gc.TINT8,
gc.OOR<<16 | gc.TUINT8,
gc.OOR<<16 | gc.TINT16,
gc.OOR<<16 | gc.TUINT16,
gc.OOR<<16 | gc.TINT32,
gc.OOR<<16 | gc.TUINT32,
gc.OOR<<16 | gc.TPTR32:
a = arm.AORR
case gc.OXOR<<16 | gc.TINT8,
gc.OXOR<<16 | gc.TUINT8,
gc.OXOR<<16 | gc.TINT16,
gc.OXOR<<16 | gc.TUINT16,
gc.OXOR<<16 | gc.TINT32,
gc.OXOR<<16 | gc.TUINT32,
gc.OXOR<<16 | gc.TPTR32:
a = arm.AEOR
case gc.OLSH<<16 | gc.TINT8,
gc.OLSH<<16 | gc.TUINT8,
gc.OLSH<<16 | gc.TINT16,
gc.OLSH<<16 | gc.TUINT16,
gc.OLSH<<16 | gc.TINT32,
gc.OLSH<<16 | gc.TUINT32,
gc.OLSH<<16 | gc.TPTR32:
a = arm.ASLL
case gc.ORSH<<16 | gc.TUINT8,
gc.ORSH<<16 | gc.TUINT16,
gc.ORSH<<16 | gc.TUINT32,
gc.ORSH<<16 | gc.TPTR32:
a = arm.ASRL
case gc.ORSH<<16 | gc.TINT8,
gc.ORSH<<16 | gc.TINT16,
gc.ORSH<<16 | gc.TINT32:
a = arm.ASRA
case gc.OMUL<<16 | gc.TUINT8,
gc.OMUL<<16 | gc.TUINT16,
gc.OMUL<<16 | gc.TUINT32,
gc.OMUL<<16 | gc.TPTR32:
a = arm.AMULU
case gc.OMUL<<16 | gc.TINT8,
gc.OMUL<<16 | gc.TINT16,
gc.OMUL<<16 | gc.TINT32:
a = arm.AMUL
case gc.OMUL<<16 | gc.TFLOAT32:
a = arm.AMULF
case gc.OMUL<<16 | gc.TFLOAT64:
a = arm.AMULD
case gc.ODIV<<16 | gc.TUINT8,
gc.ODIV<<16 | gc.TUINT16,
gc.ODIV<<16 | gc.TUINT32,
gc.ODIV<<16 | gc.TPTR32:
a = arm.ADIVU
case gc.ODIV<<16 | gc.TINT8,
gc.ODIV<<16 | gc.TINT16,
gc.ODIV<<16 | gc.TINT32:
a = arm.ADIV
case gc.OMOD<<16 | gc.TUINT8,
gc.OMOD<<16 | gc.TUINT16,
gc.OMOD<<16 | gc.TUINT32,
gc.OMOD<<16 | gc.TPTR32:
a = arm.AMODU
case gc.OMOD<<16 | gc.TINT8,
gc.OMOD<<16 | gc.TINT16,
gc.OMOD<<16 | gc.TINT32:
a = arm.AMOD
// case CASE(OEXTEND, TINT16):
// a = ACWD;
// break;
// case CASE(OEXTEND, TINT32):
// a = ACDQ;
// break;
// case CASE(OEXTEND, TINT64):
// a = ACQO;
// break;
case gc.ODIV<<16 | gc.TFLOAT32:
a = arm.ADIVF
case gc.ODIV<<16 | gc.TFLOAT64:
a = arm.ADIVD
}
return a
}
const (
ODynam = 1 << 0
OPtrto = 1 << 1
)
var clean [20]gc.Node
var cleani int = 0
func sudoclean() {
if clean[cleani-1].Op != gc.OEMPTY {
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&clean[cleani-1])
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
if clean[cleani-2].Op != gc.OEMPTY {
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regfree(&clean[cleani-2])
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
cleani -= 2
}
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
func dotaddable(n *gc.Node, n1 *gc.Node) bool {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if n.Op != gc.ODOT {
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
var oary [10]int64
var nn *gc.Node
o := gc.Dotoffset(n, oary[:], &nn)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if nn != nil && nn.Addable != 0 && o == 1 && oary[0] >= 0 {
*n1 = *nn
n1.Type = n.Type
n1.Xoffset += oary[0]
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return true
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
/*
* generate code to compute address of n,
* a reference to a (perhaps nested) field inside
* an array or struct.
* return 0 on failure, 1 on success.
* on success, leaves usable address in a.
*
* caller is responsible for calling sudoclean
* after successful sudoaddable,
* to release the register used for a.
*/
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
func sudoaddable(as int, n *gc.Node, a *obj.Addr) bool {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if n.Type == nil {
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
*a = obj.Addr{}
switch n.Op {
case gc.OLITERAL:
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
if !gc.Isconst(n, gc.CTINT) {
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
break
}
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
v := gc.Mpgetfix(n.Val.U.Xval)
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
if v >= 32000 || v <= -32000 {
break
}
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
switch as {
default:
return false
case arm.AADD,
arm.ASUB,
arm.AAND,
arm.AORR,
arm.AEOR,
arm.AMOVB,
arm.AMOVBS,
arm.AMOVBU,
arm.AMOVH,
arm.AMOVHS,
arm.AMOVHU,
arm.AMOVW:
break
}
cleani += 2
reg := &clean[cleani-1]
reg1 := &clean[cleani-2]
reg.Op = gc.OEMPTY
reg1.Op = gc.OEMPTY
cmd/internal/gc: change Naddr to take a *Addr to fill in This allows gins to let Naddr fill in p.From and p.To directly, avoiding the zeroing and copying of a temporary. Change-Id: I96d120afe266e68f94d5e82b00886bf6bd458f85 Reviewed-on: https://go-review.googlesource.com/7742 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-03-16 13:27:19 -06:00
gc.Naddr(a, n)
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
return true
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
case gc.ODOT,
gc.ODOTPTR:
cleani += 2
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
reg := &clean[cleani-1]
[dev.cc] cmd/5g etc: code cleanup: delay var decls and eliminate dead code Ran rsc.io/grind rev 6f0e601 on the source files. The cleanups move var declarations as close to the use as possible, splitting disjoint uses of the var into separate variables. They also remove dead code (especially in func sudoaddable), which helps with the var moving. There's more cleanup to come, but this alone cuts the time spent compiling html/template on my 2013 MacBook Pro from 3.1 seconds to 2.3 seconds. Change-Id: I4de499f47b1dd47a560c310bbcde6b08d425cfd6 Reviewed-on: https://go-review.googlesource.com/5637 Reviewed-by: Rob Pike <r@golang.org>
2015-02-23 14:07:24 -07:00
reg1 := &clean[cleani-2]
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
reg.Op = gc.OEMPTY
reg1.Op = gc.OEMPTY
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
var nn *gc.Node
var oary [10]int64
o := gc.Dotoffset(n, oary[:], &nn)
if nn == nil {
sudoclean()
return false
}
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
if nn.Addable != 0 && o == 1 && oary[0] >= 0 {
// directly addressable set of DOTs
n1 := *nn
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
n1.Type = n.Type
n1.Xoffset += oary[0]
cmd/internal/gc: change Naddr to take a *Addr to fill in This allows gins to let Naddr fill in p.From and p.To directly, avoiding the zeroing and copying of a temporary. Change-Id: I96d120afe266e68f94d5e82b00886bf6bd458f85 Reviewed-on: https://go-review.googlesource.com/7742 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-03-16 13:27:19 -06:00
gc.Naddr(a, &n1)
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
return true
}
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Regalloc(reg, gc.Types[gc.Tptr], nil)
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
n1 := *reg
n1.Op = gc.OINDREG
if oary[0] >= 0 {
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Agen(nn, reg)
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
n1.Xoffset = oary[0]
} else {
cmd/internal/gc: move cgen, regalloc, et al to portable code This CL moves the bulk of the code that has been copy-and-pasted since the initial 386 port back into a shared place, cutting 5 copies to 1. The motivation here is not cleanup per se but instead to reduce the cost of introducing changes in shared concepts like regalloc or general expression evaluation. For example, a change after this one will implement x.(*T) without a call into the runtime. This CL makes that followup work 5x easier. The single copy still has more special cases for architecture details than I'd like, but having them called out explicitly like this at least opens the door to generalizing the conditions and smoothing out the distinctions in the future. This is a LARGE CL. I started by trying to pull in one function at a time in a sequence of CLs and it became clear that everything was so interrelated that it had to be moved as a whole. Apologies for the size. It is not clear how many more releases this code will matter for; eventually it will be replaced by Keith's SSA work. But as noted above, the deduplication was necessary to reduce the cost of working on the current code while we have it. Passes tests on amd64, 386, arm, and ppc64le. Can build arm64 binaries but not tested there. Being able to build binaries means it is probably very close. Change-Id: I735977f04c0614f80215fb12966dfe9bbd1f5861 Reviewed-on: https://go-review.googlesource.com/7853 Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
2015-03-18 15:26:36 -06:00
gc.Cgen(nn, reg)
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
gc.Cgen_checknil(reg)
n1.Xoffset = -(oary[0] + 1)
}
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
for i := 1; i < o; i++ {
if oary[i] >= 0 {
gc.Fatal("can't happen")
}
gins(arm.AMOVW, &n1, reg)
gc.Cgen_checknil(reg)
n1.Xoffset = -(oary[i] + 1)
}
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
a.Type = obj.TYPE_NONE
a.Name = obj.NAME_NONE
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
n1.Type = n.Type
cmd/internal/gc: change Naddr to take a *Addr to fill in This allows gins to let Naddr fill in p.From and p.To directly, avoiding the zeroing and copying of a temporary. Change-Id: I96d120afe266e68f94d5e82b00886bf6bd458f85 Reviewed-on: https://go-review.googlesource.com/7742 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-03-16 13:27:19 -06:00
gc.Naddr(a, &n1)
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
return true
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
cmd/5g etc: mechanical cleanup Run rsc.io/grind rev a26569f on C->Go conversions. The new change in grind is the inlining of goto targets. If code says 'goto x' and the block starting at label x is unreachable except through that goto and the code can be moved to where the goto is without changing the meaning of its variable names, grind does that move. Simlarly, a goto to a plain return statement turns into that return statement (even if there are other paths to the return statement). Combined, these remove many long-distance gotos, which in turn makes it possible to reduce the scope of more variable declarations. (Because gotos can't jump across declarations, the gotos were keeping the declarations from moving.) Checked bit-for-bit compatibility with toolstash + buildall. Reduces compiler runtime in html/template by about 12%. Change-Id: Id727c0bd7763a61aa22f3daa00aeb8fccbc057a3 Reviewed-on: https://go-review.googlesource.com/6472 Reviewed-by: Aram Hăvărneanu <aram@mgk.ro> Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
2015-03-02 10:35:15 -07:00
case gc.OINDEX:
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
[dev.cc] cmd/internal/obj, cmd/internal/gc, new6g: reconvert Reconvert using rsc.io/c2go rev 27b3f59. Changes to converter: - fatal does not return, so no fallthrough after fatal in switch - many more function results and variables identified as bool - simplification of negated boolean expressions Change-Id: I3bc67da5e46cb7ee613e230cf7e9533036cc870b Reviewed-on: https://go-review.googlesource.com/5171 Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
2015-02-17 20:13:49 -07:00
return false
[dev.cc] cmd/internal/gc, cmd/new6g etc: convert from cmd/gc, cmd/6g etc First draft of converted Go compiler, using rsc.io/c2go rev 83d795a. Change-Id: I29f4c7010de07d2ff1947bbca9865879d83c32c3 Reviewed-on: https://go-review.googlesource.com/4851 Reviewed-by: Rob Pike <r@golang.org>
2015-02-13 12:40:36 -07:00
}
Copy Permalink